A Comparison of Multiple Algorithms for Deriving Regional-Scale Biomass Maps with Airborne Lidar Metrics and Multispectral Datasets

Aboveground biomass maps spatially exhibit the distribution of aboveground biomass of rangelands and forests. Thus, this research investigated feasible approaches to generate an aboveground biomass map of rangelands and forests. We utilized three approaches to generate regional-scale aboveground biomass maps with different combinations of airborne lidar metrics and multispectral dataset. The first approach was the cokriging interpolation algorithm which calculated the aboveground biomass based on the near-infrared band (NIR), normalized difference vegetation index (NDVI), and the in-situ aboveground biomass samples. The second approach established the regression relationship of the in-situ aboveground biomass samples with lidar metrics and a multispectral dataset to build a biomass map. The third approach imported the airborne lidar metrics and multispectral dataset into the random forest algorithm to generate an aboveground biomass map. The objective of this map comparison is to find the most suitable approach for deriving an aboveground biomass map with airborne lidar and multispectral remote sensing technologies. The results showed that the cokriging interpolation algorithm was strongly influenced by the in-situ aboveground biomass samples. The regression approach showed that the lidar metrics and multispectral dataset have limitations in explaining the variance associated with the biomass map. In summary, results proved that the random forest method is the most reliable and reasonable approach to generate an aboveground biomass map. The point density of airborne lidar data constrains the accuracy of the map and the NAIP image dataset was useful to create an acceptable aboveground biomass map when the lidar data is difficult or expensive to acquire.