Bidirectional reflectance distribution function estimation of typical ground objects in Xiamen and its availability in validation of normalized-difference vegetation index

Abstract

The bidirectional reflectance distribution function (BRDF) is necessary for studying complex land surface parameters. We implemented BRDF extraction for typical ground objects in Xiamen, China using multi-temporal HJ-1A/B satellite images in conjunction with the common Roujean kernel-driven model (RKDM). Ground objects were automatically recognized in post-processed images, and six typical ones (buildings, woods, water, farmland, roads, and sand) were extracted. We then extracted the land surface reflectance and imaging angles of the same sample points for each typical ground object from four images taken over a six-week period. We estimated the BRDF of the six ground objects using the RKDM, and analyzed the precision of the model. This led to generally accurate BRDF estimation of typical ground objects. Furthermore, we systematically analyzed BRDF availability in the validation of the normalized-difference vegetation index (NDVI). Although it was necessary to bring geometric BRDF imaging normalization into NDVI validation, problems were associated with this. Because the RKDM is statistical in nature, it cannot describe the essential characteristics of the spatial distribution of land surface reflectance, and may generate other uncertainties for NDVI validation. We intend to explore more robust and precise BRDF models in future work.