Improved multi-sensor satellite-based aboveground biomass estimation by selecting temporally stable forest inventory plots using NDVI time series

Accurate estimates of aboveground biomass (AGB) are crucial to assess terrestrial C-stocks and C-emissions as well as to develop sustainable forest management strategies. In this study we used Synthetic Aperture Radar (SAR) data acquired at L-band and the Landsat tree cover product together with Moderate Resolution Image Spectroradiometer (MODIS) normalized difference vegetation index (NDVI) time series data to improve AGB estimations over two study areas in southern Mexico. We used Mexican National Forest Inventory (INFyS) data collected between 2005 and 2011 to calibrate AGB models as well as to validate the derived AGB products. We applied MODIS NDVI time series data analysis to exclude field plots in which abrupt changes were detected. For this, we used Breaks For Additive Seasonal and Trend analysis (BFAST). We modelled AGB using an original field dataset and BFAST-filtered data. The results show higher accuracies of AGB estimations using BFAST-filtered data than using original field data in terms of R2 and root mean square error (RMSE) for both dry and humid tropical forests of southern Mexico. The best results were found in areas with high deforestation rates where the AGB models based on the BFAST-filtered data substantially outperformed those based on original field data (R2 BFAST = 0.62 vs. R2 orig = 0.45; RMSEBFAST = 28.4 t/ha vs. RMSEorig = 33.8 t/ha). We conclude that the presented method shows great potential to improve AGB estimations and can be easily and automatically implemented over large areas