基于RF和EBKRP算法的新安江流域有效土壤厚度反演  

Effective soil thickness inversion in Xinanjiang River basin based on random forest and empirical Bayesian Kriging regression prediction algorithms

作　　者：王尚晓 张晓东[1] 张明[1] 牛晓楠 周墨 唐志敏 张洁[1] 宗乐丽 徐帅 Wang Shangxiao;Zhang Xiaodong;Zhang Ming;Niu Xiaonan;Zhou Mo;Tang Zhimin;Zhang Jie;Zong Leli;Xu Shuai(Nanjing Geological Survey Center,China Geological Survey,Nanjing,Jiangsu 210016,China;School of Education(Shanwei),South China Normal University,Shanwei,Guangzhou 516600,China)

机构地区：[1]中国地质调查局南京地质调查中心,江苏南京210016 [2]华南师范大学教育科学学院(汕尾),广东汕尾516600

出　　处：《水土保持通报》2025年第1期168-177,共10页Bulletin of Soil and Water Conservation

基　　金：中国地质调查局项目“华东地区自然资源动态监测与风险评估”(DD20230103),“华东地区国土空间用途管制技术支撑与应用服务”(DD20230495);江苏省科技计划(“一带一路”创新合作项目)“黑土地土壤退化诊断评价与动态监测技术合作研发”(BZ2023003)。

摘　　要：[目的]快速、准确地获取区域有效土壤厚度,分析其空间分布特征和影响因素,为植被生长、土壤保持和粮食安全工作提供理论指导。[方法]以新安江流域为研究区,将野外调查数据、地形、岩性和气候等成土因素结合起来,采用经验贝叶斯克里金回归预测(EBKRP)和随机森林(RF)算法,得到有效土壤厚度反演结果,并分析其与环境变量之间的关系。[结果](1)区域平均有效土壤厚度为0.2~0.3 m,城镇建设集中和人类活动密集的盆地和平原区土壤厚度较高,丘陵山地区则较低。(2)从MAE(平均绝对误差)、R^(2)(判定系数)和RMSE(均方根误差)3项精度评价指标来看,RF算法的预测结果明显优于EBKRP算法,而且更能显示出土壤厚度空间异质性分布特征,在一定程度上提高了土壤厚度数字制图的效果。(3)有效土壤厚度的估算受地形和气候变量的影响较大,它们分别占变量重要性的46.77%和18.78%。[结论] RF算法能够有效实现对区域有效土壤厚度的反演,克服了土壤厚度空间异质性的特点,相较于有限采样的模型更精确,分辨率也更高。[Objective]The effective soil thickness of a region was rapidly and accurately obtained,and its spatial distribution and influencing factors was analyzed,in order to provide theoretical guidance for vegetation growth,soil conservation,and food security.[Methods]Taking the Xinanjiang River basin as the research area,combining field survey data,topography,lithology,climate,and other soil-forming factors,the empirical Bayesian Kriging regression prediction(EBKRP)and random forest(RF)algorithms were applied to obtain the effective soil thickness inversion results.The relationship between this data and environmental variables was also analyzed.[Results]①The average effective soil thickness in the region ranged from 0.2 to 0.3 m.Soil thickness was higher in basin and plain areas with concentrated urban development and intensive human activity.Meanwhile,it was lower in hilly and mountainous regions.②Based on three accuracy evaluation indicators of MAE(mean absolute error),R^(2)(coefficient of determination),and RMSE(root mean square error),the prediction results of the RF algorithm were significantly better than those of the EBKRP algorithm.It could more effectively show the spatial heterogeneity distribution of soil thickness,improving the effect of soil thickness digital mapping.③The effective soil thickness estimation was strongly influenced by topography and climate variables,which accounted for 46.77% and 18.78% of the variable importance,respectively.[Conclusion]The RF algorithm could effectively invert regional effective soil thickness,overcoming the spatial heterogeneity of soil thickness,and is more accurate and has a higher resolution compared to models with limited sampling.

关 键 词：有效土壤厚度 随机森林(RF) 土壤数字制图 经验贝叶斯克里金回归预测(EBKRP) 新安江流域 

分 类 号：S159[农业科学—土壤学]