SWRM_Data_Setail
Description
This dataset supports a study comparing two rainfall-runoff models of contrasting structure — the semi-distributed, process-based SWAT+ and the lumped conceptual IHACRES — in the Setail Watershed (278.43 km²), a monsoonal tropical catchment in Banyuwangi Regency, East Java, Indonesia. Both models were applied under a single harmonized protocol: one catchment outlet, one observed discharge record at the Kradenan gauge, one Thiessen-weighted areal rainfall series derived from eight rain gauges, a shared 2005 warm-up year, a common monthly evaluation step, and identical calibration (2006-2015) and validation (2016-2025) windows. The central question the data address is whether a model ranking established during calibration survives a change in flow regime. The validation decade proved considerably quieter than the decade the models were fitted on, with areal rainfall falling from 222.7 to 196.3 mm per month and mean monthly discharge from 6.94 to 4.45 m³/s, which makes the two windows a natural test of parameter transferability rather than a routine split-sample exercise. The deposit contains the paired monthly series of observed and simulated discharge for both models in their uncalibrated and calibrated states, covering 240 months from January 2006 to December 2025; the rain gauge network with coordinates, elevations and Thiessen weights; the full variance-based Sobol sensitivity results for all fourteen screened SWAT+ parameters, obtained from 3,000 samples with NSE on monthly flow as the objective function; one-at-a-time perturbation results at ±20% for the five adjustable IHACRES parameters; the complete calibrated parameter sets for both models, including the derived IHACRES unit-hydrograph terms; the SWAT+ project and TxtInOut directory for the calibrated configuration; the SWAT+ Toolbox v3.2 sensitivity reports; and the IHACRES project files. A short Python script using only the standard library is included. It reads the paired discharge series and regenerates every performance statistic reported in the article — R², Nash-Sutcliffe efficiency, percent bias, root mean square error, and performance retention between the two evaluation windows — so that the published results can be verified in a single command without any proprietary software.