Harmonized soil-erosion database and machine-learning erodibility predictor for overtopping dam-breach forecasting
It provides (1) a harmonized multi-device soil-erosion database — 1,146 specimen records from EFA, SETD, JET, HET and related devices (1,013 with critical shear stress and 972 with erodibility coefficient), 186 raw erosion-rate-versus-shear-stress curves with power-law fits, and built-in data-dictionary and source-summary sheets; and (2) the machine-learning pipeline that produces the calibrated critical-shear-stress and erodibility-coefficient prediction intervals used in that paper (Gaussian-process regression with split-conformal prediction under grouped cross-validation), together with trained models and a prediction interface. The compiled measurements derive from the TAMU/NCHRP erosion-test database (946 records), the authors' SETD controlled-shear device (120 records), and additional literature sources; original provenance is recorded per record. Data are released under CC-BY-4.0 and the code under the MIT license (see LICENSE). The physically based dam-breach model used in the paper is proprietary to the Nanjing Hydraulic Research Institute and is not included; its governing equations are fully described in the paper and in Zhong et al. (2019). See README.md for full contents, usage, and attribution.