Downscaling SMAP Soil Moisture Products With Convolutional Neural Network

Abstract

Soil moisture (SM) downscaling has been extensively investigated in recent years to improve coarse resolution of SM products. However, available methods for downscaling are generally based on pixel-to-pixel strategy, which ignores the information among pixels. Hence, a new downscaling method based on a convolutional neural network (CNN) is proposed to solve the problem. Furthermore, a weight layer is designed for the input, and residual SM is treated as the output of the CNN to improve the accuracy. This method is applied to downscale Soil Moisture Active Passive (SMAP) SM products (i.e., 36-km $\mathbf {L3{\_}SM{\_}P}$ and 9-km $\mathbf {L3{\_}SM{\_}P{\_}E}$) from January 1, 2018 to December 30, 2018. Compared with 9-km $\mathbf {L3{\_}SM{\_}P{\_}E}$, the 9-km downscaling result is satisfactory with obtained correlation coefficient ($R$), root mean square error (RMSE), and unbiased RMSE (ubRMSE) values of 95.81%, 2.77%, and 2.67%, respectively. Moreover, SMAP SM products (36 and 9 km) and downscaling SM (3 and 1 km) are validated by the in situ data, which are collected by the 109 stations of the Oklahoma Mesonet SM monitoring network. Mean $R$, RMSE, and ubRMSE values are 67.92%, 7.94%, and 4.87% for 36-km $\mathbf {L3{\_}SM{\_}P}$; 67.78%, 8.35%, and 4.95% for 9-km $\mathbf {L3{\_}SM{\_}P{\_}E}$; 67.28%, 8.34%, and 4.97% for 3-km downscaling SM; 65.90%, 8.40%, and 5.18% for 1-km downscaling SM, respectively. The 3-km downscaling SM generated by this method can improve the coarse resolution of 9-km $\mathbf {L3{\_}SM{\_}P{\_}E}$ while preserving its accuracy. However, error will remarkably increase in the 1-km downscaling SM. Therefore, the proposed method provides a new strategy for SM downscaling and obtains satisfactory results in practice. Additional studies can be conducted in the future.

Keywords

• CNN
• downscaling
• SMAP
• soil moisture (SM)