1. Considerations in Using Climate Change Information in Hydrologic Models and Water Resources Assessments
JISAO Center for Science in the Earth System
Climate Impacts Group
and Department of Civil and Environmental Engineering
University of Washington
November, 2003
Alan F. Hamlet
Dennis P. Lettenmaier
Philip W. Mote

2. Climate Change Scenarios
Incorporating Climate Change in Critical Period Planning
Long term planning for climate change may include a stronger emphasis on drought contingency planning, testing of preferred planning alternatives for robustness under various climate change scenarios, and increased flexibility and adaptation to climate and streamflow uncertainty.
Observed Streamflows
Planning Models
Altered Streamflows
Climate Change Scenarios
System Drivers

3. Global climate models predict temperature more accurately than precipitation at the global scale.
1 BMRC
2 CCC
3 CCSR
4 CSIRO
5 ECHAM3
6 ECHAM4
7 GFDL
8 HadCM2
9 IAP
10 MRI
11 CERFACS
12 PCM
13 GISS
14 HadCM3
15 LMD
16 CSM
Lower Predictive Skill
For Precipitation
For each of 15 climate models with CMIP runs, plus NCEP and ECMWF reanalyses, this figure shows comparisons with observations as correlation (azimuth) and ratio of simulated to observed variance. Model 3 is CCSR, 6 ECHAM4, 8 HadCM2, 10 MRI, 12 PCM, 14 HadCM3, 15 LMD, 16 NCAR
Higher Predictive Skill
For Temperature

4. Bias Correction of GCM simulations is required for hydrologic studies.
ACPI: PCM-climate change scenarios, historic simulation v air temperature observations

5. ACPI: PCM-climate change scenarios, historic simulation v precipitation observations

6. Downscaling of GCM simulations is required for hydrologic studies.
~ Five grid cells over CA
Small scale topography is not represented.

7. Overview of GCM Downscaling Techniques

8. “Delta” Method
Monthly changes in temperature and precipitation from the GCM simulation are applied uniformly to an observed temperature and precipitation record.
Advantages:
Relatively simple to interpret
Simultaneously removes GCM bias and downscales
Minimizes the effects of time series and spatial uncertainties in free running GCMs
Disadvantages:
Assumes time series and variability of monthly values like those in the historic record

9. Statistical Bias Correction and Downscaling
Advantages:
Computationally efficient in comparison with dynamic downscaling and has been shown to have comparable performance at large spatial scales when the spatial variability is principally controlled by topography.
Includes potentially significant temporal, spatial or topographic variations in temperature and precipitation (or other variables) unlike those in the historic record. (Most sophisticated methods may include variation in number of sunny days, daily precipitation variability, etc.)
Disadvantages:
Inherits additional uncertainties from the large scale GCM forcing simulation (e.g. including spatial variability from the GCM also introduces uncertainty about the accuracy of the spatial variability in the simulations.)

10. Dynamic Downscaling Using Meso-Scale Climate Models
Advantages:
Includes potentially significant temporal, spatial or topographic variations in temperature and precipitation (or other variables) unlike those in the historic record.
Spatial downscaling not required in some cases (depends on scale of interest)
Produces physically-based simulations at small spatial scales (e.g. changes to topographically driven micro-climates may be more realistically simulated).
Disadvantages:
Computationally intensive
Meso-scale simulations inherit uncertainties from the large scale GCM forcing
Adds another stage of modeling with associated increases in cumulative bias

11. Hybrid Methods
(e.g. traditional stochastic hydrology with altered probability distributions derived from GCM simulations)
Advantages:
Can potentially exploit the strengths of different methods while avoiding some pitfalls.
Disadvantages:
Objective rationale for mixing and matching results from different sources is not straight-forward to produce.

12. Conclusions
Currently one of the largest sources of uncertainty in climate change hydrologic assessments is contained in the GCM simulations themselves and in the linkages between GCMs and hydrologic models.
Bias must be removed from GCM and Meso-Scale climate simulations to produce useful hydrologic assessments, and spatial and temporal downscaling is also frequently required.
More sophisticated downscaling techniques can include more information from GCM simulations, but also introduce additional uncertainties.
No downscaling scheme is free of problems, and the techniques used to interpret GCM simulations should be chosen based on the goals of the study.