ICRC-CORDEX Sessions A: Benefits of Downscaling Session A1: Added value of downscaling Stockholm, Sweden, 18 May 2016

Transcription

1 ICRC-CORDEX Sessions A: Benefits of Downscaling Session A1: Added value of downscaling Stockholm, Sweden, 18 May 2016 Challenges in the quest for added value of climate dynamical downscaling: Evidence of added value in North American RCM simulations with increasing horizontal resolutions René Laprise and Philippe Lucas-Picher Centre ESCER (pour l étude et la simulation du climat à l échelle régionale) Département des sciences de la Terre et de l atmosphère Université du Québec à Montréal Montréal, Québec, Canada Outline Review of the challenges in identifying RCMs added value CRCM5 simulations at 12 km, 25 km and 50 km Examples of added value in simulating North American weather processes 1. Orographic precipitation 2. Snow over high topography 3. Snow belt around Great Lakes 4. North American Monsoon 5. Summer precipitation over Florida 6. Wind channelling in the St. Lawrence River Valley

2 Dynamical downscaling with nested models Drive high-resolution Regional Climate Model (RCM) with Lateral Boundary Conditions (LBC) from low-resolution Global Climate Model (GCM) or Reanalyses Grid meshes of GCM participating in centennial climate-change projections in AR5 ranged from 139 km to 476 km (average 321 km) Grid meshes of RCM: Initially (& default CORDEX) km Recent years 15 km Developmental convection-permitting 2-4 km What is expected to be gained by high resolution: Reduced numerical truncation, more accurate discretization of equations More realistic representation of surface forcings, e.g. orography, lakes, rivers, land-sea contrasts, etc. Some mesoscale weather processes explicitly resolved, e.g. sea breezes, lake-effect snowstorms, local winds, mesoscale convective systems, etc. 2

3 Dynamical downscaling with nested models Expectation of Added Value of RCMs: RCM simulations improve upon provided driving atmospheric boundary conditions But is there Added Value gained? 3

4 Empirical illustration of Added Value Once upon a time, in December, in the virtual reality of a GCM T63, 2.8 o, 317 km) and its downscaling by an RCM 0.44 o, 49 km) Maps shown at 12-h intervals

5 700 hpa Relative humidity (shading > 50%) 850 hpa Temperature (red) 1000 hpa Geopotential (blue) MPI-ESM-LR T o 317 km CRCM o 49 km 1/5

6 700 hpa Relative humidity (shading > 50%) 850 hpa Temperature (red) 1000 hpa Geopotential (blue) MPI-ESM-LR T o 317 km CRCM o 49 km 2/5

7 700 hpa Relative humidity (shading > 50%) 850 hpa Temperature (red) 1000 hpa Geopotential (blue) MPI-ESM-LR T o 317 km CRCM o 49 km 3/5

8 700 hpa Relative humidity (shading > 50%) 850 hpa Temperature (red) 1000 hpa Geopotential (blue) MPI-ESM-LR T o 317 km CRCM o 49 km 4/5

9 700 hpa Relative humidity (shading > 50%) 850 hpa Temperature (red) 1000 hpa Geopotential (blue) MPI-ESM-LR T o 317 km CRCM o 49 km 5/5

10 Empirical illustration of Added Value Maps of RCM 49 km: Look more like analysis weather maps than GCM 319 km Sharper gradients, front-like features, more intense precipitation, with increased resolution Obvious added value with increased resolution Next: Average fields over one December

11 700 hpa Relative humidity (shading > 50%) 850 hpa Temperature (red) 1000 hpa Geopotential (blue) Average fields over one December MPI-ESM-LR T o 317 km CRCM o 49 km

12 Empirical evidence of Added Value For monthly averaged or climatological fields: RCM 49 km is essentially indistinguishable from GCM 319 km No outstanding Added Value, unless one looked very closely

13 Search for RCM added value The added value in RCM climate simulations has been Hard to verify with observations Few high-density climatological networks Gridded climatology of variable quality Difficult to identify in climatological fields Time-averaging removes most of fine scales (except for stationary features) Scale decomposition tools have been used to isolate (smaller amplitude) fine scales from (larger amplitude) large scales (e.g. Feser, von Storch, Di Luca, ) Lack of clear examples where RCMs are better (except possibly when looking at intensity-frequency distributions rather than mean and variance) Until recently (including basic CORDEX), RCMs used intermediate resolution (50 km) What about added value at higher resolution? 13

14 Search for RCM added value Simulations of the Canadian Regional Climate Model version 5 (CRCM5) driven by ERA-Interim for over North American CORDEX domain Compare 3 resolutions: Simulations with grid meshes of 50 km, 25 km and 12 km Mesh size # Cells (free / total) 172x160 / 212x x320 / 380x x640 / 695x680 Time step (min.) # CPUs Computation time ~12 hr per yr ~24 hr per yr ~48 hr per yr Cost Ratio

15 Added Value in specific weather processes Canadian Regional Climate Model version 5 (CRCM5) driven by ERA-Interim for over North American CORDEX domain with grid meshes of 50, 25 and 12 km 15

16 1. Orographic precipitation Prevailing winds Pacific Ocean wet Up-lift on the windward side of mountain results in adiabatic cooling, and ultimately condensation and precipitation dry Zoom over the Rocky Mountains 50km 25km 12km At higher resolution Higher mountain peaks Stronger orographic effect More precipitation on the windward side and drier on the leeward side Measuring precipitation over rugged terrain is challenging 80km 10km 55km DJF precipitation (mm/d) Observed gridded datasets likely underestimate precipitation (stations in the valley, snowfall undercatch) 16

17 2: Snow at high elevations Higher mountains at higher resolutions 50km 25km 12km 80km With higher mountains: Colder conditions More snow can fall and remain Associated with the orographic precipitation Great challenge to measure snow Snow undercatch February Snow water equivalent (mm) 17

18 3. Snowbelts around the Great Lakes Source: Wikipedia 50km 25km 12km Zoom over the Great Lakes 80km February Snow water equivalent (mm) Lake-effect snow: Cold air masses move across open-water lakes, warming and humidifying the atmosphere, resulting in high snowfall on the downstream shores Snowbelts depend on: Ice extent in the lakes, lake temperature, land-sea mask, topography, convection scheme, surface scheme, At higher resolutions: Better snowbelts More snow in the Adirondacks and less in the Champlain River Valley at higher resolutions 18

19 4. North American Monsoon (NAM) Mean annual cycle of precipitation at Alpine City UT AZ NM CO City of Alpine, AZ NAM feature: Increased rainfall from a dry June to a rainy July over large areas of southwestern USA and northwestern Mexico Important forcings for NAM: Land-sea mask and SST CRCM CRCM CRCM ERAI 0.75 CONUS 0.06 OBS (station) Topography Mesoscale circulations More intense and larger precipitation amount at 0.11 Closer to obs. 19

20 5. Summer precipitation in Florida < > Gulf of Mexico Florida 200km CRCM CRCM CRCM ERAI 0.75 CONUS 0.06 Stage IV 4km > < Atlantic Ocean Summer late afternoon thunderstorms due to sea-breeze convergence over central Florida Better breezes and precipitation at higher resolutions Better precipitation over islands such as Cuba and Jamaica at higher resolutions Probably precipitation overshoots at JJA precipitation (mm/day) 20

21 6: Wind channelling along the St. Lawrence River Valley Topography (m) CRCM km CRCM km Wind channelling along the North shore valley (Laurentians) St. Lawrence River South shore (Appalachians) CRCM km ERAI 50 km 25 km Wind speed distribution Île d Orléans 12 km 80 km CRCM CRCM CRCM OBS (station) Wind roses of 3-hourly wind from Ile d Orléans station for DJF The St. Lawrence River Valley appears at 0.22 and 0.11 The winds tend to follow the valley Funnel effect (wind accelerates) 21

22 Conclusions Do RCMs really add value? Is it worth doing higher resolution simulations? Yes, but more obvious at 25 km and 12 km, less the case at 50 km Better simulation of fine-scale, mesoscale phenomena More realistic simulations of local climate processes Best suited to force impact or coupled models Coastal erosion (atmospheric and oceanic circulation) Hydrological model (precipitation intensity and distribution) Glacier model (precipitation distribution and temperature) Modest improvements from 0.22 to 0.11 Probably necessary to retune CRCM5 subgrid-scale parameterization In this study, RCM simulations used BCs from reanalysis Could be more challenging when BCs come from a GCM N.B.: Atmospheric lateral BC & Ocean surface BC 22

23 Paper in review Lucas-Picher P., Laprise R., Winger K. : Evidence of added value in North American regional climate model simulations using ever-increasing horizontal resolutions. Clim. Dyn. Acknowledgements

24

25

26 m Temperature bias Vs. CRU ( C) 50km 25km 12km The bias is almost the same at all resolutions 80km 10km 55km Small sensitivity to the time step 50km 25km 12km Little added value is expected at large scales of multi-year means 80km 10km 55km 26

27 Precipitation relative bias Vs. CRU 50km 25km 12km The bias is almost the same at all resolutions 80km 10km 55km Small sensitivity to the time step 50km 25km 12km Little added value is expected at large scales of multi-year means 80km 10km 55km Degradation in southeastern USA Improvement in southwestern USA 27

28