1. AMS 25th Conference on Hydrology
Seattle, WA
January 25, 2011
Probabilistic Climate Change Analysis for Stormwater Runoff In the Pacific Northwest
Gregory S. Karlovits, now with USACE
Jennifer C. Adam (presenting), Washington State
University

2. Introduction

3. Climate Change in the PNW
2045
Temperature Relative to
Larger agreement among GCMs for annual temperature than for annual precipitation
However, seasonality and extreme events also important
Precipitation Relative to
LOWESS (locally-weighted scatterplot smoothing)
Smoothed traces in temperature (top) and precipitation (bottom) for the twentieth and
twenty-first century model simulations for the PNW, relative to the 1970–1999 mean. The heavy
smooth curve for each scenario is the REA value, calculated for each year and then smoothed using
loess. The top and bottom bounds of the shaded area are the 5th and 95th percentiles of the annual
values (in a running 10-year window) from the ∼20 simulations, smoothed in the same manner as
the mean value. Mean warming rates for the twenty-first century differ substantially between the
two SRES scenarios after 2020, whereas for precipitation the range is much wider than the trend and
there is little difference between scenarios
From Mote and Salathé (2010), University of Washington Climate Impacts Group

4. Sources of Uncertainty in Predicting Stormwater Runoff under Climate Change
Future Meteorological Conditions
Future Greenhouse Gas (GHG) emissions
Global Climate Model (GCM) structure and parameterization
Downscaling to relevant scale for hydrologic modeling
Hydrologic Modeling
Hydrologic model structure, parameterization, and scale
Antecedent (Initial) Conditions
Soil moisture
Snowpack / Snow Water Equivalent (SWE)
Snowpack and soil moisture are two examples of how moisture storage may moderate the effects of climate change, but there are other sources of uncertainty as well

5. Objectives
At the regional scale, how will stormwater runoff from key design storms change due to climate change?
What is the range of uncertainty in this prediction and what are the major sources of this uncertainty?
How can we make these forecasts useful to planners and engineers?

6. Data and Methods

7. General Methodology
For key design storms, find changes in storm intensities for different emission scenarios/GCMs
Use a hydrology model to compare future projected storm runoff to historical
Use a probabilistic method to assess range and sources of uncertainty

8. Design Storms
24-hour design storms with average return intervals of 2, 25, 50 and 100 years
Statistical modeling using Generalized Extreme Value (GEV) using method of L-Moments (Rosenberg et al., 2010)
Meteorological data: from Elsner et al. (2010): gridded at 1/16th degree over PNW
Historical: 92 years of data ( )
Future: 92 realizations of 2045 climate, hybrid delta statistical downscaling

9. VIC Macroscale Hydrology Model
Variable Infiltration Capacity (VIC) Model
Process-based, distributed model run at 1/2-degree resolution
Sub-grid variability (vegetation, elevation, infiltration) handled with statistical distribution
Resolves energy and water budgets at every time step
Routing not performed for this study
Gao et al. (2010), Andreadis et al. (2009),
Cherkauer & Lettenmaier (1999), Liang et al. (1994)

10. Modeled in VIC, fit to discrete normal distribution
Monte Carlo Framework
Random Sampling from:
Future Meteorological Conditions
Future Greenhouse Gas (GHG) emissions
Global Climate Model (GCM) structure and parameterization
Downscaling to relevant scale for hydrologic modeling
Hydrologic Modeling
Hydrologic model structure, parameterization, and scale
Antecedent (Initial) Conditions
Soil moisture
Snowpack
Snowpack and soil moisture are two examples of how moisture storage may moderate the effects of climate change, but there are other sources of uncertainty as well
Modeled in VIC, fit to discrete normal distribution

11. Monte Carlo Framework, cont’d
For each return interval, 5000 combinations were selected for VIC simulation
GCM weighted by backcasting ability as quantified by Mote and Salathé (2010)
Approach based on Wilby and Harris, 2006, WRR

12. Results and Conclusions

13. Monte Carlo Results (Average of 5000 Simulations)
Historical
Future
Similar pattern historical to future.
Historical 50-year storm
Random selection of soil moisture and SWE
Future 50-year storm
Random selection of emission scenario, GCM, soil moisture and SWE

14. Monte Carlo Results, Continued
In general, increasing in intensity, some areas decreasing with intensity.
Negative areas: could be partially because of decreases in SWE with climate change.
CV is stdev of 5000 realizations over mean of 5000 realizations. Gives estimate of overall uncertainty.
Largest CVs in areas with most intense storm depths (Even when normalized by the mean)
Percent change, historical to future runoff due to 50-year storm
Coefficient of variation for runoff for 5000 simulations of 50-year storm

15. Isolation of Uncertainty due to GCM
All Sources
GCM only
Coefficient of variation due to selection of GCM only (50-year storm)
Coefficient of variation for runoff for 5000 simulations of 50-year storm

16. Uncertainty Estimation for Individual Grid Cells
Canada
Washington State
Palouse Watershed
Oregon

17. Cumulative Density Functions

18. Conclusions
Runoff is projected to increase for many places in the Pacific Northwest
Largest increases related to most uncertainty
Range and sources of uncertainty highly variable across the PNW
Probabilistic methods can improve forecasts and isolate sources of uncertainties
enables us a better understanding on where to focus resources for improved prediction
Need for more comprehensive uncertainty assessment and higher resolution studies

19. Questions?
Chehalis, WA
Photo: Bruce Ely (AP) via

21. Outline Introduction: Data, model and methods
Pacific Northwest (PNW) climate change
Sources of uncertainty in predicting hydrologic impacts
Data, model and methods
Climate data
Design storms
Hydrologic model
Monte Carlo simulation
Results and uncertainty analysis

22. Isolation of Uncertainty: Emission Scenario
Absolute Difference (A1B – B1)
As a Percentage of Historical
On left: absolute difference: largest differences in areas of large storms intensities.
On right: expressed as a percent of historical, showing more even distribution of effects between A1B and B1.
Reason for pattern on left: A1b and b1 effects polar and opposite along the coasts. Resulting in the strong polarity there. Why??
Absolute difference in runoff due to emissions scenario (A1B – B1) (mm)
Difference (A1B – B1) as a percentage of historical (%)