Study on automated detection methods of shallow surface soil water content based on GPR signal level

Abstract

The GPR-based soil moisture detection method achieves an effective balance between spatial scale coverage and detection accuracy; however, the automation level and efficiency still need improvement. This study adopts refined gradient modeling, optimizing delay, envelope amplitude area, and centroid frequency as key indicators for soil moisture prediction. Random forest feature importance analysis indicates that the selected three indicators can effectively characterize soil moisture variation at different scales. Single-factor and three-factor soil moisture prediction models were constructed, and comparisons reveal that the three-factor model significantly outperforms the single-factor model in both prediction accuracy and stability. Bayesian regression was used to assess model and data uncertainty, and the results indicate that the model exhibits low uncertainty within the existing three-factor knowledge range. To achieve automated soil moisture detection, this study proposes an error recursive optimization framework, overcoming the bottlenecks in GPR-based soil moisture automation, and significantly improving detection accuracy and efficiency.