The STARDEX project - background, challenges and successes A project within the EC 5th Framework Programme 1 February 2002 to 31 July Clare Goodess Climatic Research Unit, UEA, Norwich, UK

The STARDEX consortium

STARDEX general objectives To rigorously & systematically inter-compare & evaluate statistical and dynamical downscaling methods for the reconstruction of observed extremes & the construction of scenarios of extremes for selected European regions & Europe as a whole To identify the more robust downscaling techniques & to apply them to provide reliable & plausible future scenarios of temperature & precipitation-based extremes

Consistent approach: e.g., indices of extremes

STARDEX Diagnostic extremes indices software Fortran subroutine: –19 temperature indices –35 precipitation indices –least squares linear regression to fit linear trends & Kendall-Tau significance test Program that uses subroutine to process standard format station data User information document All available from public web site

STARDEX core indices 90th percentile of rainday amounts (mm/day) greatest 5-day total rainfall simple daily intensity (rain per rainday) max no. consecutive dry days % of total rainfall from events > long-term P90 no. events > long-term 90th percentile of raindays Tmax 90th percentile Tmin 10th percentile number of frost days Tmin < 0 degC heat wave duration

trend in frost days Days per year Blue is increasing Malcolm Haylock, UEA

trend in summer rain events > long-term 90th percentile Scale is days/year Blue is increasing Malcolm Haylock, UEA

Local scale trends in extreme heavy precipitation indices + variable French Alps ++++ Switzerland -- Greece ++-- N. Italy Germany --++ England AutumnSummerSpringWinterRegion Andras Bardossy, USTUTT-IWS

Investigation of causes, focusing on potential predictor variables e.g., SLP, 500 hPa GP, RH, SST, NAO/blocking/ cyclone indices, regional circulation indices

Winter R90N relationships with MSLP& NAO, Malcolm Haylock R = 0.64

MSLP Canonical Pattern 1. Variance = 44.4%. R90N Canonical Pattern 1. Variance = 11.3%. Winter R90N relationships with MSLP, Malcolm Haylock

Analysis of GCM/RCM output & their ability to simulate extremes and predictor variables and their relationships

Annual Cycle RCMs: HadAM3H control ( ). ERA15-driven Domain: °E, °N, All Alps Christoph Frei, ETH

SON Wet-day 90% Quantile (mm/day) RCMs: HadAM3H control ( ). Christoph Frei, ETH

Approach Use high-resolution observations to evaluate model at its grid scale How well can a GCM represent regional climate anomalies in response to changes in large-scale forcings? Use interannual variations as a surrogate forcing. Use Reanalysis as a quasi-perfect surrogate GCM. Distinguish between resolved (GCM grid- point) and unresolved (single station) scales. Christoph Frei, ETH

Study Regions England (UEA) P: per gp T: 8-30 per gp German Rhine (USTUTT) P: ~500 per gp T: ~150 per gp Greece (AUTH) P: 5-10 per gp T: 5-10 per gp Emilia-Rom. (ARPA) P: per gp T: 5-10 per gp Europe (FIC) 481 stations in total Alps (ETH) P: ~500 per gp Christoph Frei, ETH

Example: German Rhine Basin DJFJJA GCM scale Station scale Precipitation Indices Christoph Frei, ETH

Inter-comparison of improved downscaling methods with emphasis on extremes

Downscaling methods canonical correlation analysis neural networks conditional resampling regression conditional weather generator potential precipitation circulation index/critical circulation patterns Study regions

Predictor selection methods Correlation Stepwise multiple regression PCA/CCA Compositing Neural networks Genetic algorithm Weather typing Trend analysis

PREDICTAND Time series of 90 th percentile of maximum temperature (Tmax90p); 30 stations from Emilia-Romagna ( ) that were clusterised in 3 regions (Fig.2) PREDICTORS Exp 1: Seasonal mean (DJF) of first 4 PCs of Z500 over the area: 90°W-60°E, 20°N-90°N ) Exp 2: Seasonal mean (DJF) of WA, EA, EB, SCA, over the area: 90°W-60°E, 20°N-90°N Model is constructed on the period / and validated on Clusters for Tmax90p (DJF) Downscaling of Tmax90p Tomozeiu et al., ARPA-SMR

Interannual variability of downscaled, Observed and NCEP Tmax90p (DJF), Tomozeiu et al., ARPA-SMR

Downscaling of 692R90N – 2 exp. PREDICTAND Time series of observed no.of events greater than 90 th percentile of raindays (692R90N);44 stations from Emilia-Romagna ( ) that were clusterised in 5 regions (Fig.1) PREDICTORS Seasonal mean (DJF) of first 4 PCs of Z500 that covers the area: 90°W-60°E, 20°N-90°N (NCEP reanalysis,2.5°x2.5°) Downscaling of 692R90N Fig.1 Clusters for 692R90N (DJF) Model is constructed on the period / and validated on Tomozeiu et al., ARPA-SMR

Skill of the statistical downscaling model expressed as correlation coefficient between the observed and estimated 692R90N (bold-5%significance) Zone 1_forecast2_forecast3_forecast4_forecast5_forecast Emilia Romagna region forecast 1_observed _observed _observed _observed _observed 0.26 Emilia Romagna region obs 0.25 Tomozeiu et al., ARPA-SMR

Probability of precipitation at station conditioned to wet and dry CPs Andras Bardossy, USTUTT-IWS

At the end of the project (July 2005) we will have: Recommendations on the most robust downscaling methods for scenarios of extremes Downscaled scenarios of extremes for the end of the 21st century Summary of changes in extremes and comparison with past changes Assessment of uncertainties associated with the scenarios

Dissemination & communication internal web site (with MICE and PRUDENCE) public web site scientific reports and papers scientific conferences information sheets, e.g., 2002 floods, 2003 heat wave powerpoint presentations external experts within-country contacts