Preserving flow variability in watershed model calibrations

Background/Question/Methods

Although watershed modeling flow calibration techniques often emphasize a specific flow mode, ecological conditions that depend on flow-ecology relationships often emphasize a range of flow conditions. We used informal likelihood methods to investigate tradeoffs of calibrating streamflow on three standard objective functions, each representing different flow components, as well as a multiobjective function aggregating these three calibration targets to simultaneously address a range of flow conditions for watersheds in the North Carolina Piedmont.  We used Latin hypercube sampling to reduce the runs required for Monte Carlo sampling, and an iterative likelihood-weighting scheme to narrow parameter ranges for simulation sets during calibration.  The selected single objective functions (Nash-Sutcliffe, modified Nash-Sutcliffe, and standard deviation ratio) had relatively low cross-correlation and represent three distinct aspects of streamflow dynamics (high flows, low flows, and flow variability, respectively). We hypothesized that calibration targets known to emphasize certain aspects of flow (e.g., peak flows) would result in more accurate streamflow simulations for that particular response mode, while proving less accurate in others.  Furthermore, we expected to see that calibrations based on calibrated simulation sets would result in superior validations and, by extension, better predictive models and/or spatial extrapolations, than calibrations using a single best-fit approach. 

Results/Conclusions

Nash-Sutcliffe efficiency calibrations result in better matches between simulated and observed high flows than low flows, in addition, our results show that modified Nash-Sutcliffe calibrations demonstrate the lowest uncertainty within the low flow simulation sets and the central tendency of the modified Nash-Sufcliffe calibrated low flow simulation sets were closest to the observed values. However, the extent of these relationships was less than what we expected.  We used standard deviation ratio as a calibration target for flow variability, this ratio was optimized when flow ranges were greatest which inevitably favored the widest parameter ranges (typically corresponding to the first pass). Values for optimized parameters vary among calibrations using different objective functions, which underscores the importance of linking modeling objectives to selection of calibration target.  Variability among these parameters is linked to their assumed representation of watershed function. The simulation set approach yielded validated models of similar quality, as seen with a single best-fit parameter set, with the added benefit of uncertainty estimations.  Combining the simulation set approach with the multiobjective function was demonstrated to be a practicable approach for model calibration that can yield flow predictions that are more appropriate for inferring ecological conditions in watersheds.