资源环境科技发展态势分析平台

Aboveground biomass (AGB) plays an important role in carbon cycle. Assessment of AGB presents a challenge in forest management. Reported studies have explored the potential of synthetic aperture radar (SAR) and multi spectral instrument (MSI) data using random forest (RF) approach in AGB mapping. However, how AGB prediction would be affected by using data from different sources based on random forest kriging (RFK), which integrates RF and estimates residuals by ordinary kriging (OK), deserves further exploration. This study reported an assessment of multisensor data from Advanced Land Observing Satellite 2 (ALOS-2) L band and Sentinel-1C band SAR, Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM), and Sentinel-2 MSI for forest AGB mapping using RFK. The effectiveness of 97 predictor variables derived from multisensor data was evaluated for AGB prediction in a temperate continental forest site in northeastern China. The assessment was tested by field-measured data from 1167 forest plots in 2017. The results showed that the RFK model achieved the accuracy with mean error, mean absolute error, root mean square error and correlation coefficient in -0.11, 19.37, 28.15 Mg ha(-1) and 0.98, respectively. The study revealed that backscatters and texture features from ALOS-2 L band SAR and vegetation indices from Sentinel-2 MSI were primary contributors for explaining the observed variability of AGB. Topographic indices from SRTM DEM were more important than C band SAR backscatters and texture. features. The accuracy improvement on forest AGB mapping by RFK over RF was more distinguished in models using a single sensor than those using multisensors.