Classification of Forest Structural Types with LiDAR Remote Sensing Applied to Study Tree Size-Density Scaling Theories

Periodic Reporting for period 1 - LORENZLIDAR (Classification of Forest Structural Types with LiDAR Remote Sensing Applied to Study Tree Size-Density Scaling Theories)

Periodo di rendicontazione: 2016-09-01 al 2018-08-31

Light detection and ranging (LIDAR) remote sensing provides detailed measurements of vegetation height, density and spatial heterogeneity. Until recently, the acquisition of airborne LIDAR surveys was considered too costly for operational applications at broad scales. However, today a number of national surveying programmes are producing publicly-available LIDAR datasets covering entire countries. These low-density datasets are also demonstrably useful for forest inventory and ecological applications. As national LIDAR surveys are becoming more common, unique opportunities exist for generating habitat indicator variables and classifications that can be consistently obtained throughout entire countries. There is still a need to clarify and harmonize what indicator variables and classifications can be derived from LIDAR datasets and employed for transnational comparisons and monitoring.

Objectives:
LORENZLIDAR tested the feasibility of a simple methodology for classification of forest structure from low-density LIDAR datasets acquired by nation-wide programmes, adapting it to assure its validity across biogeographical regions in Europe. The methodology developed was built upon an analysis framework based on Lorenz curve analysis. The Lorenz curve is a method widely used in econometrics to measure the degree of wealth or income inequality in a society. LORENZLIDAR adapted the method to forest ecosystems and ecology, and to the assessment of its structural properties using LIDAR.

Conclusions:
LORENZLIDAR concluded with a two-tier methodology for forest structure classification. The higher tier are parameters can be extracted from a Lorenz curve constructed from the sizes of trees in a forest, and it is used to determine the shape of the tree diameter distribution. In a lower tier, mean diameter (QMD) and stem density (N) were used to discriminate young/mature and sparse/dense subtypes . The resulting forest structural types relate to tree size-density scaling relationships: metabolic scaling or demographic equilibrium. Furthermore, we used similar structural predictors derived from LIDAR to predict the forest classes, obtaining reliable classification accuracies. The simplicity of the developed two-tier approach paves the way toward transnational assessments of forest structure across bioregions.

Research work performed:

Firstly, we tested the reliability of applying the Lorenz curve method with low-density LIDAR datasets typically acquired by nation-wide programmes. We resolve the optimal scale for forest structural type determination, which was 250-450 m2 (Adnan et al., 2017). Research also resolved that LIDAR pulse density should be increased to around 3 pulses per m2, in order to reliably characterize the structure of forests on the grounds of the relationships between LIDAR metrics and Lorenz indicators.

Secondly, we controlled potential sources of variability in order to expand the definitions of forest structural types. New forest structural classes were declared according to the findings (Adnan et al., 2018), first categorizing into “even-sized” and “uneven-sized”, and then subdividing into “young / mature” and “dense / sparse”. We achieved the discrimination of forest structural types using only information derived from LIDAR, without a priori knowledge of the study area (Valbuena et al. 2017b). The methodology was adapted for a structural classification of forest ecosystems across European biogeographical regions. We analysed the variability found between the different pilot areas and study the causes for these differences. The final conclusion of LORENZLIDAR was a region-independent classification of forest structure from LIDAR data.

Results:
Adnan et al. (2018) Forest Ecology and Management [in press]
Valbuena et al. (2017a) Ecological Modelling 366:15.
Valbuena et al. (2017b) Remote Sensing of Environment 194:437.
Adnan et al. (2017) Canadian Journal of Forest Research 47:1590.

Teaching.
▪ Remote sensing and REDD+: an R code practical. University of Cambridge Conservation Research Institute.
▪ LiDAR remote sensing of forests. Department of Geography.
▪ Conservation Science. Oxbridge tutorial undergraduate group and dissertation supervision. Depts Geography and Zoology.
Many of these were done under the umbrella of Ruben Valbuena’s participation at the Teaching Associate Programme (TAP) (accredited by UKPSF), issuing an application to become a Fellow of the UK Higher Education Academy (HEA).

Supervision.
Syed Adnan (PhD, University of Eastern Finland). His work has been directly related to LORENLIDAR with some of the most relevant resulting publications.
Noemi Ammaturo (Undergraduate in Maths, University of Cambridge). Completed an internship during summer 2017 conducted by Rubén Valbuena. She worked on adaptations of the Lorenz method to forest science and ecology using mathematical development.
Marta Galluzi (PhD, Universita Studi Firenze). Completed an internship in 2018 conducted by Ruben Valbuena, working on the implications of LORENLIDAR to ICP Forests in the relationships of that data with forest fragmentation.

Communication & Outreach:
Forest Spatial Assessment Tools, ForestSAT 2018. University of Maryland. Washington DC, US.
International Union of Forest Research Organizations IUFRO Congress 2017. Leipzig, Germany

LORENZLIDAR work included a secondment at the UN Programme- World Conservation Monitoring Centre (UNEP-WCMC) with the intention to use the indicators and classifications developed in LORENZLIDAR for assisting the development of essential biodiversity variables (EBVs) from LIDAR.

Progress beyond the state-of-the-art
Results imply that ecosystem structure may be informed directly from LIDAR data with no requirement for field information. The indicators and classifications derived can be used to summarize the 3D complexity of ecosystem structure using few descriptors: vegetation height, density, and vertical heterogeneity. Vegetation structural descriptors allow analyses of long term anthropogenic and non-anthropogenic ecosystem disturbance and dynamics. They can also be employed to process regional-scale LIDAR data into categorical classes representing natural groupings of habitat structure. The independence from the use of field data provides an added value to airborne LIDAR, which can thus be directly employed in providing structural information across large areas – regional or national coverages – to directly analyse habitat extent connectivity and fragmentation.

Wider societal implications of the project:
The analysis and classification of forest structural types is of interest for the conservation and promotion of biodiversity and other ecosystem services. Nature conservation bodies, landscape planning and ecotourism stakeholders are among the activities and professionals potentially interested in the application of LORENZLIDAR research results. Thanks to its simplicity, the approach outlined by LORENZLIDAR could be beneficial for European efforts for harmonizing national forest inventories, initialized by the COST Action E43, such as ICP Forests. More globally, it could assist the development of EBVs from LIDAR, and contribute to the use of remote sensing to inform policy-makers on progress towards sustainable development goals and biodiversity targets (Convention of Biological Diversity 2020 Aichi biodiversity targets and UN 2030 Global Sustainable Development Goals).

Classification into forest structural types derived in LORENZLIDAR

Periodic Reporting for period 1 - LORENZLIDAR (Classification of Forest Structural Types with LiDAR Remote Sensing Applied to Study Tree Size-Density Scaling Theories)

Condividi questa pagina

Scarica