Projection of Changes in Extremes by Very High Resolution Atmospheric Models Akio KITOH Meteorological Research Institute, Tsukuba, Japan IPCC Extremes-SR Scoping Meeting, 23 March 2009, Oslo, Norway

Precipitation changes Precipitation increases very likely in high latitudes Decreases likely in most subtropical land regions Model agreement is not so high in other regions White: <2/3 of models agree on sign of change Stippled: >90% of models agree on sign of change IPCC AR4 CMIP3 models

Projected changes in extremes Intensity of precipitation events is projected to increase. Even in areas where mean precipitation decreases, precipitation intensity is projected to increase but there would be longer periods between rainfall events. “It rains less frequently, but when it does rain, there is more precipitation for a given event.” (Tebaldi et al. 2006) Extremes will have more impact than changes in mean climate IPCC AR4 CMIP3 models

All precip Heavy precip mm/d Light precip 1-10 mm/d Sun et al. (2007) JCLI Frequency Intensity CMIP3 models

20-yr return values of annual extremes of 24- h precipitation rates (P20) Kharin et al. (2007) JCLI 20-yr return values of 24-h precipitation CMIP3 models Waiting times for P20 Global average P20 increases about 6%/K of global warming, with the range of 4%/K-10%/K Large inter-model differences in precipitation extremes

Needs for high resolution models representation of topography depends on resolution (land-sea distribution, mountain height, snow-rain threshold, …) low resolution models often fail to reproduce precipitation systems such as tropical cyclones, stationary front systems and blocking high resolution models have better mean climate

Indian summer monsoon rainfall 20-km modelIMD observation Orographic rainfall is successfully reproduced Rajendran and Kitoh (2008) Current Science

60km mesh model 20km mesh model Typhoon track and intensity: 60km vs 20km 60-km model forecasts shallower central pressures and weaker maximum winds. 20-km model represents typhoon development closer to the observations. NWP from Nov 2007 at JMAJMA In forecast mode

model TL63(270km) model TL95(180km) model TL959(20km) obs (GPCP1DD) 1deg(100km) pavwetdayr5dcdd Comparison of extremes indices between observation and each resolution models pav: Annual mean precipitation (mm/day) wetday: Number of days (> 1 mm/day) r5d: The annual maximum 5-day total precipitation cdd: The annual maximum consecutive dry-days better Typhoons/Baiu in high-res model MRI AGCM more wetday in low- res model more bias in low- res model

There are now some high-resolution modeling results available for climate change projections Regional climate model PRUDENCE, ENSEMBLES, NARCCAP other regions follow multi-model but regional (can apply to other regions) Stretched-grid model CCAM, GEM, ARPEGE, GEOS Super-high resolution models (=NWP model) MRI/JMA 20-km mesh AGCM + 5-km RCRCM global but single-model (can apply physics/b.c. ensemble)

Example by regional climate model

Fischer and Schär (2009) ClimDyn Daily summer temperaturePRUDENCE Over France, at least every second and third summer day exceeds the 95th percentile, and a considerable number of days even the maximum values of the present

Example by global 20-km mesh AGCM tropical cyclone extratropical cyclone blocking

Kakushin Team-Extremes Time-Slice Experiments Mizuta et al. (2008) Present-day climate experiment ( ) –the observed sea surface temperature (SST) and sea- ice concentration Future climate experiment ( ) –the warming in the SST for the CMIP3 multi-model ensemble mean is added to the observed SST 20km,60km AGCM5km,2km,1km RCMCMIP3 AOGCMs SST Ocean Atmosphere Lower B.C. Projected SST SST Atmosphere Boundary condition AGCM/RCM is a climate model version of the JMA operational NWP models

Inter-annual variability of TC frequency Observation 20-km AGCM (AMIP run ) 0.48## # ###0.53### 0.04 ###:99% significance level ##:95% significance level #:90% significance level There is a skill for TC frequency interannual variability associated with ENSO

Number of TC Generated in Each Latitude Present-day(25yr) Future(25yr) Observation Latitude TC freqency ２０％ decrease Annual global average Present =82 Future =66 (20% decrease) (Observation:84)

Intensity Stronger TCs will increase Longer lived TCs will increase Duration when wind speed is over 17m/s Future Experiment Present Experiment Observation Change in TC intensity and duration

Radial Profile Change around TC ・ Large changes occur near inner-core region, 40-60% for precipitation and 15-20% for surface wind. ・ A surface wind speed increase of more than 4% can be seen up to 500 km from storm center. Surface Wind Radial Distance in km from Storm Center Precipitation Future Experiment Present Experiment Change rate

Resolution dependency of TC intensity (wind speed) In the simulation mode, TCs in the 20-km model are still weak. Lower resolution models have more difficulty to interpret results.

Lambert and Fyfe (2006) Extratropical Cyclones Present Total cyclone number“Strong” cyclone number (<970hPa) When tracking extratropical cyclones.. –Number of cyclones decreases –“Strong” cyclones increase 0 ~+10% Present CMIP3 models

Same in high-resolution models but with different threshold for + or - (60km model, 3 ensembles) Frequency of cyclones as a function of threshold pressure (20km model) [Future] / [Present] MRI AGCM Mizuta et al. (2009)

NH wintertime atmospheric blocking Matsueda et al. (2009) submitted to JGR 20km 120km180km 60km MRI AGCM higher resolution models better represents Atlantic blockings

Uncertainty in future projections of blocking frequencies of Euro-Atlantic and Pacific blockings are projected to decrease significantly. MRI AGCM Matsueda et al. (2009) 60km ensemble long duration blockings decrease The larger the warming is, the less blocking frequency Large warming Medium warming Small warming

Further regional downscaling is necessary to obtain quantitative assessment of future weather extremes

Dynamical downscaling by RCM 5-km mesh cloud resolving RCM (summer) Boundary condition from 20-km GCM High resolution climate change information 1-km mesh CRCM (heavy precip events) 2-km mesh CRCM (summer)

Three heaviest precipitation total (Ptop3) Wakazuki et al. (2008 JMSJ) Differences of 1-hour Ptop3 between 20km- AGCM and 5km-NHM are significantly large. Differences become smaller for longer accumulated precipitations. (OBS)

Three heaviest precipitation total (Ptop3) Present Future (end of 21c ） Change ratio (F-P)/P Daily maximum precipitation greater than 110 mm/day increases significantly Present Future (end of 21c) Future/Present

Observed data for extremes sufficient?

What is real precipitation? Uncertainty among observations Daily Monthly Rain-gauges Satellites Rain-gauges + Satellites The areal average precipitation (Asia Land; June-August mean) Error-bar shows amplitude of interannual variability Arakawa et al. (2009)

● Radar-AMeDAS □ GPCP-1DD × TRMM3B42 Can we use satellite-based daily precipitation data to study extreme events? Satellite-based “observation” underestimates heavy precipitation compared to rain-gauge-based observation (Radar-AMeDAS) Satellite-based rainfall estimation is not sufficient to validate extreme precipitation events simulated by high-resolution models Daily Precipitation for June to August, 2000 at Kagoshima (mm/day)

Example: Annual precipitation in 1995 at Kathmandu AP. GHCN and GSOD are 1/10 of real data. ⇒ Errors in units-of-measurement exist in widely used global datasets! GSOD (GTS base real-time data ） GHCN (a global dataset ） Directly obtained from Department of Hydrometorology, Nepal A QC tool with Google Earth in APHRODITE

Summary Resolution of climate models becomes finer; now we can use 60-km or even 20-km mesh global climate models Topography is better represented by high resolution model Large-scale features of model climate improve by increasing horizontal resolution High resolution model is needed to better represent weather extremes and tropical cyclones Resolution vs ensemble is an issue Study to interpret and connect high-resolution and lower-resolution results (e.g. scaling) is needed Further work for observation data itself Collaboration with impact assessment group is important (e.g. to avoid misuse of GCM output)