Comparing Mean or Median Gauge Performance as Calibration Objective for Hydrologic Models

This dataset was used to test the difference between spatially aggregating optimization error using the mean performance and the median performance across all gauges in the calibration of hydrologic model. Two comparative and parallel model optimization tests are included: one that optimizes mean error from all gauges, and a second that optimizes median error. The watershed model used for this research is a CHARM model of the Athabasca River basin, which is included in the dataset. Model calibrations were performed using the OSTRICH program (version 17.12.19), values for 27 parameters covering all simulated hydrological storages were calibrated using the DDS algorithm. One set of calibrations was run with the objective to maximize the mean Kling-Gupta Efficiency (KGE) for 22 gauges, and a second set was run with the objective to maximize the median KGE for the same 22 gauges. Five random seeds and initial solutions were generated and used as starting points for both sets of calibrations. In total, 10 calibrations (5 paired calibration trials) were run with 5000 iterations per calibration. Mean-calibrated models perform significantly better than median-calibrated models.

This dataset includes:
1) The complete hydrologic model used in the calibration experiment, including the calibration software and statistics calculation script
2) The calibration trial input and output files (ostIn, OstOutput) with the summary data files combining results for plotting