Climate Projections for Florida: Can we trust the models? Jayantha Obeysekera ( Obey ) Hydrologic & Environmental Systems Modeling

Transcription

1 Climate Projections for Florida: Can we trust the models? Jayantha Obeysekera ( Obey ) Hydrologic & Environmental Systems Modeling Public Water Supply Utilities Climate Impact Working Group (PWSU-CIWG) workshop October 7, 2011

2 Is Model Culling an option?

3 Outline Introduction & Approach Data General Circulation Models Statistically Downscaled Climate Data Dynamically Downscaled Climate Data Extremes Conclusions

4 Relevant Publications wmd_repository_pdf/ccireport_publicationversion_1 4jul11.pdf

5 Potential Impacts on Water Resources Management in South Florida Climate Change Drivers Natural Cycles Inter-annual (e.g. El Nino and La Nina) to Multi-decadal (e.g. AMO*) Solar, Volcanos Human Induced Land use changes Greenhouse gases Quartet of change: Stressors Rising Seas Temperature Rainfall, floods, and droughts Tropical Storms & Hurricanes Water Management Impacts Direct landscape impacts (e.g. storm surge) Water Supply (e.g. droughts, saltwater intrusion) Flood Control (e.g. urban flooding, hurricanes) Natural Systems (e.g. ecosystem impacts, both coastal and interior) *Atlantic Multi-decadal Oscillation of temperature in the Atlantic Ocean

6 Incorporation of Climate Change into Water Resources Planning GCM Downscaling Uncertainty Statistical Dynamic al Scenario Uncertainty B1 A1T B2 A1B A2 A1F1 Ice Sheet Uncer -tainty IPCC (2007) ΔT ( C) Statistical Dynamical SLR(m) ???? BCM2 Global Circulation Model Uncertainty CGHR CGMR CNCM3 CSMK3 ECHOG FGOALS GFCM20 GFCM21 GIAOM INCM3 IPCM4 MIHR MIMR MPEH5 NCCCSM NCPCM Constructed Analogues (CA) Bias Correction and Spatial Downscaling (BCSD) Weather Generators Regional Climate Models (RCMs) Climate Change Implications in Water Resources Planning: Scenario Based Approaches Risk Assessment Framework (Model Culling!) Other

7 Using Climate Change Information General Ciculation Models (GCMs) Observed Climate Data Simulation of Late 20 th Century 21 st Century Climate Projections Is there evidence that climate is changing in Florida? Downscale global information to regional information How well are south Florida s climate and teleconnections represented by climate models? How do climate projections affect water resources management?

8 Observational Datasets

9 COAPS FAWN PRISM USGS USGS map UCF - extremes

10 Satellite Based USGS Dataset Daily Minimum Temperature Daily Maximum Temperature Solar Radiation Daily Minimum Relative Humidity Daily Maximum Relative Humidity Reference ET

11 GCM Resolution in Florida Uncertainties in GCM predictions due to: Poor resolution South Florida not even modeled in some GCMs; greater errors at smaller scales From IPCC AR4-WG1, Ch. 8 - Simulation of tropical precipitation, ENSO, clouds and their response to climate change, etc.

12 GCM Skill for Florida - Temperature

13 GCM Skill for Florida Precipitation

14 Regional Climate Change Projections from Multi-Model Ensembles (Tebaldi et al., 2008) at MODEL Likelihood: Observed: X 0 ~ N[μ, λ -1 0 ] GCM (current): X i ~ N[μ, λ -1 i ] GCM(future): Y i ~ N[ν, (θλ i ) -1 ] Priors: μ, ν ~ U(-,+ ) λ i ~ Γ (a,b), θ i ~ Γ (c,d) A Bayesian approach Reward models with respect to BIAS (w.r.t. current climate) and CONVERGENCE (consensus on future projections) 23 Models, SRES scenarios A2(high), A1B (midrange), B1(low) Posterior distribution of precipitation & temperature for each season & future decades

15 Projected Temperature Change from AOGCMs (for 2050) Posterior Distribution The vertical bars correspond to the percentiles, 5% and 95% of the posterior distributions of temperature change for b1,a1b, and a2 scenarios (red, black and blue)

16 Projected Precipitation Change from AOGCMs (for 2050) - Posterior Distribution The vertical bars correspond to the percentiles, 5% and 95% of the posterior distributions of %precip change for b1,a1b, and a2 scenarios (red, black and blue)

17 Statistically Downscaled Data for US (Ed Maurer s Data, 1/8 degree) BCSD method Fernandina Beach Monticello 5 SE Madison Glen St Mary 1 W Lake City 2 E Perry Federal Point Crescent City Daytona Beach Ocala Deland 1 SSE Inverness 3 SE Brooksville Chin Hill Titusville Saint Leo Tarpon Springs Swg Plant Plant City Bartow Avon Park 2 W Ft Pierce Arcadia Moore Haven Lock 1 Belle Glade Ft Myers Page Field Ft Lauderdale Everglades

18 Ability to Reproduce Seasonality Temperature ( ) Ft Lauderdale Historical Downscaled Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Mean Monthly Temperature(deg C)

19 Ability to Reproduce Seasonality - Precipitation ( ) Ft Lauderdale Historical Downscaled Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Mean Monthly Precipitation (inches)

20 20C3M versus Historical: Monthly, A2 Scenario Titusville Annual Average deg C Year

21 Projections: compared to Ft Lauderdale Everglades 2041:2070 versus 1971: :2070 versus 1971:2000 Change in Mean Annual Temp b1 A1b A2 Change in Mean Annual Temp b1 A1b A %Change in Mean Annual Precip %Change in Mean Annual Precip.

22 Spatial Trends

23 Hotter and Longer Summer? Everglades Temperature (deg C) Month

24 Dynamical Downscaling North American Regional Climate Change Assessment Program Acknowledgement: NARCCAP is funded by the National Science Foundation (NSF), the U.S. Department of Energy (DoE), the National Oceanic and Atmospheric Administration (NOAA), and the U.S. Environmental Protection Agency Office of Research and Development (EPA)."

25 NARCCAP Scenario & Models A2 Emissions Scenario GFDL Time slice 50 km GFDL CGCM3 HADCM3 link to European Prudence CCSM CAM3 Time slice 50km current Provide boundary conditions future MM5 Iowa State/ PNNL RegCM3 UC Santa Cruz ICTP CRCM Quebec, Ouranos HADRM3 Hadley Centre RSM Scripps WRF NCAR/ PNNL

26 Validation & Projection: Tmin NARCCAP-current ( ) PRISM Future ( ) minus Current ( ) NARCCAP (CRCM) data appear to follow the general pattern of the PRISM re-analysis data. However, it seems to underestimate the PRISM by about 2 C. Net increase of temperature by about 2050 appears to be approximately 2 C. Results for Tmax and Tave were similar.

27 Projection: Tave NARCCAP ( ) NARCCAP ( ) Future ( ) minus Current ( )

28 Validation & Projection: Precipitation NARCCAP-current ( ) PRISM Future ( ) minus Current ( ) Although the general spatial pattern of Annual Precipitation amounts produced by the NARCCAP model mimics the PRISM data, there appears to be a significant underestimation (~5 inches). The projected net change is -3 to 2 inches.

29 Point Comparison - Temperature Observations NARCCAAP NARCCAP Grid Points and the Observation Locations

30 Point Comparison - Precipitation

31 Evapotranspiration derived from NARCCAP Data Penman-Monteith Equation (FAO-56) Data used: Tmax, Tmin, Tave, Solar Radiation, Pressure, Wind, Min and Max Relative Humidity

32 Validation of ET estimates USGS pixel points NARCCAP grid

33 Issues with NARCCAP Data (Could Cover?)

34 Relative Humidity Too many values > 1 SFWMD Observation Stations Observations: Is this trend real? Seasonal Pattern

35 Solar Radiation vs. Relative Humidity Joint Variable Spatial Downscaling?

36 Biases in Rs and RET Solar Radiation Reference Crop ET

37 Annual Estimates of ET NARCCAP ( ) NARCCAP ( ) Future ( ) minus Current ( ) Compared to the derived ET data from satellite and station information, NARCCAP (CRCM) model data appears to be about 5 inches lower, particularly in the south.

38 Solar Radiation (W/m^2) Comparison Cloud effect Cloud effect Cloud effect Cloud effect captured somewhat Cloud effect Cloud effect not captured

39 Average Annual Rs (MJ/m 2 /d) Comparison R2 RSM-CLARRES10 Rs ( ) USGS Rs ( ) Rs significantly overestimated. This is due in part to not capturing cloud effect during summer, but it also appears that low daily values are not captured either (variability underestimated). See USGS dataset (from GOES satellite) captures higher Rs over LOK due to decreased cloudiness

40 Average Annual Reference Grass ET (in) Comparison ERA40 RSM-CLARRES10 RET( ) USGS RET ( ) RET is significantly overestimated due to Rs significantly overestimated. This is due in part to not capturing cloud effect during summer, but it also appears that low daily values of Rs are not captured either (variability underestimated). See scatterplots. USGS dataset (from GOES satellite) captures higher RET over LOK due to decreased cloudiness (increases Rs)

41 Major findings by variable Consistent significant overestimation of relative humidity (RH) seen in other models is not an issue here Solar radiation (Rs): Cloud effect not captured at the great majority of stations in Central and Southern Florida with the exception of S65CW Significantly overestimated across the state due to cloud effect not being captured, low values not captured (underestimation of variability). Precipitation (Precip): ERA40 shows significant overestimation in the southwest mostly in summer (ASO). NDJ precipitation in the Panhandle underestimated. Both models show decreased coastal precipitation in South Florida (especially R2) whereas PRISM and SFWMM rainfall grid show the opposite along the Lower East Coast (LEC) of South Florida. Unclear as to which is correct (model or observations) Daily average temperature (Tave) is captured very well in both models. Daily maximum temperature (Tmax): Both models significantly overestimate Tmax across the state especially from Feb-Oc

42 Major findings by variable (Cont.) Daily minimum temperature (Tmin): In general, Tmin is overestimated across the state, especially in northern Florida and in the Panhandle area. In both models LOK and areas very close to the coast show higher Tmin than interior areas (higher Tmin over LOK not seen in PRISM or USGS). Models are probably correct. Daily average dew point temperature (Tdew) is captured very well in both models. Daily maximum relative humidity (RHmax): Both models significantly underestimate RHmax. Wind speed (Wind): Both models show overestimation of wind speed especially in central areas of the Panhandle and along the LEC in South Florida. Reference grass Evapotranspiration (RET) derived using FAO-56 Penman-Monteith equation (virtually equivalent to USACE standardized equation) and meteorological data from model output: Both models significantly overestimate RET mainly due to overestimating Rs since cloud effect is not being captured.

43 GEV parameters of annual maxima Location Scale Shape Duration = 6-hours Duration = 24 hours

44 All locations Location Parameter

45 Model Skill Scale Parameter

46 Model Skill Shape Parameter

47 Model Skill!

48 OLAM University of Miami, RSMAS 48 Ocean Land Atmosphere Model New, full-physics non-hydrostatic Earth System Model based on Regional Atmospheric Modeling System (RAMS) Unstructured grid (triangular or hexagonal mesh) Local mesh refinement instead of nesting: two-way seamless communication between regional/mesoscale and global portions of grid domain through conservative advection and turbulent transport

49 OLAM: Hexagonal grid cells

50 By 2050 (findings to date - may change as science evolves) Variable Global Models Statistically Downscaled Data Dynamically Downscaled Data Average Temperature 1 to 1.5ºC 1 to 2ºC 1.8 to 2.1ºC Precipitation -10% to +10% -5% to +5% -3 to 2 inches Reference Crop Evapotranspiration 3 to 6 inches

51 Spatial Distribution of Delta changes!

52 Conclusions Resolutions of GCMs inadequate to capture hydrometeorology of Florida peninsula Statistically downscaled data mimics the seasonal climatology reasonably well. Precipitation has biases Dynamically downscaled data mimics seasonal patterns of both temperature and precipitation well but there are spatial biases. Computing ET from RCMs appears to be promising but all models appear to overestimate summer solar radiation, which is likely due to inadequacy in simulating clouds. Relative Humidity is also overestimated.