Machine learning-based multilayer soil moisture datasets: SoMo.ml

Soil moisture information is valuable for a wide range of applications in various fields such as hydrology, agriculture, and climate. Although spatially continuous soil moisture data can be obtained from satellite observations or model simulations, each type of data has its own uncertainty and bias. In this study, we use machine learning as a hydrologic model and generate a gridded soil moisture dataset—SoMo.ml—that can complement existing soil moisture datasets (O and Orth, 2021). We train a Long Short-Term Memory neural network model using in-situ measurements to extrapolate daily soil moisture dynamics in space and in time. The first version of the data, SoMo.ml v1, provides multilayer soil moisture (0-10cm, 10-30cm, and 30-50 cm) at 0.25° and daily resolutions for the period 2000-2019; it has been actively used for drought analysis, data comparison, and other relevant research. The dataset is freely available from https://www.bgc-jena.mpg.de/geodb/. Given the growing needs for this unique soil moisture dataset, SoMo.ml v2 is currently under development, which aims to provide soil moisture data over Europe with a higher spatial resolution (0.1°). In this presentation, we will introduce the SoMo.ml datasets and show examples of data applications in other studies.

 

Reference: O and Orth, Global soil moisture data derived through machine learning trained with in-situ measurements. Sci Data, (2021). https://doi.org/10.1038/s41597-021-00964-1