关键词: 氢氧同位素, 岷江上游, 随机森林模型, 生长季, 生态水文

Abstract: To study the feasibility of simulating the spatial distribution of hydrogen and oxygen stable isotopes composition (δ²H and δ¹⁸O) in the surface soil based on the machine learning method and to investigate large-scale distribution of δ²H and δ¹⁸O in the upper reaches of Minjiang River, 183 soil samples were collected from the 0-10 cm soil layer. After variable selection, back propagation (BP) neural network, random forests (RF) and support vector machine (SVM) were used to model the δ²H and δ¹⁸O of the study area, with the accuracies being evaluated. The structural equation model (SEM) was used to reveal the mechanism between the auxiliary variables and the δ²H and δ¹⁸O of soil water. The results showed that the RF model had the highest prediction accuracy, and could explain 75.0% and 64.0% of the variations of δ²H and δ¹⁸O in the surface soil, respectively. In this model, soil water content was the most important auxiliary variable, contributing 48.9% and 37.4% to δ²H and δ¹⁸O. Vegetation factors had stronger influence on δ²H and δ¹⁸O in the surface soil than climate factors, and the influence of climate factors on δ²H and δ¹⁸O was media-ted by vegetation factors. Among all the auxiliary variables, hydrogen/oxygen isotope of precipitation had the lowest effect on δ²H and δ¹⁸O due to the fractionation. The δ²H and δ¹⁸O in the surface soil of the upper reaches of the Minjiang River changed significantly across different months during the growing season. The increases of δ²H and δ¹⁸O in the early growing season and the decreases in the late growing season were mainly affected by vegetation, while climate change led to a small fluctuation in the middle growing season.

Key words: hydrogen and oxygen stable isotopes, upper reaches of the Minjiang River, random forest model, growing season, ecological hydrology