Uncertainty Analysis of Climate Change Effects on Runoff for the Pacific Northwest Greg Karlovits and Jennifer Adam Department of Civil and Environmental Engineering, Box , Washington State University, Pullman, WA 991 Department of Civil and Environmental Engineering, Box , Washington State University, Pullman, WA Introduction Rainfall Statistics Acknowledgements Components of Uncertainty Conclusions The Pacific Northwest Monte Carlo Forecasts Monte Carlo Simulation For weighting VIC runoff results, 5000 random selections of emissions scenario, GCM, SWE and soil moisture were made based on a weighting scheme. Emissions scenarios had equal selection probability (p=0.5) GCMs were weighted by ability to re-create climate over the PNW SWE and soil moisture were simulated with VIC for and quantiles based on a discretized normal distribution were selected GCMT BiasP BiasRA1B PB1 P CCSM CGCM3.1_t CNRM_CM ECHAM ECHO_G HADCM HADGEM IPSL_CM MIROC_ PCM Pictured above at 1/16-degree resolution are the average annual precipitation (L) and elevation (R) for the Pacific Northwest (Elsner et al. 2010) Historical ( ) and GCM-projected (2040s) annual maximum 24-hour rainfall events were fit to the Generalized Extreme Value (GEV) distribution using the method of L-moments at 1/2 degree resolution. Design storm intensities were generated using the GEV quantile function. In general, design storms were found to increase in intensity over the PNW. The most uncertainty in projecting future runoff is due to a choice in emissions scenario. The uncertainty in this choice is amplified by the different GCM projections. Biases in the historical runs of each GCM were reflected in the future projections, with the warmest and wettest models forecasting the largest increases. Using a weighting scheme, the VIC runs were averaged to produce results reflecting the likelihood or skill of a predictor, which improves the forecasting results. For the majority of the PNW, runoff is projected to increase. Most locations with heavy precipitation demonstrate increases in runoff in the future. All locations in the Puget Sound/Olympic Peninsula region show an increase in runoff due to the higher emissions scenario, which is closer to actual emissions rates. Declining snowpack west of the Cascades is linked to increased runoff. Overall Coefficient of Variation Difference in Emissions Scenario Coefficient of Variation for GCMs Difference in Snowpack Difference in Soil Moisture Historical 50-Year StormCNRM CM3 (B1) 50-Year Storm Historical (50-Year Storm Runoff) Monte Carlo (50-Year Storm Runoff) Percent Change, Historical to Future While runoff is projected to increase due to climate change for much of the Pacific Northwest, the magnitude of that change is uncertain due to a number of factors. The built-in assumptions for the emissions scenario are already low in the 21 st century, so realistic scenarios are above the “worst case” in this study. A suite of options created by emissions scenarios and GCMs helps find a central tendency in projections, where reliance on a single scenario and GCM offers no guarantee of reliability. Additional research on downscaling techniques and finer scale simulation could offer insight into more complicated runoff interactions due to the complicated topography and climate of the Pacific Northwest, and help advise changes on a level more relevant to stormwater management. Thanks goes out to TransNOW for funding this research. This research is advised by Jennifer Adam. The Master’s thesis committee consists of Michael Barber and Liv Haselbach of WSU and Veronica Griffis of Michigan Tech. GEV Quantile Function The intensity of design storms in the Pacific Northwest is projected to increase due to climate change. Assumptions of stationarity in estimating the intensity of rainfall design events are no longer valid due to this change, and designs meant to handle runoff events estimated by these storms need to be changed. Using the Variable Infiltration Capacity (VIC) hydrology model, runoff over the Pacific Northwest for the and 2040s climate were simulated for design storms of 2, 25, 50 and 100-year average return intervals. The results were weighted using Monte Carlo simulation, with selections of uncertainty parameters made randomly for 5000 realizations. Two parameters for climate uncertainty and two for model uncertainty were modeled stochastically. The amount of uncertainty due to emissions scenario, Global Climate Model (GCM), antecedent snow water- equivalent and soil moisture were isolated for their contributions to runoff.