遥感BRDF物理模型均建立于一定的假设或基于某些理想状况，其模拟的数据与观测数据之间多少会存在一些差异（误差）。利用BRDF模型反演地表参数时，如果不加选择地使用所有观测数据，势必会影响模型参数反演的准确度。遥感反演时一般都采用代价函数进行参数拟合。经典的最小二乘（LS）拟合代价函数对正态分布误差具有一定的抗干扰性，但是当观测数据含有异常值时却会导致反演结果的不稳定。最小中值平方（LMS）方法具有鲁棒性特点，反演时若将其作为代价函数，则可以有效地检测出观测数据中含有的异常值，从而可以使模型反演准确度提高。本文以遥感BRDF物理模型——SAIL模型为例，使用模拟数据与真实地面观测数据，构建LMS与LS两种代价函数，分阶段地进行地表参数的反演方法研究。结果显示，针对具有一定误差或模型不能完全表示的观测数据，本文采用的分阶段方法可以对模型参数鲁棒地反演。

As any physically-based BRDF models were established on some assumptions,there always exist some differences between the simulated data and the measured data.When using the model to invert the ground parameters,the accuracy will be decreased if we use all measured data without distinguishing them.A merit function is usually used as the fitness of the modeled value and that of measured.The least-squares(LS) criterion,traditionally selected as the merit function,lacks the robustness when there are some stochastic errors in the measured data,though it can deal with the normal distribution errors.The least median of squares(LMS) method has the potential to find the abnormal data which belong to the stochastic errors.So we can improve the accuracy of the inversion through kicking away the abnormal data relative to the model with LMS.Using LMS and LS as the merit function separately,in this paper we take the multi-stage inversion of the SAIL model as an example to inverse the ground parameter.It has demonstrated that,toward the measured data which have some errors or can't be simulated by the model,this approach is robust to estimate the parameters.