Deep Learning Approaches to Spatial Downscaling of GRACE Terrestrial Water Storage Products Using EALCO Model Over Canada

• DOI: 10.1080/07038992.2021.1954498
• Publication Date: 2021-07-26T18:06:35Z
• Authors: Hongjie He; Ke Yang; Shusen Wang; Hasti Andon Petro...; Ming Liu; Junhua Li; José Marcato Junior; Wesley Nunes Gonç...; Lanying Wang; Jonathan Li
• Keywords: Environmental sciences; Technology; 550; T; 0207 environmental engineering; GE1-350; 02 engineering and technology; 6. Clean water; 620

Estimating terrestrial water storage (TWS) with high spatial resolution is crucial for hydrological and resource management. Comparing to traditional in-situ data measurement, observation from space borne sensor such as Gravity Recovery Climate Experiment (GRACE) satellites quite effective obtain a large-scale TWS data. However, the coarse of GRACE restricts its application at local scale. This paper presents three novel convolutional neural network (CNN) based approaches including Super-Resolution CNN (SRCNN), Very Deep (VDSR), Residual Channel Attention Networks (RCAN) downscaling monthly products using outputs Ecological Assimilation Land Observations (EALCO) model over Canada. We also compare performance CNN-based methods empirical linear regression-based method. All comparison results were evaluated by root mean square error (RMSE) between reconstructed original one. RMSEs matched pixels are 22.3, 14.4, 18.4 71.6 mm SRCNN, VDSR, RCAN method respectively. Obviously, VDSR shows best accuracy among all methods. The result super preform much better than in downscaling.