Forest aboveground biomass estimation using machine learning regression algorithm in Yok Don National Park, Vietnam

Abstract

Forest biomass is one of the key measurement for carbon budget accounting, carbon flux monitoring, and climate change studies. Hence, it is essential to develop a credible approach to estimate forest biomass and carbon stocks. Our study applied Sentinel-2 satellite imagery combined with field-measured biomass using Random Forest (RF), a machine learning regression algorithm, to estimate forest aboveground biomass (AGB) in Yok Don National Park, Vietnam. A total of 132 spectral and texture variables were extracted from Sentinel-2 imagery (February 7, 2017) to predict AGB of the National Park using RF algorithm. It was found that a combination of 132 spectral and texture variables could predict AGB with an R2 value of 0.94, RMSE of 34.5 Mgha−1 and % RMSE of 18.3%. RF regression algorithm was further used to reduce the number of variables in such a way that a minimum number of selected variables can be able to estimate AGB at a satisfactory level. A combination of 11 spectral and texture variables was identified based on out-of-bag (OOB) estimation to develop an easy-to-use model for estimating AGB. On validation, the model developed with 11 variables was able to predict AGB with R2=0.81, RMSE=36.67 Mg ha−1 and %RMSE of 19.55%. The results found in the present study demonstrated that Sentinel-2 imagery in conjunction with RF-based regression algorithm has the potential to effectively predict the spatial distribution of forest AGB with adequate accuracy.