mike dettinger usgs, la jolla, ca downscaling to local climate

29
Mike Dettinger USGS, La Jolla, CA DOWNSCALING DOWNSCALING to local climate

Upload: dorcas-irene-griffin

Post on 29-Jan-2016

228 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Mike DettingerUSGS, La Jolla, CA

DOWNSCALING DOWNSCALING to local climate

Page 2: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

The “downscaling” problem

Downscaled

Original GCM valuesGlobal Climate Model

(actually, general circulation model)

Page 3: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

(GFDL A2)

One day in the 21st Century

Downscaled

Original GCM values

Page 4: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Downscaling options:

• T and %P rescaling

• Synthetic statistical

• Deterministic statistical

• Dynamical simulation

Page 5: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

dT and %P re-scalings

ADD projected mean temperature changes to a historical record

Multiply historical record by projected mean precipitation (as fraction of historical)

Easy, maintains realistic variability, know exactly what changed

No new variability or extremes, not realistic, minimal use of GCM

Historical

Future (+3C)

Historical

Future (80%)

Page 6: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Synthetic statistical scenarios

e.g., Wilby et al., 2001

Page 7: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Synthetic statistical scenarios

e.g., Wilby et al., 2001

Explained

Explained

Precipitation

Temperature

Page 8: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Deterministic statistical

Local example

(Dettinger et al, 2004 Clim Chg)

Map GCM variables into historical distribution of variables, maintaining ranks from GCM but absolute values from historical records

Page 9: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Deterministic statistical: A continental scale example

Hidalgo, H.G., Dettinger, M.D., and Cayan, D.R., in review, Downscaling using constructed analogues daily US precipitation and temperatures: J. Climate, 24 p.

The constructed-analogs method

Given a daily GCM map to downscale,

Page 10: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Hidalgo, H.G., Dettinger, M.D., and Cayan, D.R., in review, Downscaling using constructed analogues daily US precipitation and temperatures: J. Climate, 24 p.

Page 11: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate
Page 12: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

The Bias-Correction Spatial Disaggregation Method

(e.g., Maurer & Hidalgo, 2008 HESS)

Hybrid statistical:

BCSD

Page 13: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

(e.g., Maurer & Hidalgo, 2008 HESS)

BCSD

Page 14: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

(e.g., Maurer & Hidalgo, 2008 HESS)

BCSD

Page 15: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Global Model

Regional Climate Model

Dynamical simulation for downscaling

Page 16: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

WindsTemperatures

VaporPressure levels

WindsTemperaturesVaporPressure levels

WindsTemperaturesVaporPressure levels

WindsTemperaturesVaporPressure levels

Approximately the same physics,

dynamics & drivers as in global

model

Page 17: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate
Page 18: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

What are you going to use the downscaled scenarios for?

Which climate variables/statistics matter most?

Do you need daily resolution, daily congruence among climate variable? Monthly? Long-term mean? Is interannual variability important? Is long-term evolution of impacts significant?

What spatial resolution do you really need? What resolution do you get by with now? Are key changes really as small scale as all that?

Page 19: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Downscaling “other” variables

Reanalysis: Jan 1 1950surface humidity

Dynamical:CARD10 by Kanamitsu & Kei

Statistical:Constructed analogsby Hidalgo et al

Downscaling the usual variables

Temperatures & precipitation

Probable skills:

TemperatureHumidityLongwave radSolar radPrecipitationSurface winds

Page 20: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Temporal resolution & characteristics:

• T and %P rescaling• Time scales match historical record used• Variability duplicates historical unless shuffled• Interannual etc characteristics can be lost if shuffled• Higher-order statistics duplicate historical

•Synthetic statistical• Daily harder to get right than monthly than annual

(especially for precipitation)

• Deterministic statistical • Daily is possible if daily GCM output available• Temporal characteristics drawn from GCM

• Dynamical simulation• High temporal resolutions• Long-term sequencing from GCM• Higher-order statistics can change in consistent ways• Long scenarios expensive• Model biases still need correction

Page 21: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Downscaled Projected Trends in December Precipitation by Two Approaches

(GFDL CM2.1, A2 emissions, 21st Century)

Constructed analogsBCSDBias correction & spatial downscaling, from Ed Maurer, SCU

Page 22: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

My downscaling wish list for CASCaDE:

-- Resolution: Daily time resolution, ~ 10-km spatial resolution, multiple climate variables

-- Accuracy: Reproduction of high-resolution historical records

-- Feasibility: Not too computationally burdensome (?)

-- Synchronicity: Downscaled weather synchronized with climate model weather (just a Delta SLR-floods thing?)

-- Theoretical: Doesn't constrain future higher-order stats to be same as historical

-- Aesthetic: Climate (& trends) arise from entire weather field rather than being imposed GCM-grid cell by grid cell

-- Practical: Ability to downscale to grids + stations at same time, maintaining internal consistency

Page 23: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Skill of downscaling as indicated by application of method to historical OBSERVATIONS

Skill at monthly average scale

Page 24: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Application of method to historical OBSERVATIONS shows that even extremes are captured accurately

Hidalgo, H.G., Dettinger, M.D., and Cayan, D.R., in review, Downscaling using constructed analogues daily US precipitation and temperatures: J. Climate, 24 p.

Distributions of daily precip at selected sites

Page 25: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Trends, annual precipitation (GFDL-A2 scenario)

Page 26: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Some PROs:

dT %P: Easy, know exactly what you changed,realistic variability & geography, no biases

Statistical synthesis: Can be easy (not always),draws on best parts of GCMs, computationallyquick, multiple realizations, biases handled

Deterministic statistical: Computational middle ground,uses variability from GCM (thus can reflect

changing climate modes), whole distributions can be preserved (high-order debiasing), realistic geography

Dynamical (regional) simulation: Draws large scale variability from GCM, first-principles physics, all variables downscaled

Page 27: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Some CONs:

dT %P: No new variability, only minimal info from GCM being used, changes not realistic

Statistical synthesis: Difficult to retain spatial & variable inter-relations, climate variability is not really simple noise, some variables purely random

Deterministic statistical: Computational middle ground,assumes local climate responses to large-scale climate are stationary, single climate realization

Dynamical (regional) simulation: Computationally burdensome, bias correction still needed, relatively short simulations provided

Page 28: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

What is in a typical set of downscaled scenariosthese days?

1. Projections by multiple climate models

2. Projections under various greenhouse-gas emission scenarios

3. Historical-period simulations for use in calibrations/testinge.g., 1950-99

along with the 21st Century climate projections

4. Daily to monthly time steps (over, say, 150 years)

5. 10- to 12-km spatial resolution (statistical) or 20-30 km (dynamical), regional to continental scales

6. Precipitation & temperatures (just beginning other variables)

Page 29: Mike Dettinger USGS, La Jolla, CA DOWNSCALING to local climate

Some available downscaled projections:

• BCSD—dozens of scenarios at 12 km, monthly

http://gdo-dcp.ucllnl.org/downscaled_cmip3_projections/

• ConstrAnalogs—several scenarios at 12 km, daily

http:// cascade.wr.usgs.gov/data/Task1-climate/

• Numerical regional model projections—multiple models, not so much off the shelf

http://www.narccap.ucar.edu/

An upcoming (statistical) downscaling improvement:BCCA—best of BCSD & CA methods!