4.3 Probabilistic climate change analysis for stormwater runoff in the Pacific Northwest

Tuesday, 25 January 2011: 9:00 AM
611 (Washington State Convention Center)
Gregory S. Karlovits, University of Iowa, Iowa City, IA; and J. C. Adam

Climate change has the potential to intensify storm events which would affect design storms currently used in engineering that are based on historical, stationary data. Since the loss of stationarity due to climate change decreases the ability to accurately predict the magnitude of stormwater runoff due to extreme precipitation events, a framework for assessing the range of possibilities becomes necessary. The resulting change in design storms affects the anticipated amount of stormwater runoff and therefore the riskiness of existing hydraulic structures constructed to handle it. This paper presents a framework for assessing the risk associated with predicting stormwater runoff in the face of climate change.

Historical and future climate scenarios in the Pacific Northwest were modeled using the Generalized Extreme Value (GEV) distribution, which was fit to the annual maximum 24-hour precipitation event for gridded data at 1/16 degree resolution using the method of L-moments. The intensity of the design storms for key return intervals was determined for the 1916-2006 historical climate and a number of future climate scenarios for the 30-year period around 2045, which encompassed three SRES emissions scenarios, ten GCMs, and three downscaling methods. It was found that over the Pacific Northwest, on average the intensity of storm events was projected to increase. The less-frequent storms (such as the 100-year 24-hour storm) were found to increase in intensity proportionally more than more frequent storms.

In order to determine the posterior distribution of runoff volumes as a result of climate change, Markov chain Monte Carlo (MCMC) simulation coupled with the Variable Infiltration Capacity (VIC) macroscale hydrology model was employed over the Pacific Northwest at 1/2 degree resolution. For the Monte Carlo simulation equal probabilities were assigned to the occurrence of each emissions scenario. Each GCM was weighted by its ability to re-produce 20th century climate over the Pacific Northwest. Downscaling methods were weighted similarly. For each return interval, a “storm scenario” based on the combined probability of the emissions scenario, GCM and downscaling method was selected at random for a large number of realizations. The VIC model was then run with the corresponding scenario and a gridded runoff volume was produced. From the resulting number of realizations a confidence interval was constructed for each of the precipitation event return intervals over the entire domain.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner