Performance Evaluation of Rain-on-Grid Flood Simulations with a Collocated Co-kriging Reference in a Mamminasata Urban Watershed, Indonesia

This repository provides the reference flood–depth surface and Rain-on-Grid (RoG) model outputs used to evaluate urban flood simulations in the Mamminasata metropolitan area, Indonesia, with a focus on the Tallo Watershed. The data set corresponds to the 11 February 2025 flood event and was developed to assess the influence of DEM selection and parameter calibration on two-dimensional flood modelling and flood hazard mapping. The reference flood-depth surface was generated using Collocated Co-Kriging (CCoK) with field-surveyed flood depths as the primary variable and topographic parameters as secondary variables. Secondary covariates derived from the selected DEM include elevation, slope, Height Above Nearest Drainage (HAND), and distance to river. Experimental variograms and cross-variograms were used to characterise spatial dependence between variables, and candidate variogram models (spherical, exponential, Gaussian) were calibrated. Model performance was evaluated with a blocked cross-validation scheme and compared against Ordinary Kriging (OK), Inverse Distance Weighting (IDW) and kriging with external drift (KED) using RMSE and MAE. The final CCoK surface provides a spatially continuous benchmark of flood depth for the Tallo Watershed. The Rain-on-Grid simulations were forced by bias-corrected GPM-IMERG rainfall that was applied directly to grid cells, with overland flow routed using the two-dimensional shallow-water equations. Several DEM products (DEMNAS, SRTM, MERIT, FABDEM, and an Enhanced DEM combining DeltaDTM and FABDEM) were used separately in the model under a common initial parameter configuration. Each DEM was hydrologically conditioned (depression filling and channel correction) prior to simulation. Land-cover and soil information were used to assign initial Manning roughness and Curve Number values, which were then iteratively calibrated by comparing simulated flood depth and inundation extent with the CCoK reference surface using RMSE, MAE, and Probability of Detection (POD). The DEM and parameter set that achieved the best compromise between these metrics is included as the recommended configuration. As an extended application, the repository also contains derived flood hazard products based on modelled flood depth (d), the depth–velocity product (d·v), and flood duration (T). These hydraulic variables were classified using threshold values and combined through a simple decision rule to obtain a five-class hazard index (very low, low, moderate, high, very high) for Mamminasata. The hazard maps are provided for research and demonstration purposes and should not be interpreted as official flood hazard zoning. The repository includes: Gridded reference flood-depth surface from the CCoK interpolation for the 11 February 2025 event in the Tallo Watershed. Rain-on-Grid model outputs for each DEM scenario, including flood depth and, where available, flow velocity and flood duration. Calibrated model configuration for the selected DEM and parameter set. Flood hazard index maps derived from depth, depth–velocity, and duration. Supporting metadata describing spatial reference, units, and file structure.