Derivation of biophysical parameters from fused remote sensing data

Abstract

Recent launches of optical space-borne remote sensing systems with high spatial resolution, high temporal revisit frequency and constant viewing angles (e.g.: Venµs, Sentinel-2) will fortify the potential of remote sensing applications in the context of agricultural monitoring (e.g.: the derivation of biophysical parameters). The practicality of this kind of remote sensing data is limited by tile and cloud coverage. Spatial temporal image fusion techniques (e.g: STARFM) can be used to combine the data from different remote sensing sensor systems to overcome these challenges. In order to investigate the reliability of synthesized remote sensing data in agricultural monitoring, we evaluated the quality of the prediction of FPAR and LAI on maize. In this context, we used synthetic daily Landsat-like data and a RandomForest model to predict FPAR and LAI for the entire growing period in 2015. The evaluation of the biophysical time series was concluded using a weekly to bi weekly ground measurements in different phenological stages of the maize plant. The quality assessment of the entire growing period revealed the high potential of synthetic remote sensing data for agricultural monitoring. The quality of results range between R² = 0.68; RMSE = 0.79 (LAI) and R² = 0.76; RMSE = 0.12 (FPAR).