Atlantic Hurricanes and Climate Change

Transcription

1 Atlantic Hurricanes and Climate Change Tom Knutson Geophysical Fluid Dynamics Lab/NOAA Princeton, New Jersey Hurricane Katrina, Aug GFDL model simulation of Atlantic hurricane activity

2 There is some recent evidence that overall Atlantic hurricane activity may have increased since in the 1950s and 60s in association with increasing sea surface temperatures Source: Kerry Emanuel, J. Climate (2007). PDI is proportional to the time integral of the cube of the surface wind speeds accumulated across all storms over their entire life cycles.

3 What are the implications of pronounced future warming for Atlantic Power Dissipation Index (PDI)? Observations Power Dissipation Index vs SST IPCC AR4 Climate Models ~ ~

4 Attribution of hurricane changes to human-induced climate change? Detection: is there an observed change that exceeds internal variability? Attribution: is the observed change consistent with expected anthropogenic influence? And inconsistent with alternative explanations? Models/theory must reconcile with observations Observations must be assessed for false trends based on evolving observational capabilities

5

6 The frequency of recorded storms (low-pass filtered) in the Atlantic basin is well-correlated with tropical Atlantic SSTs But is the storm record reliable enough for this? Source: Emanuel (2006); Mann and Emanuel (2006) EOS. See also Holland and Webster (2007) Phil. Trans. R. Soc. A

7 +1.60 storms/century ( ) storms/century ( ) Trend from : Not significant (p=0.05, 2-sided tests, computed p-val ~0.2) Trend from : Is significant at p=0.05 level Source: Vecchi and Knutson, J. Climate, in press.

8 TK storms/century ( ) storms/century ( ) Trend from : Not significant (p=0.05, 2-sided tests, computed p-val ~0.2) Trend from : Is significant at p=0.05 level Source: Vecchi and Knutson, J. Climate, in press.

9 TK 4 Global Mean Temperature U.S. Landfalling vs basin-wide activity Tropical Atlantic Sea Surface Temp. Atlantic Tropical Storm Counts (unadj.) Adj. Atlantic Trop. Storm Counts (Vecchi/Knutson) U.S. Landfalling Tropical Storms (unadj.) U.S. Landfalling Hurricanes (unadj.) Storm Track Density Linear Trend ( ) Unadjusted Adjusted Note: All time series are low-pass filtered (5-yr mean) and normalized to unit standard deviation (y-axis tic marks: 1 st. dev). Source: Vecchi and Knutson, J. Climate, accepted for publication.

10 GFDL Zetac Model: A new high-resolution regional model for Atlantic hurricane season simulations The model runs for entire hurricane seasons. The model generates its own sample of hurricanes during each season. These experiments push the limits of available computing resources.

11 Prototype model for future global hurricane forecasting system

12 The model captures both the increase in hurricane activity since the 1980s and the year-by-year fluctuations. Note: Model uses large-scale interior nudging to NCEP Reanalysis. Aug.-Oct. season only. Source: Updated from Knutson et al (Bull. Amer. Meteor. Soc.)

13 Other hurricane metrics (ACE, PDI) are simulated fairly well, but major hurricanes are under-simulated in number (though still well-correlated) ACE: Correlations: Model1: 0.72 Model2: 0.68 Ensem: 0.77 PDI: Correlations: Model1: 0.70 Model2: 0.62 Ensem: 0.73 Cat 3-5: Correlations: Model1: 0.64 Model2: 0.51 Ensem: 0.69

14 Zetac Regional Model Downscaling: geographical distribution of storms Tropical Storm Formation Tropical Storm Occurrence Hurricane Occurrence a) Observed b) Observed c) Observed d) Simulated e) Simulated f) Simulated Note: Model uses large-scale interior nudging to NCEP Reanalysis

15 Large-scale tropical Atlantic climate changes projected for late 21 st century by CMIP3 models (A1B scenario). Average SST change in MDR is 1.72 o C with warming near 4 o C in the upper troposphere.

16 Late 21st Century projections: increased vertical wind shear may lead to fewer Atlantic hurricanes Average of 18 models, Jun-Nov storm-friendly storm-hostile Source: Vecchi and Soden, Geophys. Res. Lett., (2007)

17 -27% Projected changes in Atlantic hurricane/tropical storm numbers: Late 21 st century; Zetac regional model downscaling of CMIP3 multi-model ensemble climate change signal. -18% Note: U.S. Landfalling hurricanes: -30% -8% Source: Knutson et al., 2008, submitted.

18 The 26.5 o C threshhold temperature for tropical storm formation: a climate dependent threshhold

19 The model provides projections of Atlantic hurricane and tropical storm frequency changes for late 21 st century, downscaled fromm a multi-model ensemble climate change (IPCC A1B scenario): 1) Decreased frequency of tropical storms (-27%) and hurricanes (-18%). Storm Intensities (Normalized by frequency) 2) Increased frequency and intensity of the strongest hurricanes (5 12) 4) A more consistent intensity increase is apparent after adjusting for decreased frequency 3) Caveat: this model does not simulate hurricanes as strong as those observed. Source: Knutson et al., 2008, submitted.

20 The new model simulates increased hurricane rainfall rates in the warmer climate (late 21 st century, A1B scenario) consistent with previous studies Present Climate Warm Climate Rainfall Rates (mm/day) Warm Climate Present Climate Avg. Rainfall Rate Increases: 50 km radius: +37% 100 km radius: +23% 150 km radius: +17% 400 km radius: +10% Average Warming: 1.72 o C

21 The control model reproduces the observed close relationship between SST and hurricane frequency ( ), but this statistical relationship does not hold for future human-caused warming in the model. Hurricane frequency actually decreases by 18% in the warm climate case although the model does not simulate hurricanes as intense as observed. Lesson: Caution using correlations from the present climate to make future climate projections Source: Knutson et al., submitted (2008).

22 Next Steps: Examining Model Dependence: The regional details of future projections matter (and these are not highly confident features of future climate projections). Major hurricane count sensitivity varies (-8% to +~70% so far in very preliminary testing). Changes in Atlantic hurricane activity as downscaled from GFDL CM2.1 climate model: Hurricane Frequency: +3% Major Hurricane Frequency: +70% +70% Late 21 st century Current Climate

23 Courtesy Morris Bender and Tim Marchok, NOAA/GFDL

24 Hurricane models project increasing hurricane intensities and rainfall rates with climate warming but probably not detectable at present. Hurricane Intensity Hurricane Rainfall Rates Current climate ~Late 21 st century Current climate ~Late 21 st century 6-hr accumulated rainfall [cm] within ~100 km of storm center. Sensitivity: ~4% increase in wind speed per o C SST increase Sensitivity: ~12% increase in near-storm rainfall per o C SST increase Sources: Knutson and Tuleya, J. Climate, 2004 (left); Knutson and Tuleya (2008) Cambridge Univ Press (right). See also Bengtsson et al. (Tellus 2007) and Oouchi et (J. Meteor. Soc. Japan, 2006); Walsh et al. (2004) Stowasser et al. (2007).

25 Alternative Downscaling Approach: Emanuel et al. (2008)

26 Alternative Downscaling Approach: Emanuel et al. (2008) Emanuel Obs. Last 25 yr: +250% (Atl.) +47% (global) K08: -27% (? To -8%) Atl. Trop. Storms -18% (? To +7%) Atl. Hurricanes -8% (? To +70%) Major Hurricanes (Tropical Storms) K08: -25% (Atl., weak hurricanes) KT04: +6% K08: 2-3% (Atl.) with some selected rough comparisons to other modeling studies

27 How sensitive is hurricane intensity to greenhouse gas-induced warming? ~8 m/s/ o C Estimated Sensitivity of TC intensity to large-scale SST Observations Projected 21 st Century Changes in Thermodynamic Potential Intensity of Hurricanes Average of 18 models, Jun-Nov Model/Theory From Vecchi and Soden (2007, GRL) Source for TC intensity/sst plots: Jim Kossin, personal communication, 2007.

28 Climate Change and Tropical Cyclones: Impediments to Scientific Understanding 1. Inhomogeneities in climate records, insufficient length of climate records related to tropical cyclones 2. Uncertainties in global climate projections (emissions, transient sensitivity) 3. Uncertainties in regional projections of climate models (wind shear, lapse rate, SST patterns, easterly waves) 4. Uncertainties in response of tropical cyclones to regional changes in climate (frequency, intensity) 5. For decadal predictions: Radiative forced response (Greenhouse gases, aerosols) vs internal climate variability (AMO)

29 Concluding Remarks: Where Do We Stand At This Point? 1. Based on the current state of models and ongoing data concerns, it is not appropriate at this time to make a likelihood statement attributing past changes in hurricane activity to increasing greenhouse gases or other human-caused factors. 2. Some statistical studies, particularly using data from recent decades, suggest the possibility of a human influence on hurricanes. Atlantic SST and hurricane power dissipation are wellcorrelated on low frequencies since ~1950. Studies attribute part of the Atlantic SST warming since 1950 to increasing greenhouse gases. 3. However, longer (~ yr) tropical cyclone records, including consideration of data problems, give conflicting indications on whether there have been significant increases in Atlantic tropical storm and hurricane numbers. U.S. landfalling hurricanes show no increase. Basin-wide statistics are much more uncertain for these longer records than for the more recent decades. 4. Models: The recent increase in Atlantic hurricane numbers ( ) is reproduced by models, and is strongly correlated to SSTs in both models and observations. However, the much larger projected greenhouse warming (late 21 st century, A1B scenario) leads to decreased hurricane numbers in our model. We infer that the detailed structure of a tropical SST changes (whether it be due to internal variability, greenhouse gases, aerosols, etc.) has a major influence on how those SST changes will influence hurricanes. 5. Model dependence: While the average model projection calls for a slight decrease (-8%) in major hurricane counts by the late 21 st century, at least one model (GFDL CM2.1) projects a large increase (+70%) in major hurricane counts. This indicates that further studies are needed to narrow the uncertainty in important details of future regional SSTs which may impact Atlantic hurricane activity. 6. Intensity: models simulate increased average hurricane wind speeds and rainfall rates in response to a general greenhouse gas-induced warming of tropical SSTs. Rough range of sensitivity: 1 to 8%/ deg C for wind speed; ~12%/deg C for near-storm rainfall. Significant increases of intensity have not yet been detected in observations (but stay tuned ). Attribution?

30

31 GFDL Regional Model Simulation of the 2005 Atlantic Hurricane Season Regional model low-level winds (388 m); Red: >40 m/s (~ hurricane strength) Reference: T. Knutson et al., Bulletin of the American Meteorological Society, October 2007.

32 Developments in high-resolution hurricane modeling studies for greenhouse warming: Broccoli and Manabe (1990): high resolution was ~200 km grid model. Decreased tropical storm frequency, but physics dependence of result. Bengtsson et al. (1996): ~100 km grid model. Decreased tropical storm frequency. Still little credible intensity information. Knutson et al. ( ): GFDL hurricane forecast model (18 km, later 9 km grid). Case study approach: stronger hurricanes with higher rainfall rates, even with ocean coupling. No frequency information, and wind shear effects remain a concern. Oouchi et al. (2006): Japan s Earth Simulator allows 20 km grid global simulation. Frequency decreases globally but increases in Atlantic. Limited sampling of decadal variability (20 yr SST averages) raises questions about robustness of Atlantic result. Bengtsson et al. (2007): Range of global model resolutions from ~150 km to ~30km. Intensity increase seen at higher resolution only. A 20 th century simulation (a unique experiment!) shows no significant tropical cyclone changes. Knutson et al. (2008): 18-km nested Atlantic model with interior nudging reproduces both the recent rise in Atlantic hurricane activity ( ) and interannual (El Nino) variations. Cat 4-5 intensities are not simulated. New climate change experiment (based on multi-member ensemble of climate models) simulates decreased Atlantic hurricane numbers in the 21st century.

33 What type of SST anthropogenic signal should we look for in the tropical Atlantic? Anthropogenic forcing: Linear trend-like, but the caveat of no indirect aerosol forcing yet. Natural + Anthropogenic forcing: Volcanic/solar/anthropogenic forcing gives a closer fit to observations Residual (blue black) = AMO?? Sources: Vecchi and Knutson, J. Climate, accepted for publication. See also Knutson et al. J. Climate, 2006

34 Internal Climate Variability vs Radiative Forcing: Detrended NH Mean Temperatures Model: Radiative forcing and ocean circulation changes both provide a plausible explanation for the fluctuations in 20 th century Northern Hemisphere mean temperature (detrended) Model: Some tentative evidence in support of a significant role for ocean circulation changes: surface and sub-surface fluctuations are out of phase in observations and modeled AMO-like change. Sources: Zhang, Delworth and Held (2006) Zhang (2007) GRL papers.

35 A measure of annual U.S. landfalling hurricane activity shows no clear long-term trend since 1900 Source: Chris Landsea, NOAA/NHC

36 Summary Statement on Tropical Cyclones and Climate Change WMO International Workshop on Tropical Cyclones VI Highlights (excerpts) of Consensus Statements as of December 2006: No firm conclusion yet on whether there is a detectable anthropogenic signal in hurricane activity. Detection of trends is made more difficult by changes in hurricane observation methods over time and by strong multi-decadal variability in hurricane activity. Some increase in hurricane peak wind speed (and hurricane rainfall) is likely if the climate continues to warm. Projected magnitude: 3-5% increase in wind speed per degree Celsius sea surface temperature increase. Vulnerability to hurricane storm-surge flooding will increase if the projected rise in sea level due to global warming occurs. Source: WMO (World Meteorological Organization)

37 Influence of pronounced greenhouse warming on distribution of hurricane tracks:

38 Source: Knutson et al (in review) r=0.77 r=0.69 r=0.73 r=0.73 Model performance: various Atlantic hurricane measures r=0.57 r=0.41 r=0.87 r=0.16 Note: Model uses large-scale interior nudging to NCEP Reanalysis

39

40 Summary and Conclusions 1. Atlantic SST and hurricane power dissipation are well-correlated on low frequencies since ~1950. Since studies attribute part of the Atlantic SST warming since 1950 to increasing greenhouse gases, this suggests the possibility of a human influence on hurricanes. 2. Longer (~ yr) tropical cyclone records, including consideration of data problems, give conflicting indications on whether there have been significant increases in Atlantic tropical storm and hurricane numbers. U.S. landfalling hurricanes show no increase. 3. In a new Atlantic nested regional hurricane model SSTs are strongly correlated with hurricane numbers ( ), as in observations. However, a much larger simulated GHG-induced SST warming (late 21 st century, A1B scenario) leads to decreased frequency of hurricanes in our model. This strongly suggests that the detailed structure of a climate warming (whether it be due to natural or anthropogenic causes) has a major influence on how a given warming will influence hurricanes. 4. Models simulate increased average hurricane wind speeds and rainfall rates in response to greenhouse gas-induced warming. 5. Model dependence of results is being examined using individual climate models in addition to the multi-model ensemble mean. Simulation results so far for major hurricanes: -8% (using multimodel ensemble climate change) vs. ~+70% (using GFDL CM2.1 model climate change). Further study is ongoing using other climate models and downscaling models. 6. Based on the current state of models and ongoing data concerns, it is not appropriate at this time to make a likelihood statement attributing past changes in hurricane activity to increasing greenhouse gases or other human-caused factors.

41 MODEL DEPENDENCE: Changes in Atlantic hurricane activity as downscaled from GFDL CM2.1 model -20% Late 21 st century +70% Current Climate

42 Model Dependence: Important Results To Date: The regional details of future projections matter (and these are not highly confident features of future climate projections). Major hurricane count sensitivity varies (-8% to +~70% so far in very preliminary testing).

43 Zetac Regional Model Downscaling: Mean Precipitation and Tropical Storm Genesis Observed (Aug.-Oct.) mm/day Simulated (Aug.-Oct.) mm/day Note: Model uses large-scale interior nudging to NCEP Reanalysis

44 Unadjusted tropical storm counts show a significant increasing trend But is the storm record reliable enough for this? Source: Vecchi and Knutson, J. Climate, accepted for publication.

45 Reconstructing past tropical cyclone counts Satellite-era ( ) storm tracks assumed perfect. Apply satellite-era storm tracks to documented ship tracks (ICOADS). Storm detected if 2 ship observations within radius of tropical storm force winds (17 m/s). First detection must occur equatorward of 40N. Monte Carlo simulation, varying storm radii within reasonable bounds. All land assumed to be perfect detector of tropical storms (equatorward of 40N) planned to further test Assume all relevant ship tracks are in data base plan to further test with additional tracks. (First will look for evidence of storms in new ship data.) Source: Vecchi and Knutson (2007) J. Climate, accepted.

46 Source: G. Vecchi, personal communtication Adjusted Annual Count of Hurricanes (Cat 1-5)

47 SST is still correlated with our adjusted series, though not quite as closely Trend from : Not significant (p=0.05, 2-sided tests, computed p-val ~0.2) Trend from : Is significant at p=0.05 level Source: Vecchi and Knutson, J. Climate, in press.

48 Simulated Hurricane Structure 18 km grid regional model Sample model precipitation snapshot mm/day Observed (Frank, 1977) Simulated hurricane composite Warm-core structure: Deg C

49 Showing the individual ensemble members (n=2). North Atlantic Basin (August-October) Hurricane Frequency Correlations vs. Obs: Model1: 0.76 Model2: 0.76 Source: Knutson et al (BAMS) Note: Model uses large-scale interior nudging to

50 But the model does not correlate as well with observed US landfall statistics as with basin-wide statistics...especially US hurricane landfalls Correlations: Model1: 0.36 Model2: 0.51 Ensem: 0.57 Correlations: Model1: 0.30 Model2: 0.32 Ensem: 0.41

51 The model also reproduces the observed reduction of Atlantic hurricane activity during El Niño events fairly well Source: Knutson et al. (2007) Bull. Amer. Meteor. Soc.

52 Other hurricane metrics (ACE, PDI) are simulated fairly well, but major hurricanes are under-simulated in number (though still well-correlated) ACE: Correlations: Model1: 0.72 Model2: 0.68 Ensem: 0.77 PDI: Correlations: Model1: 0.70 Model2: 0.62 Ensem: 0.73 Cat 3-5: Correlations: Model1: 0.64 Model2: 0.51 Ensem: 0.69

53 The Zetac regional model does only a fair job at capturing the observed dependence of maximum Atlantic TC intensity on SST (the simulated dependence is too weak)

54 Large-scale tropical Atlantic climate changes projected for late 21 st century by CMIP3 models (A1B scenario). Average SST change in MDR is 1.72 o C with warming near 4 o C in the upper troposphere.

55 and the model s poor wind-pressure relation at lower central pressures leads to maximum nearsurface winds of ~47 m/s, considerably lower than observed

56 Note: Excessive simulated core season activity during last two seasons (2006, 2007). Why?

57 a) Histogram of minimum central pressure b) Normalized histogram of minimum central pressure

58 The control model reproduces the observed close relationship between SST and PDI ( ), but this strong relationship does not hold for future human-caused warming in the model. PDI actually decreases by 24% in the warm climate case although the model doesn t simulate the hurricanes as intense as observed Lesson: Caution using correlations from the present climate to make future climate projections

59 Influence of pronounced greenhouse warming on distribution of hurricane occurrence: Some biases in hurricane occurrence may distort climate change projection A hint of an eastward shift in hurricane occurrence

60 CMIP3 Multi-model ensemble down-scaling results: Warm climate vs. Control U.S. landfalling hurricanes decrease by 30%, compared with an 18% decrease basin-wide.

61 MODEL DEPENDENCE: Late 21 st century Atlantic hurricane simulations: (CMIP3 Multi-model ensemble vs. GFDL CM2.1) Decreased tropical storm frequency in warm climate using either model Key: Black: Control Red: Multi-model Blue: GFDL CM2.1 Increased occurrence rate for strongest hurricanes is more pronounced using GFDL CM2.1 than for the multi-model ensemble e.g. +70% for major hurricane counts Caveat: Regional model under-predicts major hurricane counts by 50% in the present climate simulation, but correlates at 0.7.

62 but some storms may have been missed and not recorded in the database. Pre-satellite era- 77% strike land Source: Chris Landsea, NHC/NOAA. See EOS (2007) for final versions. Source: Chris Landsea, NOAA/NHC

63 Landsea: No significant trend from AMO warm phase to warm phase, or cold phase to cold phase. Holland: Assumption of constant landfalling fraction not justified. Multi-decadal variability of fraction?? Source: Adapted from Landsea (2007) EOS. Note: includes 5 of 13 additional early 20 th century storms recently identified by Landsea et al.

64 TK 8 Application of Vecchi/Knutson approach to hurricanes Sources: Richard Smith, UNC, personal communication (left); Gabe Vecchi, GFDL, personal communication (right)

65 Statistical significance testing Method 1: Linear least-squares regression on annual storm count series. Adjust degrees of freedom for two-sided t-test based on lag- 1 autocorrelation. Method 2: Same as Method 1, but for the ranks rather than the original series. Addresses issue of skewness in storm count annual data. Method 3: Bootstrap resampling (with replacement) of series sugsegments of length L. Compute linear trends of resampled data sets as a control comparison. L values in range of 2-8 tested. (Recommended value of 2-3 based on Wilks text.) The three methods give roughly similar results here, although Method 3 appears a bit more conservative (at least for some L values in range of 2-8).

66 One possible reason for the difference between US landfalling and basin-wide trends: storm occurrence may have decreased preferentially in the western part of the basin Source: Vecchi and Knutson, J. Climate, accepted for publication.

67 A comparison of several climate change metrics: Global Mean Temperature Tropical Atlantic Sea Surface Temperature Atlantic Tropical Storm Counts (unadj.) Atlantic Trop. Storm Counts (Vecchi/Knut. Adj.) U.S. Landfalling Tropical Storms (unadj.) U.S. Landfalling Hurricanes (unadj.) Note: All time series are low-pass filtered (5-yr mean) and normalized to unit standard deviation (y-axis tic marks: 1 st. dev). Source: Vecchi and Knutson, J. Climate, in press.

68 Source: Vecchi and Knutson, J. Climate, accepted for publication.

69 The model provides projections of Atlantic hurricane tropical storm frequency changes for late 21 st century A1B warming scenario: 1) Decreased frequency of tropical storms (-27%) and hurricanes (-18%). 2) Increased frequency and intensity of the strongest hurricanes (5 12) or +140% 3) Caveat: this model does not simulate hurricanes as strong as those observed.

70 Application of the Vecchi/Knutson approach to Atlantic Hurricane Counts Source: G. Vecchi, personal communtication

71 Research needs for this problem: Improved observations: particularly reanalysis efforts to produce climate quality data on hurricanes for detection/attribution studies (e.g., PDI error? Adjusted storm counts?) Improved hurricane modeling: Higher resolution models, retrospective studies of 20 th century hurricane variability that can simulate or infer past hurricane activity Observational and modeling studies of Atlantic multidecadal variability (e.g., the Atlantic Multidecadal Oscillation AMO, and forced climate change (past and future) affecting SST, shear, etc. in the basin Paleoclimate studies, both for the AMO and for pre-historic hurricane activity. Convergence of statistical models (correlations) and dynamical models (hurricane simulation) More confident projections of future Atlantic large-scale environment (e.g., reexamination of Vecchi/Soden with future models, improved aerosols, etc.)

72 Sea surface temperatures have increased in the region where Atlantic hurricanes form and intensify, and they are projected to increase much more during the 21 st century

73 Future Atlantic Hurricane Activity? Models indicate increased hurricane intensities with warmer ocean temperatures. Increased vertical wind shear may reduce hurricane activity in the Atlantic. How do we assess which of these effects will win out?

74 Projected 21 st Century Changes in Thermodynamic Potential Intensity of Hurricanes Average of 18 models, Jun-Nov From Vecchi and Soden (2007, GRL) storm-hostile storm-friendly

75 (Alternative) Conclusions Observed data, including consideration of data problems, give conflicting indications on whether there have been significant increases in Atlantic tropical storm and hurricane numbers. High resolution models consistently project increasing hurricane intensities and rainfall rates for the late 21 st century, but whether there will be more or fewer hurricanes remains uncertain. A new modeling approach reproduces many important aspects of Atlantic hurricane activity observed since 1980, and thus shows promise as a tool for both understanding past variations and for making more reliable projections of future hurricane activity.

76 Other slides (not used in presentation for now)

77 Correlations vs. Obs: Model Ensemble: 0.73

78 Atlantic Major Hurricane counts (basin-wide) since the mid-1940s: no long-term trend Source: Chris Landsea, NOAA/NHC

79

80

81