基于ROC曲线的玛纳斯河灌区土壤盐分空间分析  

Spatial Analysis of Soil Salinity in Irrigation Areas Field of Manas River Basin Based on ROC Curve

作　　者：辛明亮[1,2] 吕廷波[1,2] 何新林[1,2] 曹玉斌 王萌萌[1,2] 

机构地区：[1]石河子大学水利建筑工程学院,新疆石河子832000 [2]现代节水灌溉兵团重点实验室,新疆石河子832000

出　　处：《灌溉排水学报》2016年第12期45-50,共6页Journal of Irrigation and Drainage

基　　金：国家自然科学基金-新疆联合基金资助项目(U1203282);石河子大学优秀青年项目(2013ZRKXYQ17)

摘　　要：选择我国干旱内陆河玛纳斯河流域平原灌区为研究对象,基于区域采样点数据,应用试验诊断评价方法——受试者工作特征曲线(ROC曲线)分析了不同采样时间各土层土壤含盐量的变化以及对多种空间插值方法进行了评价,最后选择估值效果较好的空间插值方法绘制了土壤盐分空间分布图,并与传统评价方法进行了对比。结果表明,基于ROC曲线,4月、10月表层土壤盐分变化剧烈,其变化敏感于其他土层,盐分存在表聚现象,ROC曲线下面积最大且贡献均为正;传统交叉检验和ROC曲线分析均得出利用经验贝叶斯克里格方法进行表层土壤含盐量空间插值效果最优;研究区表层土壤盐分具有明显的连续变化和存在局部聚集的现象;不同采样时间土壤盐分空间分布差异显著,总体表现为低值区位于南部山区和东北部,高值区主要分布在平原水库周围以及下游地势低处。ROC曲线能够很好地应用于土壤盐分空间分析领域。In Manas River basin plain irrigation areas which belong to special mountain-basin system in the north- west arid zone of China, based on the 64 sampling points and 1 408 soil samples in April （before the spring plough） and October （after the autumn harvest） of 2014, the ROC curve was used as testing evaluation method to analyze the corresponding relation of the sampling time and different soil depth＇ s soil salinity, and the accura- cy of several spatial interpolation methods was evaluated. Finally, choosing which space interpolation method＇ s valuation effect was the best to make a spatial distribution map of soil salinity. The main results indicated that sur- face soil＇ s saltiness changed drastically and had appears accumulation in topsoil phenomenon, the surface soil sa- linity A UC value was 0.599, which was bigger than others, and it was sensitive to others. 5 interpolation methods interpolating surface soil salinity were selected, including Inverse Distance Weighting （IDW）, Local Polynomial （LPI）, Ordinary Kriging （OK）, Universal Kriging （UK） and Empirical Bayesian Kliging （EBK）. As a result, EBK was more accurate than others by Cross-validation and ROC curve analysis. Because of intensely human activi- ties and special natural conditions, the distribution of the surface soil salinity was different that sampling in April and October. Low value area was located in the southem mountains region and northeast of the research region where groundwater depth was deeper and irrigation water quality was better. High value area was mainly distrib- uted around the plain reservoir and basin＇ s downstream lowlands. ROC curve was above the reference line which created positive contribution in soil salinity＇ s spatial analysis domain.

关 键 词：受试者工作特征曲线 土壤盐分 空间分析 插值 经验贝叶斯克里格 

分 类 号：S153[农业科学—土壤学]