Improved Crop Height Estimation of Green Gram and Wheat Using Sentinel-1 SAR Time Series and Machine Learning Algorithms

Crop height is an important agronomic indicator that plays a crucial role in assessing production levels, providing essential information about the crop health. This study demonstrates the application of machine learning models for estimating crop height of green gram and wheat using dual-polarimetric Sentinel-1 Synthetic Aperture Radar (SAR) datasets. Multi-temporal Sentinel-1A dual pol data obtained over agricultural areas in India during the green gram and wheat growing seasons in 2023 and 2024, along with comprehensive ground truth surveys. Fifteen key dual-polarimetric SAR observables were extracted and evaluated for their sensitivity to crop height in both crops. The most influential parameters identified for each crop were subsequently used as input features in several machine learning models. Results indicate that Random Forest Regressor (RFR) and Extreme Gradient Boosting (XGBoost) models outperformed other models in estimating crop height. Specifically, for green gram, XGBoost ( \(\hbox {R}^{2}= 0.98\) ${\text{R}}^2=0.98$ , RMSE = 2.5 cm) and RFR ( \(\hbox {R}^{2} = 0.85\) ${\text{R}}^2=0.85$ , RMSE = 6.38 cm) demonstrated high modeling accuracy. For wheat, RFR achieved a high modeling accuracy ( \(\hbox {R}^{2} = 0.8\) ${\text{R}}^2=0.8$ , RMSE = 12.87 cm). These findings underscore the use of Sentinel-1 SAR data for precise crop height retrieval, suggesting its potential for operational applications.