FAPAR Retrieval from GF-5 Hyperspectral Images Based on Unified BRDF Model
TIAN Dingfang,YANG Siqi,XU Dawei,REN Huazhong,FAN Wenjie,LIU Rongyuan
DOI: https://doi.org/10.11834/jrs.20210097
2023-01-01
Abstract:The Fraction of Absorbed Photosynthetically Active Radiation（FAPAR） is a key parameter in characterizing the photosynthesis process of vegetation and widely used in many study areas, such as vegetation monitoring, NPP estimation, and global change. Remote sensing is the only way to obtain FAPAR at large scales. Compared with the multispectral instrument, the hyperspectral instrument has an advantage in analyzing the canopy reflectance and absorption on the basis of the high accuracy of the spectrum measurement, which is important in FAPAR retrieval. This study developed a new FAPAR retrieval algorithm for the Chinese GF-5 Visible-shortwave Infrared Advanced Hyperspectral Imager（AHSI） data on the basis of BRDF unified model and the neural network（NNT）. The validation was performed in Hulun Buir Xeltala, which is a grassland and farming-pastoral area in Inner Mongolia. First, the simulated GF-5 AHSI reflectance-FAPAR datasets were generated by the BRDF unified model, and the characteristics of the data set were analyzed. Five groups of NNT input bands were selected based on the Optimal Index Factor（OIF） and a new factor OIFR, which was modified by the relevance of the band reflectance and the FAPAR. Different groups of bands were used to build the NNT, and the results were assessed by a test set in the simulation dataset. Finally, the best feature bands and the NNT were selected to generate the FAPAR map of the study area from the GF-5AHSI image. Validation with in-situ observations was made. Overall, the new factor OIFR is more efficient than the origin factor OIF in band selection. As the amount of input bands increases, the NNT accuracy gradually increases, but the trend stops when the amount reaches a certain level. Considering both band information and instrument noise, 8 bands were selected as the feature bands of FAPAR retrieval with the FAPAR RMSE of NNT is 0.014. The FAPAR map of the study area was generated, and the comparison with in-situ FAPAR shows the applicability of the method with RMSE=0.048. The reflectance and absorption by the hyperspectral data when the NNT reduced the middle term and parameters of the traditional methods simultaneously can be analyzed, presenting a new approach to the surface parameters retrieval of domestic satellite hyperspectral instruments.