CORDIS - EU research results

Integration of innovative remote sensing techniques for optimum modelling of tropical forest primate habitat and carbon storage

Periodic Reporting for period 1 - FOREST 3D - ECOCARB (Integration of innovative remote sensing techniques for optimum modelling of tropical forest primate habitat and carbon storage)

Reporting period: 2015-10-01 to 2017-09-30

Tropical forests are being destroyed at a rate of 1.5 acres every second due to human activities , thereby accelerating climate change through impacts on the carbon cycle, and causing the extinction of species dependent on these habitats. There is a need to develop methods that can rapidly assess tropical forest structure and relate this to carbon stocks stored in tree biomass and to habitat quality for keystone species, like primates.

The aim of this project was to utilise innovative methods of acquiring detailed 3-dimensional data of tropical forests at a landscape-scale, using remote sensing systems on aircraft and unmanned aerial vehicles (UAVs), to model primate habitat and to measure forest carbon stocks. The objectives of the project were to link forest structure in 3-D directly to primate behaviour and forest use for the first time, and to develop cost-effective remote sensing methods using UAVs for monitoring changes in habitats and forest carbon stocks. This project was part of the Landscape Ecology and Primatology (LEAP) project at Bournemouth University (BU), and ran from Oct 2015 to Sept. 2017.
Progress in Achieving Research Objectives
Data from Airborne Laser Scanners (ALS) for Batang Toru in Sumatra, were provided by the Sumatran Orangutan Conservation Program (SOCP). Data from Unmanned Aerial Vehicles (UAVs) were also provided by SOCP for Batang Toru and Sikundur; whilst primate data have been collected by SOCP and post-graduate students from BU in Sikundur (Fig. 1).

1. Work Package 1: Airborne Data Processing and Analysis
1.1 ALS Data Processing
ALS data points from the ground were processed using FUSION software to generate a Digital Terrain Model (DTM) with a cell-size of 1 m. The DTM made it possible to estimate the above-ground height of vegetation points, and a Canopy Height Model (CHM) was generated for the entire site (162 km2). More than 40 variables at a cell size of 20 m were generated from the ALS data for modelling the forest structure in the area.
• The DTM and the CHM were provided to SOCP as tools for the development of an Environmental Management Plan for Batang Toru.

1.2 Identifying homogeneous forest patches based on individual trees
A history of logging and clearing of land for agro-forestry, selective logging to establish ‘forest gardens’, and natural dynamics has created a mosaic of forest patches in the study area. Patches of similar composition of trees could be visually identified in the CHM, based mainly on the heights and density of trees. Individual canopy trees were detected from the CHM in FUSION software. A method to group trees with similar attributes into forest patches based on Thiessen polygons and k-medoids clustering was developed (Figs 2 & 3), combining the advantages of both raster and individual tree–based methods.
• The work was presented at the annual conference of the Remote Sensing and Photogrammetry Society (RSPSoc), UK in September 2016; winning one of the two Merit Awards
• The results were published in the International Journal of Earth Observation and Geo-information (Volume 55, March 2017, Pages 68–72).
• The results were discussed with researchers at SOCP while the Fellow was in Indonesia in January 2017.

1.3 Habitat Classification
In addition to the method based on Thiessen polygons, a grid-based solution (Fig 4) was also explored to delineate the forest patches over a large area, based on canopy cover and canopy height. The patches were clustered to identify natural forest classes in the study area (Fig 5); field data collected by SOCP were used for labelling the clusters.
• The initial results were shared with SOCP and, through them, with Sarulla Operations Ltd., a geo-thermal company that commissioned the collection of ALS data, for the preparation of an Environmental Management Plan. The habitat map was finalised during a visit to Indonesia in January 2017, after discussions with researchers at SOCP.
• A manuscript based on this work will be submitted to Remote Sensing in the last quarter of 2017.

2 Work Package 2: Primate Modelling
2.1 Generation of forest structure variables from UAV data
Three UAV datasets from Sikundur were acquired by Graham Usher (SOCP). The datasets were processed in Agisoft PhotoScan software to generate a Digital Surface Model (Fig 6). Variables related to canopy roughness were generated within 25×25 m grids. These were used by C. Marsh, a PhD student at BU, to explore the relationship between the variables generated from UAV data and field data on vegetation and micro-climate. A strong relationship was found between micro-climate data and UAV-based variables. Micro-climate is considered to be a determinant of the level of activity of orang-utans throughout the day. Variables within 10×10 m grids were also generated which will be used by J. Abernethy, a PhD student at LJMU, for his work on orang-utan density modelling.
• This work was presented by C. Marsh at the joint meeting of the International Primatological Society and the American Society of Primatologists in August 2016, in Chicago.
• This will be written up for submission to Behavioural Ecology.
• A workflow for converting photographs acquired by UAVs to variables that could be used for primate modelling was developed, and made available to primatologists and other researchers through the LEAP website.

2.2 Detection of emergent trees from UAV data
Arboreal primates spend a significant part of their days moving through the canopy and their nights sleeping in trees. Primate species such as gibbons and siamangs select sleeping sites in emergent trees, which are taller than surrounding trees with exposed crowns. Traditional plot-based ground surveys have limitations in detecting and mapping these trees at the landscape level, especially in dense tropical forests. A method was therefore developed to detect emergent and potential sleeping trees in a tropical rainforest using data from UAVs (Fig 7), after discussions with E. Hankinson, a PhD student at BU experienced in field-based vegetation and primate studies in Sumatra.
• A manuscript based on the results will be submitted to the International Journal of Applied Earth Observation and Geoinformation in the last quarter of 2017.

3 Work Package 3: Carbon Modelling
Tree or canopy height is an important attribute for carbon stock estimation, forest management and habitat quality assessment. ALS data have advantages over other remote sensing techniques for describing the structure of forests, however sloped terrain can be challenging for accurate estimation of tree locations and heights based on a Canopy Height Model generated from ALS data. The influence of terrain slope and crown characteristics on the detection of treetops and estimation of tree heights was assessed using the ALS data in Batang Toru (Fig 8). A model was developed for idealised tree crowns, which showed that errors occur only when terrain angle exceeds the crown angle, with the horizontal displacement equal to the crown radius. The results are especially relevant for biomass and carbon stock estimations in tropical forests where there are trees with large crown radii on slopes.
• A research paper based on this work has been published online in the International Journal of Applied Earth Observation and Geoinformation (Volume 65, March 2018, Pages 105–113).