Integrating ChloroNet and XAI for accurate SPAD prediction in rice

Remote sensing technology in precision agriculture enables extensive monitoring of crop health and environmental conditions. Sentinel-1 SAR imagery is highly effective for vegetation monitoring due to its ability to penetrate cloud cover and function under diverse meteorological circumstances. This research introduces ChloroNet, a novel Artificial neural network (ANN) model that utilizes Sentinel-1 SAR-derived indices and field data to predict rice chlorophyll (SPAD, Soil and Plant Analysis Development) levels. We evaluate ChloroNet in comparison to TabNet, a prominent Deep Neural Network for tabular data, using two datasets from rice cultivation regions in China: Tianjin and Linhai. ChloroNet exhibits superior accuracy compared to TabNet. The Tianjin dataset had a mean squared error (MSE) of 16.31, a root mean squared error (RMSE) of 4.04, a mean absolute error (MAE) of 2.96, and a coefficient of determination (R 2 ) of 0.90 during training. The testing yielded a Mean Squared Error (MSE) of 12.67, a Root Mean Squared Error (RMSE) of 3.56, a Mean Absolute Error (MAE) of 2.49, and a coefficient of determination (R 2 ) of 0.92. ChloroNet exhibited superior generalization on the Linhai dataset, which was not utilized for training (MSE 40.49, RMSE 6.36, MAE 5.73, R 2 0.69), compared to TabNet (MSE 46.53, RMSE 6.68, MAE 5.94, R 2 0.40). The SHAP study demonstrated that SAR indicators (PD, Cross-Ratio) and seasonal factors (month_sin, month_cos) were significant predictors. The results indicate that ChloroNet may serve as an effective instrument for monitoring chlorophyll in precision agriculture, particularly in areas with limited data availability.