Generalized spatio-temporal-spectral integrated fusion for soil moisture downscaling

Soil moisture (SM) is one of the key land surface parameters, but the coarse spatial resolution of the passive microwave SM products constrains the precise monitoring of surface changes. The existing SM downscaling methods typically either utilize spatio-temporal information or leverage auxiliary parameters, without fully mining the complementary information between them. In this paper, a generalized spatio-temporal-spectral integrated fusion-based downscaling method is proposed to fully utilize the complementary features between multi-source auxiliary parameters and multi-temporal SM data. Specifically, we define the spectral characteristic of geographic objects as an assemblage of diverse attribute characteristics at specific spatio-temporal locations and scales. Based on this, the SM-related auxiliary parameter data can be treated as the generalized spectral characteristics of SM, and a generalized spatio-temporal-spectral integrated fusion framework is proposed to integrate the spatio-temporal features of the SM products and the generalized spectral features from the auxiliary parameters to generate fine spatial resolution SM data with high quality. In addition, considering the high heterogeneity of multi-source data, the proposed framework is based on a spatio-temporal constrained cycle generative adversarial network (STC-CycleGAN). The proposed STC-CycleGAN network comprises a forward integrated fusion stage and a backward spatio-temporal constraint stage, between which spatio-temporal cycle-consistent constraints are formed. Numerous experiments were conducted on Soil Moisture Active Passive (SMAP) SM products. The qualitative, quantitative, and in-situ site verification results demonstrate the capability of the proposed method to mine the complementary information of multi-source data and achieve high-accuracy downscaling of global daily SM data from 36 km to 9 km.