Integrating Sentinel-1 and 2 with LiDAR data to estimate aboveground biomass of subtropical forests in northeast Guangdong, China

Abstract

Accurate monitoring of forest aboveground biomass (AGB) is vital for sustainable forest management. Generally, the AGB is estimated by combining satellite images and field measurements. Nevertheless, field plots are inaccessible in complex forest areas. Due to the limitations related to either optical or SAR data alone, combining the two data types is indispensable. Here, we explored the LiDAR-derived biomass as training/testing samples for the combination of yearly time-series of Sentinel-1 and Sentinel-2 images to estimate AGB in north-eastern Conghua, Guangdong province, China. The AGB reference map derived from airborne LiDAR data and field plots could provide more samples for satellite-based AGB estimation with a limited amount of sampling plots. We designed four groups of experiments based on diverse Sentinel-1 and Sentinel-2 variables, including backscatter, backscatter indices, texture features, spectral bands, vegetation indices, and biophysical features. Four different prediction methods (support vector regression, multi-layer perceptron neural network, K-nearest neighbour, and random forest (RF)) were separately used to estimate AGB. Results showed that the RF model achieved the best accuracy for AGB mapping. All Sentinel-1 and Sentinel-2 experiment (R2: 0.72, RMSEr: 17.65%) performed better than the monthly complementary experiment (R2: 0.66, RMSEr: 19.01%). Additionally, the arid period images were observed to be sensitive for estimating AGB. The most contributing Sentinels predictors were determined via a sequential forward selection. Consequently, the proposed methodology has great potential for low-cost, large-scale, and high-precision AGB estimation.