Above ground biomass estimation across forest types at different degradation levels in Central Kalimantan using LiDAR data

The quantification of tropical forest carbon stocks is a key challenge in creating a basic methodology for REDD (reducing emissions from deforestation and degradation in developing countries) projects. Small-footprint LiDAR (light detection and ranging) systems have proven to successfully correlate to above ground biomass (AGB) estimates in boreal and temperate forests. Their applicability to two different tropical rainforest types (lowland dipterocarp and peat swamp forest) in Central Kalimantan, Indonesia, was tested by developing multiple regression models at plot level using full waveform LiDAR point cloud characteristics. Forest inventory data is barely available for Central Kalimantan's forests. In order to sample a high number of field plots the angle count method was applied which allows fast sampling. More laborious fixed-area plots (three nests of circular shape) were used as a control and approved the use of the angle count method. AGB values, calculated by using existing allometric models, were in the range of 15–547 Mg ha−1 depending on forest type, degradation level and the model used for calculation. As expected, logging resulted in significant AGB losses in all forest types. AGB-prediction models were established for each forest type using statistical values of the LiDAR point clouds and the forest inventory plots. These regression models were then applied to six LiDAR tracks (altogether with a size of 5241 ha) covering unlogged, logged and burned lowland dipterocarp and peat swamp forest. The regression analysis showed that the 45th and 65th percentiles and the standard error of the mean explain 83% of the variation in lowland dipterocarp forest plots (RMSE = 21.37%). The best model for peat swamp forest could only explain 32% of the AGB variation (RMSE = 41.02%). Taking both forest types together explained 71% (RMSE = 33.85%). Calculating AGB for whole LiDAR tracks demonstrated the ability of this approach to quantify not only deforestation but also especially forest degradation and its spatial variability in terms of biomass change in different forest ecosystems using LiDAR transects. Concluding it can be stated that the combined approach of extensive field sampling and LiDAR point cloud analysis have high potential to significantly improve current estimates of carbon stocks across different forest types and degradation levels and its spatial variation in highly inaccessible tropical rainforests in the framework of REDD.

► Field measured above ground biomass (AGB) estimates of Central Kalimantan's forest ranges from 15 to 547 Mg ha−1 depending on forest type and degradation level.
► The angle count sampling method was tested to be adequate for fast sampling in tropical forest.
► AGB-predicting regression models for peat swamp forest and lowland dipterocarp forest could be developed by linking field inventory and LiDAR data within 1-ha-plots.
► Models explain 83% of the variation in lowland dipterocarp forest plots (RMSE = 21.37%), 32% in peat swamp forest plots (RMSE = 41.02%) and 71% taking both types together (RMSE = 33.85%).
► Quantifying above ground biomass of whole LiDAR tracks showed the ability of our approach to extract spatial biomass variability from LiDAR data. The detection of forest degradation, i.e. logging could be improved compared to the analysis of Landsat imagery.