High-resolution soil moisture mapping in northern boreal forests using SMAP data and downscaling techniques

Abstract

Soil moisture plays an important part in predicting different forest-related phenomena, such as tree growth or forest fire risk. As they influence the carbon storage capacity of boreal forest ecosystems, it is crucial to provide soil moisture information at high spatio-temporal scales. Current satellite-based soil moisture products often have high temporal resolution at the expense of spatial resolution. Therefore, we developed a machine-learning-based model to estimate soil moisture at high spatial resolution over boreal forested areas for the annual time period from May to October, while retaining the high temporal resolution. The basis data of the model is the 36 km spatial resolution soil moisture data from the Soil Moisture Active Passive (SMAP) mission. Additionally, vegetation properties, weather-related parameters, and measured in situ soil moisture data are used to guide the model construction process. The analysis of the developed model shows that it retains the temporal and large-scale spatial variability of SMAP soil moisture. Furthermore, comparisons with the independent in situ soil moisture data indicate that the model's predictions align more closely with in situ values than SMAP soil moisture, as RMSE decreases from 0.103 to 0.092 m3m-3 , and correlation increases from 0.46 to 0.55 over forest sites. Therefore, this machine-learning-based model can be used to predict high-resolution soil moisture over boreal forested areas.