4. Two Major Downscaling Approaches
• Statistical – Estimate of local series based on
empirical /stochastic relationship
• Dynamic – Estimate of local series based on
process-based model output
5. Downscaling Types
• Simple Step Change (‘Delta Method’)
• Synthetic Weather Generator
• Statistical Modeling
• Dynamic Downscaling (RCM)
• Combinations of all the above
6. ‘Delta Method’
• Future projection is based on the historical series
‘adjusted’ by a GCM-based step change:
TF = TH + Δe
• Simplest of all approaches, can be carried out
very rapidly.
• Approach does not allow for any changes in
variability, projections statistically constrained to
historical climate
7. Synthetic Weather Generator Method
• Perturbed series are based on weather generator
output based on adjusted distributions from GCM
output (usually monthly)
• TF = f(rand. draw from pert. dist.), f(var. interrelations)
• Provides more realistic series than step change, can
allow for changes in variability. Possible to generate
many years of projected series.
• Assumes historical relationships between variables will
hold in future. Can be time- consuming. Typically does
not simulate frequency and severity of extremes well.
8. Statistical Modeling Approach
• Perturbed series are developed as statistically
based on historical series with technique such as
multiple regression or canonical correlation, e.g.
TF = f(free atmosphere variablesH,F)
• Allows realistic, physically-based projections
• Model output constrained by limits of the original
input data.
• Can be very time consuming, expensive.
9. NCAR
ECHAM
Canadian 4 GCMS, 32 daily
2 Emission temperature,
Hadley Scenarios, precipitation
4 Downscaling scenarios for
A2, B2 techniques 1990-2100
multiple
downscaling
methodologie
s
12. Dynamical Downscaling
• Perturbed series is based on the output of a process-
based model (RCM) initiated with GCM projections.
Mt. Kenya
• Since it is process-based, method should work for
almost any future scenarios. Theoretically, the best
approach.
• Models may not accurately simulate some variables,
e.g. precipitation, clouds. Based on gridded output,
series are still areal averages. Can be very time, labor,
intensive and require special computational
infrastructure.
13. Summary
• The vast majority of impact assessment
research requires downscaled climate series.
• The choice of downscaling strategy depends
on the type of application, ultimate research
objectives, and available project resources.
• Safe to assume that some type of downscaling
will be needed for impact assessment well
into the future.