Error correction of precipitation extremes: performance and implications for scenarios Matthias Themeßl, Thomas Mendlik, Andreas Gobiet Wegener Center for Climate and Global Change and Inst. for Geophysics, Astrophysics, and Meteorology (IGAM)/ Inst. of Physics, University of Graz, Austria AMS, Jan 26 2012, New Orleans
Contents Motivation and introduction Downscaling and error correction - Quantile mapping Application to daily precipitation amount and derived extremes Take home
Contents Motivation and introduction Downscaling and error correction - Quantile mapping Application to daily precipitation amount and derived extremes Take home
Motivation Provide user-tailored, accurate and useful climate information for climate change impact research and decision making
Introduction to regional climate modelling Overview of concepts
Introduction to regional climate modelling Overview of concepts
Introduction to regional climate modelling State of the art RCM errors REMO57-ERA40 2m temperature (a) and precipitation amount (b) compared to EOBS (v1) 1961-1990 HUN ROM BUL Standard RCM error for temperature: Several K Standard RCM error for precipitation: ±50 % However, the minimum standard for any useful downscaling procedure for hydrological applications is that the historic (observed) conditions must be reproducible (Wood et al., 2004).
Contents Motivation and introduction Downscaling and error correction - Quantile mapping Application to daily precipitation amount and derived extremes Take home
Quantile Mapping 2 4 3 1 Implementation For each station/grid cell seperately Applied for each day seperately; calibrated for doys (using moving window) to encounter for seasonality Based on empirical cumulative distributions thus in theory applicable to any meteorological parameter without major adaptations
Applicability of QM precipitation amount CCLM control run 1961-2000 compared to E-OBS; cross validation First results: Uncorrected Corrected QM removes seasonal and regional varying bias
Applicability of QM precipitation amount CCLM control run 1961-2000 compared to E-OBS; cross validation First results: Pdf Pdf-difference QM adapts modelled probability distribution until 99th percentile (~30 mm/d) nearly perfectly, including drizzling effect
Applicability of QM to derived extremes CCLM control run 1961-2000 compared to E-OBS; cross validation Uncorrected Corrected Pdf Precip intensity Days with 10 mm/d Max. 1-day precip Although absolute extremes (max 1-d precip) are more prone to remaining errors, overall one order of magnitude of error can be removed
Applicability of QM to derived extremes CCLM hindcast (ERA40) 1961-2000 compared to E-OBS; cross validation Range of daily errors is unchanged Correlation of corrected data to observation (also on daily basis) is unchanged, unless e.g. index is treshold related and uncorrected model shows significant distributional shift
Applicability of QM to NEW extremes Facing global warming, it is very liekly that new extremes will occur By definition statistical models are only valid for the range they are calibrated on 2 extrapolations of error correction (EC) functions are tested for this purpose (based on Déqué, 2007)
Applicability of QM to NEW extremes typical EC function EC function shows strong dependency on precip intenisty EC functions show rather stable behaviour of 99% of data, but also severe breakpoints at the extremes
Applicability of QM to NEW extremes typical EC function EC functions breakpoints obviously at 99.7th percentile
Applicability of QM to NEW extremes 5 yearly maxima for each grid cell over entire Europe First results: between 1961 and 2000; split sample RCM strongly overestimates new extremes QMv0 underestimates new extremes as expected (limited to calibration) Both extrapolations are able to produce new extremes, although in this case QMv1a show better performance than QMv1b
Contents Motivation and introduction Downscaling and error correction - Quantile mapping Application to daily precipitation amount and derived extremes Take home
Take home Error correction is inevitable for climate impact assesment Uncorrected data may lead to serious misestimations in climate impact studies QM is recommendable (simple and non-parametric) Errors in mean, variability and extremes are reduced/removed (rule of thumb: reduction of bias of order of magnitude can be expected) QM is regionally transferable as well as applicable to different parameters and models QM does not change the correlation of the time series QM can provide new extremes outside its calibration range with similar skill as RCMs and thus gives confidence in its application also to future scenarios
Thank You Themeßl M, A Gobiet, G Heinrich Empirical-statistical downscaling and error correction of daily precipitation of regional climate models and its impact on the climate change signal. Clim. Change, doi: 10.1007/s10584-011-0224-4, 2011 Themeßl M, M Suklitsch, A Gobiet Coping with daily precipitation errors of regional climate models. In: Precipitation: Prediction, Formation and Environmental Impact (eds.: Henry Dohring and Jeremy Dixon), Nova Science Publishers, ISBN: 978-1-62100-447-9. 2011