High-Resolution Digital Soil Maps of Forest Soil Nitrogen across South Korea Using Three Machine Learning Algorithms

Yoosoon An, Woojin Shim, Gwanyong Jeong

Reliable estimation of the forest soil nitrogen spatial distribution is necessary for effective forest ecosystem management. This study aimed to develop high-resolution digital soil maps of forest soil nitrogen across South Korea using three powerful machine learning methods to better understand the spatial variations of forest soil nitrogen and its environmental drivers. To achieve this, the study used national-level forest soil nitrogen data and environmental data to construct various geographic and environmental variables including geological, topographic, and vegetation factors for digital soil mapping. The results show that of the machine learning methods, the random forest model had the best performance at predicting total soil nitrogen in the A and B horizons, closely followed by the extreme gradient-boosting model. The most critical predictors were found to be geographic variables, quantitatively confirming the significant role of spatial autocorrelation in predicting soil nitrogen. The digital soil maps revealed that areas with high elevation, concave slopes, and deciduous forests had high nitrogen contents. This finding highlights the potential usefulness of digital soil maps in supporting forest management decision-making and identifying the environmental drivers of forest soil nitrogen distribution.