CONUS soil moisture and attribute dataset

This dataset compiles in-situ soil moisture observations from the International Soil Moisture Network across the contiguous United States (CONUS). A total of 1,017 locations were selected, yielding approximately 3,200 soil moisture time series spanning depths from 0.05 to 1 m. Sites include both single-depth and multi-depth profiles (up to five layers). Temporal coverage extends from January 1, 2000 to December 31, 2020, with record lengths varying from short campaigns (3–6 months) to near-continuous 21-year observations. Dynamic meteorological forcings were derived from ERA5-Land, DayMet, and NLDAS, enabling assessment of the impact of spatial resolution, temporal frequency, and data provenance on model performance and uncertainty. Static physiographic attributes were obtained from HydroATLAS. All gridded inputs were extracted at exact station coordinates using a modified version of the Caravan data pipeline, adapted from watershed-based aggregation to point-scale data extraction suitable for soil moisture modeling. The resulting dataset provides a diverse and well-documented benchmark for training and evaluating data-driven soil moisture models across a wide range of hydroclimatic conditions in CONUS.