中文摘要：

空间异质性的存在，会导致水质参数遥感反演中的尺度效应，影响反演精度，因此通过分析水质参数空间异质性，对于选择适当分辨率的遥感影像，提高反演精度具有重要意义．通过2008年10月在太湖布置的3个样方，利用GIS地统计学原理和分形维数的方法，对水质遥感反演中的三要素浓度，包括叶绿素a（Chl．a）、总悬浮物（TSM）和溶解有机碳（DOC）的空间异质性及其可能产生的尺度效应进行了研究．结果表明：太湖水体的三要素浓度在不同样方单元中变异系数相差较大，存在着明显的尺度效应；三个样方内Chl．a变异函数曲线斜率在变程范围内变化都较为剧烈，分形维数较高，说明太湖水体Chl．a受到某种起主导作用的生态过程的影响和控制；Chl．a和TSM的空间结构比例都在90％左右，有较强的空间相关性，表明其空间异质性的产生主要是由于结构性因素引起的，随机性因素作用微弱；DOC空间结构比例较小，说明随机性因素对其空间异质性的产生起了主导作用．三个样方中Chl．a的变程分别为147．3m、129．3m和115．0m，TSM的变程分别为1131．7m、130．6m、149．1m，因此在遥感反演中可选择TM影像，选择5×5窗口，以150m×150m作为基本单元；而DOC的变程分别为34．3m、38．5m、26．4m，表明其自相关距离较小，建议直接选择分辨率为30m的TM影像，使实际测量值与遥感影像最小单元相对应，消除反演过程中的尺度效应带来的误差．该研究也表明，MODIS的像元尺寸（250、500、1000m）明显偏大，在太湖水体三要素反演过程中，由于空间异质性引起的尺度效应，会造成一定的误差．

英文摘要：

Spatial heterogeneity results in marked scaling effect and influences retrieval precision in the retrieval of water from re- mote sensing image, so the research on the spatial heterogeneity benefit to choose suitable resolution remote sensing image and increase retrieval precision. A study was conduced to set sampling points of water in Lake Taihu in October, 2008. The research analyzed three essential factors＇ spatial heterogeneity and scaling effect in the water quality retrieval, that is, chlorophyn-a （ Chl. a） , the total suspended matter （TSM） and the dissolved organic carbon （DOC） using the Geostatistics principle and the fractal dimension method. The result showed that three essential factors of Lake Taihu had different coefficient of variation in different sampling areas and had marked scaling effect. Furthermore, the semi-variogram curves of Chl. a in three sampling areas changed fiercely in range and the fraetal dimension was high, which shows Chl. a in Lake Taihu was possibly controlled by some leadership ecological processes. Spatial structure proportion of Chl. a and TSM was higher, about 90%, and had strongly spatial relevance, which indicated that the spatial heterogeneity was mainly caused by the constitutive factors, while random factors played less role. DOC had lower spatial structure proportion, which indicated it was mainly affected by random factors. In three sampling areas, the range ofChl. a was 147.3m, 129.3m, 115.0m, and TSM was 1131.7m, 130.6m, 149. l m, so 150m ~ 150m as the smallest unit and 5 by 5 of TM was appropriate to estimate them during the retrieval process. The range of DOC was 34.3m, 38.5m, 26.4m, which indicated that the autocorrelation distance was small, so 30m × 30m as the smallest sampling unit and choose TM image which resolution was 30m to correspond observed values with the smallest unit of remote sensing image. The research also indicated the MODIS pixel size （ 250,500, 1000m） was so large in the retrieval of Lake Taihu, and it may bring retrieval