基于软分类的太湖水体叶绿素a浓度遥感反演与时空变化分析

近岸和内陆Ⅱ类水体光学特性复杂而且变化较大，许多Ⅱ类水体叶绿素a反演方法都具有很强的区域和季节局限性。实际上，影响反演方法适用性的主要是更上一层的反演策略问题。传统的反演策略通常是一个区域和一个季节的水体对应于一种反演模型，但是由于相同区域、相同季节水体的光学特性仍然存在较大变化，导致反演方法适用性不高。目前一些先分类再反演的反演策略能够一定程度上提高反演方法的适用性；但是由于通常采用硬分类方法直接将水体分到一个类别，忽略了同一类别中不同水体的差别，导致反演方法仍然存在很大局限性。本项目首先构建基于软分类（模糊分类）的叶绿素a反演策略，利用水体与每个类别的隶属度对每个类别反演结果进行加权平均，试图从根本上提高反演方法的适用性。然后以太湖为例，分别利用水面光谱和遥感图像对软分类反演策略进行检验。最后基于经过检验的软分类反演策略生产太湖水体叶绿素a时序产品，并分析其时空变化规律及主要影响因素。

The optical properties of coastal and inland case Ⅱ waters are complicated and vary with regions and seasons, many chlorophyll-a retrieving approaches have strong limitations for specific region and season. Actually, the main factor affecting the applicability of chlorophyll-a retrieving approach is retrieving strategy. Traditional retrieving strategy usually uses one approach for each area and each season. Since the optical properties of water in the same area and season may also vary largely, the traditional strategy has much limitation. Presently there are some retrieving strategies which retrieve chlorophyll-a based on classification, and can improve the applicability of chlorophyll-a retrieving approach. However, theses retrieving strategies use hard-classification method to directly put water into classes. Therefore, the differences of water within one class are ignored, which lead to the limitations of the retrieving approach. Our research studies chlorophyll-a retrieving strategy based on soft-classification (fuzzy-classification), the chlorophyll-a concentrations retrieved by the retrieving approach of each class are weighted averaged by the subordinate degree of each class. This strategy can fundamentally improve the applicability of chlorophyll-a retrieving approach. Our research take the turbid eutrophicate Taihu Lake as the study area, and validate the soft-classification based chlorophyll-a retrieving strategy based on water surface reflectance spectra and satellite images. Then, long time series of chlorophyll-a concentration products of Taihu Lake are produced, and the spatial-temporal distribution rules and their main affecting factors are analyzed.