Enhancing Soil Moisture Estimation: Exploring the Synergy of Optical Trapezoid and Deep Learning models

This study aimed to propose an effective model for estimating soil moisture by integrating the optical trapezoid method with a deep learning Long Short-Term Memory (LSTM) model. The performance of the proposed model was compared with two other methods, i.e., Partial Least Squares (PLS) regression and Group Method of Data Handling (GMDH) multivariate neural network.

This study combined the optical trapezoid method with deep learning models to propose an effective model for soil moisture estimation in the Maragheh watershed. A total of 499 in-situ soil moisture data were collected. Relative moisture content was calculated using the optical trapezoid method and imported into the LSTM model, along with other inputs such as spectral indices and DEM-based derived variables. The performance of the mentioned models was assessed both with and without the optical trapezoid method to evaluate its efficacy on the performance of AI models.

The results demonstrate that the combined model of deep learning LSTM and the optical trapezoid method achieves satisfactory performance, with an R2 of 0.95 and a RMSE of 1.7%. The PLS and GMDH methods performed moderately, both without the involvement of the optical trapezoid method and in the combined mode.

This study shows that the optical trapezoid method can improve the performance of deep-learning models in estimating soil moisture. However, considering the significant difference in computational costs among these models, choosing the appropriate model depends on the user's objectives and desired level of accuracy and precision.