报告选择泾河流域作为研究区域，利用MODIS卫星影像作为地表植被覆被监测的数据源，从降水-植被-水土流失的空间和时间两个尺度分析泾河流域植被-降水-水土流失间的耦合关系。在空间尺度上旨在阐明一定的降水条件下植被的空间分布特征和形态特征，并分析不同植被格局下区域水土流失的差异；在时间尺度上，探讨了区域降水-水土流失年际变化特征以及降水-植被-水土流失年内变化过程。报告主要研究内容、研究方法及结论主要有：（1）进行了泾河流域地表覆被的遥感分类。在非监督分类的基础上，将非监督分类结果按照像元进行细分，并运用地形限制条件对细分后的像元进行二次判别分类，最终将研究区域的地表覆盖划定为9种类型。结果表明，分类精度明显提高，其中，农田和居民点分类精度的提高最为明显。这一方法使得完全同质的单元可以进行属性的变更，改善了像元空间分辨率差造成的误差；而地形限制特征的引入在减少遥感分类不确定性，使模糊区域的分类有了较为明确的区分特征，提高了分类的精度。（2）从空间尺度分析了泾河流域年降水量对植被分布的影响以及植被空间分布对水土流失的影响。研究发现，泾河流域降水存在着较显著的空间差异，降水的空间差异决定了区域植被组成和结构都存在着明显的区域分异特征。由东南到西北，植被类型又森林、农田过渡为半荒漠草地。通过对若干流域植被组成-水土流失关系的对比分析发现，以森林植被为主的流域具有较小的土壤侵蚀，而以半荒漠草地为主的流域具有较大的土壤侵蚀，流域植被组成与水土流失的关系是：森林〈 灌丛〈 高、中盖度草地〈 农田〈 低盖度草地 〈 半荒漠草地。（3）从年际尺度分析了泾河支流和马莲河支流所在流域40年来的降水和水土流失变化特征。研究发现，泾河子流域40年来降水-水土流失保持一种相对稳定的状态，而马莲河流域40年来呈现降雨减少、水土流失加剧的状态。通过对这两个流域水土流失时间过程分析，发现在不同的时间段两个流域表现出不同的水土流失特征。但自20世纪90年代以来，两个流域的水土流失均呈现微弱的下降趋势或稳定状态。（4）从年内尺度分析了降水-植被-水土流失的季节动态特征。研究发现，泾河流域降水-植被-水土流失年内变化呈现高度相关的耦合关系。植被季节性变化在降水-水土流失耦合过程中的作用以雨季为界可以分为两个阶段，雨季来临前的植被状态将直接决定雨季时区域水土流失的程度。

This research has analyzed the relationship between precipitation, vegetation and soil erosion from spatial and temporal scale respectively at Jinghe watershed, Loess plateau, China. Temporal series MODIS images from Marland University were used to classify land cover at research region in 2002. At spatial scale, influence of precipitation on vegetation distribution and contribution of vegetation to soil erosion were analyzed. At temporal scale, the seasonal dynamics of precipitation, vegetation and soil erosion were analyzed to find the interrelationship between precipitation, vegetation and soil erosion.The main method and conclusion of this study included:(1) Based on the result of non-supervised classification method, a new method, namely the Topographical Features Based Method, was presented to classify remotely sensed images. Finally, nine land cover types were obtained from MODIS image at Jinhe watershed, in Loess plateau. The accuracy of this classification method increases from 71.88%, the accuracy of traditional non-supervised classification, to 84.09%, especially the accuracy of cropland and urban increase more apparently. The method of sub-division of cell avoids the shortcoming of traditional classification, which based on raw cell of image, and makes it probably that land cover types at homogenous cells can be classified again. Topographical features make fuzzy land cover types distinguishable. To sum up, the Topographical Features Based Method has more obvious advantage than traditional non-supervised classification.(2) The precipitation has affected the vegetation distribution, species composition and physignomical characteristics. The vegetation distribution form southeast to northwest at Jinghe watershed is following: forest and cropland → shrubland → grassland → semi-desert grassland, and cover degree and height of vegetation decrease accordingly. The different vegetation has result in different soil erosion grades, watershed in which forest is main land cover type has little soil erosion than one with other types, and watershed with semi desert grassland has maximal soil erosion.(3) The analysis on interannual variability between precipitation and soil erosion indicates the soil erosion has been a balance condition in the past 40 years at jinghe sub-watershed and the soil erosion present an increased behavior in the past 40 years. It is also found that there were different characteristics of soil erosion at different temporal segmentation.(4)The analysis on innerannual dynamics of precipitation, vegetation and soil erosion indicates that there is high correlation among them. Seasonal dynamics of vegetation plays an important role on the soil erosion, especially the vegetation condition before rainy period.