遥感数据误差对地表参数定量反演可靠性的影响——以太湖叶绿素a反演为例

地表参数的遥感反演误差的大小由遥感数据误差和反演模型误差共同构成，数据误差不是简单地“加减”到反演误差中，而是经过反演模型改造后融入到反演误差中。因此，在地表参数定量反演过程中，用回归系数最大或均方根误差最佳代价函数来描述地表参数与遥感反射光谱之间的关系将不太可靠。从理论上指出了最高回归系数或最小均方根误差评价方法失效的根源在于反演模型的形式，以2003年10月27日太湖实测数据为例进行了论证。研究表明，虽然TM2/TM3算法比TM2/TM1算法的回归系数高，但其对数据误差的放大效果是TM2/TM算法的2.28倍，这导致了反演结果的均方根误差比TM2/TM1算法大了7.938 5 μg·L-1;另外从定量反演结果来看，基于TM2/TM3算法和基于TM2/TM1算法的反演结果完全相反，与以往研究成果对比可知，基于TM2/TM1算法的反演结果更符合实际。因此，数据误差应该作为一个约束条件，加入到代价函数的求解过程中，才能增加反演结果的可靠性。

Effect of Remotely Sensed Data Errors on the Retrieving Accuracy of Territorial Parameters--a Case Study on Chlorophyll a Concentration Inversion of Taihu Lake

The errors of the territorial parameters retrieved from remote sensing are decided by the data error and the model error.The data error is not simply added to the total errors of retrieval results.It would be reformed by the quantitative inversion model,and then,combined with the model errors and melts into the totals errors.Accordingly,during the quantitative process,taking advantage of the highest correlation coefficient or the least root mean square error as assessment standard for describing the chlorophyll a concentration vs remote sensing parameters is not reasonable.Focusing on the above problem,the study pointed out that the reason why the result of the optimized cost function is contrary with the practical is that different model has different influence on data errors.Combined with the in situ measurements of Taihu Lake,in October,2003,it is known that due to the error magnification phenomena (TM2/TM3 algorithm is 2.28 times more than TM2/TM1 algorithm),although the regression coefficient of TM2/TM3 algorithm is higher than TM2/TM1 algorithm,the quantitative errors of TM2/TM3 algorithm are 7.938 5 μg·L-1 more than TM2/TM1 algorithm.Moreover,the retrieval results show that distribution pattern of the resuits of TM2/TM3 algorithm is completely opposite to the TM2/TM1 algorithm.According to the former research achievements,the results of TM2/TM1 algorithm would be more reasonable.In summary,only when that the factor of data error is added to the optimized cost function is taken as a constrain condition in search for the optimal solution of the quantitative models,would the retrieval results be more reliable.