Quantitative remote sensing inversion is ill-posed. The Moderate Resolution Imaging Spectroradiometer at 250 m resolution （MODIS_250m） contains two bands. To deal with this ill-posed inversion of MODIS_250m data, we propose a framework, the Multi-scale, Multi-stage, Sample-direction Dependent, Target-decisions （Multi-scale MSDT） inversion method, based on spa- tial knowledge. First, MODIS images （1 km, 500 m, 250 m） are used to extract multi-scale spatial knowledge. The inversion accuracy of MODIS_lkm data is improved by reducing the impact of spatial heterogeneity. Then, coarse-scale inversion is taken as prior knowledge for the fine scale, again by inversion. The prior knowledge is updated after each inversion step. At each scale, MODIS_lkm to MODIS_250m, the inversion is directed by the Uncertainty and Sensitivity Matrix （USM）, and the most uncertain parameters are inversed by the most sensitive data. All remote sensing data are involved in the inversion, during which multi-scale spatial knowledge is introduced, to reduce the impact of spatial heterogeneity. The USM analysis is used to implement a reasonable allocation of limited remote sensing data in the model space. In the entire multi-scale inversion process field data, spatial knowledge and multi-scale remote sensing data are all involved. As the multi-scale, multi-stage inversion is gradually refined, initial expectations of parameters become more reasonable and their uncertainty range is effectively reduced, so that the inversion becomes increasingly targeted. Finally, the method is tested by retrieving the Leaf Area Index （LAI） of the crop canopy in the Heihe River Basin. The results show that the proposed method is reliable.
Leaf area index (LAI) is an important parameter in monitoring crop growth. One of the methods for retrieving LAI from remotely sensed observations is through inversion of canopy reflectance models. Many model inversion methods fail to account for variable LAI values at different crop growth stages. In this research, we use the crop growth model to describe the LAI changes with crop growth, and consider a priori LAI values at different crop growth stages as constraint information. The key approach of this research is to assimilate multiple canopy reflectance values observed at different growth stages and a priori LAI values into a coupled crop growth and radiative transfer model sequentially using a variational data assimilation algorithm. Adjoint method is used to minimize the cost function. Any other information source can be easily incorporated into the inversion cost function. The validation results show that the time series of MODIS canopy reflectance can greatly reduce the uncertainty of the inverted LAI values. Compared with MODIS LAI product at Changping and Shunyi Counties of Beijing, this method has significantly improved the estimated LAI temporal profile.
WANG DongWei1,2,3, WANG JinDi1,2 & LIANG ShunLin4 1 State Key Laboratory of Remote Sensing Science, Jointly Sponsored by Beijing Normal University and Institute of Remote Sensing Applications, CAS, Beijing 100875, China
