基于GBDT的望奎县农田土壤有机碳主控因子研究

Spatial distribution and main controlling factors of soil organic carbon under cultivated land based on GBDT model in black soil region of Northeast China

利用多源公开数据,针对于2012年搜集到的农田土壤有机碳(SOC)数据,通过梯度增强决策树(GBDT)探究SOC的主控因子,揭示SOC空间分异机理.选择国家商品粮食生产基地县和全国产量大县-望奎县作为研究区域,结果表明,GBDT模型相较于一元线性回归模型在所有的自变量上均具有更好的预测效果,总体的R^(2)为0.958,表明该模型能够很好地解释目标变量的变异;碱解氮、有效磷、气温、土壤微生物多样性、速效钾、土壤pH和降水量依次是影响有机碳含量最为显著的几个因素,其中碱解氮以33.85%的影响权重位居最高;7个环境协变量均观察到对SOC的阈值效应,且均具有两个阈值.研究发现任意两个变量都不是独立的,均存在相互作用的负值效应.表明环境变量对SOC含量的影响是一个复杂的交互作用,而不是简单的叠加.侧面证明了过度施肥并不会达到增加土壤肥力的作用,而会造成资源浪费和农田生态污染问题.

The nonlinear relationships between soil organic carbon(SOC)and environmental covariates have been extensively proved,the threshold and interaction effects of environmental covariates on SOC content are often ignored.This study,founded upon the SOC data collected in 2012,employs the Gradient Boosting Decision Tree(GBDT)to delve into the thresholds and interactive effects of SOC.It aims to discern the main controlling factors of SOC and unveil the spatial differentiation mechanism of SOC.The national commodity grain production base county,Wangkui County,which is also one of the high-yield counties in China,was selected as the study area.The GBDT model can more flexibly fit the complex nonlinear relationship between SOC and environmental variables,and shows superior predicting performance under all independent variables to the simple linear regression model.The model achieves an overall R^(2)value of 0.958,indicating its ability to effectively explain the variations of the target variable.Alkali-hydrolyzed nitrogen,available phosphorus,temperature,soil microbial system diversity,available potassium,pH and precipitation in turn are the most significant factors influencing SOC content.Alkali-hydrolyzed nitrogen holds the highest influence weight of 33.85%.All 7environmental covariates had threshold effects on SOC with two distinct thresholds.It was found that any two covariates are not independent but exhibit interactive negative effects,indicating that the influence of environmental covariates on SOC concentration is a complex interaction rather than a simple superposition.The above results emphasize that threshold and interaction effects between environment variables should be considered in determining the effective range of environmental variables,the potential process of SOC spatial differentiation,and improving the explanatory power of SOC change in farmland.It indirectly proves the fact that excessive fertilization fails to enhance soil fertility but instead leads to resource wastage and the ecological pollution of farmlands.