The new German project KLIWEX-MED: Changes in weather and climate extremes in the Mediterranean basin Andreas Paxian, University of Würzburg MedCLIVAR.

The new German projectKLIWEX-MED:

Changes in weather and climate extremes in the Mediterranean basin

Andreas Paxian, University of Würzburg

MedCLIVAR Workshop 2008, Trieste

Group members

• University of Würzburg:– Heiko Paeth ([email protected])– Andreas Paxian– Gernot Vogt

• University of Augsburg:– Jucundus Jacobeit ([email protected]

augsburg.de)– Elke Hertig– Stefanie Seubert

Project overview

• Main goal:Detection of climate change and extreme events in the Mediterranean basin and probabilistic quantification of uncertainties

• Broad spectrum of different methods:

• Comparison of different scenarios, methods and models evaluation with observations

-

Global model

simulations (direct)

Regional model

simulations (direct)

Statistical down-scaling

(indirect)

Weather generator (indirect)

Project overview

multi -model ensembles fromglobal climate models (AR4)

ensembles froma regional climate model (REMO)

probabilisticassessmentof climate

change

impact studies / plannings

evaluation (space -time )observed

station data


change

dynamicalextreme

valueestimates

dynamicalextreme

valueestimates

large -scaleweather

types

meso -scaleweather

types

weathergenerator

statisticalextreme

valueestimates

statisticalextreme

valueestimates

physicalstochastic

extreme valueestimates

Observations




change



station data


change

dynamicalextreme

valueestimates

dynamicalextreme

valueestimates

large -scaleweather

types

meso -scaleweather

types

weathergenerator

statisticalextreme

valueestimates

statisticalextreme

valueestimates

physicalstochastic


Observations

• Analysis of observation data for evaluation of present-day results of different methods

• Station time series:– EMULATE [MOBERG et al. 2006]– assistance of MedCLIVAR and CIRCE community

• Aggregated grid box data:– CRU [NEW et al. 2000, MITCHELL & JONES 2005]– VasCLIMO [BECK et al. 2007]

Observations

• Analysis:– Quality control (no inhomogeneous time series)– Mean value statistics, seasonal cycle, trends– Extreme value statistics:

• appropriate extreme valuedistribution (General Pareto) return times and values

• uncertainty of assessmentand significance of variability(Monte Carlo sampling approach)

0

1

RV

RT

RV

f

present-dayclimate forced

climate

Global model simulations




change



station data


change

dynamicalextreme

valueestimates

dynamicalextreme

valueestimates

large -scaleweather

types

meso -scaleweather

types

weathergenerator

statisticalextreme

valueestimates

statisticalextreme

valueestimates

physicalstochastic


• IPCC 2007 AR4 multi-model ensembles (1870-2100 observed GHG and A1b, A2, B1 scenarios)

• Variables: temperature, precipitation, wind• Analysis:

– Mean value statistics, seasonal cycle and trends (analogue to observations)

– Extreme value statistics (analogue to observations)– Probabilistic assessment of climate change:

statistical confidenceintervals due to differentensemble members

1%

10%

90%

99%

x=50%

s+=84%

s-=16%

2000 2050


– Analysis of variance:detection of climate change signals in multi-model ensemble results

-

Temperature

internal model

variability

differences between

ensemble members

external model

variability

trend of ensemble

mean

model uncertainty





change



station data


change

dynamicalextreme

valueestimates

dynamicalextreme

valueestimates

large -scaleweather

types

meso -scaleweather

types

weathergenerator

statisticalextreme

valueestimates

statisticalextreme

valueestimates

physicalstochastic


Regional model simulations

• REMO from MPI Hamburg (JACOB et al., 2001) dynamical downscaling:– high spatial resolution (0.5°) detailed climate

change fingerprints (impact modelling)– orography, land-sea contrast and synoptic processes

improved extreme value statistics

• REMO simulations for IMPETUS project:– 1960-2000 observed GHG– 2001-2050 A1b, B1 and FAO

land degradation scenarios– 1979-2003 simulations

successfully validated


• REMO precipitation post-processing with MOS (correction of systematic model errors):– extraction of orthogonal predictors (stepwise

multiple regression, principal component analysis)– statistical transfer functions between predictors and

target variable (training over past 25 years)– implementation of transfer functions on simulations

of 1960-2050

• Analysis (analogue to global model simulations)


Statistical downscaling




change



station data


change

dynamicalextreme

valueestimates

dynamicalextreme

valueestimates

large -scaleweather

types

meso -scaleweather

types

weathergenerator

statisticalextreme

valueestimates

statisticalextreme

valueestimates

physicalstochastic


• Transfer functions from observations:– classification of Mediterranean large-scale

circulation and meso-scale weather types in ERA40/ NCEP reanalysis data

– extreme values from daily EMULATE temperature and precipitation time series

– statistical transfer functions between circulation types and extreme values (canonical correlation analysis, multiple regressions)

– cross validation to test robustness of scale transfer

large-scalecirculation patterns statistical transfer functions

localextreme values


• Implementation of transfer functions to models:– classification of large-scale circulation and meso-

scale weather types in global and regional models– implementation of transfer functions to models

local station extreme values (past and future) probabilistic assessment of extreme value uncertainty (model ensembles)

• Linear statistical downscaling underestimation of climate system complexity


Weather generator




change



station data


change

dynamicalextreme

valueestimates

dynamicalextreme

valueestimates

large -scaleweather

types

meso -scaleweather

types

weathergenerator

statisticalextreme

valueestimates

statisticalextreme

valueestimates

physicalstochastic


Weather generator

• Creation of virtual precipitation station time series from global and regional model grid box data improved comparison to observations:

virtual stationprecipitation

Physical part: coastal distance, sub-scale synoptic processes, orography (DEM, station meta data)

Dynamical part: REMO or AR4 model grid box data

Stochastic part: stochastic spread within model grid box (station time series)

climate models:area-mean

precipitation

observations:local

station dataproblems of comparison

Weather generator

• Correction algorithm: final adjustment of simulated precipitation sums or daily distribution functions to observation data in training period

• Cross validation: evaluation of results• Probabilistic assessment of uncertainty (model

ensembles)

• Good results of weather generator for analysis of IMPETUS simulations in tropical Africa (input for hydrological impact modeling)

Evaluation and Comparison




change


evaluation (space -time)observed

station data


change

dynamicalextreme

valueestimates

dynamicalextreme

valueestimates

large -scaleweather

types

meso -scaleweather

types

weathergenerator

statisticalextreme

valueestimates

statisticalextreme

valueestimates

physicalstochastic


Summary




change



station data


change

dynamicalextreme

valueestimates

dynamicalextreme

valueestimates

large -scaleweather

types

meso -scaleweather

types

weathergenerator

statisticalextreme

valueestimates

statisticalextreme

valueestimates

physicalstochastic


The new German project KLIWEX-MED: Changes in weather and climate extremes in the Mediterranean basin Andreas Paxian, University of Würzburg MedCLIVAR.

Documents

trieste slide

weather generator slide

project overview slide

observations analysis

ar4 multimodel ensembles

local station extreme

analysis of observation

extreme events