Estimating tree aboveground biomass using multispectral satellite-based data in Mediterranean agroforestry system using random forest algorithm

Abstract

Forest aboveground biomass (AGB) is a key biophysical variable to assess and monitor the spatio-temporal changes of forest ecosystems. AGB should be accurately and timely estimated through remote sensing to provide valuable information to better support sustainable forest management strategies. QuickBird and WorldView-2 satellites data and Random Forest (RF) regression model were used to estimate tree AGB in Mediterranean agroforestry systems. Spectral bands, vegetation indices and Grey-Level Co-occurrence Matrix (GLCM) texture features of 140 plots with and without vegetation mask were used as independent variables, while total of AGB per plot was used as dependent variable. A model with good performance was obtained for a complex agroforestry system, with an R2 of 82.0% and RMSE of 10.5 t/ha (22.6%). The top 11 most important variables have 80.3% of total relative importance, with 59.6% of GLCM textural features, 12.3% of vegetation indices and 8.4% of spectral bands. The results highlight the importance of the variable GLCM texture, and the use of vegetation mask and RF regression model to collect accurate spatial information on key crown cover attributes, by excluding the spectral contribution of understory vegetation and soil characteristic, of Mediterranean agroforestry systems.