HydroAI Data
Data products and observation networks distributed by the HydroAI Lab. Select a dataset to explore.
South Korea’s first Core Validation Site (CVS), at Hampyeong: one 1 km grid cell measured from the ground, from drones and from satellites, to check the soil moisture maps of NASA SMAP and NASA-ISRO NISAR.
Prithvi-OpenET: monthly evapotranspiration at 30 m from Landsat and Sentinel-2 imagery, estimated with a geospatial foundation model. The Korean Peninsula and France are open now as a preview; other regions are produced on request.
L-band soil moisture that retrieves the vegetation scattering albedo instead of assuming it: HDCA solves soil moisture, VOD and a time-varying effective scattering albedo together from SMAP brightness temperatures, with first-order canopy scattering, so dense forests no longer read as persistently wet. Over the Amazon basin the record follows the wet and dry seasons, as the picture shows.
How wrong is each SMAP soil-moisture value? The random error of every retrieval from triple collocation with ASCAT and GLDAS in a moving window, and, where triple collocation fails, predicted from SMAP’s own variables by LGBM and DNN models, with flags that say which is which and why.
Six water quality parameters for the rivers and reservoirs of South Korea on a 1 km grid, each value with a calibrated uncertainty: water temperature, suspended solids, chlorophyll-a, dissolved oxygen, total nitrogen and total phosphorus, estimated from Sentinel-2 imagery, weather reanalysis and national monitoring records. The portal says what each layer can and cannot be used for.