Improving the accuracy of digital soil mapping using remote sensing and topography covariates in the Central Dry Zone of Myanmar

Remote sensing has been extensively applied in Digital Soil Mapping to support land resource management and environmental monitoring. However, in humid tropical regions, the effectiveness of optical sensors is often constrained by persistent cloud cover. As a result, many studies in these areas rely primarily on topographic variables as covariates. This study aims to achieve high-accuracy mapping by combining remote sensing images and topographic variables in the Pyawbwe township of the Central Dry Zone of Myanmar, covering 1,650 km². This study explored (1) the correlation between soil properties and remote sensing predictors retrieved from monthly composite images of Landsat 5 bands on the accuracy of soil property predictions and (2) the improved prediction accuracy of Digital Soil Mapping when topographic variables were combined with Landsat 5 images. Correlation and regression analyses show significant relationships between spectral bands and soil properties. Integrating Landsat imagery with topographic data consistently improved the prediction of soil properties using Random Forest (RF) and Cubist models, with R² values more than double in some cases compared to models using topography alone. Particularly for clay content (R² improved from 0.27 to 0.54 for RF and 0.14 to 0.57 for Cubist), effective cation exchange capacity (from 0.17 to 0.48 for RF and 0.07 to 0.42 for Cubist), and sand content (from 0.15 to 0.46 for RF and 0.10 to 0.50 for Cubist). Results from linear correlation analysis show that clay, silt, sand, effective cation exchange capacity, and exchangeable magnesium display the highest correlations with near-infrared and shortwave infrared bands during the fallow period in March, May, and December. The results suggest the potential for using remote sensing to interpret soil fertility, texture, and nutrient-supplying capacity during these periods.