Transcript Statistical downscaling of future climate change scenarios

Statistical Projection of Global
Climate Change Scenarios onto
Hawaiian Rainfall
Oliver Timm,
International Pacific Research Center, SOEST, University of
Hawai'i at Manoa
Henry Diaz, NOAA/ESRL/CIRES, Boulder, Colorado
Climate Change in the News

Hawaii researchers to look at effect of
global warming on the islands,
USA TODAY, Aug, 14, 2006

UH to study how global warming affects isles,
Star*Bulletin, Aug, 13, 2006

Floods, hotter climate in Isles likely by 2090,
Honolulu Advertiser, Feb., 25, 2007
Presentation overview

Introduction

What is the present knowledge of Hawaii's rainfall
changes during the 21st century?


Uncertainty in future climate change projections

The idea behind statistical downscaling
Results from the statistical downscaling

Connection between large-scale circulation changes
and regional precipitation

Discussion & Outlook
Introduction:
What is the scientific information
behind our present understanding of rainfall changes
over Hawaii?
Introduction:
Changes in atmospheric
Uncertainty in the scenarios
Greenhouse gas concentrations
CO2 emission 2000-2100
CO2 during the last 1000
years
Introduction:
Changes in atmospheric
Uncertainty in the anthropogenic
Greenhouse gas concentrations climate forcing
CO2 concentrations 2000-2100
Introduction:
Changes in atmospheric
Uncertainty in the global
Greenhouse gas concentrations
temperature increase
CO2 concentrations 2000-2100
A1B Scenario:
2-4.5 deg C warming
(3.6-8F)
Introduction:
Uncertainties in regional projections of climate change
Dynamical or statistical
Greenhouse
downscaling methods
gas emission
Introduction:

IPCC's Fourth Assessment Report, 2007:
(more than 20 climate models took part)

precipitation change: likely to decrease

but for Hawaii, no robust signals
Models show a drier
climate
Models show a wetter
No significant change
climate
Most models: wetter
Most models: drier climate Models results inconsistent
climate
Introduction:
Uncertainties in regional projections of climate change
Differences among
climate models
Dynamical or statistical
downscaling
methods
Greenhouse
gas emission
Introduction:
Uncertainties in regional projections of climate change
Sampling
(statistical) error
Differences among
climate models
Dynamical or statistical
downscaling
methods
Greenhouse
gas emission
Introduction:
Linkage between large-scale
and regional climate changes
Downscaling
uncertainty
Sampling
(statistical) error
Differences among
climate models
Dynamical or statistical
downscaling
methods
Greenhouse
gas emission
Introduction:
Goal of downscaling procedure: Reducing the
uncertainties of projected regional climate change
Statistical/dynamical/expert
information
Ad hoc (unguided)
downscaling
downscaling
uncertainty
uncertainty
Introduction:
What is the scientific information
behind our present understanding of rainfall changes
over Hawaii?
Statistical,
+
dynamical,
and elaborated
experts' estimates
Regional downscaling projects:
The Prediction of Regional scenarios and Uncertainties
for Defining Euorpean Climate change
risks and Effects (PRUDENCE)
Their goal: Provide a dynamically downscaled scenario for Europe
Huge project > 20 research groups!
Key steps in downscaling procedure:
1) Investigating the physical links between Hawaiian rainfall and
large-scale climate variability (diagnostic analysis)
2) Building a statistical transfer-model
3) Analysing the IPCC models (model analysis)
a) Comparison models' 20th century simulations with observations
b) Identification of circulation changes around Hawaii
c) Robustness of the projected changes
4) Application of the statistical transfer-model to the IPCC scenarios
(Statistical downscaling)
Results:
Mean surface pressure pattern during the
wet season (Nov-Apr), 1970-2000
ERA-40
H
Prevailing NE trade winds
with showers on the
windward sites
L
Data ERA-40 data avaiable at IPRC's
Asia-Pacific Data-Research Center
http://apdrc.soest.hawaii.edu/
Results: Previous diagnostic climate studies of Hawaiian Rainfall
Dry minus wet composite
Strong dependence on
El Nino-Southern Oscillation
and the
Pacific Decadal Oscillation
(P.-S. Chu and Chen, Journal of
Climate, 18,4796- 4813, 2007)
El Nino/+PDO minus La Nina/-PDO
Models project more La Nina
and more El Nino-like
tropical Pacific climate
G.A. Vecchi, A. Clement,
B.J. Sodon, EOS,89(9),81-82,2008
Results:
Months with high/low precipitation
in Hilo site of Big Island (region #5)
[ERA-40 sea level pressure, Nov-Apr, 1970-2000
High Preciptation
Low Preciptation
H
H
Results:
2) Developing a statistical transfer model:
Hawaiian Rainfall as a function of large-
scale circulation changes
Results:
Selection of circulation pattern associated with
rainfall variability over the Hawaiian Islands
Linear regression of surface wind field onto regional rainfall
[ERA-40, 1000 hPa winds, Nov-Apr, 1970-2000, n=186]
‘Trade Wind’ pattern
‘Kona Low’ pattern
Results:
Selection of circulation pattern associated with
rainfall variability over the Hawaiian Islands
Maximum Covariance Analysis of surface wind field and the regional rainfall
Results:
Selection of circulation pattern associated with
rainfall variability over the Hawaiian Islands
Maximum Covariance Analysis of sea level pressure and the regional rainfall
For region (#5)
Results:
Statistical transfer-model projects circulation
anomaly onto the 'template'
=> rainfall projection index
Observed sea level pressure
anomaly in year t
< SLP(t) , E >
y(t)
Results:
2) How well do the IPCC models reproduce
the natural variability?
- Mean sea level pressure fields
- Decompostion of the interannual sea level pressure variability
into its dominant modes (Principal Component Analysis)
[ERA-40, Nov-Apr, 1970-2000, region 10S-40N/180W-120W]
Results:
Analysis of IPCC models
Comparison of the observed
mean sea level pressure field
(wet season) with control
simulations of the IPCC models
Blue low pressure
[ERA-40 reanalysis 1970-2000]
Control simulation model #18
Orange high pressure
Control simulation model #15
Results:
Dominant pattern of observed sea level pressure
variability (1970-2000, winter seasons)
ERA-40
Anomalies (with respect to a climatological mean)
Results:
Dominant pattern of observed sea level pressure
variability (control simulation, 1970-2000, wet season)
Model #16
Results:
Dominant pattern of observed sea level pressure
variability (control simulation, 1970-2000, wet season)
Model #18
Results:
Dominant pattern of observed sea level pressure
variability (control simulation)
Model #22
Results: Finding objective criterions to select
the ‘most reliable’ models
Similarity of the dominant climate variability pattern:
Observation vs control simulation.
EOF pattern 1-10 in observation
Model #18
Model #22
EOF pattern 1-10 in simulation
0
EOF pattern 1-10 in simulation
correlation
1
Results:
Changes in the mean sea level pressure
2061-2099 – Control simulation
Model #1
Model #38
Model #28
Model #40
Model #30
Model #53
Results:
4) Application of the transfer model
downscaled projection of rainfall changes
Results:
Statistical transfer-model projects circulation
changes onto the 'template'
=> rainfall projection index
Sea level pressure
anomaly (SLPA) 2061-2099
Projection template pattern (E)
for Hilo area rainfall (wet season)
< SLPA , E >
y
Preliminary results for the Hilo area
on the Big Island
Projected changes in the wet season
(November-April) mean rainfall:

1 inch/month more rainfall

large spread among models
Summary

rainfall in different Hawaiian regions are connected different large-scale
circulation pattern (‘Trade wind’, ’Kona Low’ pattern)

Statistical downscaling of sea level pressure allows first estimates for rainfall
changes

On average, small positive rainfall changes are associated with trade wind
changes

IPCC model uncertainty for Hawaii region is very large
=> downscaled uncertainty is also very large.
Future Research/Improvements

Refining the regional structure of our diagnostic studies

Including other large-scale circulation information to improve the statistical
transfer model (e.g. wind field, stratification of the lower atmosphere)

Using model-weighted ensemble averages

Investigating changes in the extreme precipitation
(using daily data, instead of monthly /seasonal means)

Developing spatial maps of rainfall changes with confidence intervals.