Final Report Summary - THERMOLIDAR (A new tool for sustainable forest management based on LiDAR (Laser Imaging Detection And Ranging) and THERMAL data integration)
Executive Summary:
The THERMOLIDAR project examines technological and methodological developments that may improve efficiency in the process of monitoring forest health using cost-effective alternatives to current field-based methods. On this context, the integrated estimation of structural and physiological variables on forest inventories using remote sensing tools is a huge challenge to European forest management.
The develop of new operational forest inventory methodologies and their integration at management and final users levels can lead to technically adapted proposals for forest needs. Field data collection aims to support remote sensing analysis (thermal and LiDAR), gain a better understanding on tree physiology and the disturbance mechanisms that interfere normal growth increments as dictated by site indices and, once those mechanisms are understood, infer methods for its detection over extensive areas.
The development of any forest management planning requires accurate information of forest composition, structure, vegetative condition and spatial and temporal dynamic. Impacts of climate change on trees and forests will interact to affect future trees and forests. New forest inventory data are needed in a context of global change to minimize anthropogenic stressors (pollution, forest decline, erosion and altered hydrology, loss of soil fertility, etc.), and to implement feasible environmental adaptation strategies.
The main THERMOLIDAR result is to combine airborne LiDAR with thermal remote sensing data sets to determine structural information of tree, canopy organization and physiological variables obtained at the scale of individual trees. This approach is essential in a context of climate change to obtain accurate forest inventory variables and evaluate forest management practices used in forest predictive models. The THERMOLIDAR project focus on the integration of LIDAR-thermal remote sensing technology and data processing to evaluate structural and physiological variables on forest ecosystems in software capable of processing both LiDAR and thermal imagery, as an operational tool, has also been developed.
The main results achieved on ThermoLiDAR project can be summarized in the THERMOLIDAR software, which is made up of four packages:
A. Processing. This package includes tools for conducting the processing of raw Thermal and LiDAR data in order to obtain the products required to achieve the parametric analysis of forest health assessment.
A.1. LiDAR processing:
A.2. Thermal image processing..
A.3. Ortho-rectification and the integration of LiDAR and Thermal data..
B. Forest Health Assessment. This package includes four tools for conducting the simultaneous analysis of thermal and Lidar data related to field data to evaluate the state and trends of forest health. A schematic description of data inputs and processes applied is included in Fig. 2
B.1. Forest Stand Segmentation (FSS).
B.2. Structurally homogeneous forest units (SHFU).
B.3. Health condition levels (HCL).
B.4. Forest Health Monitoring (FHM).
C. Data Analysis package: This section provides the tools for analysis and interpretation of results. Once thermal and LiDAR data have been processed in the previous sections, the user can generate the mapping needed to interpret the physiological state of the forest mass analysed.
Thermolidar partners have been active in preparing scientific publications and delivering conference presentations since December 2013 and the activities will continue after completion of the project period. One journal article is currently under per review and 4 other articles are still planned to be prepared for submission by the end of year 2014. Besides total 5 Master’s thesis works have been initiated and are close to be competed with support of the Thermolidar project. Altogether six oral and two poster presentations have been delivered by the end of November 2014.
Project Context and Objectives:
Project context:
The Forest management Plans are technical documents describing forest management objectives, strategies and commitments. It identifies intended methods of cutting, reforesting, and managing timber resources within the defined area of responsibility. The first step on a forest management plan is forest inventory: Field information collected to plan and reach the management goals".
Forests Inventories determine the present situation of forest stands and are essential to quantify and obtain indicators of biomass stocks, regeneration, protected flora and fauna, fire risks, stocking and overgrown landform, erosion problems, forest condition, water resources and infrastructures. Furthermore, through forest inventories we can obtain information to determine the role of forests such as carbon sinks (CO2), information extremely important to be taken into account in the development of carbon offset plans. Forests are the only sink of terrestrial carbon that can be managed. Therefore, the opportunity to improve its management is the only proven, safe and cheaper alternative to reduce the carbon of the atmosphere in a natural way.
Current forest inventories in Europe involve a large numbers of variables measured in the field; usually at least between 10 and 40 variables concerning e.g. site, volume and increment of growing stock, density, general capacity, forest damage and forest biodiversity. This information needs to be accurate and up to date in order to support wise decisions.
Forest inventory attributes have been typically measured in the field using hand-held equipment. However, despite the fact that field-based methods are regarded as highly accurate, they are extremely costly and time-consuming and sometimes affected by human induced errors. Moreover, traditional forests inventories become limited when collecting timely information about natural processes over large areas.
The collection of information over large areas is facilitated by the use of remote sensing techniques. For instance, Light Detection and Ranging (LiDAR) emerges as a valuable technique capable to provide a unique insight into forest structure in most forest inventories in Europe.
Discrete return small footprint LiDAR, is commercially available and used operationally in several EU countries. Increasingly, data providers in Europe have built the capability to facilitate data collection when required by the user by having more control on the date of the flight, the configuration of the sensor and the area to be covered. Research has been taking place in this area for at least 10 years, and has provided a suite of commercially available tools for forest inventory applications.
The focus of forest inventory has been on the estimation of basic measurement variables, such as volumes by tree species, basal areas, age and mean breast height diameter of stand. However, in the context of climate change, which is altering the physiological status of forests ecosystems, these traditional forestry measures become limited when confronted with the evolving nature of climatic change and the impact of disturbances. Now, the main problems that forests are facing are the increase of temperatures and climatic extreme events, like heat waves, droughts, floods and storms.
These extreme events increase the risk of forest fires and the susceptibility to pests and diseases, causing general forest decline. As a result, Sustainable Forest Planning requires not only information about forest composition and its structure but also about vegetation condition in a spatial and temporal context. Damage caused by biotic or abiotic agents, changes in nutrient balance and acidity and other diseases are being measured in a highly subjective way based on general tables or visual field observations.
The physiological status of forest ecosystems can be evaluated using variables like water stress, chlorophyll pigments, stomata conductance, photosyntesis efficiency, etc.
A Stress indicator suggested in several studies proposed by Jackson et al, (1977) is the temperature of the canopy as an indicator of tree transpiration (stomata conductance rate). Thermal remote sensing of water stress has been successfully applied to tree crop canopies based on high resolution thermal remote sensing imagery (Berni et al., 2009), airborne thermal imagery (Sepulcre-Cantó et al., 2007) and satellite thermal information in combination with 3D radiative transfer models to understand the effects of scene thermal components on large ASTER pixels (Sepulcre-Cantó et al., 2009).
At this point, there is a fundamental advantage in the early detection of physiological processes before their visual manifestation takes place (e.g. defoliation) for a quick intervention. To detect early critical physiological processes before its visual manifestations helps forest manager to take appropriate decisions to mitigate or reduce damage’s impact.
Forestry managers need new forest inventory information to respond to the new challenges and risks produced by anthropogenic stressors (pollution, acid rain, erosion, altered hydrology, loss of soil and fertility). It is necessary that the estimation of structural variables will be complemented with the monitoring of physiological processes in order to describe the complexity of the forest ecosystems and their response to external stressors.
THERMOLIDAR project aims to resolve these problems developing a forest management tool that, first of all reduces the amount of time dedicated to field data collections and the costs associated to traditional forest inventories. Then, inventories should also provide information about physiological variables at a forest stand level in the early stages in order to improve decision making. THERMOLiDAR will help to evaluate forest management practices used in forest predictive models.
To date, there is not any company in the world able to integrate structural and physiological information of forest stands in a reliable and competitive way.
There is a clear trend worldwide to use remote sensing for forest inventories. This is evident from the latest public tenders for forest management plans and forest inventories in several European countries. There is also a sharp increase in demand for LiDAR surveys by public administrations and forest companies alike. Although some companies in Europe provide a suite of commercial LiDAR tools for forest applications, there are still few algorithms capable of rendering valuable information about forest structure (both vertical and horizontal) that may be linked to fundamental physiological processes.
OBJECTIVES:
The main objective of THERMOLIDAR is to combine airborne structural variables derived from LiDAR with physiological information produced by thermal sensors. This innovative approach will enhance the capabilities of SMEs to provide commercial services to companies and government agencies managing European forests.
Overarching objectives
- Identify the obstacles for up-scaling structural and physiological variables to address the requirements of forest management at stand and landscape scales.
- Develop a scientifically sound data fusion between LiDAR and thermal data sets capable of integrating the synoptic view about current forest resources provided by a forest inventory with the dynamic perspective of the physiological processes.
- The assessment of those key structural and physiological tree variables that can be detected by remote sensing and are paramount for monitoring forest health and sustainable production.
- The creation of thematic cartography that spatially locate each variable. This information will be complemented by uncertainty values for each estimate also spatially located.
- Integrate derived cartography and field collected biophysical data for each study area into an accessible data repository.
- Design a protocol for image data acquisition with the specific technological details of the flights, the sensors applied, and the data capture for validation and calibration of the images.
- Establish a protocol for field measurements adapted in intensity and number of variables to airborne LiDAR and Thermal sensors requirements.
- Co-registration of the images acquired from different sensors into a stack. Orthorectification, radiometric and atmospheric corrections. Integration of image data into the geodatabase.
- Integration of biophysical data and remote sensing data. Relate and integrate the biophysical data collected at crown level with remote sensing data at different spatial levels based on scaling uptechniques.
- The development of algorithms based on the relationships between biophysical data and remote sensing data.
- Development of operational software tools capable of processing both LiDAR and Thermal imagery and generating spatially located datasets fully integrated to corporate GIS.
- SMEs training in the use of THERMOLiDAR tools.
Project Results:
The overall design of the project is to combine airborne structural variables derived from LiDAR with physiological information produced by thermal sensors.
To achieve the scientific and technological objectives for THERMOLIDAR project, the first 4 work packages were design:
1) Field data acquisition
2) Data analysis
3) Software development
4) Validation of thermal and LiDAR data application .
1) . Field data acquisition and airborne campaigns
WP 1 aims to support remote sensing analysis (thermal and LiDAR), gain a better understanding on tree physiology and the disturbance mechanisms that interfere normal growth increments as dictated by site indices and, once those mechanisms are understood, infer methods for its detection over extensive areas. On this WP an accurate methodology for field data collection was provided. These main highlights obtained were:
1. - Field data collected methodology oriented to physiological and structural variables related to forest damage. The proposed plots are oriented to capture inventory data for training and testing methods that estimate forest inventory variables using remote sensing data (LiDAR and thermal).
2. - Methods and instruments to be used in each country, divided by physiological and structural plots.
3. - Reviews of current knowledge related to forest damage in each study area.
One of the objectives of Thermolidar project is to design a protocol for image data acquisition with the specific technological details of the flights, the sensors applied, and the data capture for validation and calibration of the images, and also to establish a protocol for field measurements adapted in intensity and number of variables to airborne LiDAR and Thermal sensors requirements.
To achieve this goal an “Airborne image acquisition guideline” was done. (See attachment).
GEODATABASE:
The first stage for field data acquisition was the creation of a geo-database by gathering existing cartographic and biophysical information in each study area complemented with other datasets shared by each group during the life of this project, subjected to partial restrictions. Background information for each study area has been pre-analysed in order to get a preliminary knowledge about the forest condition at each location. The analysis has focused on two main issues: background information about previous datasets, such as LiDAR, thermal data and plot data and background information about forest management and forest health condition of each study area.
A review of the main spatial Database Management Systems (sDBMS) was performed in order to select an appropriate system that suits the main requirements of the project.
The implementation of the INSPIRE Directive in May 2007 was the first step towards the creation of a pan-European spatial data infrastructure to support the Community environmental policies. The Directive addresses 34 spatial data themes needed for environmental applications, with key components specified through technical implementation rules. This makes INSPIRE a unique example of a legislative “regional” approach.
The implementation of INSPIRE protocol is carried out based on the “Technical Guidelines” for Thermolidar project with the following results:
- Data is collected only once, kept where it can be maintained most effectively, and be readily and transparently available.
- It is possible to combine seamless spatial information from different sources across Europe and share it with many users and applications.
- It is easy to find what geographic information is available, how it can be used to meet a particular need, and under which conditions can be acquired and used.
(Table 1; Metadata protocol established for Thermolidar)
METADATA MANAGEMENT FOR THERMOLIDAR PROJECT:
Metadata can be defined as data about data (Pelkki and Arthaud, 2008). Information is provided which render the data useful, providing context, data properties (e.g. resolution, units of measurement), defining terms, noting any conditions, quality assessment and explaining applications. A good Metadata record enables the user of a dataset to understand the content they are reviewing, its potential and its limitations. Metadata describe the summary information or characteristics of a set of data.
The objective of this task is to define a protocol to consistently document datasets within the Thermolidar project using international Metadata standards. A suitable metadata structure is paramount within the Thermolidar project in order to assist with data interoperability and the exchange of information between international partners. This will help to add value to the project outcomes by facilitating access to deliverables.
Furthermore in order to support dissemination, data must be represented using commonly-understood terminology and containing information which will allow the user to fully understand where the data originated, what degree of processing has been undertaken and who can be contacted in case of queries.
For Thermolidar project metadata generated for the three principal dataset types (Raster, Lidar point clouds and Features data) have been integrated within the project Geo-Database which has been designed to manage this spatial information.
Within each group, the geo-database includes a specific metadata protocol (D 3.2).
LIDAR: Name, format, file tipe, processing level, processind method, projection system, date, resolution, sensor, scanning angle.
RASTER: File name, format file, file types, processing level, processing method, projection system, date, resolution (spectral/spatial), sensor.
FEATURES: File name, format file, files types or feature data, projection system, source, land cover type, field data type, sampling methodology, variables, object level.
For each study area a geodatabase is created following the same structure.
The CatMDEdit metadata editor tool has been selected to meet the metadata documentation needs of the Thermolidar project (http://catmdedit.sourceforge.net/). This open source software was developed by contributors at the Instituto Geográfico Nacional, the Advanced Information Systems Laboratory of the University of Zaragoza, and the GeoSpatiumLab S.L. all based in Spain.
STUDY AREAS:
Four study areas were selected located in Czech Republic (CZ), Finland (FI), United Kingdom (UK) and Spain (ES). The areas present different forest health condition and cover different types of forest species such coniferous species (mainly composed by Picea and Pinus) and Broadleaved species (mainly composed by Quercus and Fagus:
(FIG 1. WP1 location of the study plots)
The experimental scheme has been described independently for each study area based on the nature of the analysis. Early detection analysis of forest decline affected by different biotic threats was assessed on study areas located in UK, FI and ES. The areas selected have similar species than in UK and ES, such us Pinus sylvestris and, therefore, this will allow further research of forest dynamics affected by biotic threats.
(FIG 2 study area in Spain)
Spatial data background (field data, images, Lidar data, and terrestrial LiDAR) of each study area was collected as a start point for further measurements.
The use of LiDAR data to simulate forest Net Primary Production (NPP) was performed on the study area located in Bily Kriz (Czech Republic). The experimental design in this area is focused on the prediction of net primary production NPP of by integrating conventional and remote-sensing (LiDAR) data. These analyses were performed combining structural information retrieve form LiDAR data and models to simulate water and carbon fluxes such us, C-Fix or BIOME-BGC. Specifically, results obtained from the analysis of LiDAR data (estimates of stem volume) and tree age were assessed to predict the NPP of the examined ecosystems.
In Finland, UK and Spain growth models based on time series LiDAR data and thermal information were performed. The aim of this approach is to analyse the effect of biotic and abiotic factors on forest stands production, specifically on Picea abies forest affected by Typographus in Finland, and Pinus sylvestris affected by Dothystroma in UK and the same specie affected by water stress in Spain locations. Analysing growth models requires a modelling approach that relates the growth of individual trees to the resources available to them and thus, the required model simulate each individual tree as a basic unit and sum the resulting estimates to produce stand level values. Growth refers to the increase in dimensions of one or more individuals in a forest stand over a given period of time (e.g. volume growth in m-2 ha1y1). A growth equation might predict annual increment of diameter, basal area or volume in units per annum as a function of age and other stand characteristics, whereas a yield equation would predict the diameter, stand basal area or total volume production attained at a specific age. Forest productivity has been calculated using LiDAR analysis calibrated by field data inventories. The results have been compared to stand model predictions in order to highlight growth anomalies that can be attributed to forest decline processes.
Early detection of forest decline was performed on the Spanish and UK study areas. In UK For example Two pine stands were monitored in the field during an entire growing season. A 30 years old Lodgepine plantation affected by Dothistroma Red Band (Dothistroma septospora) and a 11 years old Scots pine stand affected by Lophodermella (Lophodermium seditiosum) was used to set up field instruments recording the physiological activity of individual trees with different levels of affection. Unaffected trees in the neighbourhood were used as control.
FIG. 3 field data collection in UK
In Spain the monitored trees consisted of 36 Pinus nigra and 36 Pinus sylvestris, located in three study areas (12 trees per study areas) in Filabres and 15 Oak trees in Huelva.
FIG 4 field data collection in Spain
Lidar and Thermal data was also acquired in each study area in order to obtain results on early detection of forest decline for Pinus and Quercus.
Fig 5. Sample of Lidar Data from Sierra de Filabres
Based on the analysis of field data, Lidar and Thermal data simple linear and multiple regression models were applied to evaluate the accuracy of DAI- and ALS-based metrics separately in retrieving in situ tree height. Overall, simple linear regression model results showed that the performance of most metrics extracted from Lidar point data was better than that of metrics extracted from photogrammetric DAI in estimating canopy height.
In order to analyze the factors that affected the above-mentioned relationships, we explored the effects produced by the percentage of crown overlapping and slope.
According to results, model performance was not significantly affected by the percentage of slope or crown overlapping. This result is relevant to demonstrate that the proposed method was not affected by the structure of the vegetation canopy or ground unevenness.
In Thermolidar project we propose a new approach to generate a canopy height model (CHM) from thermal imagery acquired with low-cost commercial color infrared (CIR) cameras using the structure from motion (SfM) method as an alternative to using more sophisticated technologies, such as Lidar, in the context of complex forest canopies. Both single-tree height and effective leaf area index (LAIe) at the crown level were estimated from DAI and ALS data and compared to field data. Robust models were developed for both variables with a set of metrics derived from ALS and DAI data independently. When estimating tree height, the reliability obtained with ALS-based Percentile 90 (P90) was slightly higher than that obtained with DAI-based metrics (P90 and minimum height (Hmin)), with a relative root mean square error (RRMSE) difference of 3% (yielding an RMSE = 0.51 m and an RMSE = 0.71 m, respectively).
FIG 6 Object-based delineation of crown trees based on Lidar and Thermal data
These results represent progress in the validation of 3D photogrammetric models applied to forest inventory development. In the case of LAI estimation, the reliability obtained with ALS-based metrics was slightly higher than that obtained with DAI-derived Normalized Difference Vegetation Index (NDVI) metrics (DAI NDVI P99), with an R-RMSE difference of 1.96% Our results demonstrated that the estimated error obtained for both the tree height and LAI parameters using low-cost airborne digital imagery was not significantly affected by the slope or the percentage of crown overlapping typically observed in an oak forest canopy.
This notwithstanding, the successful retrieval of single-tree and forest-stand biophysical variables using low-cost digital airborne imagery in other canopy types and terrain characteristics should be further analyzed and validated to assess issues related to canopy heterogeneity, crown dimensions and tree shape.
Some other conclusions related to forest decline based on Thermal and LiDAR data information are shown in fig. 7.
FIG. 7 Levels of affectation defined for Thermolidar project
First approximation to forest health condition analysis is the detection of decline processes with visual assessment of the damage. Visual assessment standards were adapted for both types of affectations, water stress detection on conifer forest and holm oak forest affected by Phytophthora. In order to quantify biomass defoliation LAI measurements were also performed with different sensors.
Although, traditional visual assessments can be subjective and therefore, it is not a direct measure of tree vigor. In contrast, a nonvisual method that allows tracking of pigment concentrations (e.g. chlorophyll) may provide an objective, early warning indicator of stands requiring remedial or salvage action before damage is visible and, potentially, before biomass loss occurs. Physiological parameters measured from the selected trees in Thermolidar project were total concentration of chlorophyll, needle water content and dry mass, stomatal conductance and crown temperature. These data were averaged from four measurements per tree during each period at the time of the Thermal imagery acquisition (12:00, GMT). Field Gas exchange measurements were performed in attached leaves at controlled CO2 external concentration (Ca = 350 ppm) and ambient relative humidity. Stomatal conductance (Gs) was estimated using gas exchange data and the total needle area exposed obtained from photos taken for each measurement.
Apart of these physiological measurements, continuous data was obtained from canopy temperature, photosynthetic activity, soil moisture, tree growth, sap flow and air temperature and humidity.
Analysis of forest condition.
Forest health was analysed based on the integration of Thermal and LiDAR data and forest health indicators collected during the field work. Forest structure was analysed based on data from 2010 and 2013. Dominant-height curves of Norway spruce was applied to analyse forest dynamics. A secondary step was to study the relationship between the survival and height increment of Norway spruce in relation to stand temperature. In order to accomplish this step, it is necessary to calculate the difference between stand temperature minus air temperature (Ts-Ta) and them analyses the spatial patterns described by (Ts-Ta) in relation to the spatial structure of forest stands. Thermal data was trained based on field data information of forest condition temperature. In order to accomplish this step, it is necessary to calculate the difference between stand temperature minus air temperature (Ts-Ta) and them analyses the spatial patterns described by (Ts-Ta) in relation to the spatial structure of forest stands. Thermal data was trained based on field data information of forest condition.
In UK the main objective was the analysis of pine forest affected by Dothistroma Red Band (Dothistroma septospora) based on LiDAR and Thermal data integration.
For Thermolidar two types of field data collection were planned for the validation and calibration of airborne sensors analysis:
-Plot data for validation of area-based analysis using airborne LiDAR
-Tree physiology for the monitoring of plant stress using Thermal imagery
This research is looking into an alternative approach to the area-based methods that relies on an effective combination of LiDAR metrics, existing sub-compartment data and stand models. This hybrid approach aims at the feasibility of replacing, or at least minimising, field data collection by using local stand models in combination with data inputs to this model estimated directly by LiDAR with good levels of accuracy (e.g. Top Height).
The aims of the proposed field data collection are:
1) The development of a low-cost solution for forest inventory that is robust and competitive in terms of cost and quality compared to current field data collection methods.
2) The production of a suite of multi-scale cartographic products fit for purpose, capable of addressing different management needs and fully integrated to local sub-compartment data and corporative GIS.
Related to Tree physiology for the monitoring of plant stress using Thermal imagery the main hypothesis is that it is possible to implement a robust monitoring system capable of detecting parts of the woodland under stress based on temperature differences observed at the top of the canopy. Therefore field data collection focused on a correct interpretation of fundamental physiological processes at the tree level that may infer that the plant is under stress. Once a process of stress has been identified, it is necessary to measure the magnitude of affection and the subsequent response at the top of the canopy. This analysis allows us to estimate the likelihood of a stress process being detected. Finally, once this process is well understood it is possible to calibrate a method of detection based on airborne Thermal imagery.
Two sample plots were monitored in the field during an entire growing season. Within each plot, four trees had a set of sensors attached to them following the experimental design outlined in Deliverible D.1.1. The selection criteria will follow a combination of dominance status and level of affection still to be determined.
Plot 1. Lodgepole pine plantation, > 50 years old. Height range 12 to 37 m. This area is affected by Dothistroma Red Band. The level of defoliation varies between 10 to 50%. This location has also being affected by windthrow in the storm that affected the Forest District in January 2012.
Plot 2. Scots pine plantation. This area was planted for recreation and landscaping 11 years ago. The tallest trees are 3-4 m tall. The location has serious problems of Sitka spruce encroachment that require a frequent elimination of spruce seedlings. These trees are affected by Lophodermella. Some individuals are starting to be affected by Dothistroma, which is unusual.
FIG 8. Dendrometer and thermistors installation
There is evidence to suggest that the Thermal properties of forest canopies could be used to determine stress, other than water stress, through their effects on tree physiology. Such effects include their influence on movement of water through the vascular system and hence on rates of evapotranspiration from leaf surfaces. However, not all stresses may result in warmer foliage temperatures and it may not always be possible to differentiate between water and other stresses using Thermal data alone. The ability to measure Thermal properties of plant canopies will of course be dependent on the degree of coupling between leaf and air temperatures and thus the tendency of wind and atmospheric moisture conditions to equalise the two. For example, although failed to observe Thermal differences between control and herbicide-treated forest canopies, they were attempting to compare differences in autumn and winter, when temperature differences may be less than maximal. Further research may be warranted into Thermal detection of stress in plant canopies and to the optimal timing of such measurements.
The algorithms for interpreting Thermal information may need to be reasonably sophisticated to compensate for emissions from non-plant surfaces but this could be aided by the combined use of Thermal and LiDAR analysis. This would require a good method of overlapping of these two datasets capable of seamlessly integrating the polar perspective of LiDAR point clouds with the perspective geometry obtained by Thermal imagery.
FIG 9. Rain gauge and air temperature-humidity sensor to calibrate leaf and air temperature
In order to calibrate a robust method for the detection of stress in the forest canopy based on Thermal differences between healthy and affected trees, field data collection was undertaken in a group of selected individuals in the monitoring area. The fundamental principles followed in this study is using the pipe model theory, which relates active live foliar biomass in the canopy with water and sap conductance at the stem. Each one of the trees being instrumented had their canopies measured. Then, sapwood depth was estimated using a model described by Beauchamp.
It is not entirely clear how a reduction of leaf area index as a result of a defoliation process may affect the formation of heartwood. As such its formation must be carefully regulated temporally and spatially, however the regulation of this process is not understood. Heartwood is thought to form from mid-summer, at the end of the growth season, and continues throughout the dormant period, as long as temperatures are maintained above 5°C.
Canopy temperature was constantly monitored using a Campbell IR 120 narrow field infra-red sensor. This sensor was mounted on a pole or a tower at less than 1 meter distance from the canopy. The sensor measured the average temperature of the top part of the canopy during the photosynthetic active part of the day. Data obtained at the canopy top was also used to calibrate the signal from the airborne Thermal measurements.
Physiological variables e.g. PRI, sap flow, stomatal conductance and transpiration was associated with canopy temperature collected from the tree samples. Stress dates were identified and differences between infected and healthy, young and old trees compared.
TDR sensors measured the water content in the soil. This information was used to measure water shortages during the growing season that may introduce anomalies in the sap conductance activity. The installation protocols followed the same procedures than the ones in the Spanish study areas.
One Skye 2 sensor per plot to continually measure PRI with wavelengths centred at 531nm and 570nm with a 5nm bandwidth. PRI sensors developed by Skye Instruments measured incident and reflected light simultaneously for accurate readings in all daylight conditions. This sensor provided information about the amount of light energy available for photosynthesis.
Main scientific and technical results for field data acquisition and airborne campaigns in Spain:
The experimental area is located in Sierra de Filabres (Almeria province, southeastern Spain) the driest region in Western Europe. The vegetation consists of a 40-year-old mixed pine afforestation of Pinus nigra Arnold and P. sylvestris L. A progressive decline episode of Scots pine stands has been observed since 2002, to such an extent that there is damage in 66% of the 6000 ha occupied by this species.
For Thermolidar project the monitored trees consisted of 36 Pinus nigra and 36 Pinus sylvestris, located in three study areas (12 trees per study areas). Results obtained in monitored trees related to structural and physiological variables obtained from the different levels of damage are shown in Fig 9.1. and 9.2.
(FIG 9.1 Structural parameters in pinus study area)
(FIG 9.2 physiological variables obtained in pinus study area)
The variables measured showed significant differences in the physiological status for each study area (p<0.05). As result of this step, three different groups with similar health conditions (damage levels) were defined.
To validate and work with this data airborne campaigns were conducted during spring, summer and autumn 2013 with two different sensors and acquisition settings. Thermal image acquisition was performed using the thermal sensor (FLIR SC660). During the same period, LiDAR data was also acquired from an ALS60 with a mean resolution of 4 points m-2.
(FIG 9.3 Thermal data acquired over conifer forest)
LiDAR data were classified as above-ground vegetation and ground terrain. A threshold value of 1 m above the ground surface was used to separate canopy echoes from echoes below the canopy to produce the LiDAR based canopy height model. Then, statistical metrics were calculated from the CHM. LiDAR metrics were calculated to estimate biophysical characteristics related to the structure of the canopy such as canopy cover, basal area, LAI, volume and biomass The following metrics were extracted: minimum, maximum, mean, median, standard deviation, variance, coefficient of variation, interquartile distance, skewness, kurtosis, AAD (average absolute deviation), L-moment (1-4), and percentile values (P5 to P95 in 5-unit intervals and P99).
The final step of data processing was the calculation of an integrated product for the correction and transformation of raw LiDAR and thermal data into the cross-georeferenced products: canopy height model and ΔTª (Temperature of the crown-Temperature of the air).
The second step was Data Analysis for forest health assessment.
Conifer forest health assessment was addressed by THsf using an object-based approach
(FIG 9.4 Forest stands delineation)
This is a very critical step in order to reliably relate thermal information with tree physiology minimizing the effects produced by soil background, vegetation structure and vegetation mixture. A previous step to the integration of LiDAR and thermal data was the definition of stand maps based on species and forest stand structure to determine their magnitude and spatial distribution over forested areas.
The results obtained from the stand delineation are presented in Fig. 9.4. Forest stands delineation was estimated at different scale levels using LIDAR data and object-oriented image segmentation. Multi-scale approach enables the production of forest health status and forest productivity cartographies at different spatial resolutions.
Related to Structurally homogeneous forest units the results provide the classification of forest stands in units (or objects) structurally different. Input data needed to run this process was obtained from LiDAR data. Fig. 9.5 shows an example of the SHFU defined for a conifer forest in Almería (Spain) The classification was based on the canopy height model (CHM) and was conducted using a k-means clustering. Forest stands (objects) were classified in three groups (cluster-SHFU, grey scale) using the tree heights derived from the CHM to define the thresholds between classes. The stand structural classification mapping was produced using a cluster approach for improving estimation and to produce wall-to-wall structural homogeneous object map. Several features were used for this classification, including: several LiDAR metric (derived from the CHM). The algorithm used was k-means clustering to partition n objects into k clusters (SHFU) in which each object belongs to the cluster with the nearest mean.
(FIG 9.5 SHFU mapping of the conifer forest)
Forest Health Monitoring cartographies were produced integrating the spatial information derived from previous steps:
1. Stand maps based on species and forest stand structure to determine their magnitude and spatial distribution over forested areas based on pre-defined object feature or a semiautomatic segmentation tool (FSS).
2. A (statistical) classification of objects based on forest structural variables and associated characteristics by using a cluster approach for improving estimation and to produce wall-towall structural homogeneous cover type map (vector file with structurally homogeneous stands SHFU).
3. A classification of forest health condition (groups with similar damage levels, HLC) based on qualitative methodologies as defoliation or physiological indicators.
Using this information, health vegetation condition differences at the stand level was estimated using two different types of physiological indicators: the clusters defined from water potential and pigment composition, related with the current physiological status of the vegetation (Fig. 9.6) and the leaf area index (LAI) related with the level of defoliation of conifer forest stands (Fig. 9.7).
(FIG 9.6 Mapping forest decline levels)
The algorithms applied for the estimation of forest health monitoring worked using an object oriented approach modeling the ΔT of each SHFU units and based on the HCL defined by users in previous stages. In the first case, the prediction of damage distribution based on levels of defoliation was based on the relationship between LAI measurements and LiDARbased metrics. In the second case, the prediction of damage distribution was based on the relationship between ΔTª (independent variable) and the classification of forest health condition at object level (dependent variable) by using a regression approach. It should be highlighted, that at this point of the analysis, SHFU mapping are products standardized by forest stands units defined from LiDAR-based metrics.
(FIG 9.7 measurement of forest stands defoliation)
2) Data analysis
The objectives of this WP were divided into four tasks:
1. Field data processing. This task included the processing of physiological and structural data in support of airborne data analysis. Deliverable D2.1.
2. Image data processing. Activities in this WP included the preprocessing and analysis of thermal and LiDAR datasets. Deliverable D2.2.
3. Data fusion. Deliverables in this section aimed at the development of a methodology for the analysis of airborne images based on data fusion techniques. Deliverable D2.3.
4. Algorithm development. This task aimed at the optimization of the input data for the retrieval of biophysical estimates.
FIELD DATA PROCESSING REPORT
The first deliverable D2.1 reported on field data capture and processing methods. A suite of sensors were attached to individual trees to measure their physiological cycles. The methodology outlined in the D2.1 report described the observed differences between healthy and infected trees. Some of the results obtained in the field were:
1.1. Water Potential indicates the capacity of the plants to conduct water from soil to atmosphere. This was measured using the pressure-bomb technique with needles, as suggested by Scholander et al.1965. Measurements were undertaken at the beginning, middle and the end of a growing season in Spain and the UK. No measurements were undertaken in Finland and Czech Republic. According to Fereres et al., (1999) the largest differences were supposed to happen in mid-summer, right at the peak of the photosynthetic activity.
(FIG 9.8 water potential measured)
However, the mean observations (in MPa) did not show substantial differences between healthy and infected trees.
(FIG 9.9 Average water potential measured)
The analysis of samples at different levels across the vertical profile of the canopy did not render conclusive results (Figure 9.10). Infected trees showed slightly larger water potential values than healthy ones. However, the variability of the measurements did not produce conclusive results.
(FIG 9.10 water potential differences)
Water potential measured in Aberfoyle, plot 1, at the end of July 2014. Measurements were taken in the low, middle and top canopy. The trees depictured in the figure were from 1 to 4 (left to right). Three measurements were done by taking samples at each canopy level for each individual tree.
1.2 Canopy temperature
Canopy temperature was measured in the field using Infra-red canopy temperature sensors (IR120). A sensor was attached to each tree being monitored in each plot at 1 m distance from the canopy, taking permanent measurements all year round every 10 mins.
The results showed no distinctive differences between healthy and infected trees (Figure 9.11).
(FIG 9.11 Canopy temperature differences)
In order to eliminate sources of noise (i.e. air temperature), a detrending technique was applied to the time-series observations of temperature (Figure 9.12). Detrending is a statistical or mathematical operation that removes trends from the series. Detrending is often applied to remove a feature thought to distort or obscure the relationships of interest. Detrending is also sometimes used as a pre-processing step to prepare time series for analysis by methods that assume stationarity. Many alternative methods are available for detrending. A simple linear trend in mean can be removed by subtracting a least-squares-fit straight line. More complicated trends might require different procedures.
(FIG 9.12 . Detrending of canopy temperatures in healthy and infected trees)
Once detrended, there were no statistically significant between canopy temperatures. The main conclusion was that sensors measuring temperatures from tree canopies at a distance are measuring an integral of foliage, branches, shadows and air boundary layer. This was also the same conclusion observed in the analysis of the thermal images. Therefore, it must be concluded that at the moment all the techniques developed in this project to measure canopy temperatures failed to obtain good measurements that can be used to infer the physiological status of the vegetation. More work is needed to be able to differentiation signal to noise ratios.
1.3 Dendrometers
Dendrometers were installed at breast height. Bark was removed without causing a lot of damage to the cambium. Metal rods were fitted on the trunk and securely tighten making sure they are level. Sensor was pressed against the bare surface of the trunk half way through allowing the sensor to either expand or shrink.
No differences were observed between healthy and infected trees in any of the plots (Figure 9.13 Y 9.14).
(FIG 9.13 Mean daily trunk shrinkage in Aberfoyle, plot 1)
(FIG 9.14 Mean daily trunk shrinkage in Aberfoyle, plot 2)
1.4 Leaf Area Index
Leaf area index (LAI) was measured in plots with Hemispherical photos and LAI-2000. The results were correlated to the visual estimation of defoliation as described by the Visual Assessment of Crown condition protocol described in Deliverable D1.2. The results showed that substantial levels of defoliation in the canopies of infected trees. As infected trees develop smaller canopies, growth is substantially delayed and this is also shown in the volume increments. Unfortunately, no destructive sampling was conducted in any of the study areas to quantify the extent of growth reduction by different defoliation levels.
IMAGE DATA PROCESSING
The activities in this task were described in Deliverables 2.2 and 2.3.
Thermal images were atmospherically corrected according to the DiStasio and Resmini (2010) method. The orthorectification processing of the thermal images was a necessary step to overlap them with the LiDAR products. This work was undertaken in ENVI.
A suite of tools already implement methods for data conversion, representation and management of LiDAR data. Other tools allow users to filter point clouds and classify ground and vegetation hits. Classified points can be converted into interpolated surfaces such as Canopy Height Models, Digital Terrain Models and Digital Surface Models. Point clouds can also be analysed to produce metrics of interest for the implementation of the Naesset method, extensively used in Nordic countries for forest inventory, and the hybrid method, proposed at Forest Research that combines LiDAR metrics with stand models (described in Deliverable D1.3). An example of cartographic products produced by the software is depicted in Figure 9.15.
(FIG 9.15 Estimations of top height for Stika spruce stands for Thermolidar software)
The analysis of Thermal and LiDAR images were implemented in a set of routines using Open Source libraries like SPDLib and Orfeo. The Thermolidar software implemented a set of routines in QGIS that gave SMEs the functional capability to deliver cartographic products mapping structural and physiological characteristics of the forest canopy being studied. A synoptic view of the LiDAR processing capabilities is described in Figure 9.16.
(FIG 9.16 Thermolidar software. Processing of LiDAR data)
DATA FUSION
The fusion between Thermal and LiDAR data were explained in Deliverable D2.3 D4.2 and the training materials described in D5.4. The method described in Figure 9.17 and 9.18 used LiDAR data to construct a set of homogenous objects using the routines already contained in the Orfeo toolbox that was integrated in the Thermolidar software.
(Figure 9.17. Forest Health Assessment)
(Figure 9.18. Forest Health Assessment2)
This method created thematic cartography showing the spatial distribution of Forest Health Classes, arbitrary defined by the user.
ALGORITHM DEVELOPMENT
Alternatively, another method developed by Forest Research was also implemented in the suite of tools contained in Thermolidar. The method is an adaptation of the model presented by Blonquist et al. (2009) that estimates canopy stomatal conductance by using a series of standard meteorological and airborne thermal data. The model calculates canopy stomatal conductance (gC , mol m−2 ground area s−1) from canopy temperature, air temperature, barometric pressure, relative humidity, net radiation, wind speed and plant canopy height. All these variables are being collected and used in the ThermoLiDAR project. The objective is to calculate canopy stomatal conductance in line with fundamental tree physiological principles, combining the information collected from airborne and field campaigns but also utilising other modelled data inputs. The target is to provide an estimation of canopy stomatal conductance in areas where direct measurements of canopy temperature may be seriously interfered by wind speed, humidity and clouds. For more details on the model and its assumptions see Blonquist et al. (2009) and the description of the method in D4.1.
The model was used to perform a sensitive analysis of the input variables used to calculate stomatal conductance across the study area. Figure 9.19 describes the relationship between healthy and infected crowns in relationship to air temperature, as already implemented in the Thermolidar software. Infected canopies, in this case Scots pine affected by Dothistroma Needle Blithe, show an average of two degrees difference after noon. Differences are difficult to distinguish before and after 4 p.m.
Figure 9.19. Canopy temperatures in Tree Crowns (Tc) of healthy and infected trees compared to Air Temperature (Ta).
The model uses temperature and other climatic and topographic variables to estimate stomatal conductance values (Figure 9.20). The model shows an even more clear average differences between trees around noon and until sunset.
(Figure 9.20. Modelled estimations of Canopy Stomatal Conductance)
This model was tested in the monitoring plots of the Aberfoyle study area, where the estimates were compared against real measurements of stomatal conductance after up-scaling measurements from leaf to canopy level (Figure 9.21). The comparisons proved inconclusive. Canopy temperature, as measured by the airborne systems, has some caveats. First of all, the sensor is gathering information about the green material on top of the canopy but interfered by other surfaces. As a result, every pixel contains information about non-photosynthetic material such as branches, twigs or stems, shadows, and even the temperatures of lower parts of the canopy and ground. Therefore, the temperature of the canopies is far lower than they should be and, according to the model, all trees are functionally dead. Secondly, wind is another important factor altering the measurements of the canopy temperatures. Despite all the flights were done during days of low wind speed, air mixing is inevitable and canopy temperature values oscillate greatly according to the wind flow. Finally, it is rather difficult, if not impossible, the measurement of large areas at the same time or at the time of maximum stomatal activity. The model showed important differences in terms of stomatal activities during the day. Therefore, it was concluded that thermal measurements should be restricted to small areas that can be flown in less than 1 hour after noon.
(FIG 9.21 Stomata conductance in Plot 1, Aberfoyle calculated by Thermolidar)
The FR-Hybrid Model for the estimation of forest inventory parameters was developed as an alternative to mainstream area-based methods. This method relies more heavily of available stand methods and minimises the need to undertake costly and extensive field data collection.
Data were measured in the field in a set of plots and compared to the biophysical parameters calculated using LiDAR metrics and the model, as described in D4.1. The best relationships were found using equations for intermediate thinning. These models eliminate the need for entering initial spacing and they have been derived from the original models in Edwards and Christie (1981). Results are presented as options for three different species: Sitka spruce, Norway spruce and Scots pine.
3) Software development
One of the main objectives of Thermolidar project is the development of operational software capable of processing both LiDAR and Thermal imagery and generating spatially located datasets fully integrated to corporate GIS.
The main structure of the software is summarized in this figure:
(FIG 10. general flowchart Thermolidar software)
ThermoLiDAR software is designed to provide the required tools to integrate Thermal and LiDAR information for the assessment of forest health and production. The software includes different modules and several settings offering different levels of processing to suit each needs. The main functionalities of the software are:
Tools for raw LiDAR and Thermal data processing, data exploring, quality assessment and visualisation.
Tools to integrate both data sources using suitable data fusion techniques.
Spatial statistical tools to analyse biophysical variables of the vegetation based on the integration of LiDAR and Thermal data integration.
Tools for mapping forest health condition and forest production dynamics and producing accuracy assessment.
The software uses field data measurements for calibration and validation, broaden the user choice to work at different levels that ranges from a total supervised analysis to a total unsupervised analysis. The THERMOLIDAR software comprises a selection of specialized tools for the assessment of the efficiency and productivity of forest ecosystems integrated on an Open Source Geographic Information System (GIS) platform.
The accuracy and optimization in the application of THERMOLiDAR tools depends on various factors:
1. The quality of the thermal imaging and Lidar data acquisitions (see D 1.3 and D 2.2)
2. The selection of key indicators to measure on the field (see D 1.3).
3. The accuracy of field data measurements (see D 1.3).
4. Forest health condition and type of forest damage (see D 1.3).
5. Forest heterogeneity and composition (see D. 1.1)
6. Accuracy required by the user (D 3.3).
ThermoLiDAR software has been developed as a QGIS plug-in and, once installed, it becomes part of the Processing Toolbox of QGIS. This makes possible that final users get free access to lots of GIS capabilities. The QGIS version currently supported by the ThermoLiDAR plug-in is QGIS 2.4.0 Chugiak.
ThermoLiDAR software has been divided into three different packages: Processing, Forest Health Assessment and Data Analysis.
3.1) Processing package
The Processing package includes tools for conducting the processing of raw Thermal and LiDAR data in order to obtain the products required to achieve the parametric analysis of forest health assessment.
A.1. Lidar processing. LiDAR data tool set for the generation of DSMs, DTMs, DVMs and vegetation statistical derivatives thereof.
LiDAR data is often provided as a number of tiles or flight lines. Depending on computers capacity, in some cases it might be convenient to merge flight files into a single file, or to divide data in overlapping tiles with an appropriate size.
Due to ThermoLiDAR software uses SPDLib suite for LiDAR data manipulation, data has to be converted into the native SPDLib file format. This means that LiDAR data must be converted into Sorted Pulse Data format (SPD) from LAS files, which is the file standard for the interchange of LASer data recommended by the American Society for Photogrammetry and Remote Sensing (ASPRS).
Noisy data can be eventually removed before being processed. Once the SPD files have been obtained, ground points are classified and then point heights relative to the ground are inferred. Ground and no-ground points are interpolated to generate DTM, DSM and CHM. From height information a range of metrics mainly applied to forestry applications -but not only- can be derived. Here below this workflow is depicted.
(FIG 12 WorkFlow lidar procedure)
Since LiDAR processing modules make use of SPDLib tools, the first step is to convert the input dataset into SPD files. There are two types of SPD files, non-indexed and indexed. A format translation module has been included in QGIS to this purpose. The module allows the conversion between different formats and it is also used to re-project data.
Thermolidar Software includes a merging module. In some situations it might be convenient to merge various files into a single SPD file. The merging module merges compatible files into a single non-indexed SPD file.
LiDAR data is supplied as flight lines or tiles with different shapes and sizes. It is always useful to divide laser data into equally sized square tiles, though. This helps to store, manage and access data easily. Single tiles should meet memory requirements in order to reduce computational times, which determines the maximum size of each file given an average point density. Overlapping zones between tiles help to prevent border errors and guarantee continuous raster models. ThermoLiDAR software has a built-in tool to create tiles given tiles size and the overlap. The module supports to create single tiles (SINGLE option) by supplying row and column, or to generate the complete set of tiles (ALL option).
(FIG 13 Split data into tiles)
Many factors may introduce errors in LiDAR point clouds, including water vapour clouds, multipath, poor equipment calibration, or even a flock of birds. In order to avoid further errors and artefacts in final digital models and poor assess of height metrics, those points have to be removed. Thermolidar project includes a module that removes vertical noise from LiDAR datasets by means of three different. Upper and lower absolute thresholds will clip the file to fit these values. Relative threshold will remove, for each bin within a SPD file, points outside the upper and lower values relative to the median height. Whilst global threshold will use the whole SPD file to calculate the median height and remove points relative to it.
Another tool is for point classification. Two different classification algorithms have been implemented into the plug-in. These algorithms also called filters allow the classification of the LiDAR points, identifying which point belong to the ground.
Related to normalise heights SPD files supports both elevation corresponding to a vertical datum and an above-ground height for each discrete return. Before data can be used for generating a Canopy Height Model (CHM) or any height related metric, height field has to be populated. This can be done in two ways. The simplest way is to use a DTM of the same resolution as the SPD file bin size (Image option). The disadvantage of using a DTM is that it is if the DTM is not accurate it can introduce some artefacts. Using this method the only parameters are the input files, both LiDAR file and the DTM, and an output file. The raster DTM needs to the same resolution as the SPD grid and it can be any raster format supported by the GDAL library.
For interpolation Thermolidar software includes a module to create Digital Terrain Models (DTMs), Digital Surface Models (DSMs) and Canopy Height Models (CHMs). To produce those products, it is necessary to interpolate a raster surface from the classified ground returns and top surface points. Create Digital Model within the ThermoLiDAR plug-in permits to generate these products by choosing model option.
(FIG 14 Example of a CHM generated with ThermoLiDAR)
Related to metrics ThermoLiDAR is able to calculate different metrics at the same time. Metrics can be simple statistical moments, percentiles of point heights, or even count ratios. Mathematical operators can also be applied to either other metrics or operators. The module supports different output data formats, that is, raster (all GDAL formats) and vector. Raster option extends the output to the entire input file, assessing the metrics for each pixel the final raster output and creating as many bands as metrics have been defined within the XML file. Vector option requires an input shapefile containing polygon entities. The output shapefile database will be populated with the metrics computed inside the polygons.
Software offers the possibility to create the file with a single metric selected from a list. However, the list also accept the options ALL, FOREST and PERCENTILES:
ALL: Selects all the metrics listed in the option list.
FOREST: Includes some metrics that are commonly used in forest applications. This metrics includes percentiles from 99th to 50th, groundCover, canopyCover, maxHeight and meanHeight.
PERCENTILES: Includes all percentiles from 99th to 10th.
A.2. Thermal image processing
Thermal processing tools data allows a user to perform thermal imaging calibration using the Emissive Empirical Line Method (EELM) which is a common method for airborne thermal data processing based on the ‘In-scene atmospheric correction methods. These approaches were developed to remove atmospheric effects from hyper-spectral imaging data allowing the user to utilize similar conditions to the atmosphere state. The advantage of using this type of methods over model-based methods based in radiative transfer theory is that they capture the true state of the atmosphere at the time of data collection and the relative low computational efforts required to perform the corrections (in comparison with radiative transfer models approaches). The main difficulty for in-scene method is getting correctly the field measurements parameters required for the correction algorithm.
This tool will also allow the user to calculate the difference between Crown Temperature minus Air temperature (Tc-Ta). This indicator has been widely demonstrated in Thermolidar project to be related with different physiological indicators such as stem water potential, stomatal conductance or sap flow rate. User need to select as input the thermal imaging and the air temperature collected in the same time of the airborne imaging acquisitions.
(FIG 11 workflow thermal procedure)
(FIG 11. 1 Thermal calibration output)
3.2) Forest Health Assessment
Forest Health assessment module consists of four different tools: Forest Stand Segmentation (FSS), Health Condition Level (HCL), Structurally Homogeneous Forest Units (SHFU) and Forest Health Monitoring (FHM). The following figure summarizes the main structure of this module and the input required throughout the process.
(FIG 15 Forest Health assessment module)
B.1. Forest Stand Segmentation (FSS). The processed image is decomposed into regions or objects. Object based delineation algorithms are applied with this tool to define forest stands units for further study.
Within OD tools, users are willing to choose between developing a semi-automatic segmentation and using a pre-defined object feature. Segmentation tools are based on algorithms that segment an image into areas of connected pixels based on the pixel DN value. ThermoLiDAR image segmentation tools will be based on region growing algorithms. The basic approach of a region growing algorithm is to start from a seed region (typically one or more pixels) that are considered to be inside the object to be segmented. The pixels neighbouring this region are evaluated to determine if they should also be considered part of the object. If so, they are added to the region and the process continues as long as new pixels are added to the region. Region growing algorithms vary depending on the criteria used to decide whether a pixel should be included in the region or not, the type connectivity used to determine neighbours, and the strategy used to visit neighbouring pixels. Image segmentation is a crucial step within the object-based remote sensing information retrieval process. As a step prior to classification the quality assessment of the segmentation result is of fundamental significance for the recognition process as well as for choosing the appropriate approach and parameters for a given segmentation task. Alternatively, user could be interested on using a pre-defined object feature. This object feature could be a segmentation shape file provided from other source or any other land cover mapping. Also, the user can use a pre-defined regular object, defining the size of the square to be used previously.
(FIG 16 Segmentation example)
B.2. Forest Health condition levels (HCL). Different physiological indicators from field data measurements are processed with this tool to define the ground truth condition of forest status. Health condition levels are statistically generated based in clustering and subsequently validated by ANOVA.
The most critical part in applying forest health condition indicators is the user’s accuracy defining forest degradation levels. Besides user’s training another critical factor is to select under analysis a robust physiological indicator and to carry out an accurate field measurements campaign.
Potential physiological indicators of forest decline such us pigment concentration, photosynthesis, respiration and transpiration rate holds great potential to shed light on the mechanisms and processes that occur as a result of drought stress. In the short-term, climate can change the physiological conditions of the forest resulting in acute damage, but chronic exposure usually results in cumulative effects on physiological process. These factors effects on the plants light reactions or enzymatic functions and increased respiration from reparative activities. Gradual decreases in photosynthesis, stomatal conductance, carbon fixation, water use efficiency, resistance to insect and cold resistance were found in most of trees which are very typical symptom of stress conditions.
Long-term exposure of water stress to a combination of high light levels and high temperatures causes a depression of photosynthesis and photosystem II efficiency that is not easily reversed, even for water-stress-resistant forest species.
One of the most widely physiological indicator applied in the analysis of long-term effect on forest health condition is de Leaf Area Index (LAI). The following is an example of the statistical analysis performs on LAI values measured from an Oak forest inventoried in the framework of THERMOLIDAR project.
(FIG 17 example LAI values oak forest Spain)
B.3. Structurally homogeneous forest units (SHFU). This option provides the tools required for the classification of forest stands structurally different. The data applied to perform this classification is defined by the user. In this report the main average height of the trees estimated based on LiDAR data has been applied as input data. Alternatively, user can provide external forest maps in a .shp format type with an attribute of the number of class.
The following figure shows an example of the units defined for the oak forest under analysis. Using a grey scale, trees were grouped in 3 classes with significant differences in terms of structural composition.
(FIG 18 structurally homogeneous forest units)
B.4. Forest Health Monitoring (FHM).
The main function of this tool is defining health condition differences in the vegetation at the stand level. Input parameters defined by users should mainly contain: thermal imaging data and the FSC Polygons (vector file with structurally homogeneous stands. Forest stands included in this analysis should be carried specifically based on one species. The user can perform a supervised or an unsupervised classification depending of the availability of field data measurements to define training areas.
It should be highlight, that at this point of the analysis, users are willing to obtain an integrated mapping of forest health distribution levels based on thermal data but also standardized by forest stands units defined from lidar-based metrics. The following figure, shows an example of the units defined for assessment of the status of forest condition. Using a colour palette, trees were grouped in different classes with significant differences in terms of structural composition and physiological status. The colour palette ranges from red to green, where red colour is relate with trees with high level of damage and green colour represents trees with optimum health condition.
(FIG 19 Structural composition and physiological status)
3.3. The Data Analysis package
This section provides the tools for analysis and interpretation of results. Once thermal and LiDAR data have been processed in the previous sections, the user can generate the mapping needed to interpret the physiological state of the forest mass analysed.
First, the user has a set of data obtained in the field of physiology which are analysed and grouped, using the tools available at the module Health Condition Level.
From the tools available in the structurally homogeneous forest units module, the user can perform a preliminary classification of the stands, based on structural homogeneity. This factor is important because thermal values behave differently according to the structure of objects.
Finally, from the Forest Heath classification tools the user has the necessary tools to perform a classification based on the thermal values for the various homogeneous units. To improve the classification, the user can define training plots according to data collected in the field of physiology, visually are established different levels of affection. Ç
(FIG 20 workflow data Analysis Package)
Health Condition Levels
Before proceeding with the classification of items by level of damage according to several variables taken in the field, we verify that the set of physiological variables follow a normal distribution. For this we use the Shapiro test.
Thermolidar software includes a Clustering tool. This tool allows us to group one or more physiological variables according to their degree of similarity between individuals in the sample. So, the goal of clustering is to determine the intrinsic grouping in a set of unlabelled data. But how to decide what constitutes a good clustering? It can be shown that there is no absolute best criterion which would be independent of the final aim of the clustering. Consequently, it is the user which must supply this criterion, in such a way that the result of the clustering will suit their needs.
To make this tool has been chosen by a hierarchical approach. The user-supplied items are categorized into levels and sublevels within a class hierarchy, forming a hierarchical tree structure.
(FIG 20.1 LAI measurements in Huelva)
After that an ANOVA test are conducted for each variable to indicate how well the variable discriminates between clusters.
The hypothesis is tested in the ANOVA is that the population means (the average of the dependent variable at each level of the independent variable) are equal. If the population means are equal, it means that the groups did not differ in the dependent variable, and consequently, the independent variable is independent of the dependent variable.
(FIG 20.2 clustering ANOVA)
Structurally Homogeneous Forest Units
Structurally Homogeneous Forest Units tool allows to minimize the effects of structure on the thermal information, and therefore allow us to obtain related health outcomes woodland. To do this, the software allows the user to define the structure of the stand from the height data. The calculation of uniformity is therefore a function of two variables directly derived from LiDAR data the 95th percentile obtained from MDV and the penetration rate, obtained from density points that penetrate the forest canopy.
(FIG 21 Homogeneous forest units)
Forest Health Classification
Unsupervised Pixel-based Classification: The user has a stratification of the study area, and classified based on the structure. Through this tool, a classification of the pixels of temperatures will be performed, based on the classification of defined structure. Thus, the output will be a raster temperature for each of the groups of homogeneity.
Unsupervised Object-oriented Classification: The user has a stratification of the study area, and classified based on the structure. Through this tool, a classification of the objects of temperatures mean will be performed, based on the classification of defined structure. Thus, the output will be a raster temperature for each of the groups of homogeneity.
(FIG 22 _unsupervised_object CLASIFICATION)
Supervised Pixel-based Classification: Similarly as in the previous point, the user can perform a classification of the temperature response to the stand of homogeneity. Unlike supervised classification, the user has a number of AOIs that guide the classification process.
Forest Models
To measure the field biological and physical properties (e.g. dominant height, mean diameter, stem number, basal area, timber volume, etc...) throughout entire woodlands is impossible. However, due to the characteristics of LiDAR technology it is possible to assess these properties in wide areas.
Two different models have been implemented in the ThermoLiDAR software to this purpose. The first one implements an empirical method where field measurements are correlated with LiDAR data. In this case, only a few sample plots are commonly measured in field in order to relate these measurements to canopy height metrics derived from LiDAR data. These relationships are then used to estimate and extend those characteristics to the area covered by LiDAR creating forest inventory cartography. This method is commonly called Scandinavian method.
The second is a hybrid method where some equation relating LiDAR canopy height and forest variables have been defined a priori. This method does not require in the field measurement and once is calibrated, it can be used anywhere else.
It is expected that TLDSf will improve the capabilities of SMEs and commercial opportunity on the field of forest health diagnosis and management and put them in a position where they can initiate a business activity.
4. Validation of thermal and LiDAR data application.
Once Thermolidar software was done a validation of each of the software tool components was performed. These validation tasks have focused on verifying the proper operation of each of the modules as well as a control study times for these.
SMEs are currently testing tools and checking the modules with Thermolidar data to verify the proper functioning of the software.
To perform validation testing of each of the software tools developed in the structure we focus on the processes defined in the following modules of the software.
1. Processing. The quality control tests focus the processing of raw LiDAR data and Thermal. The algorithms focus on generating a new corrected and ready to be used in the following product modules.
1.1. LiDAR processing. LiDAR data management and classification, generation of Digital Surface Models (DSMs), Digital Terrain Models (DTMs), Canopy Height Models (CHMs) and vegetation statistical derivatives thereof.
1.2. Thermal Image processing Thermal data tools set for the calibration of airborne thermal imaging.
2. Analysis. Quality control is aimed to verify a control time performance in generating maps that allow us to establish and define the various levels of health of forests with due regard to the initial information (temperature and heights LiDAR).
Potential Impact:
ThermoLiDAR Integrated Services has been conceived as an innovative service provisioning in the fields of LiDAR & Thermal data acquisition and subsequent analyst; in order to optimize and facilitate the decision making by forest managers in charge of large forest areas.
The use of LiDAR has been substantially increased during the last decade due to the savings which it represents when comparing with tradition and on-ground inventory practices. Different sources of literature show about 20% of savings for areas exceeding 300 Ha. in the case of Spain (Applications of LiDAR Technology in the Forestry Sector, Fabra et. Al., 2012), between 10-30% in United Kingdom (An industrial cost-benefit comparison of LIDAR derived forest inventory to a traditional inventory for harvesting, Murray Woods, et al., 2012), and it is assumed up to 60% savings for Finland in the case of private forests (TEKES - The NATIONAL Technology Agency of Finland, Maltamo et al., 2007).
Other important aspect to take into consideration is the precision reached by the two methods of inventories (LiDAR Vs traditional) and the costs it involves. Number of research or inventory projects (i.e. National Forest Inventories) might request high precision (i.e. 5% at the 95% confidence level) for variables such volume, biomass, and carbon. Reaching this precision by using traditional methods of inventory can cost up to 50 EUR/ha while using LiDAR it costs 35 EUR/ha (for areas larger than 300 ha.), and these costs dramatically reduce with the increment of the areas -6 EUR/ha for areas equal or higher than 6,000 EUR; and even 3 EUR/ha for areas higher than 10,000- (Estimation of the cost for inventories with and without LiDAR, 6th Spanish Forestry Congress, Ortuno et al., 2013).
The use of Thermal information together with Airborne data has proven to be an effective way to assess the forest. However, there has not been so far, any company with the ambition and vision to exploit it commercially. Therefore, we do believe that creating a comprehensive tool (ThermoLiDAR) which allows the optimal integration of LiDAR and Thermal imagery will give us (to the SMEs) the chance to provide to the final user (forest managers) a wider spectrum of precise and updated information not only structural (as current service providers do) but also physiological. All in one same package, at lower costs.
For the reasons described above, we strongly believe that ThermoLiDAR Integrated Services in an enterprise with a huge potential and high probabilities of success in the middle term.
ThermoLiDAR technological tool will certainly bring us competitive advantages and let us expand our market, first in Europe and later on to other regions of the globe.
Related to dissemination activities consortium has developed a corporate image providing website design (www.thermolidar.com) and templates for posters and pamphlets. We have actively disseminating ThermoLidar project, progress and results through various media including websites (different to ThermoLidar’s one), articles, technical journals, professional meetings (i.e. ThermoLidar took and important part on the International Field-Map Users Conference hold in Czech Republic and at the ForestSAT carried out in Italy this year).
Related to ForestSAT “ForestSAT2014: a bridge between forest sciences, remote sensing and geo-spatial applications”, Forest Research and Arbonaut participated in the ForestSAT 2014 conference held in Riva del Garda (Italy) the 4-7 November 2014. The conference attracted 380 participants from 45 countries; a double number of participants compared to the previous event in Corvallis, OR (USA) in 2012. More info at: www.forestsat2014.com.
Forest Research presented an oral presentation and a poster and chaired an invited session on Forest Health and Forest Decline, presenting Thermolidar project.
(See attachments)
The poster attracted the attention of software developers and other research organisations with interest in the monitoring of tree health and forest inventory using LiDAR.
As product of our dissemination and exploitation effors we have completed a database with over 100 contacts (forest managers/owners) and starting cultivating business relationships with a dozen of those. A ThermoLidar business administration system was created to keep a dinamic follow up of business opportunites. We count now on a detailed business plan and marketing analyst which constitute the groundwork and starting point for our future busines activities. SMEPs have been trained by RTDs in the operation of the ThermoLidar technological tool. It is highly important issue when comes to the advisory that SMEs will give to their future customers. These contacts corresponds to previously gathered ones by the SMEs in their previous years of work to ThermoLiDAR; plus new ones obtained through the promotional, dissemination activities and market search. ThermoLidar Integrated Services will target primarily to the European market (75 potential customers, highly concentrated in Central Europe) due than the current process and tools have been developed in the context of Europe.
In within the market, we have found 9 segments as follows:
- Timber owners (owners of vast areas of forest for timber and logging)
- Control Agencies (Audit organisms).
- Forest Assessment Entities (carrying out commercial and non-commercial forest inventories).
- Forest Management (Administrative entities).
- Higher Education Providers (Forest related Colleges, Universities).
- Research Institutes.
- Policy Makers.
- Service Providers.
- Mixed (When activities carried out by the company are equally (bringing relatively the same revenue) belonging to different Market Segments)
We have service commitments for 3 clients (SMEs their selves which are the first beneficiaries of the ThermoLiDAR technology) and plan to substantially increase our customer database through trade fair exhibitions, pilot experiments, academic events, marketing campaigns, mail advertising and phone follow up.
Being offered by Forest Companies, ThermoLidar Integrated Services, can foreseen the needs of the forest managers and develop flexible ThermoLiDAR tools and methods that better fits to forest client requirements.
Thermolidar has turned in the cutting age technology for field data gathering of large forest stands. This is the reason why we see the state & private forest administrative entities as our biggest and most important potential clients.
ThermoLiDAR service provisioning is a highly innovative business and a new alternative of income for the SMEs. It renews the vision and increases the added value in the different targeted Market Segments.
Thermolidar has served as a first step to define barriers that prevent joint development of study of structural and physiological variables by aerial techniques, to provide reliable information to decision makers for the management of ecosystems and forest landscapes. With the execution of the project has achieved a series of results presented below:
- We have developed a scientifically sound data fusion between Lidar and thermal data sets capable of integrating the synoptic view about current forest resources provided by a forest inventory with the dynamic perspective of the physiological processes.
- We analysed and evaluated the structural and physiological data that can be captured remotely in order to create models of forest growth and evaluation of forest health
- We analysed and evaluated the structural and physiological data can be captured remotely in order to create models of forest growth and health assessment of forests, and we have the possibility to create thematic maps related to health forest and its structural state for different European ecosystems
- We have compiled the written procedures for the image data acquisition with the particular technology details of the flights, equipment used, and the information captured for the validation and calibration of the images, and we have defined a protocol for field data collection, modified in intensity and in number of parameters, to airborne LiDAR and Thermal sensors requirements
- Compilation of physiological data and remote captured information. Establish relationships between biophysical data collected at crown level with remote data at different spatial levels based on scaling up-techniques.
- Development of mathematic proceeding based on the relationships between biophysical data and remote sensing data.
- We have created operational software applications able to process both the LiDAR and Thermal imagery and producing spatially located datasets fully integrated into corporate GIS.
Other impact of the project is that it has served as a basis for the development of two university theses within the framework of cooperation between companies and research centers. Thermolidar has facilitated relationships company- research center and has served so that there is close cooperation, not only in Thermolidar and its future, but in their own areas of work of each company and outside the project, have established partnerships, collaboration agreements for undergraduate fellows, Specific agreements for the development of activities of interest to the Sme, etc... (Example: See attached Poster other collaboration UCO-AB).
The impact on employment is expected to have a reality in developing and achieving the objectives of the business plan, which will ensure employment in the three companies that are part of the consortium. Indirectly, the impacts on the European forest sector will facilitate data collection, and extend the range of information that can be obtained, and this will encourage the development of forestry and its related industries.
List of Websites:
www.thermolidar.com
The THERMOLIDAR project examines technological and methodological developments that may improve efficiency in the process of monitoring forest health using cost-effective alternatives to current field-based methods. On this context, the integrated estimation of structural and physiological variables on forest inventories using remote sensing tools is a huge challenge to European forest management.
The develop of new operational forest inventory methodologies and their integration at management and final users levels can lead to technically adapted proposals for forest needs. Field data collection aims to support remote sensing analysis (thermal and LiDAR), gain a better understanding on tree physiology and the disturbance mechanisms that interfere normal growth increments as dictated by site indices and, once those mechanisms are understood, infer methods for its detection over extensive areas.
The development of any forest management planning requires accurate information of forest composition, structure, vegetative condition and spatial and temporal dynamic. Impacts of climate change on trees and forests will interact to affect future trees and forests. New forest inventory data are needed in a context of global change to minimize anthropogenic stressors (pollution, forest decline, erosion and altered hydrology, loss of soil fertility, etc.), and to implement feasible environmental adaptation strategies.
The main THERMOLIDAR result is to combine airborne LiDAR with thermal remote sensing data sets to determine structural information of tree, canopy organization and physiological variables obtained at the scale of individual trees. This approach is essential in a context of climate change to obtain accurate forest inventory variables and evaluate forest management practices used in forest predictive models. The THERMOLIDAR project focus on the integration of LIDAR-thermal remote sensing technology and data processing to evaluate structural and physiological variables on forest ecosystems in software capable of processing both LiDAR and thermal imagery, as an operational tool, has also been developed.
The main results achieved on ThermoLiDAR project can be summarized in the THERMOLIDAR software, which is made up of four packages:
A. Processing. This package includes tools for conducting the processing of raw Thermal and LiDAR data in order to obtain the products required to achieve the parametric analysis of forest health assessment.
A.1. LiDAR processing:
A.2. Thermal image processing..
A.3. Ortho-rectification and the integration of LiDAR and Thermal data..
B. Forest Health Assessment. This package includes four tools for conducting the simultaneous analysis of thermal and Lidar data related to field data to evaluate the state and trends of forest health. A schematic description of data inputs and processes applied is included in Fig. 2
B.1. Forest Stand Segmentation (FSS).
B.2. Structurally homogeneous forest units (SHFU).
B.3. Health condition levels (HCL).
B.4. Forest Health Monitoring (FHM).
C. Data Analysis package: This section provides the tools for analysis and interpretation of results. Once thermal and LiDAR data have been processed in the previous sections, the user can generate the mapping needed to interpret the physiological state of the forest mass analysed.
Thermolidar partners have been active in preparing scientific publications and delivering conference presentations since December 2013 and the activities will continue after completion of the project period. One journal article is currently under per review and 4 other articles are still planned to be prepared for submission by the end of year 2014. Besides total 5 Master’s thesis works have been initiated and are close to be competed with support of the Thermolidar project. Altogether six oral and two poster presentations have been delivered by the end of November 2014.
Project Context and Objectives:
Project context:
The Forest management Plans are technical documents describing forest management objectives, strategies and commitments. It identifies intended methods of cutting, reforesting, and managing timber resources within the defined area of responsibility. The first step on a forest management plan is forest inventory: Field information collected to plan and reach the management goals".
Forests Inventories determine the present situation of forest stands and are essential to quantify and obtain indicators of biomass stocks, regeneration, protected flora and fauna, fire risks, stocking and overgrown landform, erosion problems, forest condition, water resources and infrastructures. Furthermore, through forest inventories we can obtain information to determine the role of forests such as carbon sinks (CO2), information extremely important to be taken into account in the development of carbon offset plans. Forests are the only sink of terrestrial carbon that can be managed. Therefore, the opportunity to improve its management is the only proven, safe and cheaper alternative to reduce the carbon of the atmosphere in a natural way.
Current forest inventories in Europe involve a large numbers of variables measured in the field; usually at least between 10 and 40 variables concerning e.g. site, volume and increment of growing stock, density, general capacity, forest damage and forest biodiversity. This information needs to be accurate and up to date in order to support wise decisions.
Forest inventory attributes have been typically measured in the field using hand-held equipment. However, despite the fact that field-based methods are regarded as highly accurate, they are extremely costly and time-consuming and sometimes affected by human induced errors. Moreover, traditional forests inventories become limited when collecting timely information about natural processes over large areas.
The collection of information over large areas is facilitated by the use of remote sensing techniques. For instance, Light Detection and Ranging (LiDAR) emerges as a valuable technique capable to provide a unique insight into forest structure in most forest inventories in Europe.
Discrete return small footprint LiDAR, is commercially available and used operationally in several EU countries. Increasingly, data providers in Europe have built the capability to facilitate data collection when required by the user by having more control on the date of the flight, the configuration of the sensor and the area to be covered. Research has been taking place in this area for at least 10 years, and has provided a suite of commercially available tools for forest inventory applications.
The focus of forest inventory has been on the estimation of basic measurement variables, such as volumes by tree species, basal areas, age and mean breast height diameter of stand. However, in the context of climate change, which is altering the physiological status of forests ecosystems, these traditional forestry measures become limited when confronted with the evolving nature of climatic change and the impact of disturbances. Now, the main problems that forests are facing are the increase of temperatures and climatic extreme events, like heat waves, droughts, floods and storms.
These extreme events increase the risk of forest fires and the susceptibility to pests and diseases, causing general forest decline. As a result, Sustainable Forest Planning requires not only information about forest composition and its structure but also about vegetation condition in a spatial and temporal context. Damage caused by biotic or abiotic agents, changes in nutrient balance and acidity and other diseases are being measured in a highly subjective way based on general tables or visual field observations.
The physiological status of forest ecosystems can be evaluated using variables like water stress, chlorophyll pigments, stomata conductance, photosyntesis efficiency, etc.
A Stress indicator suggested in several studies proposed by Jackson et al, (1977) is the temperature of the canopy as an indicator of tree transpiration (stomata conductance rate). Thermal remote sensing of water stress has been successfully applied to tree crop canopies based on high resolution thermal remote sensing imagery (Berni et al., 2009), airborne thermal imagery (Sepulcre-Cantó et al., 2007) and satellite thermal information in combination with 3D radiative transfer models to understand the effects of scene thermal components on large ASTER pixels (Sepulcre-Cantó et al., 2009).
At this point, there is a fundamental advantage in the early detection of physiological processes before their visual manifestation takes place (e.g. defoliation) for a quick intervention. To detect early critical physiological processes before its visual manifestations helps forest manager to take appropriate decisions to mitigate or reduce damage’s impact.
Forestry managers need new forest inventory information to respond to the new challenges and risks produced by anthropogenic stressors (pollution, acid rain, erosion, altered hydrology, loss of soil and fertility). It is necessary that the estimation of structural variables will be complemented with the monitoring of physiological processes in order to describe the complexity of the forest ecosystems and their response to external stressors.
THERMOLIDAR project aims to resolve these problems developing a forest management tool that, first of all reduces the amount of time dedicated to field data collections and the costs associated to traditional forest inventories. Then, inventories should also provide information about physiological variables at a forest stand level in the early stages in order to improve decision making. THERMOLiDAR will help to evaluate forest management practices used in forest predictive models.
To date, there is not any company in the world able to integrate structural and physiological information of forest stands in a reliable and competitive way.
There is a clear trend worldwide to use remote sensing for forest inventories. This is evident from the latest public tenders for forest management plans and forest inventories in several European countries. There is also a sharp increase in demand for LiDAR surveys by public administrations and forest companies alike. Although some companies in Europe provide a suite of commercial LiDAR tools for forest applications, there are still few algorithms capable of rendering valuable information about forest structure (both vertical and horizontal) that may be linked to fundamental physiological processes.
OBJECTIVES:
The main objective of THERMOLIDAR is to combine airborne structural variables derived from LiDAR with physiological information produced by thermal sensors. This innovative approach will enhance the capabilities of SMEs to provide commercial services to companies and government agencies managing European forests.
Overarching objectives
- Identify the obstacles for up-scaling structural and physiological variables to address the requirements of forest management at stand and landscape scales.
- Develop a scientifically sound data fusion between LiDAR and thermal data sets capable of integrating the synoptic view about current forest resources provided by a forest inventory with the dynamic perspective of the physiological processes.
- The assessment of those key structural and physiological tree variables that can be detected by remote sensing and are paramount for monitoring forest health and sustainable production.
- The creation of thematic cartography that spatially locate each variable. This information will be complemented by uncertainty values for each estimate also spatially located.
- Integrate derived cartography and field collected biophysical data for each study area into an accessible data repository.
- Design a protocol for image data acquisition with the specific technological details of the flights, the sensors applied, and the data capture for validation and calibration of the images.
- Establish a protocol for field measurements adapted in intensity and number of variables to airborne LiDAR and Thermal sensors requirements.
- Co-registration of the images acquired from different sensors into a stack. Orthorectification, radiometric and atmospheric corrections. Integration of image data into the geodatabase.
- Integration of biophysical data and remote sensing data. Relate and integrate the biophysical data collected at crown level with remote sensing data at different spatial levels based on scaling uptechniques.
- The development of algorithms based on the relationships between biophysical data and remote sensing data.
- Development of operational software tools capable of processing both LiDAR and Thermal imagery and generating spatially located datasets fully integrated to corporate GIS.
- SMEs training in the use of THERMOLiDAR tools.
Project Results:
The overall design of the project is to combine airborne structural variables derived from LiDAR with physiological information produced by thermal sensors.
To achieve the scientific and technological objectives for THERMOLIDAR project, the first 4 work packages were design:
1) Field data acquisition
2) Data analysis
3) Software development
4) Validation of thermal and LiDAR data application .
1) . Field data acquisition and airborne campaigns
WP 1 aims to support remote sensing analysis (thermal and LiDAR), gain a better understanding on tree physiology and the disturbance mechanisms that interfere normal growth increments as dictated by site indices and, once those mechanisms are understood, infer methods for its detection over extensive areas. On this WP an accurate methodology for field data collection was provided. These main highlights obtained were:
1. - Field data collected methodology oriented to physiological and structural variables related to forest damage. The proposed plots are oriented to capture inventory data for training and testing methods that estimate forest inventory variables using remote sensing data (LiDAR and thermal).
2. - Methods and instruments to be used in each country, divided by physiological and structural plots.
3. - Reviews of current knowledge related to forest damage in each study area.
One of the objectives of Thermolidar project is to design a protocol for image data acquisition with the specific technological details of the flights, the sensors applied, and the data capture for validation and calibration of the images, and also to establish a protocol for field measurements adapted in intensity and number of variables to airborne LiDAR and Thermal sensors requirements.
To achieve this goal an “Airborne image acquisition guideline” was done. (See attachment).
GEODATABASE:
The first stage for field data acquisition was the creation of a geo-database by gathering existing cartographic and biophysical information in each study area complemented with other datasets shared by each group during the life of this project, subjected to partial restrictions. Background information for each study area has been pre-analysed in order to get a preliminary knowledge about the forest condition at each location. The analysis has focused on two main issues: background information about previous datasets, such as LiDAR, thermal data and plot data and background information about forest management and forest health condition of each study area.
A review of the main spatial Database Management Systems (sDBMS) was performed in order to select an appropriate system that suits the main requirements of the project.
The implementation of the INSPIRE Directive in May 2007 was the first step towards the creation of a pan-European spatial data infrastructure to support the Community environmental policies. The Directive addresses 34 spatial data themes needed for environmental applications, with key components specified through technical implementation rules. This makes INSPIRE a unique example of a legislative “regional” approach.
The implementation of INSPIRE protocol is carried out based on the “Technical Guidelines” for Thermolidar project with the following results:
- Data is collected only once, kept where it can be maintained most effectively, and be readily and transparently available.
- It is possible to combine seamless spatial information from different sources across Europe and share it with many users and applications.
- It is easy to find what geographic information is available, how it can be used to meet a particular need, and under which conditions can be acquired and used.
(Table 1; Metadata protocol established for Thermolidar)
METADATA MANAGEMENT FOR THERMOLIDAR PROJECT:
Metadata can be defined as data about data (Pelkki and Arthaud, 2008). Information is provided which render the data useful, providing context, data properties (e.g. resolution, units of measurement), defining terms, noting any conditions, quality assessment and explaining applications. A good Metadata record enables the user of a dataset to understand the content they are reviewing, its potential and its limitations. Metadata describe the summary information or characteristics of a set of data.
The objective of this task is to define a protocol to consistently document datasets within the Thermolidar project using international Metadata standards. A suitable metadata structure is paramount within the Thermolidar project in order to assist with data interoperability and the exchange of information between international partners. This will help to add value to the project outcomes by facilitating access to deliverables.
Furthermore in order to support dissemination, data must be represented using commonly-understood terminology and containing information which will allow the user to fully understand where the data originated, what degree of processing has been undertaken and who can be contacted in case of queries.
For Thermolidar project metadata generated for the three principal dataset types (Raster, Lidar point clouds and Features data) have been integrated within the project Geo-Database which has been designed to manage this spatial information.
Within each group, the geo-database includes a specific metadata protocol (D 3.2).
LIDAR: Name, format, file tipe, processing level, processind method, projection system, date, resolution, sensor, scanning angle.
RASTER: File name, format file, file types, processing level, processing method, projection system, date, resolution (spectral/spatial), sensor.
FEATURES: File name, format file, files types or feature data, projection system, source, land cover type, field data type, sampling methodology, variables, object level.
For each study area a geodatabase is created following the same structure.
The CatMDEdit metadata editor tool has been selected to meet the metadata documentation needs of the Thermolidar project (http://catmdedit.sourceforge.net/). This open source software was developed by contributors at the Instituto Geográfico Nacional, the Advanced Information Systems Laboratory of the University of Zaragoza, and the GeoSpatiumLab S.L. all based in Spain.
STUDY AREAS:
Four study areas were selected located in Czech Republic (CZ), Finland (FI), United Kingdom (UK) and Spain (ES). The areas present different forest health condition and cover different types of forest species such coniferous species (mainly composed by Picea and Pinus) and Broadleaved species (mainly composed by Quercus and Fagus:
(FIG 1. WP1 location of the study plots)
The experimental scheme has been described independently for each study area based on the nature of the analysis. Early detection analysis of forest decline affected by different biotic threats was assessed on study areas located in UK, FI and ES. The areas selected have similar species than in UK and ES, such us Pinus sylvestris and, therefore, this will allow further research of forest dynamics affected by biotic threats.
(FIG 2 study area in Spain)
Spatial data background (field data, images, Lidar data, and terrestrial LiDAR) of each study area was collected as a start point for further measurements.
The use of LiDAR data to simulate forest Net Primary Production (NPP) was performed on the study area located in Bily Kriz (Czech Republic). The experimental design in this area is focused on the prediction of net primary production NPP of by integrating conventional and remote-sensing (LiDAR) data. These analyses were performed combining structural information retrieve form LiDAR data and models to simulate water and carbon fluxes such us, C-Fix or BIOME-BGC. Specifically, results obtained from the analysis of LiDAR data (estimates of stem volume) and tree age were assessed to predict the NPP of the examined ecosystems.
In Finland, UK and Spain growth models based on time series LiDAR data and thermal information were performed. The aim of this approach is to analyse the effect of biotic and abiotic factors on forest stands production, specifically on Picea abies forest affected by Typographus in Finland, and Pinus sylvestris affected by Dothystroma in UK and the same specie affected by water stress in Spain locations. Analysing growth models requires a modelling approach that relates the growth of individual trees to the resources available to them and thus, the required model simulate each individual tree as a basic unit and sum the resulting estimates to produce stand level values. Growth refers to the increase in dimensions of one or more individuals in a forest stand over a given period of time (e.g. volume growth in m-2 ha1y1). A growth equation might predict annual increment of diameter, basal area or volume in units per annum as a function of age and other stand characteristics, whereas a yield equation would predict the diameter, stand basal area or total volume production attained at a specific age. Forest productivity has been calculated using LiDAR analysis calibrated by field data inventories. The results have been compared to stand model predictions in order to highlight growth anomalies that can be attributed to forest decline processes.
Early detection of forest decline was performed on the Spanish and UK study areas. In UK For example Two pine stands were monitored in the field during an entire growing season. A 30 years old Lodgepine plantation affected by Dothistroma Red Band (Dothistroma septospora) and a 11 years old Scots pine stand affected by Lophodermella (Lophodermium seditiosum) was used to set up field instruments recording the physiological activity of individual trees with different levels of affection. Unaffected trees in the neighbourhood were used as control.
FIG. 3 field data collection in UK
In Spain the monitored trees consisted of 36 Pinus nigra and 36 Pinus sylvestris, located in three study areas (12 trees per study areas) in Filabres and 15 Oak trees in Huelva.
FIG 4 field data collection in Spain
Lidar and Thermal data was also acquired in each study area in order to obtain results on early detection of forest decline for Pinus and Quercus.
Fig 5. Sample of Lidar Data from Sierra de Filabres
Based on the analysis of field data, Lidar and Thermal data simple linear and multiple regression models were applied to evaluate the accuracy of DAI- and ALS-based metrics separately in retrieving in situ tree height. Overall, simple linear regression model results showed that the performance of most metrics extracted from Lidar point data was better than that of metrics extracted from photogrammetric DAI in estimating canopy height.
In order to analyze the factors that affected the above-mentioned relationships, we explored the effects produced by the percentage of crown overlapping and slope.
According to results, model performance was not significantly affected by the percentage of slope or crown overlapping. This result is relevant to demonstrate that the proposed method was not affected by the structure of the vegetation canopy or ground unevenness.
In Thermolidar project we propose a new approach to generate a canopy height model (CHM) from thermal imagery acquired with low-cost commercial color infrared (CIR) cameras using the structure from motion (SfM) method as an alternative to using more sophisticated technologies, such as Lidar, in the context of complex forest canopies. Both single-tree height and effective leaf area index (LAIe) at the crown level were estimated from DAI and ALS data and compared to field data. Robust models were developed for both variables with a set of metrics derived from ALS and DAI data independently. When estimating tree height, the reliability obtained with ALS-based Percentile 90 (P90) was slightly higher than that obtained with DAI-based metrics (P90 and minimum height (Hmin)), with a relative root mean square error (RRMSE) difference of 3% (yielding an RMSE = 0.51 m and an RMSE = 0.71 m, respectively).
FIG 6 Object-based delineation of crown trees based on Lidar and Thermal data
These results represent progress in the validation of 3D photogrammetric models applied to forest inventory development. In the case of LAI estimation, the reliability obtained with ALS-based metrics was slightly higher than that obtained with DAI-derived Normalized Difference Vegetation Index (NDVI) metrics (DAI NDVI P99), with an R-RMSE difference of 1.96% Our results demonstrated that the estimated error obtained for both the tree height and LAI parameters using low-cost airborne digital imagery was not significantly affected by the slope or the percentage of crown overlapping typically observed in an oak forest canopy.
This notwithstanding, the successful retrieval of single-tree and forest-stand biophysical variables using low-cost digital airborne imagery in other canopy types and terrain characteristics should be further analyzed and validated to assess issues related to canopy heterogeneity, crown dimensions and tree shape.
Some other conclusions related to forest decline based on Thermal and LiDAR data information are shown in fig. 7.
FIG. 7 Levels of affectation defined for Thermolidar project
First approximation to forest health condition analysis is the detection of decline processes with visual assessment of the damage. Visual assessment standards were adapted for both types of affectations, water stress detection on conifer forest and holm oak forest affected by Phytophthora. In order to quantify biomass defoliation LAI measurements were also performed with different sensors.
Although, traditional visual assessments can be subjective and therefore, it is not a direct measure of tree vigor. In contrast, a nonvisual method that allows tracking of pigment concentrations (e.g. chlorophyll) may provide an objective, early warning indicator of stands requiring remedial or salvage action before damage is visible and, potentially, before biomass loss occurs. Physiological parameters measured from the selected trees in Thermolidar project were total concentration of chlorophyll, needle water content and dry mass, stomatal conductance and crown temperature. These data were averaged from four measurements per tree during each period at the time of the Thermal imagery acquisition (12:00, GMT). Field Gas exchange measurements were performed in attached leaves at controlled CO2 external concentration (Ca = 350 ppm) and ambient relative humidity. Stomatal conductance (Gs) was estimated using gas exchange data and the total needle area exposed obtained from photos taken for each measurement.
Apart of these physiological measurements, continuous data was obtained from canopy temperature, photosynthetic activity, soil moisture, tree growth, sap flow and air temperature and humidity.
Analysis of forest condition.
Forest health was analysed based on the integration of Thermal and LiDAR data and forest health indicators collected during the field work. Forest structure was analysed based on data from 2010 and 2013. Dominant-height curves of Norway spruce was applied to analyse forest dynamics. A secondary step was to study the relationship between the survival and height increment of Norway spruce in relation to stand temperature. In order to accomplish this step, it is necessary to calculate the difference between stand temperature minus air temperature (Ts-Ta) and them analyses the spatial patterns described by (Ts-Ta) in relation to the spatial structure of forest stands. Thermal data was trained based on field data information of forest condition temperature. In order to accomplish this step, it is necessary to calculate the difference between stand temperature minus air temperature (Ts-Ta) and them analyses the spatial patterns described by (Ts-Ta) in relation to the spatial structure of forest stands. Thermal data was trained based on field data information of forest condition.
In UK the main objective was the analysis of pine forest affected by Dothistroma Red Band (Dothistroma septospora) based on LiDAR and Thermal data integration.
For Thermolidar two types of field data collection were planned for the validation and calibration of airborne sensors analysis:
-Plot data for validation of area-based analysis using airborne LiDAR
-Tree physiology for the monitoring of plant stress using Thermal imagery
This research is looking into an alternative approach to the area-based methods that relies on an effective combination of LiDAR metrics, existing sub-compartment data and stand models. This hybrid approach aims at the feasibility of replacing, or at least minimising, field data collection by using local stand models in combination with data inputs to this model estimated directly by LiDAR with good levels of accuracy (e.g. Top Height).
The aims of the proposed field data collection are:
1) The development of a low-cost solution for forest inventory that is robust and competitive in terms of cost and quality compared to current field data collection methods.
2) The production of a suite of multi-scale cartographic products fit for purpose, capable of addressing different management needs and fully integrated to local sub-compartment data and corporative GIS.
Related to Tree physiology for the monitoring of plant stress using Thermal imagery the main hypothesis is that it is possible to implement a robust monitoring system capable of detecting parts of the woodland under stress based on temperature differences observed at the top of the canopy. Therefore field data collection focused on a correct interpretation of fundamental physiological processes at the tree level that may infer that the plant is under stress. Once a process of stress has been identified, it is necessary to measure the magnitude of affection and the subsequent response at the top of the canopy. This analysis allows us to estimate the likelihood of a stress process being detected. Finally, once this process is well understood it is possible to calibrate a method of detection based on airborne Thermal imagery.
Two sample plots were monitored in the field during an entire growing season. Within each plot, four trees had a set of sensors attached to them following the experimental design outlined in Deliverible D.1.1. The selection criteria will follow a combination of dominance status and level of affection still to be determined.
Plot 1. Lodgepole pine plantation, > 50 years old. Height range 12 to 37 m. This area is affected by Dothistroma Red Band. The level of defoliation varies between 10 to 50%. This location has also being affected by windthrow in the storm that affected the Forest District in January 2012.
Plot 2. Scots pine plantation. This area was planted for recreation and landscaping 11 years ago. The tallest trees are 3-4 m tall. The location has serious problems of Sitka spruce encroachment that require a frequent elimination of spruce seedlings. These trees are affected by Lophodermella. Some individuals are starting to be affected by Dothistroma, which is unusual.
FIG 8. Dendrometer and thermistors installation
There is evidence to suggest that the Thermal properties of forest canopies could be used to determine stress, other than water stress, through their effects on tree physiology. Such effects include their influence on movement of water through the vascular system and hence on rates of evapotranspiration from leaf surfaces. However, not all stresses may result in warmer foliage temperatures and it may not always be possible to differentiate between water and other stresses using Thermal data alone. The ability to measure Thermal properties of plant canopies will of course be dependent on the degree of coupling between leaf and air temperatures and thus the tendency of wind and atmospheric moisture conditions to equalise the two. For example, although failed to observe Thermal differences between control and herbicide-treated forest canopies, they were attempting to compare differences in autumn and winter, when temperature differences may be less than maximal. Further research may be warranted into Thermal detection of stress in plant canopies and to the optimal timing of such measurements.
The algorithms for interpreting Thermal information may need to be reasonably sophisticated to compensate for emissions from non-plant surfaces but this could be aided by the combined use of Thermal and LiDAR analysis. This would require a good method of overlapping of these two datasets capable of seamlessly integrating the polar perspective of LiDAR point clouds with the perspective geometry obtained by Thermal imagery.
FIG 9. Rain gauge and air temperature-humidity sensor to calibrate leaf and air temperature
In order to calibrate a robust method for the detection of stress in the forest canopy based on Thermal differences between healthy and affected trees, field data collection was undertaken in a group of selected individuals in the monitoring area. The fundamental principles followed in this study is using the pipe model theory, which relates active live foliar biomass in the canopy with water and sap conductance at the stem. Each one of the trees being instrumented had their canopies measured. Then, sapwood depth was estimated using a model described by Beauchamp.
It is not entirely clear how a reduction of leaf area index as a result of a defoliation process may affect the formation of heartwood. As such its formation must be carefully regulated temporally and spatially, however the regulation of this process is not understood. Heartwood is thought to form from mid-summer, at the end of the growth season, and continues throughout the dormant period, as long as temperatures are maintained above 5°C.
Canopy temperature was constantly monitored using a Campbell IR 120 narrow field infra-red sensor. This sensor was mounted on a pole or a tower at less than 1 meter distance from the canopy. The sensor measured the average temperature of the top part of the canopy during the photosynthetic active part of the day. Data obtained at the canopy top was also used to calibrate the signal from the airborne Thermal measurements.
Physiological variables e.g. PRI, sap flow, stomatal conductance and transpiration was associated with canopy temperature collected from the tree samples. Stress dates were identified and differences between infected and healthy, young and old trees compared.
TDR sensors measured the water content in the soil. This information was used to measure water shortages during the growing season that may introduce anomalies in the sap conductance activity. The installation protocols followed the same procedures than the ones in the Spanish study areas.
One Skye 2 sensor per plot to continually measure PRI with wavelengths centred at 531nm and 570nm with a 5nm bandwidth. PRI sensors developed by Skye Instruments measured incident and reflected light simultaneously for accurate readings in all daylight conditions. This sensor provided information about the amount of light energy available for photosynthesis.
Main scientific and technical results for field data acquisition and airborne campaigns in Spain:
The experimental area is located in Sierra de Filabres (Almeria province, southeastern Spain) the driest region in Western Europe. The vegetation consists of a 40-year-old mixed pine afforestation of Pinus nigra Arnold and P. sylvestris L. A progressive decline episode of Scots pine stands has been observed since 2002, to such an extent that there is damage in 66% of the 6000 ha occupied by this species.
For Thermolidar project the monitored trees consisted of 36 Pinus nigra and 36 Pinus sylvestris, located in three study areas (12 trees per study areas). Results obtained in monitored trees related to structural and physiological variables obtained from the different levels of damage are shown in Fig 9.1. and 9.2.
(FIG 9.1 Structural parameters in pinus study area)
(FIG 9.2 physiological variables obtained in pinus study area)
The variables measured showed significant differences in the physiological status for each study area (p<0.05). As result of this step, three different groups with similar health conditions (damage levels) were defined.
To validate and work with this data airborne campaigns were conducted during spring, summer and autumn 2013 with two different sensors and acquisition settings. Thermal image acquisition was performed using the thermal sensor (FLIR SC660). During the same period, LiDAR data was also acquired from an ALS60 with a mean resolution of 4 points m-2.
(FIG 9.3 Thermal data acquired over conifer forest)
LiDAR data were classified as above-ground vegetation and ground terrain. A threshold value of 1 m above the ground surface was used to separate canopy echoes from echoes below the canopy to produce the LiDAR based canopy height model. Then, statistical metrics were calculated from the CHM. LiDAR metrics were calculated to estimate biophysical characteristics related to the structure of the canopy such as canopy cover, basal area, LAI, volume and biomass The following metrics were extracted: minimum, maximum, mean, median, standard deviation, variance, coefficient of variation, interquartile distance, skewness, kurtosis, AAD (average absolute deviation), L-moment (1-4), and percentile values (P5 to P95 in 5-unit intervals and P99).
The final step of data processing was the calculation of an integrated product for the correction and transformation of raw LiDAR and thermal data into the cross-georeferenced products: canopy height model and ΔTª (Temperature of the crown-Temperature of the air).
The second step was Data Analysis for forest health assessment.
Conifer forest health assessment was addressed by THsf using an object-based approach
(FIG 9.4 Forest stands delineation)
This is a very critical step in order to reliably relate thermal information with tree physiology minimizing the effects produced by soil background, vegetation structure and vegetation mixture. A previous step to the integration of LiDAR and thermal data was the definition of stand maps based on species and forest stand structure to determine their magnitude and spatial distribution over forested areas.
The results obtained from the stand delineation are presented in Fig. 9.4. Forest stands delineation was estimated at different scale levels using LIDAR data and object-oriented image segmentation. Multi-scale approach enables the production of forest health status and forest productivity cartographies at different spatial resolutions.
Related to Structurally homogeneous forest units the results provide the classification of forest stands in units (or objects) structurally different. Input data needed to run this process was obtained from LiDAR data. Fig. 9.5 shows an example of the SHFU defined for a conifer forest in Almería (Spain) The classification was based on the canopy height model (CHM) and was conducted using a k-means clustering. Forest stands (objects) were classified in three groups (cluster-SHFU, grey scale) using the tree heights derived from the CHM to define the thresholds between classes. The stand structural classification mapping was produced using a cluster approach for improving estimation and to produce wall-to-wall structural homogeneous object map. Several features were used for this classification, including: several LiDAR metric (derived from the CHM). The algorithm used was k-means clustering to partition n objects into k clusters (SHFU) in which each object belongs to the cluster with the nearest mean.
(FIG 9.5 SHFU mapping of the conifer forest)
Forest Health Monitoring cartographies were produced integrating the spatial information derived from previous steps:
1. Stand maps based on species and forest stand structure to determine their magnitude and spatial distribution over forested areas based on pre-defined object feature or a semiautomatic segmentation tool (FSS).
2. A (statistical) classification of objects based on forest structural variables and associated characteristics by using a cluster approach for improving estimation and to produce wall-towall structural homogeneous cover type map (vector file with structurally homogeneous stands SHFU).
3. A classification of forest health condition (groups with similar damage levels, HLC) based on qualitative methodologies as defoliation or physiological indicators.
Using this information, health vegetation condition differences at the stand level was estimated using two different types of physiological indicators: the clusters defined from water potential and pigment composition, related with the current physiological status of the vegetation (Fig. 9.6) and the leaf area index (LAI) related with the level of defoliation of conifer forest stands (Fig. 9.7).
(FIG 9.6 Mapping forest decline levels)
The algorithms applied for the estimation of forest health monitoring worked using an object oriented approach modeling the ΔT of each SHFU units and based on the HCL defined by users in previous stages. In the first case, the prediction of damage distribution based on levels of defoliation was based on the relationship between LAI measurements and LiDARbased metrics. In the second case, the prediction of damage distribution was based on the relationship between ΔTª (independent variable) and the classification of forest health condition at object level (dependent variable) by using a regression approach. It should be highlighted, that at this point of the analysis, SHFU mapping are products standardized by forest stands units defined from LiDAR-based metrics.
(FIG 9.7 measurement of forest stands defoliation)
2) Data analysis
The objectives of this WP were divided into four tasks:
1. Field data processing. This task included the processing of physiological and structural data in support of airborne data analysis. Deliverable D2.1.
2. Image data processing. Activities in this WP included the preprocessing and analysis of thermal and LiDAR datasets. Deliverable D2.2.
3. Data fusion. Deliverables in this section aimed at the development of a methodology for the analysis of airborne images based on data fusion techniques. Deliverable D2.3.
4. Algorithm development. This task aimed at the optimization of the input data for the retrieval of biophysical estimates.
FIELD DATA PROCESSING REPORT
The first deliverable D2.1 reported on field data capture and processing methods. A suite of sensors were attached to individual trees to measure their physiological cycles. The methodology outlined in the D2.1 report described the observed differences between healthy and infected trees. Some of the results obtained in the field were:
1.1. Water Potential indicates the capacity of the plants to conduct water from soil to atmosphere. This was measured using the pressure-bomb technique with needles, as suggested by Scholander et al.1965. Measurements were undertaken at the beginning, middle and the end of a growing season in Spain and the UK. No measurements were undertaken in Finland and Czech Republic. According to Fereres et al., (1999) the largest differences were supposed to happen in mid-summer, right at the peak of the photosynthetic activity.
(FIG 9.8 water potential measured)
However, the mean observations (in MPa) did not show substantial differences between healthy and infected trees.
(FIG 9.9 Average water potential measured)
The analysis of samples at different levels across the vertical profile of the canopy did not render conclusive results (Figure 9.10). Infected trees showed slightly larger water potential values than healthy ones. However, the variability of the measurements did not produce conclusive results.
(FIG 9.10 water potential differences)
Water potential measured in Aberfoyle, plot 1, at the end of July 2014. Measurements were taken in the low, middle and top canopy. The trees depictured in the figure were from 1 to 4 (left to right). Three measurements were done by taking samples at each canopy level for each individual tree.
1.2 Canopy temperature
Canopy temperature was measured in the field using Infra-red canopy temperature sensors (IR120). A sensor was attached to each tree being monitored in each plot at 1 m distance from the canopy, taking permanent measurements all year round every 10 mins.
The results showed no distinctive differences between healthy and infected trees (Figure 9.11).
(FIG 9.11 Canopy temperature differences)
In order to eliminate sources of noise (i.e. air temperature), a detrending technique was applied to the time-series observations of temperature (Figure 9.12). Detrending is a statistical or mathematical operation that removes trends from the series. Detrending is often applied to remove a feature thought to distort or obscure the relationships of interest. Detrending is also sometimes used as a pre-processing step to prepare time series for analysis by methods that assume stationarity. Many alternative methods are available for detrending. A simple linear trend in mean can be removed by subtracting a least-squares-fit straight line. More complicated trends might require different procedures.
(FIG 9.12 . Detrending of canopy temperatures in healthy and infected trees)
Once detrended, there were no statistically significant between canopy temperatures. The main conclusion was that sensors measuring temperatures from tree canopies at a distance are measuring an integral of foliage, branches, shadows and air boundary layer. This was also the same conclusion observed in the analysis of the thermal images. Therefore, it must be concluded that at the moment all the techniques developed in this project to measure canopy temperatures failed to obtain good measurements that can be used to infer the physiological status of the vegetation. More work is needed to be able to differentiation signal to noise ratios.
1.3 Dendrometers
Dendrometers were installed at breast height. Bark was removed without causing a lot of damage to the cambium. Metal rods were fitted on the trunk and securely tighten making sure they are level. Sensor was pressed against the bare surface of the trunk half way through allowing the sensor to either expand or shrink.
No differences were observed between healthy and infected trees in any of the plots (Figure 9.13 Y 9.14).
(FIG 9.13 Mean daily trunk shrinkage in Aberfoyle, plot 1)
(FIG 9.14 Mean daily trunk shrinkage in Aberfoyle, plot 2)
1.4 Leaf Area Index
Leaf area index (LAI) was measured in plots with Hemispherical photos and LAI-2000. The results were correlated to the visual estimation of defoliation as described by the Visual Assessment of Crown condition protocol described in Deliverable D1.2. The results showed that substantial levels of defoliation in the canopies of infected trees. As infected trees develop smaller canopies, growth is substantially delayed and this is also shown in the volume increments. Unfortunately, no destructive sampling was conducted in any of the study areas to quantify the extent of growth reduction by different defoliation levels.
IMAGE DATA PROCESSING
The activities in this task were described in Deliverables 2.2 and 2.3.
Thermal images were atmospherically corrected according to the DiStasio and Resmini (2010) method. The orthorectification processing of the thermal images was a necessary step to overlap them with the LiDAR products. This work was undertaken in ENVI.
A suite of tools already implement methods for data conversion, representation and management of LiDAR data. Other tools allow users to filter point clouds and classify ground and vegetation hits. Classified points can be converted into interpolated surfaces such as Canopy Height Models, Digital Terrain Models and Digital Surface Models. Point clouds can also be analysed to produce metrics of interest for the implementation of the Naesset method, extensively used in Nordic countries for forest inventory, and the hybrid method, proposed at Forest Research that combines LiDAR metrics with stand models (described in Deliverable D1.3). An example of cartographic products produced by the software is depicted in Figure 9.15.
(FIG 9.15 Estimations of top height for Stika spruce stands for Thermolidar software)
The analysis of Thermal and LiDAR images were implemented in a set of routines using Open Source libraries like SPDLib and Orfeo. The Thermolidar software implemented a set of routines in QGIS that gave SMEs the functional capability to deliver cartographic products mapping structural and physiological characteristics of the forest canopy being studied. A synoptic view of the LiDAR processing capabilities is described in Figure 9.16.
(FIG 9.16 Thermolidar software. Processing of LiDAR data)
DATA FUSION
The fusion between Thermal and LiDAR data were explained in Deliverable D2.3 D4.2 and the training materials described in D5.4. The method described in Figure 9.17 and 9.18 used LiDAR data to construct a set of homogenous objects using the routines already contained in the Orfeo toolbox that was integrated in the Thermolidar software.
(Figure 9.17. Forest Health Assessment)
(Figure 9.18. Forest Health Assessment2)
This method created thematic cartography showing the spatial distribution of Forest Health Classes, arbitrary defined by the user.
ALGORITHM DEVELOPMENT
Alternatively, another method developed by Forest Research was also implemented in the suite of tools contained in Thermolidar. The method is an adaptation of the model presented by Blonquist et al. (2009) that estimates canopy stomatal conductance by using a series of standard meteorological and airborne thermal data. The model calculates canopy stomatal conductance (gC , mol m−2 ground area s−1) from canopy temperature, air temperature, barometric pressure, relative humidity, net radiation, wind speed and plant canopy height. All these variables are being collected and used in the ThermoLiDAR project. The objective is to calculate canopy stomatal conductance in line with fundamental tree physiological principles, combining the information collected from airborne and field campaigns but also utilising other modelled data inputs. The target is to provide an estimation of canopy stomatal conductance in areas where direct measurements of canopy temperature may be seriously interfered by wind speed, humidity and clouds. For more details on the model and its assumptions see Blonquist et al. (2009) and the description of the method in D4.1.
The model was used to perform a sensitive analysis of the input variables used to calculate stomatal conductance across the study area. Figure 9.19 describes the relationship between healthy and infected crowns in relationship to air temperature, as already implemented in the Thermolidar software. Infected canopies, in this case Scots pine affected by Dothistroma Needle Blithe, show an average of two degrees difference after noon. Differences are difficult to distinguish before and after 4 p.m.
Figure 9.19. Canopy temperatures in Tree Crowns (Tc) of healthy and infected trees compared to Air Temperature (Ta).
The model uses temperature and other climatic and topographic variables to estimate stomatal conductance values (Figure 9.20). The model shows an even more clear average differences between trees around noon and until sunset.
(Figure 9.20. Modelled estimations of Canopy Stomatal Conductance)
This model was tested in the monitoring plots of the Aberfoyle study area, where the estimates were compared against real measurements of stomatal conductance after up-scaling measurements from leaf to canopy level (Figure 9.21). The comparisons proved inconclusive. Canopy temperature, as measured by the airborne systems, has some caveats. First of all, the sensor is gathering information about the green material on top of the canopy but interfered by other surfaces. As a result, every pixel contains information about non-photosynthetic material such as branches, twigs or stems, shadows, and even the temperatures of lower parts of the canopy and ground. Therefore, the temperature of the canopies is far lower than they should be and, according to the model, all trees are functionally dead. Secondly, wind is another important factor altering the measurements of the canopy temperatures. Despite all the flights were done during days of low wind speed, air mixing is inevitable and canopy temperature values oscillate greatly according to the wind flow. Finally, it is rather difficult, if not impossible, the measurement of large areas at the same time or at the time of maximum stomatal activity. The model showed important differences in terms of stomatal activities during the day. Therefore, it was concluded that thermal measurements should be restricted to small areas that can be flown in less than 1 hour after noon.
(FIG 9.21 Stomata conductance in Plot 1, Aberfoyle calculated by Thermolidar)
The FR-Hybrid Model for the estimation of forest inventory parameters was developed as an alternative to mainstream area-based methods. This method relies more heavily of available stand methods and minimises the need to undertake costly and extensive field data collection.
Data were measured in the field in a set of plots and compared to the biophysical parameters calculated using LiDAR metrics and the model, as described in D4.1. The best relationships were found using equations for intermediate thinning. These models eliminate the need for entering initial spacing and they have been derived from the original models in Edwards and Christie (1981). Results are presented as options for three different species: Sitka spruce, Norway spruce and Scots pine.
3) Software development
One of the main objectives of Thermolidar project is the development of operational software capable of processing both LiDAR and Thermal imagery and generating spatially located datasets fully integrated to corporate GIS.
The main structure of the software is summarized in this figure:
(FIG 10. general flowchart Thermolidar software)
ThermoLiDAR software is designed to provide the required tools to integrate Thermal and LiDAR information for the assessment of forest health and production. The software includes different modules and several settings offering different levels of processing to suit each needs. The main functionalities of the software are:
Tools for raw LiDAR and Thermal data processing, data exploring, quality assessment and visualisation.
Tools to integrate both data sources using suitable data fusion techniques.
Spatial statistical tools to analyse biophysical variables of the vegetation based on the integration of LiDAR and Thermal data integration.
Tools for mapping forest health condition and forest production dynamics and producing accuracy assessment.
The software uses field data measurements for calibration and validation, broaden the user choice to work at different levels that ranges from a total supervised analysis to a total unsupervised analysis. The THERMOLIDAR software comprises a selection of specialized tools for the assessment of the efficiency and productivity of forest ecosystems integrated on an Open Source Geographic Information System (GIS) platform.
The accuracy and optimization in the application of THERMOLiDAR tools depends on various factors:
1. The quality of the thermal imaging and Lidar data acquisitions (see D 1.3 and D 2.2)
2. The selection of key indicators to measure on the field (see D 1.3).
3. The accuracy of field data measurements (see D 1.3).
4. Forest health condition and type of forest damage (see D 1.3).
5. Forest heterogeneity and composition (see D. 1.1)
6. Accuracy required by the user (D 3.3).
ThermoLiDAR software has been developed as a QGIS plug-in and, once installed, it becomes part of the Processing Toolbox of QGIS. This makes possible that final users get free access to lots of GIS capabilities. The QGIS version currently supported by the ThermoLiDAR plug-in is QGIS 2.4.0 Chugiak.
ThermoLiDAR software has been divided into three different packages: Processing, Forest Health Assessment and Data Analysis.
3.1) Processing package
The Processing package includes tools for conducting the processing of raw Thermal and LiDAR data in order to obtain the products required to achieve the parametric analysis of forest health assessment.
A.1. Lidar processing. LiDAR data tool set for the generation of DSMs, DTMs, DVMs and vegetation statistical derivatives thereof.
LiDAR data is often provided as a number of tiles or flight lines. Depending on computers capacity, in some cases it might be convenient to merge flight files into a single file, or to divide data in overlapping tiles with an appropriate size.
Due to ThermoLiDAR software uses SPDLib suite for LiDAR data manipulation, data has to be converted into the native SPDLib file format. This means that LiDAR data must be converted into Sorted Pulse Data format (SPD) from LAS files, which is the file standard for the interchange of LASer data recommended by the American Society for Photogrammetry and Remote Sensing (ASPRS).
Noisy data can be eventually removed before being processed. Once the SPD files have been obtained, ground points are classified and then point heights relative to the ground are inferred. Ground and no-ground points are interpolated to generate DTM, DSM and CHM. From height information a range of metrics mainly applied to forestry applications -but not only- can be derived. Here below this workflow is depicted.
(FIG 12 WorkFlow lidar procedure)
Since LiDAR processing modules make use of SPDLib tools, the first step is to convert the input dataset into SPD files. There are two types of SPD files, non-indexed and indexed. A format translation module has been included in QGIS to this purpose. The module allows the conversion between different formats and it is also used to re-project data.
Thermolidar Software includes a merging module. In some situations it might be convenient to merge various files into a single SPD file. The merging module merges compatible files into a single non-indexed SPD file.
LiDAR data is supplied as flight lines or tiles with different shapes and sizes. It is always useful to divide laser data into equally sized square tiles, though. This helps to store, manage and access data easily. Single tiles should meet memory requirements in order to reduce computational times, which determines the maximum size of each file given an average point density. Overlapping zones between tiles help to prevent border errors and guarantee continuous raster models. ThermoLiDAR software has a built-in tool to create tiles given tiles size and the overlap. The module supports to create single tiles (SINGLE option) by supplying row and column, or to generate the complete set of tiles (ALL option).
(FIG 13 Split data into tiles)
Many factors may introduce errors in LiDAR point clouds, including water vapour clouds, multipath, poor equipment calibration, or even a flock of birds. In order to avoid further errors and artefacts in final digital models and poor assess of height metrics, those points have to be removed. Thermolidar project includes a module that removes vertical noise from LiDAR datasets by means of three different. Upper and lower absolute thresholds will clip the file to fit these values. Relative threshold will remove, for each bin within a SPD file, points outside the upper and lower values relative to the median height. Whilst global threshold will use the whole SPD file to calculate the median height and remove points relative to it.
Another tool is for point classification. Two different classification algorithms have been implemented into the plug-in. These algorithms also called filters allow the classification of the LiDAR points, identifying which point belong to the ground.
Related to normalise heights SPD files supports both elevation corresponding to a vertical datum and an above-ground height for each discrete return. Before data can be used for generating a Canopy Height Model (CHM) or any height related metric, height field has to be populated. This can be done in two ways. The simplest way is to use a DTM of the same resolution as the SPD file bin size (Image option). The disadvantage of using a DTM is that it is if the DTM is not accurate it can introduce some artefacts. Using this method the only parameters are the input files, both LiDAR file and the DTM, and an output file. The raster DTM needs to the same resolution as the SPD grid and it can be any raster format supported by the GDAL library.
For interpolation Thermolidar software includes a module to create Digital Terrain Models (DTMs), Digital Surface Models (DSMs) and Canopy Height Models (CHMs). To produce those products, it is necessary to interpolate a raster surface from the classified ground returns and top surface points. Create Digital Model within the ThermoLiDAR plug-in permits to generate these products by choosing model option.
(FIG 14 Example of a CHM generated with ThermoLiDAR)
Related to metrics ThermoLiDAR is able to calculate different metrics at the same time. Metrics can be simple statistical moments, percentiles of point heights, or even count ratios. Mathematical operators can also be applied to either other metrics or operators. The module supports different output data formats, that is, raster (all GDAL formats) and vector. Raster option extends the output to the entire input file, assessing the metrics for each pixel the final raster output and creating as many bands as metrics have been defined within the XML file. Vector option requires an input shapefile containing polygon entities. The output shapefile database will be populated with the metrics computed inside the polygons.
Software offers the possibility to create the file with a single metric selected from a list. However, the list also accept the options ALL, FOREST and PERCENTILES:
ALL: Selects all the metrics listed in the option list.
FOREST: Includes some metrics that are commonly used in forest applications. This metrics includes percentiles from 99th to 50th, groundCover, canopyCover, maxHeight and meanHeight.
PERCENTILES: Includes all percentiles from 99th to 10th.
A.2. Thermal image processing
Thermal processing tools data allows a user to perform thermal imaging calibration using the Emissive Empirical Line Method (EELM) which is a common method for airborne thermal data processing based on the ‘In-scene atmospheric correction methods. These approaches were developed to remove atmospheric effects from hyper-spectral imaging data allowing the user to utilize similar conditions to the atmosphere state. The advantage of using this type of methods over model-based methods based in radiative transfer theory is that they capture the true state of the atmosphere at the time of data collection and the relative low computational efforts required to perform the corrections (in comparison with radiative transfer models approaches). The main difficulty for in-scene method is getting correctly the field measurements parameters required for the correction algorithm.
This tool will also allow the user to calculate the difference between Crown Temperature minus Air temperature (Tc-Ta). This indicator has been widely demonstrated in Thermolidar project to be related with different physiological indicators such as stem water potential, stomatal conductance or sap flow rate. User need to select as input the thermal imaging and the air temperature collected in the same time of the airborne imaging acquisitions.
(FIG 11 workflow thermal procedure)
(FIG 11. 1 Thermal calibration output)
3.2) Forest Health Assessment
Forest Health assessment module consists of four different tools: Forest Stand Segmentation (FSS), Health Condition Level (HCL), Structurally Homogeneous Forest Units (SHFU) and Forest Health Monitoring (FHM). The following figure summarizes the main structure of this module and the input required throughout the process.
(FIG 15 Forest Health assessment module)
B.1. Forest Stand Segmentation (FSS). The processed image is decomposed into regions or objects. Object based delineation algorithms are applied with this tool to define forest stands units for further study.
Within OD tools, users are willing to choose between developing a semi-automatic segmentation and using a pre-defined object feature. Segmentation tools are based on algorithms that segment an image into areas of connected pixels based on the pixel DN value. ThermoLiDAR image segmentation tools will be based on region growing algorithms. The basic approach of a region growing algorithm is to start from a seed region (typically one or more pixels) that are considered to be inside the object to be segmented. The pixels neighbouring this region are evaluated to determine if they should also be considered part of the object. If so, they are added to the region and the process continues as long as new pixels are added to the region. Region growing algorithms vary depending on the criteria used to decide whether a pixel should be included in the region or not, the type connectivity used to determine neighbours, and the strategy used to visit neighbouring pixels. Image segmentation is a crucial step within the object-based remote sensing information retrieval process. As a step prior to classification the quality assessment of the segmentation result is of fundamental significance for the recognition process as well as for choosing the appropriate approach and parameters for a given segmentation task. Alternatively, user could be interested on using a pre-defined object feature. This object feature could be a segmentation shape file provided from other source or any other land cover mapping. Also, the user can use a pre-defined regular object, defining the size of the square to be used previously.
(FIG 16 Segmentation example)
B.2. Forest Health condition levels (HCL). Different physiological indicators from field data measurements are processed with this tool to define the ground truth condition of forest status. Health condition levels are statistically generated based in clustering and subsequently validated by ANOVA.
The most critical part in applying forest health condition indicators is the user’s accuracy defining forest degradation levels. Besides user’s training another critical factor is to select under analysis a robust physiological indicator and to carry out an accurate field measurements campaign.
Potential physiological indicators of forest decline such us pigment concentration, photosynthesis, respiration and transpiration rate holds great potential to shed light on the mechanisms and processes that occur as a result of drought stress. In the short-term, climate can change the physiological conditions of the forest resulting in acute damage, but chronic exposure usually results in cumulative effects on physiological process. These factors effects on the plants light reactions or enzymatic functions and increased respiration from reparative activities. Gradual decreases in photosynthesis, stomatal conductance, carbon fixation, water use efficiency, resistance to insect and cold resistance were found in most of trees which are very typical symptom of stress conditions.
Long-term exposure of water stress to a combination of high light levels and high temperatures causes a depression of photosynthesis and photosystem II efficiency that is not easily reversed, even for water-stress-resistant forest species.
One of the most widely physiological indicator applied in the analysis of long-term effect on forest health condition is de Leaf Area Index (LAI). The following is an example of the statistical analysis performs on LAI values measured from an Oak forest inventoried in the framework of THERMOLIDAR project.
(FIG 17 example LAI values oak forest Spain)
B.3. Structurally homogeneous forest units (SHFU). This option provides the tools required for the classification of forest stands structurally different. The data applied to perform this classification is defined by the user. In this report the main average height of the trees estimated based on LiDAR data has been applied as input data. Alternatively, user can provide external forest maps in a .shp format type with an attribute of the number of class.
The following figure shows an example of the units defined for the oak forest under analysis. Using a grey scale, trees were grouped in 3 classes with significant differences in terms of structural composition.
(FIG 18 structurally homogeneous forest units)
B.4. Forest Health Monitoring (FHM).
The main function of this tool is defining health condition differences in the vegetation at the stand level. Input parameters defined by users should mainly contain: thermal imaging data and the FSC Polygons (vector file with structurally homogeneous stands. Forest stands included in this analysis should be carried specifically based on one species. The user can perform a supervised or an unsupervised classification depending of the availability of field data measurements to define training areas.
It should be highlight, that at this point of the analysis, users are willing to obtain an integrated mapping of forest health distribution levels based on thermal data but also standardized by forest stands units defined from lidar-based metrics. The following figure, shows an example of the units defined for assessment of the status of forest condition. Using a colour palette, trees were grouped in different classes with significant differences in terms of structural composition and physiological status. The colour palette ranges from red to green, where red colour is relate with trees with high level of damage and green colour represents trees with optimum health condition.
(FIG 19 Structural composition and physiological status)
3.3. The Data Analysis package
This section provides the tools for analysis and interpretation of results. Once thermal and LiDAR data have been processed in the previous sections, the user can generate the mapping needed to interpret the physiological state of the forest mass analysed.
First, the user has a set of data obtained in the field of physiology which are analysed and grouped, using the tools available at the module Health Condition Level.
From the tools available in the structurally homogeneous forest units module, the user can perform a preliminary classification of the stands, based on structural homogeneity. This factor is important because thermal values behave differently according to the structure of objects.
Finally, from the Forest Heath classification tools the user has the necessary tools to perform a classification based on the thermal values for the various homogeneous units. To improve the classification, the user can define training plots according to data collected in the field of physiology, visually are established different levels of affection. Ç
(FIG 20 workflow data Analysis Package)
Health Condition Levels
Before proceeding with the classification of items by level of damage according to several variables taken in the field, we verify that the set of physiological variables follow a normal distribution. For this we use the Shapiro test.
Thermolidar software includes a Clustering tool. This tool allows us to group one or more physiological variables according to their degree of similarity between individuals in the sample. So, the goal of clustering is to determine the intrinsic grouping in a set of unlabelled data. But how to decide what constitutes a good clustering? It can be shown that there is no absolute best criterion which would be independent of the final aim of the clustering. Consequently, it is the user which must supply this criterion, in such a way that the result of the clustering will suit their needs.
To make this tool has been chosen by a hierarchical approach. The user-supplied items are categorized into levels and sublevels within a class hierarchy, forming a hierarchical tree structure.
(FIG 20.1 LAI measurements in Huelva)
After that an ANOVA test are conducted for each variable to indicate how well the variable discriminates between clusters.
The hypothesis is tested in the ANOVA is that the population means (the average of the dependent variable at each level of the independent variable) are equal. If the population means are equal, it means that the groups did not differ in the dependent variable, and consequently, the independent variable is independent of the dependent variable.
(FIG 20.2 clustering ANOVA)
Structurally Homogeneous Forest Units
Structurally Homogeneous Forest Units tool allows to minimize the effects of structure on the thermal information, and therefore allow us to obtain related health outcomes woodland. To do this, the software allows the user to define the structure of the stand from the height data. The calculation of uniformity is therefore a function of two variables directly derived from LiDAR data the 95th percentile obtained from MDV and the penetration rate, obtained from density points that penetrate the forest canopy.
(FIG 21 Homogeneous forest units)
Forest Health Classification
Unsupervised Pixel-based Classification: The user has a stratification of the study area, and classified based on the structure. Through this tool, a classification of the pixels of temperatures will be performed, based on the classification of defined structure. Thus, the output will be a raster temperature for each of the groups of homogeneity.
Unsupervised Object-oriented Classification: The user has a stratification of the study area, and classified based on the structure. Through this tool, a classification of the objects of temperatures mean will be performed, based on the classification of defined structure. Thus, the output will be a raster temperature for each of the groups of homogeneity.
(FIG 22 _unsupervised_object CLASIFICATION)
Supervised Pixel-based Classification: Similarly as in the previous point, the user can perform a classification of the temperature response to the stand of homogeneity. Unlike supervised classification, the user has a number of AOIs that guide the classification process.
Forest Models
To measure the field biological and physical properties (e.g. dominant height, mean diameter, stem number, basal area, timber volume, etc...) throughout entire woodlands is impossible. However, due to the characteristics of LiDAR technology it is possible to assess these properties in wide areas.
Two different models have been implemented in the ThermoLiDAR software to this purpose. The first one implements an empirical method where field measurements are correlated with LiDAR data. In this case, only a few sample plots are commonly measured in field in order to relate these measurements to canopy height metrics derived from LiDAR data. These relationships are then used to estimate and extend those characteristics to the area covered by LiDAR creating forest inventory cartography. This method is commonly called Scandinavian method.
The second is a hybrid method where some equation relating LiDAR canopy height and forest variables have been defined a priori. This method does not require in the field measurement and once is calibrated, it can be used anywhere else.
It is expected that TLDSf will improve the capabilities of SMEs and commercial opportunity on the field of forest health diagnosis and management and put them in a position where they can initiate a business activity.
4. Validation of thermal and LiDAR data application.
Once Thermolidar software was done a validation of each of the software tool components was performed. These validation tasks have focused on verifying the proper operation of each of the modules as well as a control study times for these.
SMEs are currently testing tools and checking the modules with Thermolidar data to verify the proper functioning of the software.
To perform validation testing of each of the software tools developed in the structure we focus on the processes defined in the following modules of the software.
1. Processing. The quality control tests focus the processing of raw LiDAR data and Thermal. The algorithms focus on generating a new corrected and ready to be used in the following product modules.
1.1. LiDAR processing. LiDAR data management and classification, generation of Digital Surface Models (DSMs), Digital Terrain Models (DTMs), Canopy Height Models (CHMs) and vegetation statistical derivatives thereof.
1.2. Thermal Image processing Thermal data tools set for the calibration of airborne thermal imaging.
2. Analysis. Quality control is aimed to verify a control time performance in generating maps that allow us to establish and define the various levels of health of forests with due regard to the initial information (temperature and heights LiDAR).
Potential Impact:
ThermoLiDAR Integrated Services has been conceived as an innovative service provisioning in the fields of LiDAR & Thermal data acquisition and subsequent analyst; in order to optimize and facilitate the decision making by forest managers in charge of large forest areas.
The use of LiDAR has been substantially increased during the last decade due to the savings which it represents when comparing with tradition and on-ground inventory practices. Different sources of literature show about 20% of savings for areas exceeding 300 Ha. in the case of Spain (Applications of LiDAR Technology in the Forestry Sector, Fabra et. Al., 2012), between 10-30% in United Kingdom (An industrial cost-benefit comparison of LIDAR derived forest inventory to a traditional inventory for harvesting, Murray Woods, et al., 2012), and it is assumed up to 60% savings for Finland in the case of private forests (TEKES - The NATIONAL Technology Agency of Finland, Maltamo et al., 2007).
Other important aspect to take into consideration is the precision reached by the two methods of inventories (LiDAR Vs traditional) and the costs it involves. Number of research or inventory projects (i.e. National Forest Inventories) might request high precision (i.e. 5% at the 95% confidence level) for variables such volume, biomass, and carbon. Reaching this precision by using traditional methods of inventory can cost up to 50 EUR/ha while using LiDAR it costs 35 EUR/ha (for areas larger than 300 ha.), and these costs dramatically reduce with the increment of the areas -6 EUR/ha for areas equal or higher than 6,000 EUR; and even 3 EUR/ha for areas higher than 10,000- (Estimation of the cost for inventories with and without LiDAR, 6th Spanish Forestry Congress, Ortuno et al., 2013).
The use of Thermal information together with Airborne data has proven to be an effective way to assess the forest. However, there has not been so far, any company with the ambition and vision to exploit it commercially. Therefore, we do believe that creating a comprehensive tool (ThermoLiDAR) which allows the optimal integration of LiDAR and Thermal imagery will give us (to the SMEs) the chance to provide to the final user (forest managers) a wider spectrum of precise and updated information not only structural (as current service providers do) but also physiological. All in one same package, at lower costs.
For the reasons described above, we strongly believe that ThermoLiDAR Integrated Services in an enterprise with a huge potential and high probabilities of success in the middle term.
ThermoLiDAR technological tool will certainly bring us competitive advantages and let us expand our market, first in Europe and later on to other regions of the globe.
Related to dissemination activities consortium has developed a corporate image providing website design (www.thermolidar.com) and templates for posters and pamphlets. We have actively disseminating ThermoLidar project, progress and results through various media including websites (different to ThermoLidar’s one), articles, technical journals, professional meetings (i.e. ThermoLidar took and important part on the International Field-Map Users Conference hold in Czech Republic and at the ForestSAT carried out in Italy this year).
Related to ForestSAT “ForestSAT2014: a bridge between forest sciences, remote sensing and geo-spatial applications”, Forest Research and Arbonaut participated in the ForestSAT 2014 conference held in Riva del Garda (Italy) the 4-7 November 2014. The conference attracted 380 participants from 45 countries; a double number of participants compared to the previous event in Corvallis, OR (USA) in 2012. More info at: www.forestsat2014.com.
Forest Research presented an oral presentation and a poster and chaired an invited session on Forest Health and Forest Decline, presenting Thermolidar project.
(See attachments)
The poster attracted the attention of software developers and other research organisations with interest in the monitoring of tree health and forest inventory using LiDAR.
As product of our dissemination and exploitation effors we have completed a database with over 100 contacts (forest managers/owners) and starting cultivating business relationships with a dozen of those. A ThermoLidar business administration system was created to keep a dinamic follow up of business opportunites. We count now on a detailed business plan and marketing analyst which constitute the groundwork and starting point for our future busines activities. SMEPs have been trained by RTDs in the operation of the ThermoLidar technological tool. It is highly important issue when comes to the advisory that SMEs will give to their future customers. These contacts corresponds to previously gathered ones by the SMEs in their previous years of work to ThermoLiDAR; plus new ones obtained through the promotional, dissemination activities and market search. ThermoLidar Integrated Services will target primarily to the European market (75 potential customers, highly concentrated in Central Europe) due than the current process and tools have been developed in the context of Europe.
In within the market, we have found 9 segments as follows:
- Timber owners (owners of vast areas of forest for timber and logging)
- Control Agencies (Audit organisms).
- Forest Assessment Entities (carrying out commercial and non-commercial forest inventories).
- Forest Management (Administrative entities).
- Higher Education Providers (Forest related Colleges, Universities).
- Research Institutes.
- Policy Makers.
- Service Providers.
- Mixed (When activities carried out by the company are equally (bringing relatively the same revenue) belonging to different Market Segments)
We have service commitments for 3 clients (SMEs their selves which are the first beneficiaries of the ThermoLiDAR technology) and plan to substantially increase our customer database through trade fair exhibitions, pilot experiments, academic events, marketing campaigns, mail advertising and phone follow up.
Being offered by Forest Companies, ThermoLidar Integrated Services, can foreseen the needs of the forest managers and develop flexible ThermoLiDAR tools and methods that better fits to forest client requirements.
Thermolidar has turned in the cutting age technology for field data gathering of large forest stands. This is the reason why we see the state & private forest administrative entities as our biggest and most important potential clients.
ThermoLiDAR service provisioning is a highly innovative business and a new alternative of income for the SMEs. It renews the vision and increases the added value in the different targeted Market Segments.
Thermolidar has served as a first step to define barriers that prevent joint development of study of structural and physiological variables by aerial techniques, to provide reliable information to decision makers for the management of ecosystems and forest landscapes. With the execution of the project has achieved a series of results presented below:
- We have developed a scientifically sound data fusion between Lidar and thermal data sets capable of integrating the synoptic view about current forest resources provided by a forest inventory with the dynamic perspective of the physiological processes.
- We analysed and evaluated the structural and physiological data that can be captured remotely in order to create models of forest growth and evaluation of forest health
- We analysed and evaluated the structural and physiological data can be captured remotely in order to create models of forest growth and health assessment of forests, and we have the possibility to create thematic maps related to health forest and its structural state for different European ecosystems
- We have compiled the written procedures for the image data acquisition with the particular technology details of the flights, equipment used, and the information captured for the validation and calibration of the images, and we have defined a protocol for field data collection, modified in intensity and in number of parameters, to airborne LiDAR and Thermal sensors requirements
- Compilation of physiological data and remote captured information. Establish relationships between biophysical data collected at crown level with remote data at different spatial levels based on scaling up-techniques.
- Development of mathematic proceeding based on the relationships between biophysical data and remote sensing data.
- We have created operational software applications able to process both the LiDAR and Thermal imagery and producing spatially located datasets fully integrated into corporate GIS.
Other impact of the project is that it has served as a basis for the development of two university theses within the framework of cooperation between companies and research centers. Thermolidar has facilitated relationships company- research center and has served so that there is close cooperation, not only in Thermolidar and its future, but in their own areas of work of each company and outside the project, have established partnerships, collaboration agreements for undergraduate fellows, Specific agreements for the development of activities of interest to the Sme, etc... (Example: See attached Poster other collaboration UCO-AB).
The impact on employment is expected to have a reality in developing and achieving the objectives of the business plan, which will ensure employment in the three companies that are part of the consortium. Indirectly, the impacts on the European forest sector will facilitate data collection, and extend the range of information that can be obtained, and this will encourage the development of forestry and its related industries.
List of Websites:
www.thermolidar.com
