Cloud-covered MODIS LST reconstruction by combining assimilation data and remote sensing data through a nonlocality-reinforced network

• DOI: 10.1016/j.jag.2023.103195
• Publication Date: 2023-01-25T12:47:58Z
• Authors: Yuting Gong; Huifang Li; Huanfeng Shen; Chunlei Meng; Penghai Wu
• Keywords: Moderate-resolution imaging spectroradiometer; Root mean square

Reconstruction of cloud-covered thermal infrared land surface temperature (LST) is vital for the measurement physical properties in at regional and global scales. In this paper, a novel reconstruction method Moderate Resolution Imaging Spectroradiometer (MODIS) LST data with 1-km spatial resolution proposed by combining assimilation remote sensing through nonlocality-reinforced network (NRN) model. Firstly, grading criterion introduced to evaluate importance various datasets, forming four combinations multi-modal datasets training testing NRN Secondly, model multiscale encoding–decoding structure considering module reconstruction. The results suggest that can precisely reconstruct LST, mean absolute error (MAE) less than 0.8 K, even when no auxiliary are used (Combination 1). best result full combination 4), which coefficient determination 0.8956, MAE 0.5219 root-mean-square 0.7622 K. Compared traditional harmonic analysis time series method, improved enhanced temporal adaptive reflectance fusion feature connected convolutional neural reconstruction, achieve superior results. Combination 1 has been implemented daily study area 2019. Referring meteorological station observations, reconstructed bias value indicating very effective valid