A Remote Sensing Driven Soil Moisture Estimator: Downscaling with Geostatistically Based Use of Ancillary Data

Soil moisture (SM) is a key component of ecological and hydrological models, and it's also one of the best indicators of climate change impacts. It represents the flow of water, electricity, and food, as well as being a critical component of carbon exchange between the ground, soil, and atmosphere. Despite its significance and high value, SM data is not commonly available due to the high cost of field sampling and the absence of a custodian, unlike precipitation and streamflow measurements, in many developing countries. Therefore, the accurate spatial distribution of this hydrological variable is not available for the entire catchment. In recent years, remotely sensed satellites including soil moisture active passive (SMAP) continuously monitor SM of the earth. However, the spatial resolution of the remote-sensed SM is not sufficient for hydrological applications and water resource management. In this paper, a data-driven model is utilized to capture the interaction between SM and ancillary data (AD) including normalized difference vegetation index (NDVI), altitude, slope, cumulative precipitation, and air temperature. These data along with measured SM from 158 monitoring stations in the state of Utah are collected and an artificial neural network (ANN) platform is developed for predictive measures and estimation of SM. By co-regionalization of the estimated SM from the neural network and large-scale information from the remote sensing, SM in any parts of the study area was estimated more accurately and on a finer resolution (1-km). Finally, through a linear relationship between observational SM with the ancillary up-to-date data and the downscaled satellite SM, the resulting SM values ​​were adjusted and presented by adding a stochastic term. On average, the unbiased Root Mean Square Error (UnbRMSE) between the adjusted downscaled SMAP SM data and in-situ SM observations adequately met the SMAP SM retrieval accuracy requirement of 0.04 m³/m³. In addition, the results of SM validation show that the use of this technique is able to better estimate SM for remote parts of the study area.