Enhancing field soil moisture content monitoring using laboratory-based soil spectral measurements and radiative transfer models

Abstract

Accurate information on the soil moisture content in croplands is essential for monitoring crop growth conditions. This study aimed to enhance soil moisture monitoring by employing laboratory-based soil spectral measurements and radiative transfer models. This study comprised three main components: (1) Utilizing laboratory-measured soil spectra to investigate the influence of soil moisture content on soil spectral properties (n ​= ​178), and describing the impact of canopy coverage on the mixed spectra of wheat and soil in croplands using a radiative transfer model (RTM) (n ​= ​144, 180); (2) employing a deep learning model trained on extensive simulated datasets to estimate soil moisture beneath the canopy from wheat‒soil mixed spectra (n ​= ​200); and (3) comparing the performance of deep learning model with statistical regression techniques based on soil moisture spectral index (SI) for estimating wheat fractional vegetation cover (FVC) and relative soil moisture content (RMC) under medium to low canopy coverage. The conclusions of this study were as follows: (1) Compared with the conventional statistical regression approaches, the deep learning model exhibited superior accuracy in estimating RMC across all levels of normalized difference vegetation index (NDVI). (2) By combining laboratory soil spectral measurements with an RTM, a pretrained dataset can be created. When combined with transfer learning techniques (FVC: R² ​= ​0.782, RMSE ​= ​0.107, and RMC: R² ​= ​0.825, RMSE ​= ​0.130), this approach enhanced the accuracy of estimating wheat FVC and RMC. Future research should expand experiments to include additional regions and crop types to verify the accuracy and generalizability of this method for estimating FVC and RMC under various remote sensing conditions.