[Objective] Based on the multi-scale remote sensing indicators and measured soil conductivity data,the objective of this study is to exploit the appropriate scales and indicators for inferring soil salinity in irrigation area of arid oasis,and to provide alternative schemes for rapid assessment of soil salinization in the study area.[Method] Weigan-kuqa oasis located in southern Xinjiang was selected as the study area.Soil electrical conductivity of multiple soil layers (0-10 cm,10-20 cm,20-40 cm,40-60 cm),as inference object,were analyzed.Raster resampling (30-1 000 m) and neighborhood extent (3 ×3 to 31 ×31) were employed as the method of scale transformation to generate multiple scales of environmental factors (with help of principal component analysis,tasseled cap,vegetation index,soil wet index) based on Landsat OLI image data,then,a total of 1 078 environmental variables (the former produced 352 variables,the latter produced 726 variables).On this basis,linear and nonlinear curve models were used to fit the correlation between soil salinity and environmental variables under the above mentioned method of scale transformation,and then the optimal prediction scale and environmental factor were explored.[Result] Nonlinear curve models showed a significance between soil salinity and environmental factors compared to linear model under multiple scale (raster resampling model).Secondly,the fitting accuracy decreased when the spatial resolution become coarse.In this model,the best inference scale was 30 m,and the sensitive variable was three-band maximal gradient difference (TGDVI) at depth of 40-60 cm,and the remaining depths were implied by Extended Normalized Difference Vegetation index (ENDVI).The optimal inference scale in model of neighborhood extent was 180 m (filtering scale 3×3),and the best fit variables for each layer was EEVI.Compared with the grid resampling mode,the fitting accuracy of neighborhood extent was improved by 14.60％,34.40％,32.10％ and 21.70％ (from top layer to bottom one) compared to fomer model,respectively.[Conclusion]Based on the model of neighborhood extent,the nonlinear model constructed by Extented Enhanced Vegetation Index(EEVI) (180 m,3×3) is more suitable for the prediction of soil salinity variation in the study area.%[目的]基于遥感提取的多尺度遥感指示因子和土壤实测电导率数据,借助统计分析,试图探寻适合干旱区典型绿洲灌区土壤盐度变异的最佳观测尺度和指征变量,为快速评估绿洲土壤盐渍化提供备选方案.[方法]以新疆渭干河-库车河绿洲为研究区,以野外采集的土壤盐度数据(采集0-10、10-20、20-40和40-60 cm土层土样,制备土壤饱和溶液并测试电导率(ms.cm.))并将其作为预测对象(n=87),借助Landsat 0LI遥感影像数据,利用栅格重采样(30-1 000 m)和领域滤波(原始分辨率为30、60、90、120、15 0、180、210 m,滤波尺度为3×3至31×31)两种方式,生成多个尺度若干种指示因子(主成分分析、缨帽变化、植被指数、湿度指数),共计获得1 078个(其中,栅格重采样生成352个,领域滤波生成726个)环境变量.在此基础之上,利用线性和非线性曲线模型分别拟合上述两种模式下土壤盐度和环境变量之间的相关性,进而找出最优环境因子和预测尺度.[结果]栅格重采样模式下能够较好响应各层土壤变异性的皆为非线性模式.其次,该模式下,拟合精度随着空间分辨率的降低而降低.此模式下最佳推理尺度为30 m,该尺度下最佳响应变量除了40-60 cm处为三波段差分指数(Three-band Maximal Gradient Difference,TGDVI)外,其余深度皆为扩展的归一化指数(Extended Normalized Difference Vegetation Index,ENDVI).领域滤波模式下的最佳推理尺度为180 m(滤波尺度3×3),同时,各层最佳拟合变量皆为扩展的增强型植被指数(Extented Enhanced Vegetation Index,EEVI).相比较栅格重采样模式,该模式下的拟合精度全面优于前者,各层依次提高14.60％、34.40％、32.10％和21.7 0％.[结论]基于领域滤波模式下,像元分辨率为180 m,窗口大小为3×3的ENDVI指数更适合预测本研究区土壤盐度的空间变异性.
展开▼
