Prediction of soil moisture content using genetic algorithm-optimized backpropagation algorithm from spectral data

Purpose

Accurately assessing soil moisture content (SMC) is essential for applications in agriculture and ecological sustainability. However, the dynamic monitoring and assessment of SMC presents considerable challenges due to the intricate traditional methods and the ever-evolving environmental variables. Relevant research has indicated that visible and near-infrared (vis–NIR) spectra are a practical and cost-effective alternative for accurate and convenient estimation of SMC. Advances in technology and computer hardware have enabled spectral characteristics and computer vision algorithms to show enormous potential for rapid and non-destructive characterization of soil properties. The objective of this study was to evaluate the predicted ability of SMC using vis–NIR spectral data.

Materials and methods

A total of 60 topsoil samples (0–5 cm) from the maize test field at the Shanxi Central Irrigation Test station were used as the study object. A set of four spectral parameters was derived and filtered from spectral data, and C-W and W-W models were developed using Genetic Algorithm algorithm-optimized backpropagation (GA-BP) neural networks to predict SMC based on outdoor measurements.

Results and discussion

The results showed that: (1) SMC can be successfully predicted using the spectral data through the C-W and W-W models; (2) the C-W model outperformed the W-W model, particularly in the context of deep soil, with R² ranging from 0.919 to 0.991 and corresponding RMSE values from 0.619% to 0.982%.

Conclusions

This study introduces two effective methodologies for accurate estimation of SMC at different depths using multispectral remote sensing, which showed a high degree of prediction accuracy. It further proves that GA-BP algorithm is still effective for predicting SMC in outdoor. The research result might be helpful for the further study of SMC measurement.