Dry biomass estimation of paddy rice with Sentinel-1A satellite data using machine learning regression algorithms

Abstract

Despite the growing use of Sentinel-1 in retrieving crop growth information, its potential for obtaining improved estimates of rice biophysical parameters has not been fully investigated. This study therefore assesses the capabilities of Sentinel-1A temporal datasets, of the interferometric wide-swath (IW) mode that include the vertical transmitted and horizontal received (VH), and vertical transmitted and vertical received (VV) polarizations, and the linear combination of VH and VV (VHVV), to estimate rice dry biomass over a test site located in southeast China. To this end, four machine learning regression algorithms; Random Forest (RF), Support Vector Machine (SVM), k-Nearest Neighbor (k-NN), and Gradient Boosting Decision Tree (GBDT), were applied to assess the aforementioned datasets. The results indicated that for the estimation of rice dry biomass over the entire growing season, VHVV data with RF produced the most accurate estimates with an R2 of 0.73 and an RMSE of 462.4 g/m2. However, this study suggests that a consideration of the rice growth phases could produce more accurate estimates of dry biomass as evident by an R2 of 0.72 and an RMSE of 362.4 g/m2 recorded at the reproductive phase (elongation to milking) with VH data and k-NN. Hence research works that would investigate the estimation of rice dry biomass at the vegetative (transplanting to elongation) and maturity (milking to ripening) phases are required to ascertain whether a growth-phase approach could obtain more accurate rice biomass estimates that better meet precision agriculture data needs.