Threshold optimization for forest soil organic matter prediction model: a case study of Guangdong Province of southern China

Objective This paper takes Guangdong Province of southern China as an example, and explores the applicability of optimized geographical similarity model in the prediction of forest soil organic matter at provincial scale.

Method Based on the data of 1 175 sample points from soil fertility testing system in Guangdong Province, we selected soil factors, stand factors and climate factors as environmental covariates to improve the “individual point representativeness digital soil mapping (iPSM)”. This article evaluates the applicability of the model in predicting forest soil organic matter at the provincial level using three indicators: coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE), and compares its predictive performance with that of random forest prediction model.

Result (1) The prediction accuracy of optimized iPSM was better than that of random forest prediction model, and the coefficient of determination of model can reach 0.7419. (2) The total nitrogen content and available potassium content in soil factors were the most influential environmental covariates in prediction model of forest soil organic matter in Guangdong Province.

Conclusion Compared with iPSM method, the optimized iPSM has improved the accuracy of forest soil organic matter prediction in Guangdong Province. The model has good applicability at the provincial level and can provide a new method for predicting soil organic matter at the provincial level.