Applicability of remote sensing and machine learning for predicting bulk soil electrical conductivity under different forest types in central Japan

Soil electrical conductivity (EC) is a key indicator in forest ecosystems for assessing soil nutrient availability, moisture retention capacity, ion exchange processes, and overall soil health, which are critical for tree growth, carbon sequestration, and ecosystem stability. With the growing interest in remote sensing applications, this study aimed to apply remote sensing and machine-learning (ML) models such as random forest (RF), classification and regression tree (CART), extreme gradient boosting (XGBoost), k-nearest neighbors (KNN), minimum distance (MD), and Naive Bayes (NB) for EC prediction. We aimed to propose the most suitable ML model for soil EC classification to enhance large-scale soil property assessment and improve our understanding of soil EC distribution under different forest types in central Japan. The RF model consistently outperformed others at 30 m resolution, with the image combinations of Sentinel-2 and surface soil moisture achieving the highest mean accuracy (MA) (MA = 0.926). The XGBoost model also performed strongly using the same image combinations with high mean accuracy (MA = 0.923). By demonstrating the potential of integrating remote sensing and ML, our study highlights the role of modern technologies in addressing complex ecological challenges in forestry and beyond. • Integration of multi-source remote sensing data for predicting bulk soil electrical conductivity (EC) under different forest types in central Japan. • Identify the most effective machine learning model for EC prediction. • RF and XGBoost models achieved the superior performance with highest mean accuracies for EC classification.