DONG Yadong, JIAO Ziti, ZHANG Hu, et al. Efficient algorithm for improving the hotspot effect of the operational MODIS BRDF product[J]. Journal of Remote Sensing, 2014,18(4):804-825.
DONG Yadong, JIAO Ziti, ZHANG Hu, et al. Efficient algorithm for improving the hotspot effect of the operational MODIS BRDF product[J]. Journal of Remote Sensing, 2014,18(4):804-825. DOI: 10.11834/jrs.20143229.
An algorithm for modeling bidirectional reflectance anisotropies of land surfaces has been developed as a surrogate for the operational MODIS Bidirectional Reflectance Distribution Function( BRDF) and albedo product for user community. This algorithm is a set of kernel-driven BRDF models extensively used in several space-borne remotely sensed BRDF /albedo products. A mong these models
RossThick( RT)-LiSparseR( RTLSR) has been selected as the current operational MODIS BRDF /albedo a lgorithm. However
the hotspot effect has not been considered in RT kernel. As such
the use of an RTLSR model underestimates hotspot reflectance
thereby influencing the accuracy of the retrieval of vegetation structures
such as clumping index. On the basis of Bréon’s hotspot factor
Maignan corrected RT kernel to generate a RTMaignan( RTM) kernel. For producers
a 13-year MODIS BRDF /albedo product is reprocessed using this corrected model
but this task is time consuming. For users
the direct use of this corrected model for MODIS observations is complicated because the equivalent inputs of the operational RTLSR algorithm are not easily available. In this study
a method was developed to correct the hotspot effect for the operational MODIS BRDF product
which is available for users. Based on the effective validation using POLDER-3 /BRDF data and the selected MODIS data
this study shows that( 1) an improvement of approximately 10. 12% of relative error between our method and the RTLSR model can be obtained by estimating the hotspot reflectance;( 2) a relative error of approximately 2. 10% occurs between this method and the RTM-LiSparseR( RTMLSR) model
but this difference is not significant;( 3) relative error reaches approximately 4. 99% between this method and the RTLSR model to simulate NDHD but decreases to approximately 1. 32% between this method and the R TMLSR model.