evaluating the ability of climate models to simulate extremes eric robinson natalie mclean christine...

Evaluating the ability of climate models to simulate

extremes

Eric RobinsonNatalie McLean

Christine RadermacherRoss Towe

Yushiang Tung

Project 6

Motivation

Projections depend on accurate modeling of extreme values of severe weather indicators e.g.: CAPE and shear

Extreme values of CAPE and shear have been shown to be good predictors of severe thunderstorms Hail > diameter 5cm & wind > 120km/h Significant thunderstorms (F2 or greater)

Important for the Continental United States Especially given the increased intensity of extreme weather in

recent years

Practical applications: Reinsurance firms Can use the information to minimise any pay-outs to their customers

Background

CAPE – Convective Available Potential Energy Measure of buoyancy of an air parcel Indicates atmospheric instability Summer maximum & winter minimum

Shear (vertical) Difference in wind speed between 0 and 6 km Significant wind shear aids the formation of supercells Winter maximum & late summer minimum

Background: Data sets

CCSM3 – Community Climate System Model v3 1.4o x 1.4o resolution Atmospheric, land, ocean and sea-ice models integrated

through a coupler Has positive biases for shear and negative biases for CAPE

Reanalysis – NCEP/NCAR Reanalysis Observations filled using quality control and data

assimilation methods 1.875o x 1.915o resolution CAPE and shear are Type-B variables

Determined using both actual observations and modeled data

Data Description

GCM data from CCSM3

Observation data from NCEP/NCAR Reanalysis

20 years of CAPE and shear data (1980-1999)

Maximum values extracted for the JJA season

Variables analyzed: CAPE Shear CAPE * Shear

Previous Research

Bivariate modeling has been shown to have no significant advantages

Modeling CAPE*Shear has statistical advantages Easier to quantify Resolves issues of achieving extreme weather with various combinations

of CAPE and Shear

Higher values expected over the Appalachians with lower values expected over Minnesota and North Dakota

Analysis on the entire year rather than the summer season

The use of false discovery rate was not beneficial

Methodology

Summary statistics

GEV analysis

L Moments GEV analysis

Evaluation of return values

Cluster analysis

QQ Plots to check the clustering

Pooled GEV fit

Comparison of return values

Summary Statistics

CCSM3 vs Reanalysis Extreme Value Biases

Summary Statistics

χ bar: a measure of asymptotic dependence

Asymptotic independence

Asymptotic dependence

Near extremal independence

Summary Statistics

Marginal GEV Return Values (GCM)

Marginal GEV Return Values (Reanalysis)

Cluster analysis

Clustering by return values and the shape parameter

Justification of the Cluster analysis

Results for the GCM Clusters, similar results for the Reanalysis.

Comparison of GCM and Reanalysis Clusters

Pooled GEV Return Levels (Cluster 1)

Pooled GEV Return Levels (Cluster 2)

Conclusion

Differences detected between the GCM and the reanalysis

Evidence that the GCM model is tweaked to accurately model the body rather than the tail of the distribution

L-moments improved the estimates of the GEV parameters

Cluster analysis and spatial pooling improved the parameter estimates but needs to be explored further

Future Work

Comparison with a regional climate model as well as ensembles

Investigate further the components of the CCSM3, which leads to the biases

Pooled modal GEV approach

Bootstrapping of data at each site

Introduction of covariates into the analysis

Adjust the model for temporal dependence at sites

Spatial fit to the data

More exploration into a multivariate framework

Statistical downscaling approach

Now this is EXTREME!!!!!!!!!

Nobody Canna Cross It!!!!!!!!! http://www.youtube.com/watch?v=hknVoAoyy-k

Any questions?

Thank you