Downscaling Remotely-Sensed Soil Moisture Estimates using Machine Learning Techniques

Soil moisture, a state variable constraining the fluxes at the land surface and atmosphere boundary, plays a key role in regulating the feedbacks between the terrestrial water, carbon and energy cycles. Therefore, it is necessary to characterize soil moisture at spatial scales relevant to representation of land surface processes and mesoscale processes in the atmosphere. However, current space-borne instruments have relatively coarse spatial resolution which limits their applicability in a wide range of applications including hydrologic modeling, weather forecasting, and agricultural management, among others. In this study, we develop a machine learning based algorithm to downscale remotely-sensed soil moisture observations to a fine resolution of 2.25km. The algorithm is based on an Artificial Neural Networks (ANN) and uses ancillary data from visible/infrared frequencies to capture the fine scale heterogeneity of soil moisture. We use soil moisture estimates from Soil Moisture Active Passive (SMAP) mission at two different spatial resolutions to train the downscaling algorithm. Then, we use the network with SMAP estimates at 9km resolution as input to retrieve the soil moisture at fine spatial resolution of 2.25km. Results show that ANN can successfully capture the complex relationship between the soil moisture estimates at two different spatial resolutions using the ancillary data provided.