1. Introduction
Several recent studies using atmospheric general circulation models (AGCMs) with prescribed observed sea surface temperatures (SSTs) have demonstrated the ability to simulate many observed characteristics of tropical cyclones (Bengtsson et al. 2007a,b; Oouchi et al. 2006; LaRow et al. 2008; Zhao et al. 2009; Vitart 2006; Gualdi et al. 2008). Arguably, the tropical cyclone (TC) characteristic best simulated by these AGCMs is frequency. For example, using a 50-km AGCM with prescribed observed 1982–2009 SSTs, Zhao et al. (2009) obtain a correlation of 0.78 between the simulated hurricane frequency in the North Atlantic and the observations. In the AGCMs, the greatest skill in the frequency is found in the North Atlantic with lower skills noted over the ocean basins moving westward from the North Atlantic (east Pacific, west Pacific, Southern Hemisphere Pacific, and the Indian Ocean). Coupled atmosphere–ocean models generally show lower skill in simulating TC frequency than the AGCMs (Vitart and Stockdale 2001; Vitart et al. 2007). This is due in part to drifts in the climate system and in particular the drift in the simulated SSTs resulting in large atmospheric biases. Here we show that a simple bias correction applied to a coupled model’s predicted SST greatly improves the retrospective seasonal hurricane forecast skill in the North Atlantic.
The predicted SSTs are obtained from the National Oceanic and Atmospheric Administration’s Climate Forecast System, version 1 (CFSv1), model (Saha et al. 2006) and are used as lower boundary forcing in the Florida State University/Center for Ocean–Atmospheric Prediction Studies (FSU/COAPS) atmosphere model. The FSU/COAPS model has shown high skill in simulating the North Atlantic tropical storm activity using observed SSTs (LaRow et al. 2008). This paper is divided into the following four sections. Section 2 describes the atmospheric model, the experimental setup, and the SST bias correction technique. Section 3 describes the results and section 4 presents the discussion and conclusions.
2. Experimental setup and SST bias correction
Two SST products are used to test the impact of SST bias correction on seasonal retrospective forecasts of North Atlantic hurricane activity. The first forecast SST dataset is obtained from the first of four 210-day forecasts generated by the CFSv1 model on 1 June of the respective year. The forecast SSTs are saved once per day and interpolated to the FSU/COAPS atmospheric model’s T126 horizontal resolution (Gaussian grid spacing of approximately 0.94° longitude by latitude). Throughout the rest of this paper this dataset is called the non-bias-corrected (NBC) SST.
The second SST dataset is the bias-corrected (BC) SST. The bias correction is accomplished in two steps. First, the SST anomalies are calculated from the NBC SSTs using the CFSv1 SST climatology. The CFSv1 SST climatology is calculated based on the 1 June forecasts from 1981–2006. The anomalies are then added to the Reynolds et al. (2002) SST [hereafter called the optimum interpolation, version 2 (OIv2)] climatology to form the BC SST dataset. The OIv2 climatology is based on the time period 1971–2000 and is interpolated in time to a daily series prior to adding the anomalies. The only difference between the NBC and BC SST datasets is the SST climatology. Kumar et al. (2008) propose a similar bias-correction technique. The sensitivity of the model’s hurricane counts to the choice of SST climatology is discussed in section 3a.
We note that it is generally preferable to use a model ensemble average in order to extract the signal from the noise in a climate dataset such as the SSTs. This study uses the first SST realization for each year from the CFSv1 forecasts in order to be consistent with our real-time seasonal hurricane forecasts. To fully explore the predictive nature of our atmospheric model, the use of an ensemble average SST with an ensemble of atmospheric initial conditions or using an ensemble of atmospheric initial conditions with an ensemble of SSTs is needed.
The retrospective seasonal hurricane forecasts use the NBC and BC SSTs determined for a 28-yr period (1982–2009). For each of the 28 years, four atmospheric realizations are developed using time-lagged 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40; Uppala et al. 2005) centered on 1 June of the respective year. Two ensembles are then generated using the two different SST datasets for a total of 224 experiments. The integrations are conducted for the 6-month period (June–November) to coincide with the North Atlantic hurricane season.
The objective tropical cyclone detection/tracking algorithm used is the same as in LaRow et al. (2008, 2010) and similar to that used by Zhao et al. (2009, 2010). Briefly, for a storm to be detected and tracked it must satisfy three criteria. First, the 850-hPa relative vorticity must exceed 4.5 × 10−5 s−1. Next, the relative vorticity maximum must be accompanied by a minimum in the sea level pressure field and finally a warm core must be found between 500 and 200 hPa. Once a tropical storm is detected it must last at least 2 days and have surface winds greater than or equal to 17 m s−1. In this paper, a hurricane is defined according to the detection algorithm’s 850-hPa wind magnitude. If, at any point during the storm’s lifetime, the 850-hPa wind speed becomes greater than or equal to 41 m s−1 the storm is classified a hurricane. The value of 41 m s−1 is chosen to correspond to 33 m s−1 at the surface using a reduction factor of 0.8. This reduction factor is commonly used to reduce the 850-hPa hurricane reconnaissance flight level winds to the surface (see www.nhc.noaa.gov/aboutwindprofile.shtml). The observed hurricane counts are from the International Best Track Archive for Climate Stewardship (IBTrACS; Knapp et al. 2010).
3. Results
a. SST climatology and stochastic North Atlantic hurricane variability
Because the linear bias correction technique replaces the CFSv1 SST climatology with an observed estimate of the global SST climatology, it is instructive to briefly examine the stochastic nature of hurricane activity in the FSU/COAPS model. The stochastic component is examined using the CFSv1 model-derived SST climatology (1981–2006) and two observed SST climatological estimates: the OIv2 and the Hadley Center Global Sea Ice and Sea Surface Temperature (HadISST; Rayner et al. 2003) as lower boundary conditions in the FSU/COAPS model. To ensure a close comparison with the OIv2 climatology, the HadISST SST climatology is calculated from the same years as the OIv2 SST climatology (1971–2000).
Figure 1 shows the area average monthly June–November SST climatologies from the OIv2, the HadISST, and the CFSv1 model in four domains. The four domains cover regions of the tropical oceans and give a sense of the similarity between the climatologies. The solid lines are the CFSv1 SST climatology, the dashed lines are the OIv2, and the dotted lines are the HadISST climatology. The two observed estimate (OIv2 and HadISST) are very similar in the four tropical domains while a large cold bias (approximately 2°C) is noted with the CFSv1 in the Niño-3.4 region (Fig. 1d) in September. Additionally, (Figs. 1a,b) show that the CFSv1 model has a 1-month lag in obtaining the seasonal minimum. Figure 2 shows spatially the mean SST climatological bias averaged over the August–September–October (ASO) period for the CFSv1 model. Figure 2a shows the bias with respect to the OIv2 ASO climatology and Fig. 2b shows the CFSv1 with respect to the HadISST ASO climatology.
June–November area-average climatological sea surface temperatures from the CFSv1 model (solid line), the OIv2 data (dashed line), and the HadISST (dotted line): (a) Equatorial Pacific (10°S–10°N, 120°E–90°W), (b) global tropics (30°S–30°N, 0°–360°), (c) main development region (10°–20°N, 80°–20°W), and (d) Niño-3.4 region (5°S–5°N, 170°–120°W). Units: °C.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
June–November area-average climatological sea surface temperatures from the CFSv1 model (solid line), the OIv2 data (dashed line), and the HadISST (dotted line): (a) Equatorial Pacific (10°S–10°N, 120°E–90°W), (b) global tropics (30°S–30°N, 0°–360°), (c) main development region (10°–20°N, 80°–20°W), and (d) Niño-3.4 region (5°S–5°N, 170°–120°W). Units: °C.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
June–November area-average climatological sea surface temperatures from the CFSv1 model (solid line), the OIv2 data (dashed line), and the HadISST (dotted line): (a) Equatorial Pacific (10°S–10°N, 120°E–90°W), (b) global tropics (30°S–30°N, 0°–360°), (c) main development region (10°–20°N, 80°–20°W), and (d) Niño-3.4 region (5°S–5°N, 170°–120°W). Units: °C.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Spatial plot of the mean ASO SST climatology bias: (a) CFSv1 climatology minus OIv2 climatology and (b) CFSv1 climatology minus HadISST climatology. The contour interval is 0.5°C.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Spatial plot of the mean ASO SST climatology bias: (a) CFSv1 climatology minus OIv2 climatology and (b) CFSv1 climatology minus HadISST climatology. The contour interval is 0.5°C.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Spatial plot of the mean ASO SST climatology bias: (a) CFSv1 climatology minus OIv2 climatology and (b) CFSv1 climatology minus HadISST climatology. The contour interval is 0.5°C.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Using the three SST climatologies, the FSU/COAPS atmospheric model is integrated using 15 different 1200 UTC 1 June atmospheric initial conditions obtained from ERA-40 reanalysis to form the ensembles. The integrations begin on 1 June and end on 30 November. The hurricane counts determined using the three SST climatologies along with the observed counts from the IBTrACS dataset are shown in Table 1. All three ensembles produce mean counts higher than the observed long-term mean and are also higher based on the respective time period that the SST climatologies are determined. For example, using the CFSv1 SST climatology, the model simulates a mean of 8.4 hurricanes compared to the observed mean of 6.3 hurricanes. The OIv2 climatology produces the lowest mean count (7.5 hurricanes). Based on the 15 realizations, statistical tests show that the means and the variances are not statistically different between the three ensembles. However, examination of the histograms of the hurricane frequency distribution along with the kernel density estimates (Fig. 3) reveals differences among the ensembles. A wider and flatter distribution is seen with the OIv2 climatology compared to the CFSv1 and HadISST climatologies. The differences seen in the histograms and kernel density estimates between the OIv2 and HadISST SST experiments highlight the extreme sensitivity and nonlinear nature in simulating seasonal hurricane activity given the relatively small differences between their monthly SST fields (Fig. 1). Throughout the rest of this paper, the OIv2 climatology is used in the SST bias correction since the mean hurricane count determined using the OIv2 is closer to the observed mean and the OIv2 climatology produces a larger standard deviation in hurricane counts.
Ensemble mean/median and standard deviation of hurricane counts determined using three different SST climatologies (OIv2, HadISST, and the CFSv1). The OIv2 and HadISST SST climatologies are determined from the time period 1971–2000 and the CFSv1 SST climatology is determined from the time period 1981–2006. The observed hurricane counts are from IBTrACS. Fifteen different realizations compose each ensemble.
Histogram of the hurricane frequency distribution and kernel density lines using the three climatological SSTs: (a) CFSv1 SST climatology, (b) OIv2 SST climatology, (c) HadISST SST climatology, and (d) Poisson distribution with λ = 8.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Histogram of the hurricane frequency distribution and kernel density lines using the three climatological SSTs: (a) CFSv1 SST climatology, (b) OIv2 SST climatology, (c) HadISST SST climatology, and (d) Poisson distribution with λ = 8.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Histogram of the hurricane frequency distribution and kernel density lines using the three climatological SSTs: (a) CFSv1 SST climatology, (b) OIv2 SST climatology, (c) HadISST SST climatology, and (d) Poisson distribution with λ = 8.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
b. August–September–October sea surface temperatures
Figure 4 shows the area-average ASO CFSv1 (solid line) and OIv2 (dashed line) from 1982–2009 in four regions: the Niño-3.4, the Atlantic main development region (MDR), the global tropics, and the region encompassing the Atlantic meridional mode (AMM; Vimont and Kossin 2007). The MDR is bounded by 10°–20°N, 80°–20°W; the global tropics by 30°S–30°N; the Niño-3.4 by 5°S–5°N, 170°–120°W; and the AMM by 20°S–30°N, 75°W–15°E. The left column of the figure shows the CFSv1 SSTs prior to the application of the bias correction and the right column shows the SSTs after the bias correction. The corresponding least squared linear trend lines are also shown in the figure. The Niño-3.4, MDR, and AMM regions are chosen because of their known influences on tropical cyclone activity in the North Atlantic, while the global tropic domain is shown to highlight the overall cold bias.
Mean ASO SSTs for the (a),(e) Niño-3.4 region; (b),(f) MDR; (c),(g) global tropics; and (d),(h) AMM domain. The solid lines are the observed and the dashed lines are the CFSv1 SSTs. The CFSv1 SSTs (left) before and (right) after bias correction. Linear trend lines are also shown. Units: °C.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Mean ASO SSTs for the (a),(e) Niño-3.4 region; (b),(f) MDR; (c),(g) global tropics; and (d),(h) AMM domain. The solid lines are the observed and the dashed lines are the CFSv1 SSTs. The CFSv1 SSTs (left) before and (right) after bias correction. Linear trend lines are also shown. Units: °C.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Mean ASO SSTs for the (a),(e) Niño-3.4 region; (b),(f) MDR; (c),(g) global tropics; and (d),(h) AMM domain. The solid lines are the observed and the dashed lines are the CFSv1 SSTs. The CFSv1 SSTs (left) before and (right) after bias correction. Linear trend lines are also shown. Units: °C.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Prior to the application of the bias correction, the largest root-mean-square error (RMSE) is found in the Niño-3.4 region (Fig. 4a) with an RMSE of approximately 2.0°C. The next largest RMSE is found in the global tropics (Fig. 4c). In the MDR (Fig. 4b), the CFSv1 has a warm bias of 0.26°C and an RMSE of 0.41°C averaged over the 28 years. Almost all of the positive bias in the MDR is associated with the years prior to 2000. After the application of the bias correction (Fig. 4, right column), the RMSE is substantially reduced in the Niño-3.4 domain and global tropics (Figs. 4e,g) with a reduction of approximately 1.5°C found in the Niño-3.4 domain. The CFSv1 SSTs exhibit negative least squares linear trend lines in all four domains, while the observed trends are positive. Cai et al. (2009) found that similar differences in the long-term trend in the surface air temperature in the CFSv1 model and they speculate this is due to fixing the atmospheric greenhouse gases at the 1988 level. The observed trends are significant (p < 0.01) in the MDR and global tropics. The observed warming of the Atlantic Ocean in the secular record is well known (Santer et al. 2006) and has been attributed to the increase in metrics of North Atlantic hurricane activity since 1995 (e.g., Goldenberg et al. 2001; Emanuel 2005; Landsea 2005; Hoyos et al. 2006; Saunders and Lea 2008; Holland and Webster 2007). However, the exact cause(s) for the increase in activity is still debated. In the next section we show that a statistically significant hurricane trend is achieved using the bias-corrected SSTs despite the lack of a positive trend in the MDR.
c. North Atlantic hurricane activity
Previous studies using the FSU/COAPS global model demonstrated its skill in simulating the interannual variability in total storm activity (storms of tropical storm strength or greater) in the North Atlantic using observed SSTs (LaRow et al. 2008) and using predicted SSTs from the CFSv1 climate model (LaRow et al. 2010). This section examines the difference in hurricane variability resulting from the use of the BC and NBC SSTs. The interannual variability of the hurricane counts from the IBTrACS data and from the FSU/COAPS model using the NBC SSTs is shown in Fig. 5. Figure 6 shows the hurricane counts using the BC SSTs. In both figures the solid red line shows IBTrACS data and the black line is the model’s ensemble mean. The spread of the realizations is shaded. The corresponding red and black dashed lines show the linear trend lines. The correlation of the interannual variability of the observed hurricane counts with the ensemble mean hurricane counts using the NBC SST is 0.42 (p < 0.05; Fig. 5), while using the BC SSTs the correlation is larger (r = 0.74; Fig. 6) explaining about 55% of the variance. Removal of the linear trend reduces the correlation to 0.66. The time series of both ensemble mean hurricane counts correlate well with the ENSO variability, as measured by the Niño-3.4 CFSv1 SST anomalies, although the correlation is greater with the BC SSTs (not shown). The tropical Pacific–North Atlantic hurricane connection is discussed in section 3d. In terms of trend lines, no statistically significant trend (+0.02 yr−1) is found in the NBC SST experiment; however, the BC SST experiment does produce a statistically significant trend (+0.11 yr−1; p < 0.05) in close agreement with the observed trend (+0.15 yr−1; p < 0.05) from 1982–2009.
Interannual hurricane counts from 1982–2009. The solid red line is the IBTrACS observed dataset and the solid black line is the ensemble mean. The shaded region is the ensemble spread using the NBC SSTs. The correlation coefficient is 0.42. The dashed lines are the linear trends.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Interannual hurricane counts from 1982–2009. The solid red line is the IBTrACS observed dataset and the solid black line is the ensemble mean. The shaded region is the ensemble spread using the NBC SSTs. The correlation coefficient is 0.42. The dashed lines are the linear trends.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Interannual hurricane counts from 1982–2009. The solid red line is the IBTrACS observed dataset and the solid black line is the ensemble mean. The shaded region is the ensemble spread using the NBC SSTs. The correlation coefficient is 0.42. The dashed lines are the linear trends.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
As in Fig. 5, but using the BC SSTs. The correlation coefficient is 0.74.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
As in Fig. 5, but using the BC SSTs. The correlation coefficient is 0.74.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
As in Fig. 5, but using the BC SSTs. The correlation coefficient is 0.74.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Table 2 shows the ensemble mean hurricane and named storm (hurricanes + tropical storms) counts from the BC and NBC SST experiments along with the observation for two time periods: 1982–94 and 1995–2009. The time is split into two periods to highlight the increase in hurricane activity after 1994 associated with the Atlantic multidecadal oscillation (e.g., Goldenberg et al. 2001). Using the NBC SSTs the mean hurricane counts pre-1995 and post-1994 are not statistically different (6.5 and 7.3 hurricanes), but the hurricane counts are statistically different in both the BC SST experiment and the observations. In addition, Bayesian change-point analysis (Erdman and Emerson 2008) conducted on the three hurricane time series confirms that the observed and BC SST experiment both produce a change point in the mean hurricane counts after 1994 (not shown).
Average number of hurricanes and named storms for two time periods: 1982–94 and 1995–2009. Named storms are in parentheses. Boldface font indicates mean values that are not statistically different from the observations. Significance is at the 95% level.
In terms of the number of named storms, both the BC and NBC experiments produce 20%–50% more storms than observed values during the 1982–94 retrospective forecast period. When examined over the entire retrospective forecast period, the larger number of named storms detected using the NBC SSTs (Table 2) did not translate into a greater percentage becoming hurricanes (cf. 48% and 58% using the BC SSTs). The observed percentage from 1982–2009 is 54%. Table 3 shows a summary of the hurricane counts using the NBC SSTs, BC SSTs averaged over the entire 28 years along with the observation. Both the BC and NBC SSTs experiments produce similar means (6.7 and 6.9 hurricanes) and identical median values despite the large differences between the two SST climatologies used (Fig. 1). This is consistent with the stochastic findings in Table 1. Also shown in Table 3 are the probabilities for the seasonal hurricane counts being less than or equal to 3 (a below-average season), greater than 6 (an average season), and greater than 10 (an extreme season). The NBC experiment under predicts the below-average and extreme seasons, while the BC experiment produces too few seasons with below-average activity compared to the observed.
Hurricane summary from the retrospective forecasts using the BC and NBC SSTs from 1982–2009.Observations are from IBTrACS. Also shown are the probabilities for counts less than or equal to 3 (below-average hurricane season), counts greater than 6 (average hurricane season), and counts greater than 10 (extreme hurricane season). Probabilities from the model are based on the ensemble mean.
d. Tropical Pacific–Atlantic teleconnections
It is widely recognized that the magnitude of the SSTs in the tropical ocean basins can have a pronounced influence on tropical cyclone activity in the North Atlantic on interannual time scales. For example, warm/cold ENSO events tend to suppress/enhance North Atlantic tropical cyclone activity through increase/decrease vertical shear (Gray 1984; Goldenberg and Shapiro 1996; Aiyyer and Thorncroft 2006; Camargo et al. 2007; Shaman et al. 2009). The changes in the North Atlantic vertical wind shear results, in part, from the changes in the Pacific Walker circulation. Vecchi and Soden (2007) noted that a negative phase of the relative SST index results in a weaker Pacific Walker circulation, which is found to have a negative correlation with MDR vertical wind shear. The relative SST index is defined as the average SST anomaly in the MDR relative to the tropical global average anomaly. Although not shown, the CFSv1 ASO relative SST index calculated from 1982–2009 has a 0.75 correlation with the observed relative SST index. To examine changes in the strength of the Pacific Walker circulation, scatterplots of the BC and NBC ASO Niño-3.4 SSTs and the corresponding ensemble average precipitation in the eastern equatorial Pacific (5°S–10°N, 130°–90°E) from 1982–2009 are shown in Fig. 7. The region in the eastern Pacific is chosen to show the precipitation differences in the vicinity of the descending branch of the Pacific Walker cell. The extreme warm ENSO event year of 1997 is deemed an outlier and removed from the plot. The black dots show the results using the BC SSTs and the red dots show the results using the NBC SSTs. The use of the BC SSTs results in approximately 40% more precipitation when averaged over all ASO seasons compared to the NBC SST experiment. Additionally, using the BC SSTs produces slightly more than twice the variance in precipitation. The eastern equatorial Pacific precipitation associated with the BC SST experiment has a correlation (r = 0.51, p < 0.001) with the model’s ASO vertical wind shear in the MDR. Using the NBC SSTs, the model has a negative relationship (r = −0.1) between the precipitation and the vertical wind shear. The vertical wind shear is defined as (|V200 − V850|), where V200 is the horizontal wind vector at 200 hPa and V850 is the horizontal wind vector at 850 hPa. Additionally, a strong positive correlation (0.72) is also found between the ASO ensemble average MDR vertical shear and the ASO average Niño-3.4 SSTs in the BC experiment while a negative correlation is found in the NBC SST experiment (r = −0.16). No significant correlation is found with the ASO ensemble average MDR vertical shear with the ASO SSTs in the AMM region, regardless of the choice of SSTs.
Scatterplots of the ensemble average precipitation in the eastern tropical Pacific vs the ASO Niño-3.4 SSTs. The black dots represent the BC SST experiment and the red dots represent the NBC SST experiment. The dotted lines are the linear trend lines. The extreme warm ENSO event year of 1997 is removed from the plot.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Scatterplots of the ensemble average precipitation in the eastern tropical Pacific vs the ASO Niño-3.4 SSTs. The black dots represent the BC SST experiment and the red dots represent the NBC SST experiment. The dotted lines are the linear trend lines. The extreme warm ENSO event year of 1997 is removed from the plot.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
Scatterplots of the ensemble average precipitation in the eastern tropical Pacific vs the ASO Niño-3.4 SSTs. The black dots represent the BC SST experiment and the red dots represent the NBC SST experiment. The dotted lines are the linear trend lines. The extreme warm ENSO event year of 1997 is removed from the plot.
Citation: Monthly Weather Review 141, 2; 10.1175/MWR-D-12-00152.1
4. Discussion and conclusions
In this paper, several aspects of simulated North Atlantic hurricane variability is examined by imposing different background SST climatologies onto a set of predicted sea surface temperature anomalies. First, we developed three ensembles, each comprising 15 atmospheric realizations to examine the stochastic nature of hurricane activity in the FSU/COAPS atmospheric model. The three ensembles use as lower boundary conditions two observed estimates of the SST climatology (Reynolds et al. 2002; Rayner et al. 2003), while the third ensemble uses the CFSv1 model’s climatology. Examining the annual cycle of the climatological SSTs revealed that the CFSv1 model has a strong cold bias compared to the two observed estimates in the near-equatorial region of the Pacific, while in the main development region the CFSv1 bias is positive compared to the two observed climatologies. On the monthly time scales, the two observed SST climatological estimates are almost identical in their patterns and magnitudes with differences in the magnitude generally less than 0.2°C. These small differences in the SSTs produced a difference of one hurricane in the ensemble mean and a difference in the standard deviation of 0.6 hurricanes. The hurricane count distribution using the OIv2 climatology was found to be more negatively skewed compared to the distribution using the HadISSTs.
We then developed an ensemble of retrospective forecasts from 1982–2009 of the North Atlantic hurricane seasons using two SST datasets. The first SST dataset is the predicted SSTs from the CFSv1 model. The second SST dataset is the CFSv1 bias-corrected SSTs. In terms of the ensemble mean hurricane counts averaged over the entire 28 years both the BC and NBC experiments generate statistically similar means (6.7 and 6.9 hurricanes) and are consistent with the stochastic findings using the OIv2 and CFSv1 SST climatologies (see Table 1). Examination of the interannual variability in the hurricane activity reveals the influence of the background mean SST state. The correlation of the interannual variability of the observed hurricane counts with the BC (NBC) ensemble mean counts is 0.74 (0.42). One possible reason for the differences in the hurricane variability is the intensity of the deep convective activity in the tropical Pacific, which alters the strength of the Pacific Walker cell. In the NBC experiment, the large cold SST bias in the tropical Pacific inhibits large changes in the deep convective activity on interannual time scales. The BC SST experiment produced an average of 40% more precipitation during the ASO season in the tropical Pacific than the NBC SST experiment. In addition, the variance in the ensemble average tropical Pacific precipitation using the BC SSTs is approximately twice that of the NBC SST experiment. The result is greater equatorial Pacific–Atlantic teleconnections producing enhanced variability in the MDR vertical wind shear and North Atlantic hurricane activity in the BC SST experiment.
Finally, we note that the CFSv1 climate model is no longer being run operationally and has been replaced by the CFSv2. The method of bias correction used here is an optimal choice due to the fact that the CFSv1 model has skill in forecasting the evolution of the tropical SST anomalies during the boreal summer even though the bias is large. Reducing the SST bias prior to making a dynamical seasonal hurricane forecast may be superior to postprocessing hurricane counts (or bias correcting hurricane counts) after making a seasonal forecast with a climate model that exhibits large biases in the SSTs.
Acknowledgments
All computations were performed on the FSU High Performance Computer cluster. We wish to thank Greg Holland and two anonymous reviewers for their comments and suggestions on improving the manuscript. This research was supported by DOE-CESD.
REFERENCES
Aiyyer, A. R., and C. Thorncroft, 2006: Climatology of vertical wind shear over the tropical Atlantic. J. Climate, 19, 2969–2983.
Bengtsson, L., K. I. Hodges, and M. Esch, 2007a: Tropical cyclones in a T159 resolution global climate model: Comparison with observations and re-analyses. Tellus, 59A, 396–416.
Bengtsson, L., K. I. Hodges, M. Esch, N. Keenlyside, L. Kornblueh, J.-J. Luo, and T. Yamagata, 2007b: How may tropical cyclones change in a warmer climate? Tellus, 59A, 539–561.
Cai, M., C.-S. Shin, H. M. van den Dool, W. Wang, S. Saha, and A. Kumar, 2009: The role of long-term trends in seasonal predictions: Implication of global warming in the NCEP CFS. Wea. Forecasting, 24, 965–973.
Camargo, S. J., K. A. Emanuel, and A. H. Sobel, 2007: Use of a genesis potential index to diagnose ENSO effects on tropical cyclone genesis. J. Climate, 20, 4819–4834.
Emanuel, K., 2005: Increasing destructiveness of tropical cyclones over the past 30 years. Nature, 436, 686–688.
Erdman, C., and J. W. Emerson, 2008: A fast Bayesian change point analysis for the segmentation of microarray data. Bioinformatics, 24, 2143–2148, doi:10.1093/bioinformatics/btn404.
Goldenberg, S. B., and L. Shapiro, 1996: Physical mechanisms for the association of El Niño and West African rainfall with Atlantic major hurricane activity. J. Climate, 9, 1169–1187.
Goldenberg, S. B., C. W. Landsea, A. M. Mestas-Nuñez, and W. M. Gray, 2001: The recent increase in Atlantic hurricane activity: Causes and implications. Science, 293, 474–479.
Gray, W. M., 1984: Atlantic seasonal hurricane frequency. Part I: El Niño and the 30 mb quasi-biennial oscillation influences. Mon. Wea. Rev., 112, 1649–1668.
Gualdi, S., E. Scoccimarro, and A. Navarra, 2008: Changes in tropical cyclone activity due to global warming: Results from a high-resolution coupled general circulation model. J. Climate, 21, 5204–5228.
Holland, G. J., and P. J. Webster, 2007: Heightened tropical cyclone activity in the North Atlantic: Natural variability or climate trend? Philos. Trans. Roy. Soc. London, 365A, 2695–2716.
Hoyos, C. D., P. A. Agudelo, P. J. Webster, and J. A. Curry, 2006: Deconvolution of the factors contributing to the increase in global hurricane intensity. Science, 312, 94–97.
Knapp, K. R., M. C. Kruk, D. H. Levinson, H. J. Diamond, and C. J. Neumann, 2010: The International Best Track Archive for Climate Stewardship (IBTrACS): Unifying tropical cyclone data. Bull. Amer. Meteor. Soc., 91, 363–376.
Kumar, A., Q. Zhang, J.-K. E. Schemm, M. L’Heureux, and K.-H. Seo, 2008: An assessment of errors in the simulation of atmospheric interannual variability in uncoupled AGCM simulations. J. Climate, 21, 2204–2217.
Landsea, C. W., 2005: Hurricanes and global warming. Nature, 438, E11–E12, doi:10.1038/nature04477.
LaRow, T. E., Y.-K. Lim, D. W. Shin, E. Chassignet, and S. Cocke, 2008: Atlantic basin seasonal hurricane simulations. J. Climate, 21, 3191–3206.
LaRow, T. E., L. Stefanova, D. W. Shin, and S. Cocke, 2010: Seasonal Atlantic tropical cyclone hindcasting/forecasting using two sea surface temperature datasets. Geophys. Res. Lett., 37, L02804, doi:10.1029/2009GL041459.
Oouchi, K., J. Yoshimura, H. Yoshimura, R. Mizuta, S. Kusunoki, and A. Noda, 2006: Tropical cyclone climatology in a global- warming climate as simulated in a 20 km mesh global atmospheric model: Frequency and wind intensity analyses. J. Meteor. Soc. Japan, 84, 259–276.
Rayner, N. A., D. E. Parker, E. B. Horton, C. K. Folland, L. V. Alexander, D. P. Rowell, E. C. Kent, and A. Kaplan, 2003: Global analyses of sea surface temperature, sea ice, and night marine air temperature since the late nineteenth century. J. Geophys. Res., 108, 4407, doi:10.1029/2002JD002670.
Reynolds, R. W., N. A. Rayner, T. M. Smith, D. C. Stokes, and W. Wang, 2002: An improved in situ and satellite SST analysis for climate. J. Climate, 15, 1609–1625.
Saha, S., and Coauthors, 2006: The Climate Forecast System at NCEP. J. Climate, 19, 3483–3517.
Santer, B. D., and Coauthors, 2006: Forced and unforced ocean temperature changes in Atlantic and Pacific tropical cyclogenesis regions. Proc. Natl. Acad. Sci. USA, 103, 13 905–13 910.
Saunders, M., and A. Lea, 2008: Large contribution of sea surface warming to recent increase in Atlantic hurricane activity. Nature, 451, 557–461.
Shaman, J., S. K. Esbensen, and E. D. Maloney, 2009: The dynamics of the ENSO–Atlantic hurricane teleconnection: ENSO-related changes to the North African–Asian jet affect Atlantic basin tropical cyclogenesis. J. Climate, 22, 2458–2482.
Uppala, S. M., and Coauthors, 2005: The ERA-40 Re-Analysis. Quart. J. Roy. Meteor. Soc., 131, 2961–3012.
Vecchi, G. A., and B. J. Soden, 2007: Global warming and the weakening of the tropical circulation. J. Climate, 20, 4316–4340.
Vimont, D. J., and J. P. Kossin, 2007: The Atlantic meridional mode and hurricane activity. Geophys. Res. Lett., 34, L07709, doi:10.1029/2007GL029683.
Vitart, F., 2006: Seasonal forecasting of tropical storm frequency using a multi-model ensemble. Quart. J. Roy. Meteor. Soc., 132, 647–666.
Vitart, F., and T. N. Stockdale, 2001: Seasonal forecasting of tropical storms using coupled GCM integrations. Mon. Wea. Rev., 129, 2521–2537.
Vitart, F., and Coauthors, 2007: Dynamically-based seasonal forecasts of Atlantic tropical storm activity issued in June by EUROSIP. Geophys. Res. Lett., 34, L16815, doi:10.1029/2007GL030740.
Zhao, M., I. M. Held, S.-J. Lin, and G. A. Vecchi, 2009: Simulations of global hurricane climatology, interannual variability, and response to global warming using a 50-km resolution GCM. J. Climate, 22, 6653–6678.
Zhao, M., I. M. Held, and G. A. Vecchi, 2010: Retrospective forecasts of the hurricane season using a global atmospheric model assuming persistence of SST anomalies. Mon. Wea. Rev., 138, 3858–3868.
