1. Introduction
The Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (Bernstein et al. 2008) documents a linear warming trend in global average surface temperature of 0.74° ± 0.18°C over the past century and a faster warming of 0.13° ± 0.03°C decade−1 since 1956. Furthermore, 11 of the 12 years from 1995 to 2006 ranked among the top warmest in the surface instrumental record dating since 1850 (Bernstein et al. 2008) and, including 2007, 7 of the 8 warmest years on record have occurred since 2001 (NCDC 2008).
Accompanying this warming, Wentz et al. (2007) estimate a global increasing trend in mean precipitation of 7.4% ± 2.6% °C−1 over the period 1987–2006, while Lambert et al. (2008) using a direct method note a trend of 6.7% ± 3.5% °C−1. While these global precipitation trends are primarily connected to tropical and oceanic regions, Gleason et al. (2008) find increases over the past several decades in the percentage of the United States having much greater-than-normal proportion of precipitation derived from extreme 1-day precipitation events. In particular, there has been a steady increase in this percentage from the early 1970s to the late 1990s, with elevated values continuing thereafter.
Bernstein et al. (2008) remark upon the likelihood of increased land–sea thermal contrast in the central U.S. summer and the anticipated increase in the frequency and strength of the nocturnal low-level jet and accompanying atmospheric deep moist convection (hereafter, convection). Held and Soden (2006), relying on a combination of climate model results and theoretical reasoning based upon Clausius–Clapeyron scaling, argue that an increase in droughts and floods in the extratropical land areas can be expected with continued warming. Although the Bernstein et al. (2008) conclude that it is “very likely” that heavy precipitation events will become more frequent, they note that there remains insufficient observational evidence of trends in severe convective phenomena such as high wind, lightning, hail, and tornadoes and are silent about future possibilities.
This gap in understanding has important practical implications. Severe convective weather often results in damage, injury, and death. On rare occasions, the impacts of convective activity can be catastrophic. For example, the 3 May 1999 tornado outbreak resulted in 46 fatalities (Brooks and Doswell 2002) and nearly $1 billion in damages (Thompson and Edwards 2000; Brooks and Doswell 2001). Although such outbreaks are infrequent, the impact of annual convective activity is nonetheless substantial. Annual average fatalities and injuries in the United States resulting from convection during the period 1997–2006 were 129 and 1713, respectively, with associated crop and property damages of $2.4 billion per year (see U.S. Natural Hazards Statistics at http://www.nws.noaa.gov/om/hazstats.shtml). These numbers, which exclude the impacts from flooding, represent 20% of all weather-related fatalities and 44% of all weather-related injuries during that period. A comprehensive hazard assessment of future climate change must include the effects of severe convection.
Unfortunately, at their current stage of evolution, climate models cannot realistically simulate convection. A standard grid spacing of 100 km or larger is far too coarse to resolve convective details; such models may need to be run at well below 10-km grid spacing before such systems can produce information on convective mode (defined here as the form of convection, such as supercells, multicellular clusters, or lines) and severity (Roebber and Eise 2001; Fowle and Roebber 2003; Done et al. 2004; Kain et al. 2006; Weisman et al. 2008) and finer still to provide internal consistency with the dynamics of such phenomena (Bryan et al. 2003). The nonlinear growth in required computational power as model resolution increases (e.g., Wehner et al. 2008 estimate ∼10 petaflops for global 1-km simulations), however, makes this approach infeasible in the very near future for long-term global climate studies.
In coarse grid systems like climate models and most operational weather forecast models, convection is represented through cumulus parameterizations, which are needed to account for the upscale effects of convection on the larger-scale flow. Because of the importance of convection to actual weather forecasts, however, there is a long history of attempts to predict convective details from these parameterizations. Owing to the inconsistency with the original aims of these parameterizations and related problems (e.g., Molinari 1993), however, such approaches have not proved very successful.
A further problem with using cumulus parameterizations to define convection in climate model studies is that any techniques developed in this manner will be model specific. One way to avoid this problem is by building an observation-dependant scheme based upon synoptic-scale variables that are readily obtained from climate models. Although this approach is similar to the problem of cumulus parameterization, the goal is to provide direct measures of convective activity itself and in that sense is more akin to the “perfect prog” approach that has been applied in weather forecasting (Klein et al. 1959; Marzban et al. 2006). Such systems start with the assumption that model variables can be directly exchanged with observations after development without any sort of bias removal; the advantage is that, as models are modified and presumably improved, the estimation scheme itself will also improve while avoiding the need to be updated.
There have been several recent attempts to forge such links between severe convective environments and large-scale data. Brooks et al. (2003, hereafter B03) develop a discriminant analysis technique for diagnosing severe environments from the National Centers for Environmental Prediction (NCEP) reanalysis based primarily upon convective available potential energy (CAPE) and deep-layer shear. In this approach, convection is assumed and the severity of the convection is then estimated. Brooks et al. (2007) use the same approach to examine the seasonal cycle. Del Genio et al. (2007) use an approach based on B03, but exchange cumulus updraft speed for CAPE, to diagnose severe weather environments from global climate model output. Trapp et al. (2007) apply the B03 discriminator directly to both global and regional climate model outputs while Marsh et al. (2007) apply it to output from the National Center for Atmospheric Research (NCAR) climate model.
In this study, this kind of perfect-prog approach is also utilized, but rather than assuming convection and diagnosing severity, the relative potential for surface-based convective initiation is diagnosed from the large-scale environment. Various methods can then be applied to determine the severity of the resulting convective potential. Further details can be found in Bruening (2005).
The paper is organized as follows. Section 2 describes the data and methodology employed. Section 3 provides a comparison between the estimation technique and historical convective climatologies and then applies the method to a future (business as usual) climate model run. The study results are summarized in section 4.
2. Data and methodology
To develop an observationally based estimation scheme for convective initiation, data relating synoptic-scale variables to thunderstorm occurrence are collected. Next, to apply the estimation scheme to future climate, its performance is validated using the output from historical climate simulations matched to observations. In so doing, we address the question of whether the scheme broadly reproduces the observed climate of convective occurrences from these simulations. Finally, the validated scheme is applied to a future climate scenario to provide an assessment of the potential occurrence of severe convection using diagnostic methodologies that differentiate between normal and severe convective environments.
a. Datasets
The National Climatic Data Center’s (NCDC's) TD3280 and TD3281 datasets are used to identify occurrences of thunderstorms and severe weather from June 1957 through December 1995. Data beyond 1995 are excluded from the scheme development to screen inconsistencies that would be introduced owing to the implementation of the National Oceanic and Atmospheric Administration (NOAA) Automated Surface Observation System (ASOS). L. Bosart (2004, personal communication) identifies several ASOS design flaws in shelter construction and instrumentation that lead to lower quality observations (see also Gall et al. 1992; Kessler et al. 1993). Radiosonde measurements are obtained from the NCDC/Forecast Systems Laboratory (FSL) North American radiosonde dataset (Schwartz and Govett 1992). NCAR–NCEP reanalysis data (Kalnay et al. 1996) are used to provide gridded surface wind measurements.
Historical climate simulation data for the years 1975–94 are obtained from the National Energy Research Scientific Computing Center (NERSC). The chosen model configuration is from the Parallel Climate Model (PCM), which features a coupling between the NCAR Community Climate Model, version 3 (on a T85 grid), the Los Alamos National Laboratory Parallel Ocean Program, and a sea ice model from the Naval Postgraduate School (the latter two models at approximately 1° resolution). The historical climate simulation period of record overlaps but is shorter than the observed data, owing to archive limitations at NERSC.
For future climate, this study uses the Houghton et al. (2001) 6-hourly A2 third-assessment inputs (datasets 07.53a and 07.53b for the years 2020–30 and 2045–55, respectively, from the same version of the PCM). This model run is selected because of its “business-as-usual” characteristics such as increasing population, technology, and CO2 emissions with a focus on economic growth over environmental precautions (Nakicenovic et al. 2000). Hence, it represents a likely future extreme. This study was limited to the two subsections of years within the complete A2 PCM run because of storage constraints at NERSC related to the high-frequency, multilevel data needed here. However, the diagnostic study developed here could be applied to any of the IPCC runs from any climate model.
Various methodologies exist when attempting to identify convective occurrence. Rasmussen and Blanchard (1998) define a convective event as at least 10 cloud-to-ground lightning flashes. Since such data are not readily available, however, this study defines a convective occurrence as a report of thunderstorms or severe weather at an observation station. To understand the dynamics within convective episodes, upper-air sampling is necessary; however, radiosonde measurements do not exist at every observation station. This necessitates the use of proximity soundings (Darkow 1969). Of course, spatial and temporal variability within the convective environment ensures that finding a representative proximity sounding is not trivial (e.g., Beebe 1958; Darkow 1969). To most accurately represent the upper-air environment relevant to the convective occurrence, this study defines proximity soundings following Brooks et al. (1994), in which radiosonde measurements must have been taken within 1 h before or after launch time and the sounding station must have been located no farther than 160 km from the surface station. A further restriction on the proximity soundings is to consider only those cases for which convective inhibition, a measure equivalent to the strength of the trigger needed to overcome capping in the convective environment, is less than 200 J kg−1 (or an updraft of ∼10 m s−1 or less; e.g., Bluestein 1993).
b. Variable selection
A database of convective variables is compiled for each identified convective occurrence. Convective variables are assembled based on their potential usefulness in diagnosing convective initiation and subsequent severe weather, based on the available literature. A number of candidate variables were screened—the following discussion summarizes the final selections.
It is well known that convective mode and severity are strongly dependent on the environment in which a storm develops, especially instability and vertical wind shear (e.g., Weisman and Klemp 1982; B03). CAPE is the thermodynamic instability parameter used in this study, although, as detailed by Del Genio et al. (2007), it is possible to estimate cumulus updraft speed directly and so account for additional physical constraints on storm vertical motion.
A diurnal cycle in thunderstorm occurrence has long been recognized (Means 1944; Byers and Braham 1949; Petterssen 1958; Wallace 1975), with convective events tending to occur more frequently during the late afternoon hours than any other time of day. Storms that occur at this time are dynamically distinct from the less-frequent nocturnal events, which tend to occur in association with elevated convection and lesser buoyancy (Wallace 1975; Colman 1990). Because of the suboptimal sampling interval of radiosondes (1200 UTC is after the peak time of such storms) and the lesser frequency of severe convection at these times (Grant 1995; Branick 2005; Horgan et al. 2007), only afternoon–evening storms associated with the 0000 UTC sounding environments are considered in this study. As such, only surface-based CAPE is considered. B03 have shown that reanalysis data can be used to provide proximity soundings at higher temporal frequency; such an approach would be useful in future studies to avoid some of these sampling problems.
Since only surface-based daytime convection is being considered, it is important to consider the limitations that this restricted view presents to interpretation of the results. For example, human vulnerability to nocturnal tornadoes is high (Ashley et al. 2008). Horgan et al. (2007) confirm, however, that the primary severe threat from elevated convection is large hail (59% of their cases). Further, the NOAA/Storm Prediction Center severe hail statistics used in combination with the information from Horgan et al. (2007) show that during the 5-yr period of that study, only 4% occurred in association with elevated storms. Hence, the neglect of elevated and/or nocturnal events is not considered critical. Hereafter, the term convection in this paper will be understood to refer only to surface-based daytime events.
Vertical wind shear is also derived from radiosonde measurements and is an essential diagnostic for convective storm organization (Rotunno 1981; Weisman and Klemp 1982; Weisman and Klemp 1984; B03). This study uses 0–6-km-deep layer shear, which has been shown to be a useful discriminating variable for convective severity (e.g., Rasmussen and Blanchard 1998; B03).
The significance of surface moisture convergence in convective events is documented by Hudson (1971) and Waldstreicher (1989). More recently, Banacos and Schultz (2005) show that this connection is largely achieved through the triggering influence of mesoscale convergence alone. At the climate simulation scale, such mesoscale effects cannot be resolved. Nonetheless, it may be useful to consider surface-based convergence as a proxy for the various synoptic-scale processes that produce ascent and correspondingly modify the environment in which the convection develops. For this study, surface convergence is obtained by using bilinear interpolation of surface u and v wind components from the gridded reanalysis data.
Finally, a dataset of null cases (no convective occurrence) is constructed using a random sampling of nonconvective dates within the same time period (1957–95) from which the convective cases were sampled. The frequency distributions for time of year and location of the convective event were used to ensure similar climatological distributions between the convective and null datasets.
Frequency distributions can be used to visually inspect the potential of the variables for differentiating between the convective and nonconvective samples (Table 1). The most noticeable differences appear in the CAPE and surface convergence variables, with lesser but still detectable differences present for shear. Because of these differences and the available literature suggesting their importance, each of these variables, especially when considering nonlinear interactions between them (e.g., Roebber et al. 2003), may prove useful for discriminating between convective and nonconvective cases. In nonlinear systems, between-variable interactions may accentuate the importance of variables relative to what can be seen in isolation (see Fig. 1 for evidence of this for shear). B03 showed that CAPE and deep-layer shear are useful for discriminating between severe and ordinary convective environments, which provides further motivation for retaining deep-layer shear in the present analysis.
c. Climate model data correction
The climatological distribution of parameters important to convection was compared using the actual historical data and the climate model historical simulations (Table 2). Although most variables compare well, substantial differences in CAPE are evident. Because small changes in boundary layer moisture can yield large differences in CAPE (e.g., Bluestein 1993; Crook 1996), the vertical resolution of the climate model was investigated. The model has 18 vertical levels including only 3 within the boundary layer, making it difficult to resolve near-surface details. For example, over the U.S. Great Plains, the typical difference between the lowest model pressure level and surface pressure is about 6 hPa or, hydrostatically, approximately 45 m, which in a well-mixed environment can produce temperature and dewpoint differences of near 0.5°C or approximately 500 J kg−1 of CAPE (Williams and Renno 1993).
A second error related to proper simulation of the thermal diurnal cycle exists. Dai and Trenberth (2004) studied the Community Climate System Model, version 2 (CCSM2) diurnal cycle in June, July, and August for land areas in the latitude band from 25° to 70°N and found a more than 1-h lag in the onset of the daytime temperature maximum, causing a cold temperature model bias of up to 1°C during that time. This effect is likely to be larger for specific locations because of averaging. Hence, improper timing of the daytime temperature maximum could reduce climate model CAPE in excess of 1000 J kg−1 in some circumstances (Williams and Renno 1993; Dai and Trenberth 2004).
d. Artificial neural networks
Fundamentally, artificial neural networks (ANNs) are a form of generalized regression model where inputs are mapped to outputs for all chosen training (input–output) pairs. The network model fit can be nonlinear and so ANNs are often used to model complex relationships where the actual dynamics are not understood or the equations describing them are not available. Further details on ANNs can be found in Tsonis (1992), Marzban and Witt (2001), Roebber et al. (2003), Marzban (2003), and references therein.
Here, the actual historical data are split into three component datasets: training, cross validation, and testing. The network is trained using a multilayer-perceptron (MLP) ANN architecture with two hidden layers on the input variables of CAPE, surface convergence, and deep-layer shear. Because the purpose of this study is to build a model that will establish the relative climatological potential for convection, the ANN is trained by mapping the inputs to the output (probability of convective initiation) using the information contained in the training dataset. The weights for each variable (and other network parameters, such as number of nodes) are adjusted based on the cross-validation data to prevent memorization and ensure a more general solution. Once satisfactory training is achieved, the testing dataset is used to evaluate the success of the network model.
Percentile distributions of convective and nonconvective cases reveal many differences and confirm the ANN’s ability to distinguish between the two situations (Table 3), most notably the higher probability of convective initiation at the low percentiles for the convective cases (e.g., 62% versus 16% at the 25th percentile, respectively). CAPE is notably different with the nonconvective cases, having a much smaller range and lower upper end than the convective cases, as expected.
The sensitivity of the model is gauged by evaluating its performance on a special test dataset in which all variables are held constant except for the one of interest, which is allowed to vary incrementally through a range of realistic values (Fig. 1). A strong gradient in probability as a function of shear is evident for moderate levels of CAPE (e.g., ∼500 J kg−1) and is interpreted as the result of an increasingly baroclinic environment producing stronger synoptic features and hence a more favorable environment for storm initiation for a given amount of instability. Although an increasing convective probability with increasing CAPE does not follow from traditional parcel theory (e.g., Bluestein 1993, p. 455), operationally this aspect has been acknowledged for many years through rules of thumb that link increasing thunderstorm probabilities to increasing values of the K index and other simple convective indexes (e.g., Reap and Foster 1979). Strictly, Fig. 1 indicates that this CAPE–convective initiation linkage holds mainly for moderate CAPE, while for larger values variations in deep-layer shear and convergence are more influential.
There are several reasons why one might expect the probability of convective initiation to increase with increasing CAPE for moderate values. First, data (not shown) indicate an inverse relationship between CAPE and convective inhibition. Therefore, as CAPE values increase, a weaker trigger is needed to access the instability. Generally, when a region is under the influence of the same air mass, local variations in instability, capping, and lift exist. With broadly higher CAPE and concomitant reductions in capping, however, it is likely that local differences will allow convection to initiate somewhere within that region. Second, we speculate that, on days with high CAPE and the usually attendant warm surface temperatures, boundary layer turbulence is more likely, as suggested by shallow convective parameterizations (e.g., Deng et al. 2003). Because buoyancy-driven circulations in the boundary layer (such as horizontal convective rolls) can be sufficient to initiate convection, such implicit triggers might contribute to the increasing probability.
Convergence also produces a higher probability of convective initiation for a fixed level of CAPE and deep-layer shear (Fig. 1, middle and bottom panels), as is expected from its potential as an indicator of a synoptic environment in which mesoscale triggers may be organized or otherwise more effective. For example, for CAPE equal to 750 J kg−1 and shear of 5 m s−1, the probability of convective initiation exceeds 70% (is less than 50%) for a fixed absolute amount of surface wind convergence (divergence).
3. Results
a. Climate model historical
To gauge the success of the ANN approach, observed frequencies of convection are compared to estimates of convective initiation potential derived from the historical climate model simulation. An area of higher convective initiation potential indicated by the model data exists off the southeast coast of the United States. This is consistent with recent studies that indicate a maximum number of lightning flashes off the Florida coast (e.g., Zajac and Rutledge 2001) attributable to destabilization of air masses by the warm Gulf Stream current. As is expected, because of the inability of coarse-resolution models to produce mesoscale structures like the Florida sea breeze convergence zone, such observed high-frequency areas are not evident in the simulated historical data. The minimal simulated convective initiation potential along the West Coast is consistent with observations and reflects the stabilizing influence of the cold waters of the California current. The highest convective initiation potential in the model data, however, occurs to the lee of the Rocky Mountains with a local maximum over the Mexican Plateau (Fig. 2; Table 4). Although convection is indeed common in this area [see Fig. 12 of Changnon (2001) and the climatological cloud-to-ground lightning frequencies of Murphy and Holle (2005)], the comparison with multiple sources (Table 4) reveals that this is an overforecast.
Further inspection of these data reveal that this overforecast is at the expense of a feature common to observed maps of convection: the “tornado alley” located in Kansas, Oklahoma, and the panhandle of Texas and extending north and east to the upper Midwest. Warm, moist air advecting northward from the Gulf of Mexico and the vertical wind shear associated with the prevailing westerlies aloft make severe convective weather a prominent feature in this area. The model-simulated convective initiation potential shows evidence of activity in this area but at modest levels. Gates et al. (1999) compare and discuss characteristics of climate models and show large deficiencies in warm season precipitation production within, and north and northwest of the Gulf of Mexico. Because warm season precipitation in this region is primarily the result of deep moist convection rather than stratiform rain processes, a speculation is that the coarse resolution of the climate models may be limiting simulated moisture transport from the Gulf of Mexico via the low-level jet. Although the model produces a spatial distribution of convective initiation potential that is broadly consistent with observations, the limitations noted above should be considered when interpreting the results of the A2 climate simulations.
The simulated seasonal cycle of convective initiation potential (Fig. 3) shows the highest values from Julian days 151–200, corresponding to late spring and early summer. According to observed data, the highest relative frequency of convective cases occur in July (Julian days 181–211), confirming a relative correspondence between the ANN approach and historical data.
b. Future climate
The three input variables (CAPEadj, surface convergence, and deep-layer shear) derived from the A2 business-as-usual PCM run for the years 2020–30 and 2045–55 are input to the ANN to estimate future convective initiation potential under this climate condition. As a means of comparison, difference maps expressed as the percent change in convective initiation potential (itself expressed for a given day in the range from zero to one) from the simulated historical condition as a baseline are constructed (Fig. 4). The data indicate relatively limited differences with maxima of ±4%. The most prominent differences appear over bodies of water, with a 4% decrease off the Mexican coastline and within the Gulf of Mexico and a 4% increase in New England along and just off the coast of Massachusetts. Moderate increases (∼2%) can be seen in the U.S. Midwest. To place these simulated changes into context, the future interannual variability (relative to the simulated historical baseline) is computed (not shown) and reveals that these changes fall within the range of the interannual variability, suggesting no clear signal.
The data also show a change in the annual cycle for convective initiation, with an elongation of the convective season as evidenced by increases between Julian days 61–121 and 200–271 (Fig. 5). This is consistent with research indicating warming in the summer in the northern latitudes (Houghton et al. 2001). These changes, however, likewise occur within the range of interannual variability and are therefore inconclusive.
Future changes in CAPE appear to be the largest contributor to changes in the potential for convective initiation (not shown; see Bruening 2005). For example, a 12% increase in the magnitude of CAPE coincides with the increase in convective initiation potential over southeastern Massachusetts. Likewise, areas of decreased convective initiation potential correspond to decreased amounts of CAPE (e.g., off the west coast of Mexico and within the Gulf of Mexico). Although there is some suggestion that the midwestern increases are partially linked to surface convergence, there is little absolute change from the historical baseline while deep-layer shear shows moderate changes, with increases along the West Coast and decreases in all areas east of the Rocky Mountains.
Of most relevance to future climate from a public safety perspective are changes in the distribution and/or frequency of severe convection. The relationship between severe convection in its various modes and the surrounding environment is still imperfectly understood. Using a simple severe storm criteria of CAPE greater than a specified value with associated strong updraft and downdraft potential (e.g., 1000 J kg−1) has some appeal since those cases contain the necessary energy for severe weather, regardless of storm longevity (e.g., parcel theory predicts updrafts in storms with CAPE of 1000 J kg−1 of ∼22 m s−1). Rasmussen and Blanchard (1998) examined the relationship between CAPE and deep-layer shear for a large sample of tornadic and nontornadic severe storms as well as ordinary thunderstorms. Their data indicate that cases with CAPE less than 1000 J kg−1 can still be severe, a result that is consistent with the groundbreaking work of Weisman and Klemp (1982, 1984) relating deep-layer shear and CAPE to storm organization and longevity. Trying to identify these cases is a challenge, however, because the majority of the nonsevere cases also falls below this threshold and thus provide the possibility of a high false alarm rate.
Another source of error relates to the effects of warming of the entire thermal profile on storm severity. In particular, the cap could be strengthened, resulting in less-frequent convective initiation, and where convection does initiate the potential for severe hail could be decreased as freezing levels rise. For example, the regression equation for probability of large hail of Billet et al. (1997) indicates that, for a 1°C warming in a representative hail scenario [e.g., vertically integrated liquid water (VIL) of 60 kg m−2], the probability of large hail decreases by approximately 7%. Many such complicating factors exist and are currently beyond the scope of analyses such as this study. Here, we consider any storm severe for which CAPE exceeds 1000 J kg−1 but also apply (3) for all cases regardless of CAPE for comparative purposes.
Nearly all locations that showed increases in convective initiation potential show increases in potential severity above the level of interannual variability (Figs. 4 and 6). These increases are particularly evident in the central and eastern United States, including the “tornado alley” region of the U.S. Midwest (values not shown; see Bruening 2005). Applying the B03 discriminator for all values of CAPE, increases in severe convective potential relative to interannual variability occur for all longitudinal bands across the United States, with the exception of the northeastern United States (not shown). The most prominent increase relative to the background variability, by a factor of 12.2, is found in the midwestern United States (90°–100°W), a result consistent with those found using CAPE alone. It is of interest to note that, when organized based on severity, the historical data exhibit lower surface convergence and higher deep-layer shear than the A2 PCM run, changes that are consistent with stronger average synoptic ascent but in a generally less baroclinic environment for future severe convection. There are several ways that this seemingly contradictory result might occur—for example, slower-moving synoptic-scale systems in the weaker flow—but the precise reasons for this behavior have not been studied here. The reduced deep-layer shear may also indicate some potential for reduced storm longevity, but details of such an analysis are well beyond the capabilities of this study.
Similar results for the Midwest were found by Trapp et al. (2007) based on simulated changes in the background severe weather environment, and those authors suggested that this was consistent with an interpretation that changes in instability rather than shear were the dominant influence in future severe weather. Del Genio et al. (2007) also found increases in updrafts/instability along with reduced shear. Perhaps of particular importance is their finding that high shear–high instability combinations increase east of the Rocky Mountains with the concomitant possibility of increases in the frequency of the most severe storms.
4. Summary
There is a compelling need to understand the regional impacts of global climate change. In the United States, one of the most important regional climate characteristics is the occurrence of convection and attendant severe weather. Unfortunately, future climate model simulations remain incapable of resolving convective storms and, as such, cannot provide any direct insight into such questions.
This study attempts to estimate the potential for convective occurrence and attendant severity by detecting their relationship to signals present at the scales for which the climate models are presently designed to resolve. Since the devised method is general, it can be adapted for use in determining a climatology of future convective potential from any climate model run.
The key results for the IPCC A2 PCM run from the third assessment can be succinctly summarized as follows:
the overall convective initiation potential does not change within the first half of the twenty-first century relative to interannual variability;
the overall potential for severe convection does change within the first half of the twenty-first century relative to interannual variability, east of the Rocky Mountains and most notably in the “tornado alley” region of the U.S. Midwest.
One might expect the potential for convective occurrence and severity to change in unison. This need not be the case, however, because the underlying requirements differ. Convective occurrence (corresponding climatologically to frequency) is determined by a feature triggering vertical motion sufficient to access available buoyancy. In contrast, the severity of the convection further depends on the available energy and in many cases, the ability of the deep-layer shear to organize the updrafts and downdrafts. Hence, ongoing warm season surface warming and moistening (e.g., Houghton et al. 2001; Bernstein et al. 2008) may not markedly increase the chances of convection triggering but, where it does, sufficient energy now may be available to increase the prospects for severe storms. These findings are largely consistent with studies based on examination of the future environments in which, given convection, the storms are more likely to be severe (Del Genio et al. 2007; Trapp et al. 2007).
Further improvements in such methodologies are needed. Here, we have not attempted to relate storm severity to convective mode directly, although it is well known that certain storm types are more prolific severe weather producers (e.g., Trapp et al. 2005). Additional exploration of this approach, beyond the standard instability-shear phase space, is needed. The recognition by severe storm forecasters that particular synoptic patterns are climatologically connected to higher chances for severe weather (e.g., Miller 1972) suggests that approaches that link synoptic patterns more directly to storm modes also may be a fruitful avenue to pursue in assessing the likelihood of future changes in severe convection.
Finally, a range of climate scenarios are possible. The “business as usual” PCM run investigated here has now been superseded by updates reflecting a more rapid increase in greenhouse gas concentrations (Bernstein et al. 2008) than previously projected. Mitigation scenarios should also be investigated. Continued improvements in the climate models themselves including increasing resolution and better model physics will provide a more accurate foundation on which to base these projections.
Acknowledgments
This work was supported in part by University of Wisconsin—Milwaukee RGI 01-001. Comments by Anthony Del Genio and two anonymous reviewers helped to improve this manuscript.
REFERENCES
Ashley, W. S., A. J. Krmenec, and R. Schwantes, 2008: Vulnerability due to nocturnal tomadoes. Wea. Forecasting, 23 , 795–807.
Banacos, P. C., and D. M. Schultz, 2005: The use of moisture flux convergence in forecasting convective initiation: Historical and operational perspectives. Wea. Forecasting, 20 , 351–366.
Beebe, R. G., 1958: Tornado proximity soundings. Bull. Amer. Meteor. Soc., 39 , 195–201.
Bernstein, L., and Coauthors, 2008: Summary for policymakers. Climate Change 2007: Synthesis Report, A. Allali et al., Eds., 22 pp. [Available online at http://www.ipcc.ch/pdf/assessment-report/ar4/syr/ar4_syr_spm.pdf].
Billet, J., M. DeLisi, and B. G. Smith, 1997: Use of regression techniques to predict hail size and the probability of large hail. Wea. Forecasting, 12 , 154–164.
Bluestein, H. B., 1993: Observations and Theory of Weather Systems. Vol. 2, Synoptic–Dynamic Meteorology in Midlatitudes, Oxford University Press, 593 pp.
Branick, M., 2005: A comprehensive glossary of weather terms for storm spotters. NOAA Tech. Memo. NWS SR-145. [Available online at http://www.srh.noaa.gov/oun/severewx/glossary.php].
Brooks, H. E., and C. A. Doswell III, 2001: Normalized damage from major tornadoes in the United States: 1890–1999. Wea. Forecasting, 16 , 168–176.
Brooks, H. E., and C. A. Doswell III, 2002: Deaths in the 3 May 1999 Oklahoma City tornado from a historical perspective. Wea. Forecasting, 17 , 354–361.
Brooks, H. E., C. A. Doswell III, and J. Cooper, 1994: On the environments of tornadic and nontornadic mesocyclones. Wea. Forecasting, 9 , 606–618.
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 , 73–94.
Brooks, H. E., A. R. Anderson, K. Riemann, I. Ebbers, and H. Flachs, 2007: Climatological aspects of convective parameters from the NCAR/NCEP reanalysis. Atmos. Res., 83 , 294–305.
Bruening, S. L., 2005: Assessment of future convection in the contiguous United States. M.S. thesis, Dept. of Mathematical Sciences, University of Wisconsin—Milwaukee, 54 pp. [Available from the Department of Mathematical Sciences, University of Wisconsin at Milwaukee, 3200 North Cramer Avenue, Milwaukee, WI 53211].
Bryan, G. H., J. C. Wyngaard, and J. Michael Fritsch, 2003: Resolution requirements for the simulation of deep moist convection. Mon. Wea. Rev., 131 , 2394–2416.
Byers, H. R., and R. R. Braham, 1949: The Thunderstorm. U.S. Dept. of Commerce, 287 pp.
Changnon, S. A., 2001: Thunderstorms across the Nation: An Atlas of Storms, Hail, and Their Damages in the 20th Century. Princeton Press, 93 pp.
Colman, B. R., 1990: Thunderstorms above frontal surfaces in environments without positive CAPE. Part I: A climatology. Mon. Wea. Rev., 118 , 1103–1122.
Crook, N. A., 1996: Sensitivity of moist convection forced by boundary layer processes to low-level thermodynamic fields. Mon. Wea. Rev., 124 , 1767–1785.
Dai, A., and K. E. Trenberth, 2004: The diurnal cycle and its depiction in the Community Climate System Model. J. Climate, 17 , 930–951.
Darkow, G. L., 1969: An analysis of over sixty tornado proximity soundings. Preprints, Sixth Conf. On Severe Local Storms, Chicago, IL, Amer. Meteor. Soc., 218–221.
Del Genio, A. D., M-S. Yao, and J. Jonas, 2007: Will moist convection be stronger in a warmer climate? Geophys. Res. Lett., 34 , L16703. doi:10.1029/2007GL030525.
Deng, A., N. L. Seaman, and J. S. Kain, 2003: A shallow convection parameterization for mesoscale models. Part I: Submodel description and preliminary application. J. Atmos. Sci., 60 , 34–56.
Done, J., C. Davis, and M. Weisman, 2004: The next generation of NWP: Explicit forecasts of convection using weather research and forecast (WRF) model. Atmos. Sci. Lett., 5 , 110–117.
Doswell III, C. A., H. E. Brooks, and M. P. Kay, 2005: Climatological estimates of daily local nontornadic severe thunderstorm probability for the United States. Wea. Forecasting, 20 , 577–595.
Fowle, M. A., and P. J. Roebber, 2003: Short range (0–48 h) numerical prediction of convective occurrence, mode, and location. Wea. Forecasting, 18 , 782–794.
Gall, R., K. Young, R. Schotland, and J. Schmitz, 1992: The recent maximum temperature anomalies in Tucson: Are they real or an instrumental problem? J. Climate, 5 , 657–665.
Gates, W. L., and Coauthors, 1999: An overview of the results of the Atmospheric Model Intercomparison Project (AMIP I). Bull. Amer. Meteor. Soc., 80 , 29–55.
Gleason, K. L., J. H. Lawrimore, D. H. Levinson, and T. R. Karl, 2008: A revised U.S. Climate Extremes Index. J. Climate, 21 , 2124–2137.
Grant, B. N., 1995: Elevated cold-sector severe thunderstorms: A preliminary study. Natl. Wea. Dig., 19 , 25–31.
Held, I. M., and B. J. Soden, 2006: Robust responses of the hydrological cycle to global warming. J. Climate, 19 , 5686–5699.
Horgan, K. L., D. M. Schultz, J. E. Hales Jr., and S. F. Corfidi, 2007: A five-year climatology of elevated severe convective storms in the United States east of the Rocky Mountains. Wea. Forecasting, 22 , 1031–1044.
Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der Linden, X. Dai, K. Maskell, and C. A. Johnson, 2001: Climate Change 2001: The Scientific Basis. Cambridge University Press, 881 pp.
Hudson, H. R., 1971: On the relationship between horizontal moisture convergence and convective cloud formation. J. Appl. Meteor., 10 , 755–762.
Kain, J. S., S. J. Weiss, J. J. Levit, M. E. Baldwin, and D. R. Bright, 2006: Examination of convection-allowing configurations of the WRF model for the prediction of severe convective weather: The SPC/NSSL spring program 2004. Wea. Forecasting, 21 , 167–181.
Kalnay, E., and Coauthors, 1996: The NCEP/NCAR 40-Year Reanalysis Project. Bull. Amer. Meteor. Soc., 77 , 437–471.
Kessler, R. W., L. F. Bosart, and R. S. Gaza, 1993: Recent maximum temperature anomalies at Albany, New York: Fact or fiction. Bull. Amer. Meteor. Soc., 74 , 215–226.
Klein, W. H., B. M. Lewis, and I. Enger, 1959: Objective prediction of five-day mean temperature during winter. J. Meteor., 16 , 672–682.
Lambert, F. H., A. R. Stine, N. Y. Krakauer, and J. C. H. Chiang, 2008: How much will precipitation increase with global warming? Eos, Trans. Amer. Geophys. Union, 89 , 193–195. doi:10.1029/2008EO210001.
Marsh, P. T., H. E. Brooks, and D. J. Karoly, 2007: Assessment of the severe weather environment in North America simulated by a global climate model. Atmos. Sci. Lett., 8 , 100–106.
Marzban, C., 2003: Neural networks for postprocessing model output: ARPS. Mon. Wea. Rev., 131 , 1103–1111.
Marzban, C., and A. Witt, 2001: A Bayesian neural network for severe-hail size prediction. Wea. Forecasting, 16 , 600–610.
Marzban, C., S. Sandgathe, and E. Kalnay, 2006: MOS, perfect prog and reanalysis. Mon. Wea. Rev., 134 , 657–663.
Means, L. L., 1944: The nocturnal maximum occurrence of thunderstorms in the midwestern states. Dept. of Meteorology, University of Chicago, Miscellaneous Rep. 16, 37 pp.
Miller, R. C., 1972: Notes on the analysis and severe-storm forecasting procedures of the Air Force Global Weather Central. Air Weather Service Tech. Rep. 200 (rev.), Air Weather Service, Scott Air Force Base, IL, 190 pp.
Molinari, J., 1993: An overview of cumulus parameterization in mesoscale models. The Representation of Cumulus Convection in Numerical Models, K. A. Emanuel and D. J. Raymond, Eds., Amer. Meteor. Soc., 155–158.
Murphy, M. J., and R. L. Holle, 2005: Where is the real cloud-to-ground lightning maximum in North America? Wea. Forecasting, 20 , 125–133.
Nakicenovic, N., and Coauthors, 2000: Summary for policymakers. IPCC Special Report: Emissions Scenarios, N. Nakicenovic and R. Swart, Eds., 27 pp. [Available online at http://www.ipcc.ch/pdf/special-reports/spm/sres-en.pdf].
National Climatic Data Center, cited. 2008: Climate of 2007 annual report. [Available online at http://www.ncdc.noaa.gov/oa/climate/research/2007/ann/global.html].
Petterssen, S., 1958: Introduction to Meteorology. 2nd ed. McGraw-Hill, 327 pp.
Rasmussen, E. N., and D. O. Blanchard, 1998: A baseline climatology of sounding-derived supercell and tornado forecast parameters. Wea. Forecasting, 13 , 1148–1164.
Reap, R. M., and D. S. Foster, 1979: Automated 12–36 hour probability forecasts of thunderstorms and severe local storms. J. Appl. Meteor., 18 , 1304–1315.
Roebber, P. J., and J. Eise, 2001: The 21 June 1997 flood: Storm-scale simulations and implications for operational forecasting. Wea. Forecasting, 16 , 197–218.
Roebber, P. J., S. L. Bruening, D. M. Schultz, and J. V. Cortinas, 2003: Improving snowfall forecasting by diagnosing snow density. Wea. Forecasting, 18 , 264–287.
Rotunno, R., 1981: On the evolution of thunderstorm rotation. Mon. Wea. Rev., 109 , 171–180.
Schumann, M. R., 2008: An observational analysis of the influence of synoptic forcing on convective morphology. M.S. thesis, Dept. of Mathematical Sciences, University of Wisconsin—Milwaukee, 29 pp. [Available from the Department of Mathematical Sciences, University of Wisconsin at Milwaukee, 3200 North Cramer Avenue, Milwaukee, WI 53211].
Schwartz, B. E., and M. Govett, 1992: A hydrostatically consistent North American radiosonde data base at the Forecast Systems Laboratory, 1946–present. NOAA Tech. Memo. ERL FSL-4, 81 pp. [Available from NOAA/FSL, 325 Broadway, Boulder, CO 80303].
Thompson, R. L., and R. Edwards, 2000: An overview of environmental conditions and forecast implications of the 3 May 1999 tornado outbreak. Wea. Forecasting, 15 , 682–699.
Trapp, R. J., S. A. Tessendorf, E. S. Godfrey, and H. E. Brooks, 2005: Tornadoes from squall lines and bow echoes. Part I: Climatological distribution. Wea. Forecasting, 20 , 23–34.
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 , 19719–19723.
Tsonis, A. A., 1992: Chaos: From Theory to Applications. Plenum, 274 pp.
Waldstreicher, J. S., 1989: A guide to utilizing moisture flux convergence as a predictor of convection. Natl. Wea. Dig., 14 (4) 20–35.
Wallace, J. M., 1975: Diurnal variations in precipitation and thunderstorm frequency over the conterminous United States. Mon. Wea. Rev., 103 , 406–419.
Wehner, M., L. Oliker, and J. Shalf, 2008: Towards ultra-high resolution models of climate and weather. Int. J. High Perform. Comput. Appl., 22 , 149–165.
Weisman, M. L., and J. B. Klemp, 1982: The dependence of numerically simulated convective storms on vertical wind shear and buoyancy. Mon. Wea. Rev., 110 , 504–520.
Weisman, M. L., and J. B. Klemp, 1984: The structure and classification of numerically simulated convective storms in directionally varying wind shears. Mon. Wea. Rev., 112 , 2479–2498.
Weisman, M. L., C. Davis, W. Wang, K. W. Manning, and J. B. Klemp, 2008: Experiences with 0–36-h explicit convective forecasts with the WRF-ARW model. Wea. Forecasting, 23 , 407–437.
Wentz, F. J., L. Ricciardulli, K. Hilburn, and C. Mears, 2007: How much more rain will global warming bring? Science, 317 , 233–235.
Williams, E., and N. Renno, 1993: An analysis of the conditional instability of the tropical atmosphere. Mon. Wea. Rev., 121 , 21–36.
Zajac, B. A., and S. A. Rutledge, 2001: Cloud-to-ground lightning activity in the contiguous United States from 1995 to 1999. Mon. Wea. Rev., 129 , 999–1019.
Artificial neural network response for (top) all convergence and divergence values, (middle) convergence only, and (bottom) divergence only.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Artificial neural network response for (top) all convergence and divergence values, (middle) convergence only, and (bottom) divergence only.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Artificial neural network response for (top) all convergence and divergence values, (middle) convergence only, and (bottom) divergence only.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Convective potential for climate model historical data.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Convective potential for climate model historical data.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Convective potential for climate model historical data.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Annual cycle of convective potential for climate model historical data.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Annual cycle of convective potential for climate model historical data.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Annual cycle of convective potential for climate model historical data.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Percent change in convective potential for future climate model simulation relative to the historical climate simulated baseline.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Percent change in convective potential for future climate model simulation relative to the historical climate simulated baseline.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Percent change in convective potential for future climate model simulation relative to the historical climate simulated baseline.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Percent change in future interannual variability (relative to the simulated historical baseline) as a function of day of year (dark line) along with the maximum and minimum variability values.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Percent change in future interannual variability (relative to the simulated historical baseline) as a function of day of year (dark line) along with the maximum and minimum variability values.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Percent change in future interannual variability (relative to the simulated historical baseline) as a function of day of year (dark line) along with the maximum and minimum variability values.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Changes in the severe convective potential (future, relative to historical baseline) exceeding one standard deviation. Pink (purple) indicates positive (negative) changes.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Changes in the severe convective potential (future, relative to historical baseline) exceeding one standard deviation. Pink (purple) indicates positive (negative) changes.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Changes in the severe convective potential (future, relative to historical baseline) exceeding one standard deviation. Pink (purple) indicates positive (negative) changes.
Citation: Journal of Climate 22, 12; 10.1175/2009JCLI2697.1
Percentiles for CAPE (J kg−1), convergence (10−6 s−1), and deep-layer shear (m s−1) from the historical data, stratified according to whether convection was observed (indicated by “yes”). Total number of cases is 60 521 (27 658 with observed convection).
Percentile distributions of CAPE (J kg−1), surface convergence (10−5 s−1), and deep-layer wind shear (m s−1) for the historical data and the historical climate model simulation.
Percentile distributions of historical data and associated ANN estimated potential for convection for all nonconvective and convective cases in the historical sample.
Zonal frequency of convection for the contiguous United States, for hourly station observations for 1957–95, cloud-to-ground lightning strikes from the National Lightning Detection Network for 2004–06, significant severe storm reports (Doswell et al. 2005), and the historical climate model simulation.
