A Spectral Reflectance Model of Smooth Dry Soil Surfaces for Varied Soil Properties Based on Intelligent Learning

Area of Science:

• Soil Science
• Remote Sensing
• Geospatial Analysis

Background:

• Dry soil spectral reflectance is crucial for soil attribute retrieval via remote sensing.
• Existing models often rely on empirical relationships and lack generalizability, especially concerning environmental factors.

Purpose of the Study:

• To develop an interpretable and generalizable model for dry soil spectral reflectance prediction.
• To quantitatively analyze the influence of environmental covariates on dry soil reflectance.
• To improve the accuracy and applicability of soil reflectance models in remote sensing.

Main Methods:

• Collected 700 dry soil samples with laboratory-measured spectral reflectance from Northeast China.
• Utilized the SHAP method to analyze environmental covariate contributions (soil properties, parent material, geographical location).
• Developed an environmental and edaphic-factor-driven smooth dry soil reflectance model (EEDSR) using gradient boosting regression (GBR).

Main Results:

• Dry soil reflectance (400-2500 nm) correlates with soil properties, with specific positive/negative correlations to clay, sand, silt, longitude, latitude, and parent material.
• The EEDSR model achieved high accuracy (R² = 0.93, RMSE = 0.018), with parent material and geographical factors improving prediction by 13.4%.
• The model showed good spatial consistency with satellite data and strong generalization on global datasets (R = 0.94), outperforming SOGM (R = 0.27).

Conclusions:

• The EEDSR model provides an efficient and interpretable framework for dry soil spectral reflectance modeling.
• This approach offers a robust reference for soil reflectance prediction and remote sensing-based soil property retrieval.
• The study highlights the importance of incorporating environmental factors for improved model generalizability.