River depths and widths dataset for the territory of Poland.

The accurate estimation of critical river parameters, specifically width and depth, via remote sensing and machine learning techniques requires high-quality, geographically dense reference datasets for reliable validation and training. This paper presents a new, validated, high-resolution dataset for the territory of Poland, derived from field geodesy cross-section surveys conducted across 246 rivers between 2011 and 2013. The final processed dataset contains 20,782 records, providing referential measurements of river width, depth, water level, and center point coordinates, all closely aligned with near bankfull hydrological conditions. River widths range from 0.34 m to 2,384 m, while depths range from 0.005 m to 17.62 m. The quality control excluded outlying records, incomplete data, corrupted elevations, and infrastructure features. Validation against publicly accessible datasets confirms the data's integrity: river depths align below the EU-DTM surface, river widths show strong correlation with MERIT Hydro (R2 = 0.93, RMSE = 35.1 m), and center points exhibit high positional accuracy relative to the MPHP10K map (median distance 1.8 m). This publicly available resource is applicable for training deep learning models (e.g., for river morphology estimation) and for validating large-scale hydrological and remote sensing products.