Accurate estimates of the location, timing, and severity of soil-erosion events on arable land have eluded erosion-prediction technology for decades. Here, for the first time, we demonstrate how a machine learning model parameterised with spatiotemporal covariates within a back-end infrastructure of data cubes can nowcast the occurrence and relatively rank the severity of erosion events on arable field parcels at the regional scale with high accuracy and interpretable outputs. Our findings pave the way for dynamic erosion-monitoring systems to achieve healthy soils and improve food security.
