1. Introduction
Predicting the near-future climate (i.e., the next one or two decades) is challenging because of the limitations in our understanding of climate variability, uncertainties in climate models, and the lack of long-term observational data. Climate variability on decadal time scales results from both internal variability (i.e., time-evolving natural oscillation) and external forcings (i.e., greenhouse gas effects) so that near-future prediction is a marginal time scale with application of initial values and forced boundary conditions. As such, it has been referred to as a “predictability desert” between short-range forecasting and long-term future projections (Meehl et al. 2009, 2014). At the same time, the impacts of near-future climate change have become a serious concern because of their substantial socioeconomic costs and environmental risks that directly affect humankind. Thus, studies on near-future climate are emerging and would help to formulate the long-term plans for mitigating the damages due to extreme weather phenomena, including tropical cyclones (TCs), under near-future climate conditions.
The changes of TC activity in the next several decades are among the most imminent concerns given their potential for huge damages in coastal regions. There has been a growing demand to explore the near-future changes in TC activity such as the location and frequency of TC genesis, intensity, and track patterns. Most previous studies on projecting future TC activity have focused on the genesis frequency and maximum intensity in the late twenty-first century (Bengtsson et al. 2007; Stowasser et al. 2007; Emanuel et al. 2008; Knutson et al. 2010, 2015; Murakami et al. 2012; Emanuel 2013). There have been only a few studies predicting TC tracks in the far future (Park et al. 2017) even though tracks are the most crucial factor for determining TC disasters as they are related to landfall locations. Near-future changes in TC track patterns have therefore become necessary to address TC disasters in advance, but little is known yet.
A number of studies attempted to predict TC activity for seasonal to centennial time scales using hybrid statistical–dynamical models (Wang et al. 2009; Kim and Webster 2010; Vecchi et al. 2011; Kim et al. 2012; Ho et al. 2013; Li et al. 2013; Choi et al. 2016a,b; Park et al. 2017). Hybrid methods can overcome the limitations of statistical and/or dynamical approaches by considering simultaneous relationships between predictands (i.e., TCs) and predictors (i.e., climate variables) from climate model forecasts. Recently, the hybrid method was applied to predict the spatial distribution of TC activity based on climatological TC track patterns (Kim et al. 2012; Ho et al. 2013; Choi et al. 2016a,b; Park et al. 2017). These hybrid models showed comparable or higher skills in forecasting seasonal TC activity than numerical modeling approaches thus far.
Previous studies have shown enhanced TC activity in the North Atlantic (NA) in recent years due to basinwide sea surface temperature (SST) warming and weakened vertical wind shear (VWS) over the tropical NA (Goldenberg et al. 2001; Saunders and Lea 2005; Elsner and Jagger 2006; Holland and Webster 2007; Klotzbach 2007). Among the large-scale climate variabilities, the Atlantic multidecadal oscillation (AMO) and El Niño–Southern Oscillation (ENSO) are known to strongly affect TC activity by modulating the SSTs and VWS over the tropical NA (Gray 1984; Goldenberg and Shapiro 1996; Xie et al. 2005; Vimont and Kossin 2007; Colbert and Soden 2012; Davis et al. 2015; Krishnamurthy et al. 2016). The recent positive phase of the AMO since the mid-1990s is responsible for the higher basinwide SSTs and reduced vertical wind shear over the NA, which has been attributed to the recent enhancements in TC activity there. In El Niño (La Niña) episodes, fewer (more) TCs have occurred in the NA due to stronger (weaker) VWS and greater (weaker) atmospheric stability over the Caribbean and the tropical Atlantic basin. Thus, if climatological changes in the NA basinwide SST (NASST) and ENSO occur in the near future, NA TC activity will also change in response to the altered environmental conditions (Vimont and Kossin 2007; Davis et al. 2015; Krishnamurthy et al. 2016).
This study aims to predict the NA TC activity in the near-future period 2016–30 and to explain the mechanisms related to the near-future climate condition, especially in relation to NASST and ENSO variations. To accomplish these objectives, we use a track-pattern-based hybrid prediction model developed in a previous study (Choi et al. 2016b). This model has been shown to be capable of simulating realistic seasonal NA TC activity such as track density and genesis frequency. Although this model was originally developed to predict seasonal NA TC activity, it is also applicable for near-future predictions because the present-day empirical relationships used in the model are likely to be valid for the next 15 yr. Finally, the impacts of anthropogenic forcing and natural variability on near-future climate in the NA are discussed by investigating near-future SST predictions from multimodel products.
This paper is organized as follows. The datasets and methodology used in this study are described in section 2. Section 3 presents model verification with the phases of NASST and ENSO. We provide near-future TC prediction results in section 4 and a summary and discussion are given in section 5.
2. Data and methodology
a. Data
The observed best-track data for NA TCs are from the second-generation hurricane database (HURDAT2) obtained from the National Oceanic and Atmospheric Administration National Hurricane Center. HURDAT2 provides information on the TC center position as latitude–longitude coordinates, 1-min averaged maximum sustained wind speed υmax, and the minimum sea level pressure at the center at 6-h intervals. Although the available period of TC best-track data ranges from 1851 to the present, we use the dataset since 1965 to ensure data credibility (Landsea and Franklin 2013). We investigated all the NA TC (υmax ≥ 17 m s−1) activity in this study. Regarding the observed data reliability, we have investigated with and without TCs lasting for less than two days and obtained the same conclusions. We focus on the active hurricane season [August–October (ASO)] when about 80% of the annual NA TCs are formed climatologically (Choi et al. 2016b).
To develop the hybrid prediction models, we investigate empirical relationships between the large-scale environments and TC activity using atmospheric–oceanic circulation data from the National Centers for Environmental Prediction (NCEP) Climate Forecast System Reanalysis (CFSR) (Saha et al. 2010). This product is a fully coupled global reanalysis with a spatial resolution of 0.5° × 0.5° in latitude and longitude. We analyze SST, zonal winds at the 200-hPa (U200) and 850-hPa levels (U850), VWS of zonal wind (the magnitude of difference between U200 and U850), and relative vorticity at 850 hPa (VOR850). The ASO-averaged fields from the CFSR data recorded during 1982–2015 are used for consistency with the TC datasets.
The long-term free runs of Climate Forecast System version 2 (CFSv2) in Coupled Model Intercomparison Project (CMIP) simulations up to the year 2030 are used for the near-future prediction (Saha et al. 2014). There are three ensemble members each starting from 1 January of 1988, 1996, and 2002 in the CFSv2 CMIP simulations. Although the CFSv2 CMIP simulations are initialized with actual conditions (i.e., CFSR), errors would still arise from long numerical simulation of more than several years. A previous study reported that the system is stable with no drifting caused by technical problems and all three ensembles of CFSv2 CMIP runs show reliable climate variability such as the realistic ENSO periodicity by power spectrum analysis of the Niño-3.4 index (Saha et al. 2014). Thus, we investigate the changes in the TC activity by averaging all three ensemble members for the present period of 2002–15 (P1) and for the near-future period of 2016–30 (P2), which are the overlap periods of the three ensembles. In addition, the use of CFSv2 to predict near-future TC frequency is advantageous to us because the model itself was used for developing the seasonal prediction, and it reasonably simulates the behavior of climate variations such as ENSO (Choi et al. 2016b; Saha et al. 2014). CFSv2 in the CMIP simulations has the same settings as those used for CFSv2 seasonal hindcast simulations. For the future CO2 concentration, it is extrapolated to increase by two parts per million in volume per year based on the current seasonal observations.
For comparison with the near-future SST prediction of CFSv2 CMIP simulations, we use 24 CMIP phase 5 (CMIP5) models including ACCESS1.0, BCC-CSM1.1, BNU-ESM, CanESM2, CCSM4, CESM1-BGC, CMCC-CM, CMCC-CMS, CNRM-CM5, CSIRO-Mk-3.6.0, FIO-ESM, GFDL-CM3, GFDL-ESM2G, GFDL-ESM2M, GISS-E2-H, GISS-E2-R, HadGEM2-CC, HadGEM2-ES, INM-CM4, IPSL-CM5A-LR, IPSL-CM5B-LR, MPI-ESM-LR, MPI-ESM-MR, and NorESM1-M (see http://cmip-pcmdi.llnl.gov/cmip5/availability.html for expansions and additional information). The historical simulations, prescribed with observed atmospheric composition and time-evolving land cover changes, and midrange mitigation simulations [representative concentration pathway 4.5 (RCP4.5)] corresponding to radiative forcing values of 4.5 W m−2 up to the year 2100 are investigated (Taylor et al. 2012). Because 24 CMIP5 models have different horizontal resolutions, they are interpolated into 2.5° × 2.5° grids for consistent analysis. We analyzed the historical and RCP4.5 scenario datasets for the periods 2002–05 and 2006–30, respectively.
b. Methodology
To predict near-future TC activity over the NA basin, we used the track-pattern-based model developed in Choi et al. (2016b), which divides the climatological TC tracks into four representative patterns and predicts each pattern using the simultaneous empirical relationships with large-scale environments derived from the seasonal climate forecasts. Then, the prediction results of all track patterns are combined to produce the TC forecasts over the NA basin. The detailed procedure and explanations of the model are presented in appendix A.
3. Observational responses of TC to AMOand ENSO
Before performing TC predictions, it is necessary to examine the sensitivities of the TC prediction model to the changes in NASST and ENSO phases, and whether these sensitivities are physically explainable. Although the predictors of track-pattern-based models are strongly linked to NASST and ENSO variabilities, high predictability of the TC activity responses to the phases of NASST and ENSO is not guaranteed because the model uses the area-averaged climate variables as predictors (appendix B). Thus, we need to identify the responses of NA TC activity from the model with observation during the statistical training period of this model (i.e., 1982–2015).
Figure 1 presents the differences in the gridded TC occurrences between the positive and negative phase of NASST and ENSO. It is well known that more NA TCs form during a NASST warming period associated with the positive phase of AMO years whereas fewer TCs are developed during the NASST cooling period (Fig. 1a). While 9.5 TCs occur climatologically during ASO, 1.81 more (2.95 less) TCs occur during the NASST positive (negative) phase. Similarly, 1.75 more (1.59 less) TCs occur during La Niña (El Niño) years. Some regions in the tropical NA have experienced 1.5 more TC occurrences in the NASST positive years than negative years, which means that almost 50% more (less) TCs occur in the positive (negative) NASST years compared to the climatological TC numbers. Weakened TC activity of up to one less TC occurrence in the El Niño years compared to the La Niña years is noted over the Gulf of Mexico and the East Coast of North America (Fig. 1b) due to sinking motion over the Gulf of Mexico and the Caribbean Sea induced by the upward motion in the eastern Pacific (Frank and Young 2007). The core regions of TC occurrence differences between positive and negative phases of NASST and ENSO are significant at the 95% confidence level.
Composite differences in seasonal TC occurrence for positive minus negative phase years of the (left) North Atlantic basinwide SST (NASST) and (right) El Niño–Southern Oscillation (ENSO) during the period 1982–2015. (a),(b) The best-track observations, (c),(d) CFSR reconstructions, (e),(f) reconstruction using the NASST-regressed CFSR and using the Niño-3.4-regressed CFSR, respectively. Black dots indicate that the differences are statistically significant at the 95% confidence level in each 5° × 5° latitude–longitude grid area.
Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-17-0206.1
The hindcasts of TC activity from the model using CFSR can reasonably reproduce the observed responses of TC occurrences associated with the NASST and ENSO phases despite some overestimations (underestimations) in the midlatitude NA offshore (Gulf of Mexico) for the NASST (ENSO) case (Figs. 1c,d). In addition, the model responses are significant in most parts of the NA basin. In general, this model realistically simulates the overall TC activity changes associated with NASST and ENSO variations. To verify the origin of performance, we prescribe NASST- and Niño-3.4-regressed fields to the TC prediction model. Figures 1e and 1f show TC activity responses of the model to the phase differences of NASST and ENSO with their regressed fields used as input data. They are nearly identical to the reconstructions using the original CFSR fields (Figs. 1c,d). This result implies that most of the predicted TC activity responses originate from their regressed fields (i.e., NASST and ENSO); thus, the performance of the reforecast is only due to its own variability regardless of other variabilities. For the TC activity response to NASST (ENSO), the effect of Niño-3.4- (NASST)-regressed fields is negligible (not shown). Thus, this model can reasonably predict TC activity based on independent links with NASST and ENSO.
4. TC prediction in the near-future decades
This good validation of the model performance lends credence to our predictions of the changes in TC activity for the near-future period. Here, we investigate the differences in the gridded TC occurrences between present decades (P1) and near-future decades (P2). Table 1 shows the changes for the four clusters (TC1–TC4; see Fig. A1) and their summation of seasonal TC frequencies for each ensemble during the two periods. All clusters and their summation of TC frequencies are expected to decrease in the near future except for TC2 in CFSv2 initialized in 1996 (CFS1996). Frequencies in TC2 in CFSv2 initialized in 1988 (CFS1988), TC1 in CFS1996, and TC3 and TC4 in CFSv2 initialized in 2002 (CFS2002) are expected to decrease significantly in P2 compared with that in P1. For the averages of the three ensembles, the values of TC1–TC4 during the ASO of P2 will decrease by 0.15, 0.32, 0.53, and 0.53, respectively, corresponding to a 5%–20% reduction in predicted TC frequency. The total number of seasonal TCs is reduced by 1.53 (about a 12% decrease, 11.25 in P2 and 12.78 in P1).
Seasonal TC genesis frequencies of each track pattern and their summation by the track-pattern-based model for periods 2002–15 (P1), 2016–30 (P2), and the difference between the two periods. Asterisks represent statistical significance for each ensemble member at the 95% (***), 90% (**), and 80% (*) levels.
According to these changes, TC activity is predicted to weaken over the entire NA basin in the near-future period (Fig. 2a). In particular, seasonal TC occurrences over the region of 10°–20°N, 30°–60°W and in the open ocean of the NA basin will decrease by up to 0.4. This means that about four fewer TCs per decade are anticipated over the core region of reduced TC activity. All three ensembles based on the CFSv2 in CMIP runs consistently show reduced TC activity except for the Caribbean Sea and the Gulf of Mexico. The large decrease in NA TCs is concentrated mainly in the open ocean with a maximum decrease of two TCs per decade in the 50°N region, which would likely be in the form of extratropical storms.
Ensemble-averaged differences for seasonal TC occurrence between the two periods (2016–30 minus 2002–15) from using the (a) total fields, (b) North Atlantic basinwide SST (NASST)-regressed fields, (c) Niño-3.4-regressed fields, and (d) residual fields. Black dots indicate regions in which all three ensembles of reconstruction show the same sign in each 5° × 5° latitude–longitude grid area.
Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-17-0206.1
The weakened TC activity over the NA can be explained by the effects of the phase transitions in NASST and ENSO during the near-future period. The decrease in the TC activity over the open ocean is attributed mainly to the effects of NASST (Fig. 2b). The prediction using the NASST-regressed fields explains a large portion of the reduced TC occurrences offshore. However, no region shows the same sign in all three CFSv2 CMIP ensembles. For the ENSO effects, the overall decrease in the basinwide NA TC activity is caused by changes in the ENSO effects during P2 (Fig. 2c). In particular, the remote effects of ENSO can reduce TC activity over the Caribbean Sea, the Gulf of Mexico, and the East Coast of the United States. Unlike NASST, the changes in the ENSO phase show the same TC change tendencies in the entire NA basin. Although residual fields from NASST and ENSO can lead to decreases in TC activity over the tropical region, its variation shows relatively small amplitudes compared to the TC reductions by the changes in NASST and ENSO (Fig. 2d). Summation of these three decomposed changes presents nearly the same amplitude and spatial distribution of the total TC activity decrease (not shown).
We have analyzed near-future TC changes by considering the NASST and ENSO impacts. Here, we investigate how the changes in NASST and ENSO are simulated in the CFSv2 CMIP runs, and we examine their relationship to the predicted TCs changes. Table 2 presents a comparison of the event frequency per decade for NASST and ENSO phases between P1 and P2 for each ensemble member and their averages. Calculations of the long-term NASST and ENSO events over several decades are not available because the analysis period is limited to 2002–30, the common period for all three ensembles. However, we can assess the climatological phase transitions of NASST and ENSO, and their changes in event frequency between P1 and P2. All ensembles predict more frequent NASST neutral phases and less frequent positive phases. For ENSO, all three ensembles show more frequent El Niño events in the near-future period. Thus, more frequent El Niño episodes and fewer neutral and La Niña phases are expected to occur in P2. These changes in the climate variability will become unfavorable for the TC genesis and development over the NA in the near-future period.
Event frequencies per decade for positive, neutral, and negative phases of the North Atlantic basinwide SST (NASST) and ENSO for periods 2002–15 (P1) and 2016–30 (P2) and their difference.
Combinations of the NASST and ENSO phase transitions result in large-scale environmental changes in P2. Figure 3 shows the ensemble-averaged changes for seasonal environmental fields such as SST, U200, U850, VWS, and VOR850, which are predictor variables for the TC prediction model (Fig. A1). Although the near-future changes in these three ensemble averages reveal that the SST warming will occur in the midlatitude central NA by 0.1°C, other tropical and subpolar gyre regions in the NA show 0.1° and 0.5°C SST cooling in P2, respectively (Fig. 3a). In addition, notable SST warming regions appear in the central to eastern Pacific. These results imply that a NASST phase slowdown from the positive to neutral or cold phase in conjunction with more frequent El Niño events is anticipated in P2 compared with that in P1 as discussed earlier. All atmospheric circulations are systematically organized with these SST change patterns (Figs. 3b–e). The difference between the easterlies at 850 hPa and the westerlies at 200 hPa over the tropical NA causes the VWS to increase by more than 1.4 m s−1 in P2. Both the strengthened VWS and weakened VOR850 over the tropical NA, shown in the ensemble-averaged CFSv2 CMIP simulations, lead to unfavorable conditions for TC developments.
Ensemble-averaged seasonal differences in (a) SST (°C), (b) zonal wind at 200 hPa (m s−1), (c) zonal wind at 850 hPa (m s−1), (d) vertical wind shear (m s−1), and (e) relative vorticity at 850 hPa (10−6 s−1) between the two periods (2016–30 minus 2002–15). Black dots indicate regions in which all three ensembles of CFSv2 CMIP runs show the same sign.
Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-17-0206.1
These SST changes in the NA and eastern Pacific can be understood with their own regressed fields. The SST decreases in the tropical NA and subpolar gyre occur in the changes of NASST-regressed SST (Fig. B1a). The decadal SST variability in the NA subpolar gyre is known to be closely related to AMO (Hermanson et al. 2014). Because the variability of NASST is closely related to the AMO, the subpolar gyre heat convergence induces SST cooling and the NASST phase change. The upper and lower tropospheric wind fields changes related to the NASST variability lead to an increase in VWS in the tropical Atlantic region and a decrease in VOR850 in the subtropical NA region (Figs. B1b–e). All of these variations induce decreased TC activity in the NA offshore in near-future period. The SST warming in the eastern Pacific is also found in Niño 3.4-regressed fields (Fig. B2). Enhanced VWS in tropical and midlatitude region of NA as well as weakened VOR850 near shore (e.g., Caribbean and Gulf of Mexico) associated with the ENSO variation would cause reduction of overall TC activity in the entire basin (Figs. B2b–e).
As for residual effects, many types of climate factors other than NASST and ENSO can affect NA TC activity. Considering the predictor variables and their critical domains of the model (see Fig. A1), we can infer these other factors. The change in VWS owing mainly to 200-hPa level winds over the subtropical NA could be a reason for the weakening of TC activity in the near-future period because this will have a significant impact by destroying the TC vertical structure (Figs. B3b–d). In addition, strengthening of low-level easterlies (i.e., U850) in the tropical NA appear to be suppressing factors for TC development because it can lead to the decrease in the relative vorticity (VOR850) over the main TC development region (Fig. B3e). These changes are not captured in our statistical decomposition and do not appear to be directly related to the NASST and ENSO variabilities (see Fig. B3a). However, the effects of these residual factors cancel out each other to result in small net effects (Fig. 2d).
As in Fig. 3, but for residual fields.
Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-17-0206.1
As one of the candidates for other influences, long-term external forcing effects such as greenhouse effect may be considered as residual effects. However, greenhouse effects are not expected to have a significant impact in the near-future decades. Moreover, significant differences in NA TC activity are not detected even in the prediction using environmental fields regressed onto global warming (not shown). Therefore, if we want to accurately characterize the residual effects, further investigations of various time scales and spatial scales are needed.
5. Summary and discussion
Prediction using a track-pattern-based model suggests that TC activity over the NA will decrease in the near future. This model is known to effectively simulate TC activity in the current climate (Choi et al. 2016b) and thus is expected to be useful for near-future TC prediction in conjunction with climate-model-projected atmospheric and oceanic fields. Changes in two dominant climate variabilities, the NASST and ENSO, in the upcoming decade are mainly related to the suppression of NA TC activity. The phase transition of NASST from positive to neutral and the greater frequency of El Niño events will work together to decrease SST for the overall NA basin and to strengthen VWS in the tropical NA. The impacts of changes for NASST and ENSO will result in a more unfavorable environment for TC development for tropical regions in the near future. However, the cooling of NASST in the near-future period is an interesting issue when we consider recent global warming trends induced by anthropogenic greenhouse gas emissions. As an attempt to understand this discrepancy, we analyzed the near-future SST predictions in other CMIP5 models to discuss the main driver of NASST variability and ultimately of TC activity changes in the near future.
Figure 4 shows the ensemble-mean seasonal SST differences from 24 CMIP5 models following the historical and RCP4.5 scenarios between P2 and P1. The RCP4.5 scenario data are calculated from preindustrial initial conditions and forced by anthropogenic influences. We can assess that difference in seasonal SST from the CMIP5 multimodel ensemble average is mainly attributable to the anthropogenic forcing rather than internal variability for each model. As expected, the ensemble-mean prediction shows prevailing SST warming over the Atlantic and Pacific basins in the near-future period. For ENSO, the CMIP5 models generally predict greater SST warmings in the eastern Pacific, which implies that the models simulate more frequent occurrences of El Niño events. For the SST in the NA, the warming amplitude is relatively small in the NA subpolar gyre region because of the cooling from some models, but most of the models show warming SST in the tropical NA. Overall, anthropogenic forcing acts to increase the SST over both the NA and the eastern Pacific basins in the near future. Considering that NASST warming (more frequent El Niño events) can lead to enhancement (decline) in the NA TC activity, the near-future decrease in the NA TC activity may be small or almost unchanged due to the opposite effects of the NASST and ENSO if only anthropogenic impacts affect TC changes.
Ensemble-averaged seasonal difference in SST from 24 CMIP5 models for historical and RCP4.5 scenarios between the two periods (2016–30 minus 2002–15). Black dots indicate regions in which 18 of 24 CMIP5 models show the same signs of averages in SST differences.
Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-17-0206.1
Our results show clear decreases in the NA TC activity via the changes in both NASST and ENSO. All of these results depend on the performance of the CFSv2 CMIP climate forecasts. Although we used a single climate model, CFSv2 CMIP runs are appropriate for near-future prediction because the internal variability and external forcing are considered simultaneously by initiating it with the reanalysis data and being forced by increased CO2 concentration (see section 2a). We can estimate the impact of natural variability to SST increases in the near future by comparing the CFSv2 CMIP runs and the CMIP5 multimodel products. The number of El Niño events is expected to increase due to natural variability or anthropogenic forcing. The difference between the CFSv2 and CMIP5 multimodel ensemble-mean SST predictions implies that the cooling effects of natural variability dominate the warming effects of anthropogenic forcing in determining the future NASST. Thus, this result suggests that the cooling phase of NASST in conjunction with increasing El Niño events will significantly decrease the NA TC activity in the near future.
Near-future prediction should be handled differently from long-term climate projection (e.g., late twenty-first-century prediction) that is dominated by external forcing. Our study investigates the role of natural variability and anthropogenic forcing on the near-future climate and TC activity changes. We hope that this study can address the scientific challenges and the so-called predictability desert fairly to satisfy the social needs in preparing for TC-induced disasters with long-range plans.
Acknowledgments
This study was funded by the Korea Ministry of Environment as the “Climate Change Correspondence Program.” The work of JCLC was supported by the Research Grants Council of Hong Kong General Research Fund CityU 100113. We acknowledge the critical comments from three anonymous reviewers.
APPENDIX A
Track-Pattern-Based Model
To predict near-future TC activity over the NA basin, we use the track-pattern-based model developed in a previous study (Choi et al. 2016b), which divided the climatological TC tracks into four representative patterns and predicted each pattern using the simultaneous empirical relationships with large-scale environments derived from the seasonal climate forecasts. Then, the prediction results of all track patterns were combined to produce the TC forecasts over the NA basin.
The original version in the previous study used the NCEP CFSv2 seasonal reforecasts during the period 1982–2012 and quasi-real-time CFSv2 operational forecasts. In this study, the track-pattern-based TC prediction model is developed using the NCEP CFSR datasets. We modify the statistical training period to 1982–2015 to include very recent climate variability. However, the overall structure for this model and its logical flow are the same as those in the original version.
Figure A1 shows the four representative patterns of NA TC tracks (TC1–TC4) and the associated interannual correlation maps between TC track patterns and climate predictors. We identify climatological NA TC track patterns by using the fuzzy c-means method (Choi et al. 2016b). Then, the hybrid prediction model is developed for each track pattern. The candidate variables for the predictor are SST, VWS, VOR850, and U850, which are well-known climate factors strongly related to TC activity. If the characteristics of the predictors are empirically explainable for TC activity and the corresponding relations are statistically significant, we employ their relations as predictors. The positive relationships with SST and VOR850 are used for calculating predictors; other connections of VWS and U850 are also utilized. Because significant relations are still retained despite the inclusion of recent years, the detailed domain for the predictor region and combinations of predictors are the same as those used in the original version of the model.
(a),(d),(h),(l) Gridded TC occurrences of the four track patterns for hurricane seasons (August–October) of 1965–2015 over the NA. (b),(c),(e)–(g),(i)–(k),(m),(n) Spatial correlation coefficients between observed TC frequencies and predictor variables from CFSR during 1982–2015 for each track pattern. Red (blue) shadings indicate areas with statistically significant positive (negative) correlations at the 90% confidence level. The predictor regions are presented as a colored box.
Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-17-0206.1
To reforecast TC frequencies of the four track patterns, we use a leave-one-out cross-validation method. This method has been mainly used to evaluate the skill of statistical model employing the independent predictors on different years (Kim et al. 2012; Ho et al. 2013; Choi et al. 2016a,b). When we select a certain target year for prediction, the model is optimized to produce the best-fitting result by using the remaining years. This process is iteratively performed to reforecast the predictand (i.e., TC) for the whole training period. Statistical quantities of hindcasts assure the performance of this model even with the inclusion of recent years to the training period (Table A1). All correlation coefficients between reforecasted TC and observed TC are statistically significant at the 95% confidence level. Moreover, we investigate the root mean square errors (RMSEs) and mean square skill scores (MSSSs) to objectively assess the predictability. The RMSE value is the mean error of the forecasting model; a lower RMSE value indicates better performance. The MSSS value is a measure of improvement in performance over the reference prediction; a positive (negative) MSSS value implies that the model is better (worse) than the reference prediction. The MSSS value in the case of best prediction is 1. The expected errors (i.e., RMSEs) for all track patterns are 1.07, 1.59, 1.48, and 1.38, respectively, and all of the MSSS values are positive. Therefore, the lower RMSE (less than 1.59) and positive MSSS values of all TC track patterns demonstrate the skillful performance of our model.
Correlation coefficients (CORR), root mean square errors (RMSE), and mean square skill scores (MSSS) of hindcasts using the CFSR compared with best-track observations for the period 1982–2015.
APPENDIX B
Near-Future Changes in Regressed Fields
Figures B1–B3 show near-future changes in the NASST-regressed, Niño-3.4-regressed, and residual fields.
As in Fig. 3, but for the North Atlantic basinwide SST-regressed fields.
Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-17-0206.1
As in Fig. 3, but for Niño-3.4-regressed fields.
Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-17-0206.1
REFERENCES
Bengtsson, L., K. I. Hodges, M. Esch, N. Keenlyside, L. Kornblueh, J.-J. Luo, and T. Yamagata, 2007: How may tropical cyclones change in a warmer climate? Tellus, 59A, 539–561, doi:10.1111/j.1600-0870.2007.00251.x.
Choi, W., C.-H. Ho, C.-S. Jin, J. Kim, S. Feng, D.-S. R. Park, and J.-K. E. Schemm, 2016a: Seasonal forecasting of intense tropical cyclones over the North Atlantic and the western North Pacific basins. Climate Dyn., 47, 3063–3075, doi:10.1007/s00382-016-3013-y.
Choi, W., C.-H. Ho, J. Kim, H.-S. Kim, S. Feng, and K. Kang, 2016b: A track pattern–based seasonal prediction of tropical cyclone activity over the North Atlantic. J. Climate, 29, 481–494, doi:10.1175/JCLI-D-15-0407.1.
Colbert, A. J., and B. J. Soden, 2012: Climatological variations in North Atlantic tropical cyclone tracks. J. Climate, 25, 657–673, doi:10.1175/JCLI-D-11-00034.1.
Davis, K., X. Zeng, and E. A. Ritchie, 2015: A new statistical model for predicting seasonal North Atlantic hurricane activity. Wea. Forecasting, 30, 730–741, doi:10.1175/WAF-D-14-00156.1.
Elsner, J. B., and T. H. Jagger, 2006: Prediction models for annual U.S. hurricane counts. J. Climate, 19, 2935–2952, doi:10.1175/JCLI3729.1.
Emanuel, K. A., 2013: Downscaling CMIP5 climate models shows increased tropical cyclone activity over the 21st century. Proc. Natl. Acad. Sci. USA, 110, 12 219–12 224, doi:10.1073/pnas.1301293110.
Emanuel, K. A., R. Sundararajan, and J. Williams, 2008: Hurricanes and global warming: Results from downscaling IPCC AR4 simulations. Bull. Amer. Meteor. Soc., 89, 347–367, doi:10.1175/BAMS-89-3-347.
Frank, W. M., and G. S. Young, 2007: The interannual variability of tropical cyclones. Mon. Wea. Rev., 135, 3587–3598, doi:10.1175/MWR3435.1.
Goldenberg, S. B., and L. J. Shapiro, 1996: Physical mechanisms for the association of El Niño and West African rainfall with Atlantic major hurricane activity. J. Climate, 9, 1169–1187, doi:10.1175/1520-0442(1996)009<1169:PMFTAO>2.0.CO;2.
Goldenberg, S. B., C. W. Landsea, A. M. Mestas-Nunez, and W. M. Gray, 2001: The recent increase in Atlantic hurricane activity: Causes and implications. Science, 293, 474–479, doi:10.1126/science.1060040.
Gray, W. M., 1984: Atlantic seasonal hurricane frequency: Part I: El Niño and 30-mb quasi-biennial oscillation influences. Mon. Wea. Rev., 112, 1649–1668, doi:10.1175/1520-0493(1984)112<1649:ASHFPI>2.0.CO;2.
Hermanson, L., R. Eade, N. H. Robinson, N. J. Dunstone, M. B. Andrews, J. R. Knight, A. A. Scaife, and D. M. Smith, 2014: Forecast cooling of the Atlantic subpolar gyre and associated impacts. Geophys. Res. Lett., 41, 5167–5174, doi:10.1002/2014GL060420.
Ho, C.-H., J.-H. Kim, H.-S. Kim, W. Choi, M.-H. Lee, H.-D. Yoo, T.-R. Kim, and S. Park, 2013: Technical note on a track-pattern-based model for predicting seasonal tropical activity over the western North Pacific. Adv. Atmos. Sci., 30, 1260–1274, doi:10.1007/s00376-013-2237-6.
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, doi:10.1098/rsta.2007.2083.
Kim, H.-M., and P. J. Webster, 2010: Extended-range seasonal hurricane forecasts for the North Atlantic with a hybrid dynamical-statistical model. Geophys. Res. Lett., 37, L21705, doi:10.1029/2010GL044792.
Kim, H.-S., C.-H. Ho, J.-H. Kim, and P.-S. Chu, 2012: Track-pattern-based model for predicting seasonal tropical cyclone activity in the western North Pacific. J. Climate, 25, 4660–4678, doi:10.1175/JCLI-D-11-00236.1.
Klotzbach, P. J., 2007: Recent developments in statistical prediction of seasonal Atlantic basin tropical cyclone activity. Tellus, 59A, 511–518, doi:10.1111/j.1600-0870.2007.00239.x.
Knutson, T. R., and Coauthors, 2010: Tropical cyclones and climate change. Nat. Geosci., 3, 157–163, doi:10.1038/ngeo779.
Knutson, T. R., J. J. Sirutis, M. Zhao, R. E. Tuleya, M. Bender, G. A. Vecchi, G. Villarini, and D. Chavas, 2015: Global projections of intense tropical cyclone activity for the late twenty-first century from dynamical downscaling of CMIP5/RCP4.5 scenarios. J. Climate, 28, 7203–7224, doi:10.1175/JCLI-D-15-0129.1.
Krishnamurthy, L., G. A. Vecchi, R. Msadek, H. Murakami, A. Wittenberg, and F. Zeng, 2016: Impact of strong ENSO on regional tropical cyclone activity in a high-resolution climate model in the North Pacific and North Atlantic Oceans. J. Climate, 29, 2375–2394, doi:10.1175/JCLI-D-15-0468.1.
Landsea, C. W., and J. L. Franklin, 2013: Atlantic hurricane database uncertainty and presentation of a new database format. Mon. Wea. Rev., 141, 3576–3592, doi:10.1175/MWR-D-12-00254.1.
Li, X., S. Yang, H. Wang, X. Jia, and A. Kumar, 2013: A dynamical-statistical forecast model for the annual frequency of western Pacific tropical cyclones based on the NCEP Climate Forecast System version 2. J. Geophys. Res. Atmos., 118, 12 061–12 074, doi:10.1002/2013JD020708.
Meehl, G. A., and Coauthors, 2009: Decadal prediction: Can it be skillful? Bull. Amer. Meteor. Soc., 90, 1467–1485, doi:10.1175/2009BAMS2778.1.
Meehl, G. A., and Coauthors, 2014: Decadal climate prediction: An update from the trenches. Bull. Amer. Meteor. Soc., 95, 243–267, doi:10.1175/BAMS-D-12-00241.1.
Murakami, H., R. Mizuta, and E. Shindo, 2012: Future changes in tropical cyclone activity projected by multi-physics and multi-SST ensemble experiments using the 60-km-mesh MRI-AGCM. Climate Dyn., 39, 2569–2584, doi:10.1007/s00382-011-1223-x.
Park, D.-S. R., C.-H. Ho, J. C. L. Chan, K.-J. Ha, H.-S. Kim, J. Kim, and J.-H. Kim, 2017: Asymmetric response of tropical cyclone activity to global warming over the North Atlantic and the western North Pacific from CMIP5 model projections. Sci. Rep., 7, 41354, doi:10.1038/srep41354.
Saha, S., and Coauthors, 2010: The NCEP Climate Forecast System Reanalysis. Bull. Amer. Meteor. Soc., 91, 1015–1057, doi:10.1175/2010BAMS3001.1.
Saha, S., and Coauthors, 2014: The NCEP Climate Forecast System version 2. J. Climate, 27, 2185–2208, doi:10.1175/JCLI-D-12-00823.1.
Saunders, M. A., and A. S. Lea, 2005: Seasonal prediction of hurricane activity reaching the coast of the United States. Nature, 434, 1005–1008, doi:10.1038/nature03454.
Stowasser, M., Y. Wang, and K. Hamilton, 2007: Tropical cyclone changes in the western North Pacific in a global warming scenario. J. Climate, 20, 2378–2396, doi:10.1175/JCLI4126.1.
Taylor, K. E., R. J. Stouffer, and G. A. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485–498, doi:10.1175/BAMS-D-11-00094.1.
Vecchi, G. A., M. Zhao, H. Wang, G. Villarini, A. Rosati, A. Kumar, I. M. Held, and R. Gudgel, 2011: Statistical–dynamical predictions of seasonal North Atlantic hurricane activity. Mon. Wea. Rev., 139, 1070–1082, doi:10.1175/2010MWR3499.1.
Vimont, D. J., and J. P. Kossin, 2007: The Atlantic meridional mode and hurricane activity. Geophys. Res. Lett., 34, L07709, doi:10.1029/2007GL029683.
Wang, H., J. K. E. Schemm, A. Kumar, W. Wang, L. Long, M. Chelliah, G. D. Bell, and P. Peng, 2009: A statistical forecast model for Atlantic seasonal hurricane activity based on the NCEP dynamical seasonal forecast. J. Climate, 22, 4481–4500, doi:10.1175/2009JCLI2753.1.
Xie, L., T. Yan, L. J. Pietrafesa, J. M. Morrison, and T. Karl, 2005: Climatology and interannual variability of North Atlantic hurricane tracks. J. Climate, 18, 5370–5381, doi:10.1175/JCLI3560.1.
