Fig. 1. Overview of the study area （MO09A1 2019.7.12）

Fig. 2. Image of the target area after removing pixels with abnormal reflectance （2019.7.12， Black is the valueless area marked with 0， accounting for 64.98% of the whole image）

Fig. 3. Multi-temporal auxiliary data after removing pixels with abnormal reflectance （2019.7.4， Black is the valueless area marked with 0， accounting for 54.82% of the whole image）

Fig. 4. Multi-temporal auxiliary data after removing pixels with abnormal reflectance （2018.7.12， Black is the valueless area marked with 0， accounting for 50.13% of the whole image）

Fig. 5. Land cover classification map of target area （2019 MCD12Q1-IGBP）

Fig. 6. Basic architecture of the proposed residual network in this paper

Fig. 7. Example of training sample and auxiliary data

Fig. 8. Examples of training sample augmentation （Each row corresponds to the samples of water bodies， grasslands， permanent snow and ice area respectively）

Fig. 9. Convergence of loss function in the training procedure of the proposed method

Fig. 10. Result of reflectance reconstruction of classic surface （Each row from top to bottom corresponds to barren， water bodies， grasslands， permanent snow and ice area respectively）

Fig. 11. Result of Temporal_average

Fig. 12. Result of LC_ResNet

Fig. 13. Result of Temporal_ResNet

Fig. 14. Result of proposed method