Parameters for rainfall-runoff modelling in subcatchments of the Citarum basin, Java, Indonesia

The parameter set in each row of each file was selected from 100,000 candidate sets – being the one set that maximized KGE’ between modelled and observed river flow.

First, 100,000 candidate parameter sets were generated by sampling the parameter values from uniform distributions with upper and lower limits suggested for each model parameter in the documentation of MARRMoT v1.2 (10.5194/gmd-12-2463-2019). Each model was input with one of 21 combinations of seven precipitation datasets of daily time resolution (AgMERRA, CHIRPS v2, CPC, MSWEP v2, PERSIANN-CDR, SA-OBS, Yanto (10.1038/sdata.2017.72)) and three evapotranspiration datasets of daily time resolution (AgMERRA, SA-OBS, Yanto) as climatic input data – for each catchment separately, catchment-average values were produced by identifying the catchment outline from a hydrologically corrected Digital Elevation Model (HydroSHEDS v1.1 CON 15-arcsecond) and overlaying this into the gridded precipitation and evapotranspiration products. Daily mean runoff (the model outflow) was obtained for each of the 38 catchments for 2,100,000 cases (100,000 parameter sets with 7 precipitation and 3 evapotranspiration datasets). Then, modified Kling-Gupta Efficency (KGE’) between modelled and observed daily mean flow was found, ignoring the first three years of modelled flow as “spin-up”.

Finally, the parameter set resulting in the highest KGE’ was saved for each catchment with each combination of input climatic data (38 catchments × 7 precipitation datasets × 3 evapotranspiration datasets = 798 parameter sets), for each model, resulting in the 24 files of 798 rows each that comprise this dataset.