• Allen, J., and D. Karoly, 2014: A climatology of Australian severe thunderstorm environments 1979–2011: Inter-annual variability and the ENSO influence. Int. J. Climatol., 34, 8197, doi:10.1002/joc.3667.

    • Search Google Scholar
    • Export Citation
  • Allen, J., and K. Walsh, 2014: Future Australian severe thunderstorm environments. Part I: A novel evaluation and climatology of convective parameters from two climate models for the late twentieth century. J. Climate, 27, 38273847, doi:10.1175/JCLI-D-13-00425.1.

    • Search Google Scholar
    • Export Citation
  • Allen, J., D. Karoly, and G. Mills, 2011: A severe thunderstorm climatology for Australia and associated thunderstorm environments. Aust. Meteor. Oceanogr. J., 61, 143158.

    • Search Google Scholar
    • Export Citation
  • Allen, J., D. Karoly, and K. Walsh, 2014: Future Australian severe thunderstorm environments. Part II: The influence of a strongly warming climate on convective environments. J. Climate, 27, 38483868, doi:10.1175/JCLI-D-13-00426.1.

    • Search Google Scholar
    • Export Citation
  • Anderson, B. T., K. Hayhoe, and X. Liang, 2010: Anthropogenic-induced changes in the 21st century summertime hydroclimatology of the Northeastern US. Climatic Change, 99, 403423, doi:10.1007/s10584-009-9674-3.

    • Search Google Scholar
    • Export Citation
  • Bentsen, M., and Coauthors, 2013: The Norwegian Earth System Model, NorESM1-M—Part 1: Description and basic evaluation of the physical climate. Geosci. Model Dev., 6, 687720, doi:10.5194/gmd-6-687-2013.

    • Search Google Scholar
    • Export Citation
  • Brooks, H. E., 2013: Severe thunderstorms and climate change. Atmos. Res., 123, 129138, doi:10.1016/j.atmosres.2012.04.002.

  • Brooks, H. E., J. W. Lee, and J. P. Craven, 2003: The spatial distribution of severe thunderstorm and tornado environments from global reanalysis data. Atmos. Res., 67-68, 7394, doi:10.1016/S0169-8095(03)00045-0.

    • Search Google Scholar
    • Export Citation
  • Brooks, H. E., G. W. Carbin, and P. T. Marsh, 2014: Increased variability of tornado occurrence in the United States. Science, 346, 349352, doi:10.1126/science.1257460.

    • Search Google Scholar
    • Export Citation
  • Del Genio, A. D., M. Yao, and J. Jonas, 2007: Will moist convection be stronger in a warmer climate? Geophys. Res. Lett., 34, L16703, doi:10.1029/2007GL030525.

    • Search Google Scholar
    • Export Citation
  • Diffenbaugh, N. S., M. Scherer, and R. J. Trapp, 2013: Robust increases in severe thunderstorm environments in response to greenhouse forcing. Proc. Natl. Acad. Sci. USA, 110, 16 36116 366, doi:10.1073/pnas.1307758110.

    • Search Google Scholar
    • Export Citation
  • Donner, L. J., and Coauthors, 2011: The dynamical core, physical parameterizations, and basic simulation characteristics of the atmospheric component AM3 of the GFDL global coupled model CM3. J. Climate, 24, 34843519, doi:10.1175/2011JCLI3955.1.

    • Search Google Scholar
    • Export Citation
  • Dunne, J. P., and Coauthors, 2012: GFDL’s ESM2 global coupled climate–carbon Earth system models. Part I: Physical formulation and baseline simulation characteristics. J. Climate, 25, 66466665, doi:10.1175/JCLI-D-11-00560.1.

    • Search Google Scholar
    • Export Citation
  • Dunne, J. P., and Coauthors, 2013: GFDL’s ESM2 global coupled climate–carbon Earth system models. Part II: Carbon system formulation and baseline simulation characteristics. J. Climate, 26, 22472267, doi:10.1175/JCLI-D-12-00150.1.

    • Search Google Scholar
    • Export Citation
  • Efron, B., 1987: Better bootstrap confidence intervals. J. Amer. Stat. Assoc., 82, 171185, doi:10.1080/01621459.1987.10478410.

  • Gent, P. R., and Coauthors, 2011: The Community Climate System Model version 4. J. Climate, 24, 49734991, doi:10.1175/2011JCLI4083.1.

    • Search Google Scholar
    • Export Citation
  • Hayhoe, K., and Coauthors, 2007: Past and future changes in climate and hydrological indicators in the US Northeast. Climate Dyn., 28, 381407, doi:10.1007/s00382-006-0187-8.

    • Search Google Scholar
    • Export Citation
  • Hayhoe, K., and Coauthors, 2008: Regional climate change projections for the Northeast USA. Mitig. Adapt. Strategies Global Change, 13, 425436, doi:10.1007/s11027-007-9133-2.

    • Search Google Scholar
    • Export Citation
  • Held, I. M., and B. J. Soden, 2006: Robust responses of the hydrological cycle to global warming. J. Climate, 19, 56865699, doi:10.1175/JCLI3990.1.

    • Search Google Scholar
    • Export Citation
  • Kunkel, K. E., and Coauthors, 2013: Monitoring and understanding trends in extreme storms: State of knowledge. Bull. Amer. Meteor. Soc., 94, 499514, doi:10.1175/BAMS-D-11-00262.1.

    • Search Google Scholar
    • Export Citation
  • Li, H., and B. A. Colle, 2014: Multidecadal changes in the frequency and ambient conditions of warm season convective storms over the northeastern United States. J. Climate, 27, 72857300, doi:10.1175/JCLI-D-13-00785.1.

    • Search Google Scholar
    • Export Citation
  • Liang, X., J. Pan, J. Zhu, K. E. Kunkel, J. X. Wang, and A. Dai, 2006: Regional climate model downscaling of the U.S. summer climate and future change. J. Geophys. Res., 111, D10108, doi:10.1029/2005JD006685.

    • Search Google Scholar
    • Export Citation
  • Lombardo, K., B. A. Colle, and Z. Zhang, 2015: Evaluation of historical and future cool season precipitation over the eastern United States and western Atlantic storm track using CMIP5 models. J. Climate, 28, 451467, doi:10.1175/JCLI-D-14-00343.1.

    • Search Google Scholar
    • Export Citation
  • Murray, J. C., and B. A. Colle, 2011: A climatology of convective storms over the northeast United States. Mon. Wea. Rev., 139, 9921012, doi:10.1175/2010MWR3316.1.

    • Search Google Scholar
    • Export Citation
  • Roman, J., R. Knuteson, S. Ackerman, and H. Revercomb, 2015: Predicted changes in the frequency of extreme precipitable water vapor events. J. Climate, 28, 70577070, doi:10.1175/JCLI-D-14-00679.1.

    • Search Google Scholar
    • Export Citation
  • Seeley, J. T., and D. M. Romps, 2015: The effect of global warming on severe thunderstorms in the United States. J. Climate, 28, 24432458, doi:10.1175/JCLI-D-14-00382.1.

    • Search Google Scholar
    • Export Citation
  • Semenov, V. A., and L. Bengtsson, 2002: Secular trends in daily precipitation characteristics: Greenhouse gas simulations with a coupled AOGCM. Climate Dyn., 19, 123140, doi:10.1007/s00382-001-0218-4.

    • Search Google Scholar
    • Export Citation
  • Taylor, K. E., R. J. Stouffer, and A. G. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485498, doi:10.1175/BAMS-D-11-00094.1.

    • Search Google Scholar
    • Export Citation
  • Tippett, M., J. T. Allen, V. A. Gensini, and H. E. Brooks, 2015: Climate and hazardous convective weather. Curr. Climate Change Rep., 1, 6073, doi:10.1007/s40641-015-0006-6.

    • Search Google Scholar
    • Export Citation
  • Trapp, R. J., N. S. Diffenbaugh, H. E. Brooks, M. E. Baldwin, E. D. Robinson, and J. S. Pal, 2007: Changes in severe thunderstorm environment frequency during the 21st century caused by anthropogenically enhanced global radiative forcing. Proc. Natl. Acad. Sci. USA, 104, 19 71919 723, doi:10.1073/pnas.0705494104.

    • Search Google Scholar
    • Export Citation
  • Van Klooster, S. L., and P. J. Roebber, 2009: Surface-based convective potential in the contiguous United States in a business-as-usual future climate. J. Climate, 22, 33173330, doi:10.1175/2009JCLI2697.1.

    • Search Google Scholar
    • Export Citation
  • Voldoire, A., and Coauthors, 2013: The CNRM-CM5.1 global climate model: Description and basic evaluation. Climate Dyn., 40, 20912121, doi:10.1007/s00382-011-1259-y.

    • Search Google Scholar
    • Export Citation
  • Watanabe, S., and Coauthors, 2011: MIROC-ESM 2010: Model description and basic results of CMIP5-20c3m experiments. Geosci. Model Dev., 4, 845872, doi:10.5194/gmd-4-845-2011.

    • Search Google Scholar
    • Export Citation
  • Wu, T., and Coauthors, 2014: An overview of BCC climate system model development and application for climate change studies. J. Meteor. Res., 28, 3456, doi:10.1007/s13351-014-3041-7.

    • Search Google Scholar
    • Export Citation
  • View in gallery

    The NEUS domain showing the two regions (labeled R1 and R2), in which data from individual grid points in the CMIP5 models are analyzed. The crosses show the locations of the individual model grid points used for each region. The rectangles labeled LC1 and LC2 correspond to regions 1 and 5 in LC14, respectively, and are shown for reference. Approximate surface elevation in meters from the NARR is contoured every 100 m.

  • View in gallery

    (a) The annual number of CE days according to a composite of the NARR and CFSR over the historical period (1979–2005) using the thresholds in LC14 in region LC1 (blue line), and the two thresholds in Table 1 in region 1 (red line for threshold with TDIFF58, green line for threshold with LI). (b) As in (a), but for region LC2 and region 2 thresholds.

  • View in gallery

    The annual number of warm season CE days over the historical period (1979–2005) (a) in each of the CMIP5 models in region 1 and (b) averaged across the seven CMIP5 models in region 1 compared with the NARR and CFSR data in LC1. (c) As in (a), but for region 2; (d) as in (b), but for region 2 and compared with reanalysis data in LC2. The error bars in (b) and (d) show the variation (plus or minus one standard deviation) among the seven climate models. Trend lines significant at the 0.05 level are shown.

  • View in gallery

    Scatterplots of annual warm season average PW58 for regions (a) 1 and (b) 2 vs annual warm season CE day frequency over the historical period (1979–2005) in each of the CMIP5 models and a composite of the NARR and CFSR. (c),(d) As in (a),(b), but for TDIFF58; (e),(f) as in (a),(b), but for ω. Trend lines represent relationships between the variables averaged over all CMIP5 models.

  • View in gallery

    (a) Annual average warm season TDIFF58 over the historical period (1979–2005) for each of the seven models in region 1. (b) Average of the CMIP5 models in region 1 compared to a composite of the NARR and CFSR. (c),(d) As in (a),(b), but for region 2. Error bars in (b) and (d) show the variation (plus or minus one standard deviation) among the seven climate models.

  • View in gallery

    As in Fig. 5, but for PW58.

  • View in gallery

    (a) The annual number of warm season CE days over the future period (2006–99) for each of the seven CMIP5 models in region 1. (b) Average warm season CE days for all seven models in region 1. (c),(d) As in (a),(b), but for region 2. Error bars in (b) and (d) show the variation (plus or minus one standard deviation) among the seven climate models.

  • View in gallery

    As in Fig. 4, but for the future period (2006–99).

  • View in gallery

    As in Fig. 7, but for PW58.

  • View in gallery

    (a) Warm season average TDIFF58 (shaded; K) over the period 1976–2005. (b) The percent change (color shaded) in warm season average TDIFF58 between 1976–2005 and 2066–95. (c) As in (a), but for PW58 (mm); (d) as in (b), but for percent change in warm season average PW58. The black contours show the variability among the various models (plus or minus one standard deviation).

  • View in gallery

    As in Fig. 10, but for TDIFF58.

  • View in gallery

    (a) Annual average warm season ω (−Pa s−1), 1976–2005 (shaded) with the plus or minus one standard deviation contoured. (b) The change in average warm season ω (−Pa s−1) between 1976–2005 and 2066–95.

  • View in gallery

    (a) The change in warm season average 700-hPa temperature (shaded; K) between 1976–2005 and 2066–95 averaged across the seven models. The contours show the variation (one standard deviation) among the CMIP5 models. (b) Warm season average SHEAR58 (shaded; m s−1) over the historical period (1976–2005). (c) The percent change (shaded) in warm season average SHEAR58 between 1976–2005 and 2006–95.

  • View in gallery

    The average rate of annual temperature increase (K yr−1) over region 1 during the (a) historical period and (c) future period (2066–99) at 850, 700, and 500 hPa. (b),(d) As in (a),(c), but for region 2. Error bars show a bias-corrected and accelerated bootstrapped 95% confidence interval.

  • View in gallery

    (a) Standard deviation of the average PW85 (mm) for the future period (2006–99) averaged for all CMIP5 models in this study over regions 1 and 2. (b) Estimated probability density function of daily warm season PW85 (mm) for 2006–25 and 2080–99 in region 1.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 158 109 0
PDF Downloads 98 68 0

Future Changes in Convective Storm Days over the Northeastern United States Using Linear Discriminant Analysis Applied to CMIP5 Predictions

View More View Less
  • 1 Harvard University, Cambridge, Massachusetts
  • | 2 School of Marine and Atmospheric Sciences, Stony Brook University, Stony Brook, New York
Full access

Abstract

Future changes in the frequency of environmental conditions conducive for convective storm days (“CE days”) are determined for the northeastern United States (NEUS) during the warm seasons (April–September) of the twenty-first century. Statistical relationships between historical runs of seven models in phase 5 of the Coupled Model Intercomparison Project (CMIP5) and radar-classified convective storm days are developed using linear discriminant analysis (LDA), and these relationships are then applied to analyze changes in the convective environment under the high-emissions representative concentration pathway 8.5 (RCP8.5) scenario over the period 2006–99. The 1996–2007 warm seasons are used to train the LDA thresholds using convective precipitation from two reanalysis datasets and radar data, and the 1979–95 and 2008–10 warm seasons are used to verify these thresholds. For the CMIP5 historical period (1979–2005), the frequency of warm season CE days averaged across the CMIP5 models is slightly greater than that derived using reanalysis data, although both methods indicate a slight increasing trend through the historical period. Between 2006 and 2099, warm season CE day frequency is predicted to increase substantially at an average rate of 4–5 days decade−1 (50%–80% increase over the entire period). These changes are mostly attributed to a predicted 30%–40% increase in midlevel precipitable water between the historical period and the last few decades of the twenty-first century. Consistent with previous studies, there is decreasing deep-layer vertical wind shear as a result of a weakening horizontal temperature gradient, but this is outweighed by increases in instability led by the moisture increases.

Corresponding author address: Dr. Brian A. Colle, School of Marine and Atmospheric Sciences, Stony Brook University/SUNY, Stony Brook, NY 11794-5000. E-mail: brian.colle@stonybrook.edu

Abstract

Future changes in the frequency of environmental conditions conducive for convective storm days (“CE days”) are determined for the northeastern United States (NEUS) during the warm seasons (April–September) of the twenty-first century. Statistical relationships between historical runs of seven models in phase 5 of the Coupled Model Intercomparison Project (CMIP5) and radar-classified convective storm days are developed using linear discriminant analysis (LDA), and these relationships are then applied to analyze changes in the convective environment under the high-emissions representative concentration pathway 8.5 (RCP8.5) scenario over the period 2006–99. The 1996–2007 warm seasons are used to train the LDA thresholds using convective precipitation from two reanalysis datasets and radar data, and the 1979–95 and 2008–10 warm seasons are used to verify these thresholds. For the CMIP5 historical period (1979–2005), the frequency of warm season CE days averaged across the CMIP5 models is slightly greater than that derived using reanalysis data, although both methods indicate a slight increasing trend through the historical period. Between 2006 and 2099, warm season CE day frequency is predicted to increase substantially at an average rate of 4–5 days decade−1 (50%–80% increase over the entire period). These changes are mostly attributed to a predicted 30%–40% increase in midlevel precipitable water between the historical period and the last few decades of the twenty-first century. Consistent with previous studies, there is decreasing deep-layer vertical wind shear as a result of a weakening horizontal temperature gradient, but this is outweighed by increases in instability led by the moisture increases.

Corresponding author address: Dr. Brian A. Colle, School of Marine and Atmospheric Sciences, Stony Brook University/SUNY, Stony Brook, NY 11794-5000. E-mail: brian.colle@stonybrook.edu

1. Introduction

Climate change is projected to have a substantial impact on the northeastern United States (NEUS) during the next 50–100 years. For example, using historical observations and downscaling a global climate model (GCM) with the MM5 regional climate model, Liang et al. (2006) found that between 1986–95 and 2041–50 the mean summer (JJA) surface temperature over the NEUS is expected to increase by 0.5°–1.0°C. Hayhoe et al. (2008) used statistical downscaling of three coupled atmosphere–ocean general circulation models (AOGCMs) to highlight a 3.9°–4.3°C increase in average daily high temperatures between the 1990s and 2090s over the NEUS, with the greatest increases occurring in the northernmost part of the region. Anderson et al. (2010) used a similar approach with regional model simulations driven by AOGCMs and the A1fi high-emissions scenario, which yielded a conservative estimate of a 3.5°C increase in mean summer temperatures and a 350%–400% increase in the frequency of days with a heat index over 90°F (32.2°C) over the NEUS through the end of the twenty-first century.

In particular, a substantive increase in mean temperature leads to an increase in saturation vapor pressure at a rate of approximately 7% K−1 according to the Clausius–Clapeyron relation (Held and Soden 2006), which is linked to marked increases in precipitation. For example, Semenov and Bengtsson (2002) project a 10%–20% increase in precipitation through the twenty-first century over the NEUS as a result of warming temperatures, with most of this increase expected to occur during the cool season (Hayhoe et al. 2007; Kunkel et al. 2013). Meanwhile, Roman et al. (2015) suggest that the frequency of days with extremely high precipitable water vapor over the eastern United States is expected to increase fastest during meteorological summer. Although precipitation amounts may increase during the upcoming century, using data from the GCM to create artificial neural networks Van Klooster and Roebber (2009) predicted no significant increases in the potential for overall surface-based daytime convective initiation over the entire continental United States (including the NEUS) through the end of the twenty-first century, while they expected the potential for severe convection to increase by more than one standard deviation relative to the end of the twentieth century over the NEUS over the same period, largely due to a predicted increase in convective available potential energy (CAPE).

The use of atmospheric parameters to examine and explain historical and future convective storm trends is useful due to the inability of current climate models to adequately resolve convection. For example, Brooks et al. (2003) created a linear threshold based on CAPE and deep-layer (0–6 km) vertical wind shear according to “pseudo soundings” from the National Centers for Environmental Prediction (NCEP)–National Center for Atmospheric Research (NCAR) reanalysis to predict the occurrence of severe convection. Using a statistical relationship based on the product of CAPE and 0–6-km vertical wind shear derived from this study, Trapp et al. (2007) predict a substantial increase in the frequency of days with environments conducive to severe thunderstorms over the NEUS. In particular, the frequency of severe convective environment days in New York City is expected to more than double between 1962–89 and 2072–99. This increase is largely attributed to substantial increases in CAPE from enhanced boundary layer moisture, while any decrease in deep-layer vertical shear is not sufficiently large to offset the instability increases. More recently, Diffenbaugh et al. (2013) built on this work further using the high-emissions [representative concentration pathway 8.5 (RCP8.5)] run of models from phase 5 of the Coupled Model Intercomparison Project (CMIP5; Taylor et al. 2012) to project increases in the frequency of severe thunderstorm environments over the entire United States through the end of the twenty-first century, with the NEUS in particular showing the highest increases during spring and summer. The increased favorability of the environment for severe thunderstorms is expected to be primarily driven by the approximately 100–200 J kg−1 increases in CAPE overcoming the approximately 0.5–1.5 m s−1 decrease in 0–6-km vertical shear, with most of the vertical shear decreases occurring primarily on days with low CAPE, decreasing the apparent importance of vertical shear in characterizing future convective environments. Similar results were obtained looking at future changes in severe thunderstorm environments in other regions around the world, such as Australia (Allen et al. 2014; Allen and Walsh 2014), while Allen and Karoly (2014) showed that there is relatively large interannual variability in Australian severe storms associated with ENSO.

Allen et al. (2011) used linear discriminant analysis (LDA) to distinguish between severe and nonsevere convective events in Australia also based on CAPE and 0–6-km vertical shear, finding modest accuracy improvements over the Brooks et al. (2003) threshold. Li and Colle (2014, hereafter LC14) used LDA to relate the frequency of warm season (1 April–30 September) convective storm days defined using radar reflectivity data with daily averaged parameters from the North American Regional Reanalysis (NARR) and the Climate Forecast System Reanalysis (CFSR) over five small subregions of the NEUS for a 32-yr period (1979–2010), finding probabilities of detection greater than 80% and false alarm rates less than 20% in all regions. This LDA approach predicted that convective storm days are less frequent at the coast than inland, consistent with other previous studies (e.g., Murray and Colle 2011), and that warm season convective storm day frequency has increased at the coast but decreased inland, largely due to increases in low-level instability driven by increases in boundary layer moisture.

Although convective storms can have a significant impact on life and property over the NEUS due to the region’s high population density (Del Genio et al. 2007), there have been relatively few previous studies on future changes in convective storms over the region. Many studies investigating future trends in convection have examined the full continental United States, therefore only giving limited information of how convective storms may change over the NEUS. To address this, expanding on the LDA approach used in LC14, this study uses several climate models from the CMIP5 to project future changes in warm season (1 April–30 September) convective storm day frequency over the NEUS for the business-as-usual (high emissions) scenario. Relevant information about individual model biases and intermodel variability will also be explored.

Section 2 details the data used along with the methodology for defining convective environment days. Section 3 first delineates trends in convective environment frequency over the historical runs of a set of climate models, which are compared to the reanalysis results obtained by LC14. Afterward, projected convective environment trends over the remainder of the twenty-first century using these same models are analyzed along with the associated atmospheric changes. Finally, section 4 summarizes the key results and provides directions for future research.

2. Data and methods

Seven climate models from the CMIP5 were chosen to yield a variety of horizontal resolutions based on above average performance for northeast U.S. precipitation (Lombardo et al. 2015). Specifications and references for these models can be found in Table 1. The temporal domain of this study is separated into a historical period spanning 1979–2005, for which the historical run of the models is used, and a future period spanning 2006–99, for which the RCP8.5 run is used. Spatially, the NEUS is defined in this study as the region bounded by 36°–46°N and 83°–66°W (Fig. 1). Data at model grid points located within two small regions of the NEUS (labeled 1 and 2 in Fig. 1) are used to present representative point analysis of the long-term trends over inland regions and coastal ocean, respectively. Region 1 is located immediately west of the Appalachian crest in southwestern Pennsylvania and northern West Virginia, and corresponds roughly to region 1 in LC14 (region LC1), while region 2 is located over the coastal ocean immediately east of the New Jersey coastline and south of Long Island, covering roughly the same area as region 5 in LC14 (region LC2). These two domains are chosen to provide for meaningful comparison to LC14, although it should be noted that due to the significantly coarser resolution of the CMIP5 models compared to the high-resolution reanalyses used in LC14, the regions in this study do not correspond exactly to those in LC14 (Fig. 1).

Table 1.

Information on the seven CMIP5 models used in this study. (Expansions of acronyms are available online at http://www.ametsoc.org/PubsAcronymList.)

Table 1.
Fig. 1.
Fig. 1.

The NEUS domain showing the two regions (labeled R1 and R2), in which data from individual grid points in the CMIP5 models are analyzed. The crosses show the locations of the individual model grid points used for each region. The rectangles labeled LC1 and LC2 correspond to regions 1 and 5 in LC14, respectively, and are shown for reference. Approximate surface elevation in meters from the NARR is contoured every 100 m.

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

Convective environment days (“CE days”) are defined in regions 1 and 2 (Fig. 1) by thresholds created using LDA on daily-averaged atmospheric parameters (detailed below). These thresholds attempt to separate those warm season days with significant convective precipitation from all other warm season days. To develop these thresholds, first warm season convective precipitation “counts” are defined for each warm season day from 1979–2010 in regions LC1 and LC2 by summing at 6-hourly intervals the number of grid points in each region recording appreciable convective precipitation according to accumulated convective precipitation data averaged between the NARR and CFSR. All days (starting and ending at 0000 UTC) in which this count exceeded a given threshold were classified as convective precipitation (CP) days. This threshold was determined by normalizing the count to National Operational Weather Radar (NOWrad) reflectivity data, as described in LC14. Similar to LC14, the warm seasons from 1996 to 2007 were used to train the LDA thresholds, while the warm seasons during 1979–95 and 2008–10 were used for verification. Further details on the approach for defining CE days can be found in LC14. The final thresholds used in this study are listed in Table 2, along with their accuracies in terms of false alarm rate and probability of detection. The false alarm rate is defined as the probability that a randomly selected non-CP day is classified as CE day, while the probability of detection is defined as the chance that a randomly selected CP day is also a CE day (a lower false alarm rate and a higher probability of detection indicate a more accurate threshold).

Table 2.

(top) The threshold discriminators used in regions 1 and 2 (Fig. 1) to predict convective environment days based on TDIFF58, PW58, and ω (see section 2 for an explanation of these abbreviations). (bottom) Alternative threshold discriminators replacing TDIFF58 with LI. False alarm rate and probability of detection are defined in the text.

Table 2.

The variables used in the thresholds in Table 2 are somewhat different than those used in LC14 due to limitations in the selected daily data in CMIP5. For instance, the full-atmosphere precipitable water variable used in LC14 could not be used here due to the limited vertical resolution of the daily CMIP5 dataset (there are only five levels below 500 hPa). Thus, the precipitable water term was replaced by an 850–500-hPa integrated moisture term (in mm), defined as a mass-weighted vertical integral of daily averaged specific humidity (in g kg−1) between the 850- and 500-hPa levels estimated using a midpoint Riemann sum on data at 850, 700, and 500 hPa (hereafter PW58). While previous studies including Diffenbaugh et al. (2013) and Seeley and Romps (2015) used a CAPE–shear product in attempting to relate CMIP5 model output to convection, here we replace the CAPE–shear term by an 850–500-hPa temperature difference (in K), defined as the daily average 850-hPa temperature minus the daily average 500-hPa temperature (hereafter TDIFF58). This replacement is made since the CMIP5 CAPE values in the NEUS are unrealistically low in several models (e.g., warm season average CAPE values in region 1 for the BCC_CSM1.1 during the mid-twenty-first century are well under 100 J kg−1). The reason for this discrepancy is likely due to the low vertical resolution of the CMIP5 models, which makes it necessary to estimate CAPE based on just 5–6 vertical levels of pressure and moisture data. We considered the lifted index (LI) as another possible replacement for CAPE. However, while LI has the advantage of being a more direct measure of instability than TDIFF58, replacing TDIFF58 with LI in the thresholds led to very small differences in CE day frequency in both the reanalysis (Fig. 2) and the CMIP5 (not shown). Furthermore, unlike LI, TDIFF58 is relatively uncorrelated with PW58, making TDIFF58 a better metric for more precisely attributing changes in CE day frequency to specific changes in ingredients. Nonetheless, thresholds using LI in lieu of TDIFF58 are given in Table 2 for completeness. Vertical deep-layer wind shear is not included as a parameter for simplicity, as adding it to the thresholds in region 1 led to very negligible changes. The irrelevance of vertical shear in the thresholds can be attributed to the fact that its role is to primarily strengthen or organize developing convective storms, making it much more useful in discriminating between severe and nonsevere storms, as has been done by many previous studies, such as Brooks et al. (2003). However, our study focuses on the favorability of the environment to convective storms of all intensities rather than just focusing on severe convection, which is rare over the NEUS. The pressure vertical velocity term averaged at 500 and 700 hPa (ω; negative values denote upward motion in Pa s−1) is left unchanged from LC14.

Fig. 2.
Fig. 2.

(a) The annual number of CE days according to a composite of the NARR and CFSR over the historical period (1979–2005) using the thresholds in LC14 in region LC1 (blue line), and the two thresholds in Table 1 in region 1 (red line for threshold with TDIFF58, green line for threshold with LI). (b) As in (a), but for region LC2 and region 2 thresholds.

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

Applying the parameters and thresholds in Table 2 to a composite of the NARR and CFSR data did not substantially affect the trends in CE day frequency as compared to the LC14 results over the same historical period (Fig. 2a). Although the annual frequency of CE days in region 1 averages 5%–10% greater than for the corresponding region and approach in LC14, with the difference in results increasing somewhat toward the later part of the historical period, there is a relatively robust correlation between the results obtained with the two thresholds (r = 0.84) and neither threshold shows a statistically significant trend with time (Fig. 2a). Meanwhile, in region 2 the CE day frequency is very close to LC14 over the entire historical period, with a correlation of r = 0.97 (Fig. 2b). It should be emphasized that the occurrence of a CE day in a given region should not be literally equated with the occurrence of thunderstorms in that region on that day. Comparing CE day frequency with radar reflectivity frequency data in LC14, the number of CE days is about 50% greater than the number of CP days. Therefore, CE day frequency should be interpreted as the likelihood of days with environmental conditions conducive to the development of convection rather than as a direct indicator of the actual number of days expected to record convection.

The statistical significance of annual trends in CE day frequency and other warm season averaged atmospheric variables in regions 1 and 2 is tested using bias-corrected and accelerated (BCa) bootstrapped confidence intervals with resampling size n = 2000 (Efron 1987), where the null hypothesis is that the slope of the least squares regression line predicting the given variable as a function of the year is equal to zero. Unless otherwise indicated, all trends labeled as statistically significant indicate a difference from the null hypothesis at the 0.05 significance level (two tailed), and all confidence intervals also use α = 0.05. The bootstrapping approach used is favored over higher-power parametric methods such as the Student’s t test on the slope of a linear regression because it is difficult to verify that the limited data in this study satisfy normality assumptions, especially over the historical period, which spans only 27 warm seasons (1979–2005).

3. Results

a. Historical period

During the historical period (1979–2005), the increase in annual warm season CE day frequency according to an average of the seven CMIP5 models falls just short of a statistically significant increase at the 10% level in region 1 (Fig. 3b), while the average CE day frequency across the seven models in region 2 has increased significantly at the 5% level (Fig. 3d). These results are qualitatively consistent with the results obtained in LC14 using the NARR and CFSR (Figs. 3b,d). Average annual warm season CE day frequency was 77 days yr−1 in region 1 compared to 57 days per warm season in region 2. For context, the total length of the warm season is 183 days. These values are somewhat higher than the corresponding values obtained using NARR/CFSR data (67 days yr−1 in region 1 and 51 days yr−1 in region 2).

Fig. 3.
Fig. 3.

The annual number of warm season CE days over the historical period (1979–2005) (a) in each of the CMIP5 models in region 1 and (b) averaged across the seven CMIP5 models in region 1 compared with the NARR and CFSR data in LC1. (c) As in (a), but for region 2; (d) as in (b), but for region 2 and compared with reanalysis data in LC2. The error bars in (b) and (d) show the variation (plus or minus one standard deviation) among the seven climate models. Trend lines significant at the 0.05 level are shown.

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

There is a relatively large spread in the number of CE days for the different CMIP5 models during the historical period, especially in region 1 (Fig. 3a), with the MIROC-ESM-CHEM predicting an average of 105 CE days per warm season while the GFDL CM3 predicts only 55 annual CE days on average. The spread is less in region 2 (Fig. 3c), with roughly a 40% difference between the GFDL CM3 (46 CE days yr−1) and the BCC_CSM1.1 (69 days yr−1). The variability in CE day frequency between the models can largely be attributed to significant intermodel differences in PW58 and TDIFF58. For example, annual warm season average PW58 over the historical period ranged from 11.2 mm in the GFDL CM3 to 14.3 mm in the MIROC-ESM-CHEM in region 1 (Fig. 4a) and from 10.8 mm in the GFDL-ESM2M to 13.1 mm in the MIROC-ESM-CHEM in region 2 (Fig. 4b). Similarly, annual warm season average TDIFF58 during the period 1979–2005 ranged from 24.1 K in the GFDL CM3 to 29.3 K in the MIROC-ESM-CHEM in region 1 (Fig. 4c) and from 24.5 K in the GFDL CM3 to 28.4 K in the MIROC-ESM-CHEM in region 2 (Fig. 4d). In general, those models with higher PW58 and TDIFF58 (especially the former) tend to predict a higher frequency of CE days. The frequency of warm season CE days in the CMIP5 composite is about 15% and 12% greater in regions 1 and 2, respectively, than in the reanalysis composite (Figs. 3b,d). This is related to the CMIP5 composite showing higher warm season average TDIFF58 than the NARR or CFSR (the difference is roughly 11% in region 1 and 12% in region 2) along with marginally greater PW58 (Figs. 5b,d and 6b,d).

Fig. 4.
Fig. 4.

Scatterplots of annual warm season average PW58 for regions (a) 1 and (b) 2 vs annual warm season CE day frequency over the historical period (1979–2005) in each of the CMIP5 models and a composite of the NARR and CFSR. (c),(d) As in (a),(b), but for TDIFF58; (e),(f) as in (a),(b), but for ω. Trend lines represent relationships between the variables averaged over all CMIP5 models.

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

Fig. 5.
Fig. 5.

(a) Annual average warm season TDIFF58 over the historical period (1979–2005) for each of the seven models in region 1. (b) Average of the CMIP5 models in region 1 compared to a composite of the NARR and CFSR. (c),(d) As in (a),(b), but for region 2. Error bars in (b) and (d) show the variation (plus or minus one standard deviation) among the seven climate models.

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

Fig. 6.
Fig. 6.

As in Fig. 5, but for PW58.

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

Given the relatively coarse spatial resolution of the CMIP5 models and the relatively short length of the historical period (27 warm seasons), the slight differences in the CE day frequency trends between regions 1 and 2 over the historical period cannot be substantively attributed to differences in any particular factor in the LDA thresholds, as all three of the parameters used show similar interannual trends between the two regions from 1979 to 2005. For example, warm season average TDIFF58 does not show any significant change over the historical period in either region in the CMIP5 mean (Fig. 5b), nor are there any significant trends in any individual model (Fig. 5a), while mean PW58 has increased at 0.254 ± 0.136 mm decade−1 in region 1 and at 0.224 ± 0.116 mm decade−1 in region 2 (Figs. 6b,d), and most models illustrate an increase (Figs. 6a,c). Vertical motion (ω) became significantly more upward in region 2 in the GFDL-ESM2M, but not in any other model in region 2 or in any model in region 1 (not shown). However, even with the temporal limitations associated with the historical period of this study, it is evident that the moisture term (PW58) is the dominant factor dictating trends in CE days. For example, the relationship between the annual number of warm season CE days and warm season average PW58 for each year during the historical period averaged across all models in region 1 is given by r2 = 0.81 (Fig. 4a), compared to r2 = 0.07 for TDIFF58 and r2 = 0.18 for ω (Figs. 4c,e), while in region 2 is given by r2 = 0.81 for PW58, 0.00 for TDIFF58, and 0.25 for ω (Figs. 4b,d,f). Note that the low correlation coefficients for TDIFF58 do not necessarily indicate that TDIFF58 is a poor predictor variable; they merely suggest that the interannual variability in CE day frequency over the historical period does not track the interannual variability in warm season average TDIFF58.

b. Future trends

For the future period (2006–99), annual CE day frequency increases significantly in all models in both NEUS regions (Fig. 7). The model composite change in annual CE day frequency over the period is 4.23 ± 0.25 days decade−1 in region 1 and 4.48 ± 0.25 days decade−1 in region 2, and the annual average warm season frequency of CE days between 2006 and 2099 is 100 days yr−1 in region 1 and 79 days yr−1 in region 2 (Fig. 7). As during the historical period, there is substantial spread among the different models in terms of both the magnitude of the upward frequency trend and the overall average predicted frequency of CE days. In terms of the latter, the MIROC-ESM-CHEM is a clear outlier, especially in region 1, where it predicts an average of 133 CE days per warm season (73% of the maximum days possible), while the other six models fall in the range of 83–107 CE days yr−1 (45%–58% of the maximum days possible). The MIROC-ESM-CHEM also predicts CE days with the greatest frequency out of all the models in region 2 (averaging ~100 CE days yr−1), especially during the second half of the future period, while the NorESM1-M and GFDL-ESM2M predict CE days least frequently (64 and 67 CE days yr−1, respectively). In terms of the interannual trend in CE days, the GFDL CM3 has the greatest upward trend in region 1 at 6.77 ± 0.54 days decade−1, and the MIROC-ESM-CHEM has the most substantial trend in region 2 at 6.61 ± 0.47 days decade−1.

Fig. 7.
Fig. 7.

(a) The annual number of warm season CE days over the future period (2006–99) for each of the seven CMIP5 models in region 1. (b) Average warm season CE days for all seven models in region 1. (c),(d) As in (a),(b), but for region 2. Error bars in (b) and (d) show the variation (plus or minus one standard deviation) among the seven climate models.

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

As with the historical period, differences in predicted CE day frequency among the various models over the future period can largely be attributed to differences in PW58 and TDIFF58 between the models. For example, the MIROC-ESM-CHEM is significantly more unstable and moist than the other models. Over the future period, the average warm season TDIFF58 in the MIROC-ESM-CHEM is 2.3 K greater in both regions than in the next most unstable model, the BCC_CSM1.1 (Figs. 8c,d). The average warm season PW58 in the MIROC-ESM-CHEM is 2.4 mm greater in region 1 and 1.6 mm greater in region 2 than in the BCC_CSM1.1 (Figs. 8a,b), which is also the second moistest model after the MIROC-ESM-CHEM in region 2). Similar to the historical period, differences in ω between the models are less substantial than the other parameters (Figs. 8e,f). The predicted CE day frequency trends over the future period are even more closely related to PW58 changes than over the historical period; the relationship between annual CE day frequency and annual warm season average PW58 is r2 = 0.98 in region 1, compared to 0.25 for ω and 0.00 for TDIFF58; in region 2, the variance explained is 0.98 for PW58, 0.11 for ω, and 0.00 for TDIFF58 (Fig. 8).

Fig. 8.
Fig. 8.

As in Fig. 4, but for the future period (2006–99).

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

By the end of the twenty-first century, PW58 averaged across the seven models is expected to increase by 4–5 mm over regions 1 and 2 relative to the historical period (1976–2005), which is roughly a 32%–36% increase (Figs. 9b,d and 10d). The largest percentage changes are expected in the northern part of the domain, while the absolute changes are greatest in the southern part of the domain. The MIROC-ESM-CHEM predicts the greatest PW58 change of all the models over the future period in both regions [averaging 0.90 ± 0.04 mm decade−1 in region 1 and 0.86 ± 0.04 mm decade−1 in region 2 (Figs. 9a,c), or a roughly 50% increase from the historical period to 2066–95], and the GFDL CM3 predicts the second greatest change (0.76 ± 0.05 mm decade−1 in region 1 and 0.75 ± 0.04 mm decade−1 in region 2, or about a 60% increase from the historical period to 2006–95). The other five models all predict changes of 0.40–0.50 mm decade−1 (Figs. 9a,c), although the MIROC-ESM-CHEM and GFDL CM3 bring up the composite increases to 0.57 ± 0.03 mm decade−1 in region 1 and 0.56 ± 0.02 mm decade−1 in region 2.

Fig. 9.
Fig. 9.

As in Fig. 7, but for PW58.

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

Fig. 10.
Fig. 10.

(a) Warm season average TDIFF58 (shaded; K) over the period 1976–2005. (b) The percent change (color shaded) in warm season average TDIFF58 between 1976–2005 and 2066–95. (c) As in (a), but for PW58 (mm); (d) as in (b), but for percent change in warm season average PW58. The black contours show the variability among the various models (plus or minus one standard deviation).

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

TDIFF58 does not show any significant changes over the NEUS for the CMIP5 composite mean (Figs. 10a,b). However, the MIROC-ESM-CHEM shows a significant increase in TDIFF58 in region 1 (Fig. 11a), while 500-hPa temperatures are warming significantly faster than 850-hPa temperatures in both regions in the GFDL-ESM2M, indicating a decrease in TDIFF58 (Fig. 11a). Despite the fact the MIROC-ESM-CHEM shows the greatest increase in TDIFF58 and PW58 in region 1, this model exhibits only a modest trend in CE day frequency over the future period. In particular, the GFDL CM3 shows a noticeably sharper trend in future CE day frequency in region 1, likely because this model has the second-greatest PW58 increase (after the MIROC-ESM-CHEM) combined with the only significant decrease in ω (more upward vertical motion) across all models. Figure 12 shows that in general, vertical motion (ω) changes through the end of the century are not expected to be substantial according to the models, although according to Figs. 8e,f, it seems to help explain some of the interannual variability in CE day frequency.

Fig. 11.
Fig. 11.

As in Fig. 10, but for TDIFF58.

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

Fig. 12.
Fig. 12.

(a) Annual average warm season ω (−Pa s−1), 1976–2005 (shaded) with the plus or minus one standard deviation contoured. (b) The change in average warm season ω (−Pa s−1) between 1976–2005 and 2066–95.

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

The predicted moisture increases are largely a result of temperature increases in the climate models and the Clausius–Clapeyron equation. Since relative humidity does not change substantially over the NEUS domain during the future period according to the CMIP5 models (not shown), this explains the linear relationship between the percent moisture changes (Figs. 10c,d) and the temperature changes in degrees Celsius (Fig. 13a). Temperature increases are significant in all models at all levels, and are relatively similar at all three vertical levels studied, averaging between 4 and 5 K at each level over the NEUS between the historical period and 2066–95 (Fig. 14). The greatest increases occur in the northern part of the domain, consistent with Hayhoe et al. (2008) (e.g., Fig. 13a). This indicates a decrease in the warm season average horizontal temperature gradient, and, consistent with the thermal wind relation, a decrease in the magnitude of the vector difference of horizontal wind between 850 and 500 hPa (hereafter SHEAR58) (Fig. 13c). Over the period 2006–99, SHEAR58 decreases significantly in region 1 at 0.160 ± 0.023 m s−1 decade−1 and in region 2 at 0.137 ± 0.023 m s−1 decade−1. These changes correspond to roughly a 12% decrease between the historical period and the end of the twenty-first century (Fig. 13). The uncertainty for the slope of SHEAR58 is significantly higher than for PW58, which indicates that more time will be needed for a significant trend to emerge from interannual variability. The challenge of separating physically real trends in midlevel winds (on the order of 1 m s−1 century−1) from interannual variability was highlighted by Brooks (2013). Furthermore, it has been shown in several studies that future decreases in vertical shear will likely be more than offset by increases in instability (i.e., convective available potential energy), leading to a net increase in the frequency of both severe and nonsevere convection (e.g., Trapp et al. 2007; Diffenbaugh et al. 2013). In particular, Diffenbaugh et al. (2013) suggest that decreases in average vertical shear come primarily on days with low CAPE, which would lessen the potential impact of vertical shear decreases on offsetting instability increases in the context of convection. This conclusion is supported in our analysis by the fact that on days with high (>500 J kg−1) daily CAPE (calculated using the moisture and temperature data available in the BCC_CSM1.1 at 1000, 925, 850, 700, and 500 hPa), SHEAR58 actually shows a slight increasing trend with time (not shown). Furthermore, the frequency of days with an unstable atmosphere (indicated by a negative lifted index) and relatively high SHEAR58 (greater than 15 m s−1) shows a slight upward trend with time (not shown). Thus, consistent with Diffenbaugh et al. (2013), any mitigating effect of decreasing vertical shear on increases in convective environment frequency due to instability increases may be even smaller than expected.

Fig. 13.
Fig. 13.

(a) The change in warm season average 700-hPa temperature (shaded; K) between 1976–2005 and 2066–95 averaged across the seven models. The contours show the variation (one standard deviation) among the CMIP5 models. (b) Warm season average SHEAR58 (shaded; m s−1) over the historical period (1976–2005). (c) The percent change (shaded) in warm season average SHEAR58 between 1976–2005 and 2006–95.

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

Fig. 14.
Fig. 14.

The average rate of annual temperature increase (K yr−1) over region 1 during the (a) historical period and (c) future period (2066–99) at 850, 700, and 500 hPa. (b),(d) As in (a),(c), but for region 2. Error bars show a bias-corrected and accelerated bootstrapped 95% confidence interval.

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

Brooks et al. (2014) indicated an increase in the variability of severe storm (e.g., tornado) occurrence since the 1970s. An interesting question is whether there may be increased variability in the environment favoring thunderstorms over the northeastern United States. We explored this by illustrating the variability of PW58 (averaged across all CMIP5 models) during the future period (2006–99). There is a clear upward trend in the annual standard deviation of daily PW58 over time in both regions (Fig. 15a). Figure 15b shows the probability density plots of changes in the distribution of daily PW58 between the first 20 years of the period (2006–25) and the last 20 years (2080–99) for region 1, which suggests that this trend is explained by moisture increases during relatively moist days rather than by increased variability in the drier days. This shift in the distribution is similar for region 2 (not shown).

Fig. 15.
Fig. 15.

(a) Standard deviation of the average PW85 (mm) for the future period (2006–99) averaged for all CMIP5 models in this study over regions 1 and 2. (b) Estimated probability density function of daily warm season PW85 (mm) for 2006–25 and 2080–99 in region 1.

Citation: Journal of Climate 29, 12; 10.1175/JCLI-D-14-00831.1

4. Conclusions

Changes in the frequency of warm season days with ambient conditions conducive to convection over the northeastern United States during a baseline historical period (1979–2005) as well as projected changes through the end of the twenty-first century are explored with a modified linear discriminant analysis (LDA) approach used by LC14 and data from seven models in phase 5 of the Climate Model Intercomparison Project (CMIP5). Modifications to the threshold parameters used to define convective environment days in LC14 were made here due to data limitations (temperature difference between 850 and 500 hPa replaced a CAPE–shear product and derived precipitable water between 850 and 500 hPa replaced full column PW while vertical motion averaged at 700 and 500 hPa remained unchanged), but this did not substantially decrease the effectiveness of the thresholds in capturing warm season days with convective precipitation according to radar reflectivity data.

Over the historical period (1979–2005), convective frequency has generally increased, but the interannual (and intermodel) variability is relatively high. This combined with the relatively short time span of the period (27 warm seasons) prevents trends from being statistically significant. In general, convective environment days according to the CMIP5 models over the historical period are roughly 35% more frequent west of the Appalachian Mountains than over the coastal ocean east of the New Jersey coastline, which is qualitatively consistent with LC14. With substantial increases in warm season midlevel temperatures over the NEUS predicted by all the climate models between the historical period and the end of the twenty-first century (averaging roughly 4–5 K across the models with an intermodel standard deviation of 1–1.3 K) along with a corresponding increase in midlevel atmospheric moisture of about 35% (intermodel standard deviation ~10%), warm season CE day frequency over the two regions of the NEUS in Fig. 1 is projected to increase significantly in all of the models over the future period (2006–99). Temperature increases are greatest in the northern portion of the domain, leading to a decrease in the average horizontal temperature gradient and consequently a decrease in vertical wind shear. However, the shear decrease is only a marginal factor in the frequency of convective days. The increase in warm season CE day frequency is somewhat more substantial in region 2 (coastal ocean) than in region 1 (inland NEUS), but overall the average annual frequency of such days in region 2 is still markedly less than in region 1.

The increasing trend in convective environment (CE) day frequency over the future period is mostly modulated by changes in moisture, with a very strong positive relationship between annual average warm season precipitable water within the 500–850-hPa layer and annual warm season CE day frequency, especially over the future period. On the other hand, there is little to no correlation between annual warm season average midlevel vertical temperature differences and CE day frequency. This is attributed to the fact that midlevel temperatures are expected to rise at roughly the same rate as low-level temperatures, thereby leading to no significant change in midlevel lapse rates, while steady increases in moisture drive similar increases in the frequency of CE days. While vertical motion is not expected to drastically change over the future period, it is useful in explaining some of the interannual variability in CE day frequencies over the future period. Also, although a slight decrease in the horizontal temperature gradient is expected through the end of the twenty-first century, thereby leading to a small decrease in vertical wind shear by the thermal wind relation, unstable days with relatively high shear are actually expected to increase somewhat over the future period, so any decreases in CE day frequency due to shear are very small.

One of the key limitations of this study lies in the coarse spatial resolution of the CMIP5 models. In particular, the spatial resolutions of some of the climate models are nearly 10 times as coarse as the reanalyses in LC14, which made the west–east spatial differences in convective trends across the NEUS essentially indiscernible. This problem is exacerbated by the lack of data below the 850-hPa level over land in many of the CMIP5 models (aside from surface variables), which makes it difficult to factor in the influence of the marine boundary layer over the coastal ocean in comparison to the lack of such a layer inland. Therefore, reliable downscaling methods would be useful for future studies attempting to determine future changes in convective environment frequency over the NEUS, as mentioned by Tippett et al. (2015). In addition, although using the mean of a multimodel ensemble is a common practice to decrease the uncertainty in the predictions, there is large intermodel spread present in this study associated with the individual biases of each model. While such biases make direct comparison between values of atmospheric parameters in the reanalysis and the corresponding parameters in the CMIP5 models difficult, they still permit meaningful analysis of temporal trends.

Acknowledgments

The authors acknowledge the NOAA Climate Program Office Modeling, Analysis, Predictions, and Projections (MAPP) Program as part of the CMIP5 Task Force under Grant NA11OAR431. We thank the three anonymous reviewers for their constructive comments and suggestions to improve the paper.

REFERENCES

  • Allen, J., and D. Karoly, 2014: A climatology of Australian severe thunderstorm environments 1979–2011: Inter-annual variability and the ENSO influence. Int. J. Climatol., 34, 8197, doi:10.1002/joc.3667.

    • Search Google Scholar
    • Export Citation
  • Allen, J., and K. Walsh, 2014: Future Australian severe thunderstorm environments. Part I: A novel evaluation and climatology of convective parameters from two climate models for the late twentieth century. J. Climate, 27, 38273847, doi:10.1175/JCLI-D-13-00425.1.

    • Search Google Scholar
    • Export Citation
  • Allen, J., D. Karoly, and G. Mills, 2011: A severe thunderstorm climatology for Australia and associated thunderstorm environments. Aust. Meteor. Oceanogr. J., 61, 143158.

    • Search Google Scholar
    • Export Citation
  • Allen, J., D. Karoly, and K. Walsh, 2014: Future Australian severe thunderstorm environments. Part II: The influence of a strongly warming climate on convective environments. J. Climate, 27, 38483868, doi:10.1175/JCLI-D-13-00426.1.

    • Search Google Scholar
    • Export Citation
  • Anderson, B. T., K. Hayhoe, and X. Liang, 2010: Anthropogenic-induced changes in the 21st century summertime hydroclimatology of the Northeastern US. Climatic Change, 99, 403423, doi:10.1007/s10584-009-9674-3.

    • Search Google Scholar
    • Export Citation
  • Bentsen, M., and Coauthors, 2013: The Norwegian Earth System Model, NorESM1-M—Part 1: Description and basic evaluation of the physical climate. Geosci. Model Dev., 6, 687720, doi:10.5194/gmd-6-687-2013.

    • Search Google Scholar
    • Export Citation
  • Brooks, H. E., 2013: Severe thunderstorms and climate change. Atmos. Res., 123, 129138, doi:10.1016/j.atmosres.2012.04.002.

  • Brooks, H. E., J. W. Lee, and J. P. Craven, 2003: The spatial distribution of severe thunderstorm and tornado environments from global reanalysis data. Atmos. Res., 67-68, 7394, doi:10.1016/S0169-8095(03)00045-0.

    • Search Google Scholar
    • Export Citation
  • Brooks, H. E., G. W. Carbin, and P. T. Marsh, 2014: Increased variability of tornado occurrence in the United States. Science, 346, 349352, doi:10.1126/science.1257460.

    • Search Google Scholar
    • Export Citation
  • Del Genio, A. D., M. Yao, and J. Jonas, 2007: Will moist convection be stronger in a warmer climate? Geophys. Res. Lett., 34, L16703, doi:10.1029/2007GL030525.

    • Search Google Scholar
    • Export Citation
  • Diffenbaugh, N. S., M. Scherer, and R. J. Trapp, 2013: Robust increases in severe thunderstorm environments in response to greenhouse forcing. Proc. Natl. Acad. Sci. USA, 110, 16 36116 366, doi:10.1073/pnas.1307758110.

    • Search Google Scholar
    • Export Citation
  • Donner, L. J., and Coauthors, 2011: The dynamical core, physical parameterizations, and basic simulation characteristics of the atmospheric component AM3 of the GFDL global coupled model CM3. J. Climate, 24, 34843519, doi:10.1175/2011JCLI3955.1.

    • Search Google Scholar
    • Export Citation
  • Dunne, J. P., and Coauthors, 2012: GFDL’s ESM2 global coupled climate–carbon Earth system models. Part I: Physical formulation and baseline simulation characteristics. J. Climate, 25, 66466665, doi:10.1175/JCLI-D-11-00560.1.

    • Search Google Scholar
    • Export Citation
  • Dunne, J. P., and Coauthors, 2013: GFDL’s ESM2 global coupled climate–carbon Earth system models. Part II: Carbon system formulation and baseline simulation characteristics. J. Climate, 26, 22472267, doi:10.1175/JCLI-D-12-00150.1.

    • Search Google Scholar
    • Export Citation
  • Efron, B., 1987: Better bootstrap confidence intervals. J. Amer. Stat. Assoc., 82, 171185, doi:10.1080/01621459.1987.10478410.

  • Gent, P. R., and Coauthors, 2011: The Community Climate System Model version 4. J. Climate, 24, 49734991, doi:10.1175/2011JCLI4083.1.

    • Search Google Scholar
    • Export Citation
  • Hayhoe, K., and Coauthors, 2007: Past and future changes in climate and hydrological indicators in the US Northeast. Climate Dyn., 28, 381407, doi:10.1007/s00382-006-0187-8.

    • Search Google Scholar
    • Export Citation
  • Hayhoe, K., and Coauthors, 2008: Regional climate change projections for the Northeast USA. Mitig. Adapt. Strategies Global Change, 13, 425436, doi:10.1007/s11027-007-9133-2.

    • Search Google Scholar
    • Export Citation
  • Held, I. M., and B. J. Soden, 2006: Robust responses of the hydrological cycle to global warming. J. Climate, 19, 56865699, doi:10.1175/JCLI3990.1.

    • Search Google Scholar
    • Export Citation
  • Kunkel, K. E., and Coauthors, 2013: Monitoring and understanding trends in extreme storms: State of knowledge. Bull. Amer. Meteor. Soc., 94, 499514, doi:10.1175/BAMS-D-11-00262.1.

    • Search Google Scholar
    • Export Citation
  • Li, H., and B. A. Colle, 2014: Multidecadal changes in the frequency and ambient conditions of warm season convective storms over the northeastern United States. J. Climate, 27, 72857300, doi:10.1175/JCLI-D-13-00785.1.

    • Search Google Scholar
    • Export Citation
  • Liang, X., J. Pan, J. Zhu, K. E. Kunkel, J. X. Wang, and A. Dai, 2006: Regional climate model downscaling of the U.S. summer climate and future change. J. Geophys. Res., 111, D10108, doi:10.1029/2005JD006685.

    • Search Google Scholar
    • Export Citation
  • Lombardo, K., B. A. Colle, and Z. Zhang, 2015: Evaluation of historical and future cool season precipitation over the eastern United States and western Atlantic storm track using CMIP5 models. J. Climate, 28, 451467, doi:10.1175/JCLI-D-14-00343.1.

    • Search Google Scholar
    • Export Citation
  • Murray, J. C., and B. A. Colle, 2011: A climatology of convective storms over the northeast United States. Mon. Wea. Rev., 139, 9921012, doi:10.1175/2010MWR3316.1.

    • Search Google Scholar
    • Export Citation
  • Roman, J., R. Knuteson, S. Ackerman, and H. Revercomb, 2015: Predicted changes in the frequency of extreme precipitable water vapor events. J. Climate, 28, 70577070, doi:10.1175/JCLI-D-14-00679.1.

    • Search Google Scholar
    • Export Citation
  • Seeley, J. T., and D. M. Romps, 2015: The effect of global warming on severe thunderstorms in the United States. J. Climate, 28, 24432458, doi:10.1175/JCLI-D-14-00382.1.

    • Search Google Scholar
    • Export Citation
  • Semenov, V. A., and L. Bengtsson, 2002: Secular trends in daily precipitation characteristics: Greenhouse gas simulations with a coupled AOGCM. Climate Dyn., 19, 123140, doi:10.1007/s00382-001-0218-4.

    • Search Google Scholar
    • Export Citation
  • Taylor, K. E., R. J. Stouffer, and A. G. Meehl, 2012: An overview of CMIP5 and the experiment design. Bull. Amer. Meteor. Soc., 93, 485498, doi:10.1175/BAMS-D-11-00094.1.

    • Search Google Scholar
    • Export Citation
  • Tippett, M., J. T. Allen, V. A. Gensini, and H. E. Brooks, 2015: Climate and hazardous convective weather. Curr. Climate Change Rep., 1, 6073, doi:10.1007/s40641-015-0006-6.

    • Search Google Scholar
    • Export Citation
  • Trapp, R. J., N. S. Diffenbaugh, H. E. Brooks, M. E. Baldwin, E. D. Robinson, and J. S. Pal, 2007: Changes in severe thunderstorm environment frequency during the 21st century caused by anthropogenically enhanced global radiative forcing. Proc. Natl. Acad. Sci. USA, 104, 19 71919 723, doi:10.1073/pnas.0705494104.

    • Search Google Scholar
    • Export Citation
  • Van Klooster, S. L., and P. J. Roebber, 2009: Surface-based convective potential in the contiguous United States in a business-as-usual future climate. J. Climate, 22, 33173330, doi:10.1175/2009JCLI2697.1.

    • Search Google Scholar
    • Export Citation
  • Voldoire, A., and Coauthors, 2013: The CNRM-CM5.1 global climate model: Description and basic evaluation. Climate Dyn., 40, 20912121, doi:10.1007/s00382-011-1259-y.

    • Search Google Scholar
    • Export Citation
  • Watanabe, S., and Coauthors, 2011: MIROC-ESM 2010: Model description and basic results of CMIP5-20c3m experiments. Geosci. Model Dev., 4, 845872, doi:10.5194/gmd-4-845-2011.

    • Search Google Scholar
    • Export Citation
  • Wu, T., and Coauthors, 2014: An overview of BCC climate system model development and application for climate change studies. J. Meteor. Res., 28, 3456, doi:10.1007/s13351-014-3041-7.

    • Search Google Scholar
    • Export Citation
Save