Deep Learning-Based Estimation of Crop Biophysical Parameters Using Multi-Source and Multi-Temporal Remote Sensing Observations

Remote sensing data are considered as one of the primary sources for precise agriculture. Several studies have demonstrated excellent capability radar and optical imagery crop mapping biophysical parameter estimation. This paper aims at modeling parameters, e.g., Leaf Area Index (LAI) biomass, using a combination Earth observations. We extracted several features from polarimetric Synthetic Aperture Radar (SAR) Vegetation Indices (VIs) images to model crops’ LAI dry biomass. Then, mutual correlations between these Random Forest feature importance were calculated. two scenarios estimate parameters. First, Machine Learning (ML) algorithms, Support Vector Regression (SVR), (RF), Gradient Boosting (GB), Extreme (XGB), utilized To this end, biomass estimated three input data; (1) SAR features; (2) spectral VIs; (3) integrating both features. Second, deep artificial neural network was created. These fed mentioned algorithms evaluated in-situ measurements. observations cash crops, including soybean, corn, canola, been collected over Manitoba, Canada, during Soil Moisture Active Validation Experimental 2012 (SMAPVEX-12) campaign. The results showed that GB XGB great potential in estimation remarkably improved accuracy. Our also significant improvement compared previous studies. For LAI, validation Root Mean Square Error (RMSE) reported 0.557 m2/m2 canola GB, 0.298 corn 0.233 soybean XGB. RMSE 26.29 g/m2 utilizing SVR, 57.97 RF, 5.00 GB. revealed had better parameters than ML algorithms.