Multi-Layer Soil Moisture Profiling Based on BKA-CNN by Integrating Sentinel-1/2 SAR and Multispectral Data

Soil moisture (SM) is crucial for ecosystems and agriculture. Since the root systems of plants absorb water at different depths with different intensities, monitoring multi-layer SM can better respond to the water demand of plants and offer a crucial technical backing for drought monitoring and precision irrigation. Synthetic aperture radar (SAR) and multispectral (MS) have been widely used in SM estimation; however, their combined application for multi-layer SM profiling remains underexplored. Existing research based on these two data types has primarily focused on surface soil moisture (SSM), with limited investigation into estimating SM at deeper or varying depths. Therefore, the aims of this research are to integrate Sentinel-1 SAR and Sentinel-2 MS data and employ machine learning algorithms to estimate multi-layer SM in the Shandian River Basin. The results showed that (1) MS + SAR-based SM estimation significantly outperformed single-source data (MS or SAR alone). Specifically, MS data performed better in the root-zone estimation, while SAR data showed superior performance in SSM estimation. (2) The BKA-CNN estimation accuracy significantly outperformed RF and XGBoost. The results of its five-fold cross-validation are as follows: R2 = 0.768 ± 0.011 at 3 cm, R2 = 0.777 ± 0.013 at 5 cm, R2 = 0.799 ± 0.011 at 10 cm, R2 = 0.792 ± 0.01 at 20 cm, and R2 = 0.782 ± 0.011 at 50 cm. (3) The BKA-CNN model performed better in grassland than in farmland. These findings indicate that the BKA-CNN model proposed in this study effectively improves the estimation precision of multi-layer SM by fusing SAR and MS data, demonstrating considerable generalization ability and robustness. It holds potential application value in ecological protection and agricultural water resource management.