遥感反演中先验知识空间传播模式及验证

In the general method to retrieve land surface parameters through inverting a physical-based model, only the remotely sensed data is employed. But, it should be noted that the information contained in the adjacent pixels may provide some aids to retrieve target parameters. In this article, a priori knowledge spread model is proposed to conduct the issue of information spread while retrieving vegetation key parameters using a Bayesian network method in regional scale. In our proposed algorithm, the posterior distribution of target parameters is used as the prior knowledge for its adjacent points. When using the posterior information as the spatial adjacent points’ prior information, the spatial correlation information has been added into the process of inversion, thus the inversion process is the spread of priori knowledge and new data information. In this paper, this algorithm of spatial spread of prior knowledge is abbreviated as SSPK. To validate our proposed method, the field measured data as well as ETM+ Imagery collected in Shunyi, Beijing in China dated on 17 April 2001 is employed. The experiment’s objects are focus on the winter wheat when derived the data set. Using these data, the canopy LAI (Leaf Area Index) of winter wheat is estimated using our proposed model and validated using part of the collected data. The estimated result of LAI in this region has lower RMSE for two parts of region, one is 0.5298 and another is 0.6640. The validation result shows that using the probability likelihood as the complementary measurement index of estimated LAI and the ground truth value is a suitable method. In general, the trend of probability likelihood is slightly steady. And about 77 percent of all validation results lie in the mean value plus or minus one standard variance.