Research papersEvaluation of climate modeling factors impacting the variance of streamflow

- For the first time, variance decomposition of GCM-forecasted streamflow is performed.
- Forecasted streamflow variance depends on GCM type, GCM version, and downscaling.
- Variance is reduced when streamflow forecasted with the delta change method is used.
- A balanced GCM ensemble negates the use of excessive realizations of model runs.
- Precipitation and streamflow increases of 10% and 11% were predicted.

The present contribution quantifies the relative importance of climate modeling factors and chosen response variables upon controlling the variance of streamflow forecasted with global climate model (GCM) projections, which has not been attempted in previous literature to our knowledge. We designed an experiment that varied climate modeling factors, including GCM type, project phase, emission scenario, downscaling method, and bias correction. The streamflow response variable was also varied and included forecasted streamflow and difference in forecast and hindcast streamflow predictions. GCM results and the Soil Water Assessment Tool (SWAT) were used to predict streamflow for a wet, temperate watershed in central Kentucky USA. After calibrating the streamflow model, 112 climate realizations were simulated within the streamflow model and then analyzed on a monthly basis using analysis of variance. Analysis of variance results indicate that the difference in forecast and hindcast streamflow predictions is a function of GCM type, climate model project phase, and downscaling approach. The prediction of forecasted streamflow is a function of GCM type, project phase, downscaling method, emission scenario, and bias correction method. The results indicate the relative importance of the five climate modeling factors when designing streamflow prediction ensembles and quantify the reduction in uncertainty associated with coupling the climate results with the hydrologic model when subtracting the hindcast simulations. Thereafter, analysis of streamflow prediction ensembles with different numbers of realizations show that use of all available realizations is unneeded for the study system, so long as the ensemble design is well balanced. After accounting for the factors controlling streamflow variance, results show that predicted average monthly change in streamflow tends to follow precipitation changes and result in a net increase in the average annual precipitation and streamflow by 10% and 11%, respectively, for the wet, temperate watershed.