Generating global soil moisture data using in-situ measurements and machine learning

Abstract:

Soil moisture information is essential for a wide range of applications such as flood prediction, drought monitoring, and climate research. While in situ soil moisture measurements are regarded as ground truth, spatially-continuous soil moisture data is available from satellite observations or model simulations only. Here we present a novel global, long-term dataset of soil moisture generated through machine learning, SoMo.ml. We train a Long Short-Term Memory neural network model to simulate in situ soil moisture dynamics measured across the globe using multiple meteorological forcings and static variables. Thereby, we produce multi-layer soil moisture data at 0.25° spatial and daily temporal resolution over the period 2000–2019. We find that compared with state-of-the-art reference products, SoMo.ml performs especially well in terms of temporal dynamics, making it particularly useful for applications requiring time-varying soil moisture. As an example, we will illustrate and compare soil moisture memory, including its variations in time, space, and depth, estimated from SoMo.ml and other physics-based models.