Regional soil salinity analysis using stepwise M5 decision tree.

Area of Science:

• Earth and Environmental Sciences
• Remote Sensing
• Soil Science

Background:

• Soil salinity is a major threat to agricultural productivity worldwide.
• Accurate and efficient soil salinity assessment is crucial for sustainable land management.
• Traditional methods for soil salinity assessment are often time-consuming and labor-intensive.

Purpose of the Study:

• To evaluate the efficacy of multispectral satellite imagery for soil salinity assessment.
• To compare the performance of linear multiple regression and M5 decision tree regression models.
• To identify key spectral indices for estimating soil electrical conductivity (EC).

Main Methods:

• Collection and analysis of 96 soil samples.
• Correlation of soil samples with 15 independent spectral variables and Landsat 8 indices.
• Application of linear multiple regression and M5 decision tree regression techniques.

Main Results:

• Linear regression yielded unsatisfactory results (highest R² = 58%, RMSE = 0.78) due to nonlinear relationships.
• M5 decision tree regression achieved a higher correlation coefficient (73%) and lower RMSE (0.29) by estimating the natural logarithm of EC.
• The B64, NDII, and S2 indices were identified as the most influential spectral indices.
• The M5 model demonstrated a 37.18% improvement in accuracy over multivariate linear regression.

Conclusions:

• M5 decision tree regression is a more effective method for soil salinity assessment using multispectral satellite data compared to linear regression.
• Spectral indices derived from satellite imagery can provide valuable information for estimating soil electrical conductivity.
• Factors like vegetation cover, soil moisture, and sampling inconsistencies can influence the accuracy of remote sensing-based salinity assessments.