Ice storm frequencies in a warmer climate

Kelly Klima & M. Granger Morgan

Climatic Change An Interdisciplinary, International Journal Devoted to the Description, Causes and Implications of Climatic Change

ISSN 0165-0009 Volume 133 Number 2

Climatic Change (2015) 133:209-222 DOI 10.1007/s10584-015-1460-9

1 23 Your article is protected by copyright and all rights are held exclusively by Springer Science +Business Media Dordrecht. This e-offprint is for personal use only and shall not be self- archived in electronic repositories. If you wish to self-archive your article, please use the accepted manuscript version for posting on your own website. You may further deposit the accepted manuscript version in any repository, provided it is only made publicly available 12 months after official publication or later and provided acknowledgement is given to the original source of publication and a link is inserted to the published article on Springer's website. The link must be accompanied by the following text: "The final publication is available at link.springer.com”.

1 23 Author's personal copy

Climatic Change (2015) 133:209–222 DOI 10.1007/s10584-015-1460-9

Ice storm frequencies in a warmer climate

Kelly Klima1 & M. Granger Morgan1

Received: 14 August 2014 /Accepted: 22 June 2015 /Published online: 3 July 2015 # Springer Science+Business Media Dordrecht 2015

Abstract Ice storms can produce extensive damage to physical infrastructure, cause deaths and injuries, and result in large losses through business interruption. Total costs can be billions of dollars. If society is to increase its resilience to such events, we need a better understanding of the likely frequency, intensity and geographical distribution of ice storms. Unfortunately, due to competing temperature and precipitation effects as well as surface effects, it is unclear how climate change will affect the frequency, intensity and geographical distribution of ice storms. Here we perform a simple “thought experiment” using vertical temperature profile data to explore how these might change given plausible future temperature regimes. As temperatures increase, we find a poleward shift and a shift toward winter. Furthermore, southern locations experience fewer ice storms at all times of the year, while northern areas experience fewer in the spring and fall and more in the winter. Using an approximation for surface effects, we estimate that a temperature increase will result in an increased frequency of ice storm events throughout much of the winter across eastern Canada and in the U.S. west of the Appalachian Mountains as far south as Tennessee. Future changes in variability may enhance or moderate these changes.

1 Introduction

An ice storm, sometimes called a silver storm, is defined by the American Meteorological Society as “a storm characterized by a fall of freezing liquid precipitation” (2014). Ice storms result in the formation of glaze, a coating of smooth, clear ice that is formed on exposed objects by the freezing of a film of liquid precipitation. Freezing rain and the associated ice deposits occur when warm moist air “overruns” shallow colder air resulting in a temperature inversion. Figure 1 shows the vertical temperature profile usually associated with ice storm conditions. Except for extreme southern portions of the United States, almost any area east of the Rocky Mountains can experience freezing rain. Freezing rain generally does not fall west

Electronic supplementary material The online version of this article (doi:10.1007/s10584-015-1460-9) contains supplementary material, which is available to authorized users.

* Kelly Klima [email protected]

1 Department of Engineering and Public Policy, Carnegie Mellon University, Pittsburg, PA, USA Author's personal copy

210 Climatic Change (2015) 133:209–222 of the Rockies because shallow Arctic air is unable to flow over the mountains (Shan et al. 1998; Changnon and Changnon 2002). Major ice storms can cause billions of dollars of damage and paralyze transportation, food and agricultural, water and sewers, and energy infrastructure (Changnon and Changnon 2002; Houston and Changnon 2006). Ice storms can be especially damaging to electric power grids due to the weight of ice on the pylons and poles, surrounding trees, and on the wires themselves (Druez et al. 1999; Wong and Miller 2010;Campbell2012), resulting in large-scale blackouts that take days, and in extreme cases, can take many weeks to restore (Hines et al. 2009). An important first step in reducing risk is to characterize a hazard’s frequency and intensity so policy makers can subsequently act on this information to mitigate vulnerability. This requires both a climatology and an understanding of how the climatology might change. Two widely known ice storm climatologies exist (Changnon and EPRI); other databases such as Cortinas et al. 2004 examine freezing rain climatologies. Since these databases were con- structed, more events have occurred along the northeast United States and southeast Canada border. Notable recent storms include the North American Ice Storm of 1998 (5″, nearly 40 deaths, $5–7 billion in damages) (National Weather Service 2014); the North Carolina ice storm of 2002 (1″, 2 dead) (North Carolina Public Utilities Commission 2003; Ahrens and Samson 2011), the January 2007 North American ice storm (4″, >85 dead, $380 million in damages) (Miller et al. 2011); and the December 2008 New England/New York (1″,4dead,$2–4 billion in damages) (Ahrens and Samson 2011; CBSNews 2014). Furthermore, the “polar vortex” of the 2013–2014 winter led to ice, snow, and cold spells throughout the United States, leading to a large amount of damage. It is unclear whether these recent events suggest a trend in ice storm frequency; in a 2014 Master’s thesis, Kovacik partially updates the records, however these results only span two decades and thus given natural variability it is difficult to assess trends. There has been even less work in understanding how future ice storms might change in intensity, frequency, or spatial or temporal distribution within North America. To our knowl- edge, all existing research in this area has estimated changes by downscaling general circula- tion models. Cheng et al. 2007 used a statistical downscaling method to downscale three general circulation model scenarios to 15 weather stations, using discriminant function analysis to project the occurrence of future weather types. While they predict warming temperatures may lead to more freezing rain events in the far north, they used downscaling and focused solely on southeastern Canada. In follow-on research, Cheng et al. 2011 expanded

Precipitaon starts as snow

Altude Warm layer melts snow

Refreezes Below freezing Above freezing Temperature

Fig. 1 Schematic of vertical temperature profile of conditions that typically result in freezing rain Author's personal copy

Climatic Change (2015) 133:209–222 211 their original study to each of 42 hourly observing stations over eastern Canada, but still use downscaling of the outputs of eight general circulation models. Lambert and Hansen (2011) conducted a study using the Canadian Global Circulation Model CGCM3 and then post- processed model output using the Ramer algorithm (1993). While the results show a northward migration and a decrease in freezing rain events in the U.S., the relatively course resolution of the model and its lack of topographic resolution in the Eastern United States, a factor that other studies have demonstrated to be highly important (Millward and Kraft 2004), significantly limits the study’s conclusions. However, there is much controversy over whether GCMs can replicate either vertical temperature profiles or changes in vertical temperature profiles. First, according to IPCC 2013, observations suggest substantial disagreement between available estimates as to the rate of temperature changes; while most observations suggest it is likely that the troposphere will warm more than the global mean surface temperature. The IPCC reports low confidence in the rate of change and the vertical structure. Specifically, observations show a warming in the lower troposphere of 0.0118–0.0203 °C/year and in the upper-middle troposphere of 0.0129–0.0265 °C/year averaged over 1981–2005 (in Po-Chedley and Fu 2012). Despite this high uncertainty, most models overestimate the observed warming trend in the tropical troposphere (Lanzante and Free 2008;Fuetal.2011;IPCC2013; McKitrick et al. 2010; Santer et al. 2013), suggesting an extra change of −0.002–0.006 °C/year over 1981–2008 (Fig. 2 in Po-Chedley and Fu 2012). Even with historical SSTs as a boundary condition, most atmospheric models exhibit excess warming (Po-Chedley and Fu 2012), suggesting that they may not appropri- ately replicate future vertical temperature profiles. Second, even were they to replicate coarse resolution vertical temperature profiles, GCMs lack sufficient vertical resolution for the altitudes relevant to ice storms. Commonly used ice storm parameterizations require fine resolution vertical profiles between the surface and approximately 500 hPa that are sufficient to resolve the positive/ negative area (Bourgouin 2000), identify the coldest layer (Ramer 1993; Baldwin et al. 1994), or the mean layer depth and temperature of the coldest layer (Czys et al. 1996). GCMs commonly do not have enough vertical resolution to resolve this. Even the North American Regional Reanalysis model, an effort with 17 pressure levels from 1000 to 500 mb, thus providing finer resolution than GCMs, lacks sufficient vertical temperature resolution to identify ice storm occurrences. To reduce uncer- tainty, the reanalysis employs a mini-ensemble ice storm parameterization. Without a higher resolution in the vertical temperature profile, one must necessarily use statistical downscaling to estimate changes in-between layers, thus obviating the attempt at using a dynamic method. While downscaling might produce climatic information at a sufficiently fine vertical resolution to resolve ice storm behavior, this process involves additional information, data, and assumptions, leading to further uncertainties and limitations of the results (USAID 2014), and due to the fine resolutions required in vertical profile, may be insufficient to characterize ice storms. Ice storm physics suggests three effects that are likely to alter ice storms in the future. First, as climate change increases temperatures (IPCC 2013; Melillo et al. 2014), the occurrence of ice storms will likely exhibit a poleward shift with the season shortening from October–March to December–January. Second, as climate change is expected to enhance precipitation (IPCC 2013; Melillo et al. 2014), areas already experiencing ice events might see an increase in frequency or intensity of ice storms, if conditions remain susceptible. At the same time, there will be a competing effect with temperature; as temperatures increase, the air can hold more water, thus possibly changing ice storm frequency and increasing intensity. Third, ice storms are clearly a function of local surface effects (Millward and Kraft 2004) as well as synoptics (Ressler et al. 2011;Splawinskietal.2011). For example, mountains can prevent cold fronts and shallow Arctic air from advancing, and valleys can “catch and hold” cold air masses which Author's personal copy

212 Climatic Change (2015) 133:209–222 can be overrun by warm air. Thus, while there will be general continent level trends, in some locations local characteristics may outweigh this signal. We believe that first step in fully understanding how gradual, long term climate change might affect ice storms is to understand how temperature changes might affect ice storm frequency. Here we asked how ice storm frequencies might change over a range of plausible future temperature regimes. Our thought experiment used historical atmospheric soundings at 38 weather stations in the eastern United States and Canada and three scenarios of warming of the vertical temperature changes from −0.5 to +5 °C. We then used the Czys algorithm for freezing rain events to estimate likely percent changes in the frequency of future ice storms.

2 Methods

We conducted a three-step thought experiment to understand what might happen if, over several decades or more, the entire vertical temperature profile increased. First, we obtained historical data. Second, we applied the warming scenario. Third, we applied the freezing rain algorithm.

2.1 Historical data

We obtained historical temperature profiles from 1973 to 2013 from the Wyoming Weather Web’s archive (University of Wyoming 2014). We focused on the eastern United States and southeastern Canada, where ice storms are more common. Since meteorologists commonly use the 850 hPa level to predict prevailing weather conditions (Holton 2004), we only chose lower altitude weather stations (2 to 1099 m above sea level). Data were provided in 12 h intervals (coinciding with the release of weather balloons); locations with high data loss were dropped. The black dots in Fig. 4 show the resulting 38 locations.

2.2 Warming scenarios

We examined a range of temperature changes spanning observed and plausible future changes for the study area. According to the Intergovernmental Panel on Climate Change (Figure SPM.1), the observed change in surface temperature over North America from 1901 to 2012 has ranged between −0.5 and +2.5 °C. Furthermore, by 2081–2100 for representative concentration pathway (RCP) 8.5, the study area could experience up to 5 °C warming (IPCC’s Figure SPM.8). Next, we had to decide whether to apply a uniform change over the study area, or alter areas differentially based on future climate change effects. Since individual GCM models show very large disagreement on local changes in temperature while the multi-modal ensemble in IPCC Figure SPM.8 shows fairly uniform increase in temperature over the entire eastern United States, we assumed that to first order that the temperature change would be uniform over the study area. Thus we examined a range of temperature changes from −0.5to+5°CaspicturedinFig.2. By then performing a Monte Carlo analysis across the range of future plausible temperature changes, we can identify results that are robust to the choice of temperature change. Given this range of possible changes in the surface temperature, we examined two future scenarios. The first, and most basic Scenario, investigates the basic physics behind temperature changes with altitude and how this will affect ice storms. We assume any change in temperatures will be uniformly mixed through the troposphere, and therefore examined a uniform increase at all pressure levels (altitudes) in the vertical temperature profile (Scenario A). Author's personal copy

Climatic Change (2015) 133:209–222 213 Altude

Below freezing Above freezing Temperature

Fig. 2 Schematic of vertical temperature profile. Schematic of uniform temperature profile increase

However, as noted in the introduction, historical observations show a warming in the lower troposphere of 0.0118–0.0203 °C/year and in the upper-middle troposphere of 0.0129–0.0265 °C/ year averaged over 1981–2005, or 9–10 % increase in warming from the lower troposphere to the upper-middle troposphere (Po-Chedley and Fu 2012). If we were to assume that this warming throughout the temperature profile continued linearly for all temperatures, this would suggest that a global surface temperature rise of 0.5–5 °C would result in an upper troposphere warming of 0.55–5.5 °C. Of course, it is unclear that the observed relation between the lower troposphere to the upper-middle troposphere will continue to hold as global surface temperatures increase toward 5 °C. GCM outputs for simulations considering all forcing agents (GHG, aerosols, stratospheric ozone changes, etc.), which currently suggest relatively large warming in the upper troposphere but currently overestimate temperature warming aloft, could be used to provide an upper bound on the extra warming in the upper-middle troposphere. Examining Fig. 12.12 in IPCC 2013,it appears the annual mean zonal temperature at 30 N could range from 0.5 to 4 °C at 1000 hPa and 1–7 °C at 850 hPa depending on emission scenario assumed. Thus in order to examine the full range of historic and possible future values, we tested 100 % change aloft of 0.5 to +7 °C with only 50–90 % of that warming (corresponding to 0.5–4 °C) at the surface. Since the 850 mb pressure level (approximately 1.5 km above sea level) is commonly thought of as above the mixed layer and much less affected by effects such as surface heating (Holton 2004), we assumed a temperature change of 0.5 to +7 °C above 850 hPa, and the full range of fractional reduction between the surface and the 850 hPa layer of 50–90 % (Scenario B). Scenario A and B attempt to bound the types of changes that could occur with the least extreme (uniform warming) and most extreme (step change at 850 hPa) derivatives in temperature changes that could occur. We considered creating more complex scenarios that would be a function of altitude, but due to the complexity of vertical temperature profile changes given in both the observations and the multi-modal runs, we feel this would overly complicate our model and our results would not be meaningful.

2.3 Freezing rain algorithm

To calculate the frequency of ice storm events, we applied a freezing rain algorithm. Since we hypothesize that changes to both temperature and precipitation would induce a complex, nonlinear response, as a first step to understanding the physical processes underlying ice storm Author's personal copy

214 Climatic Change (2015) 133:209–222 frequency we made a simplifying assumption and only examined temperature changes. We used the Czys et al. algorithm’s(1996) temperature component to predict whether freezing rain would occur. The Czys algorithm models the melting process of an individual hydrometeor; if a warm layer exists, the algorithm determines the average temperature of the layer and computes the amount of time (and thus depth of layer) required to melt completely the hydrometeor at that temperature. Compared to the Baldwin et al. 1994;Ramer(1993); and Bourgouin (2000) algorithms, Czys is thought to have a low bias for freezing rain frequencies and hence our predictions may be an underestimate of changes (NCEP 2013). We calculated the percent change in ice storm frequencies as a function of the warming scenario assumed. We averaged results over the entire period for each weather station, ignoring timestamps without a measurement.

3Results

Figure 3 shows the percent change in ice storm frequency for Scenario A (a uniform temperature change ranging from −0.5 to +5 °C) for three sample weather stations as a function of month.1 At southern locations such as Pittsburgh, PA (Fig. 3a), as temperatures increase the frequency of ice storm events decreases throughout the year. In more northern locationssuchasAlbany(Fig.3b), the trend is a function of month; as temperatures increase, the spring (March and April) and fall (November) ice storm frequencies decrease, but during the winter (December, January, and February) frequencies increase. At even more northern locations (e.g., Maniwaki, Quebec, Fig. 3c), the spring and fall continue to experience a decrease in ice storm events, but the trend is less strong and winter months experience even more ice storms, possibly up to a doubling in ice storm frequency. This likely occurs because as temperature warm, spring temperatures become too warm for ice storms, while winter temperatures are warming toward near 0 °C, a temperature range where ice storms are more likely. This results in a poleward shift and a shift toward winter in ice storm frequencies. To understand whether these results hold generally, we examined contour plots of cross- sections of the full data set. Figure 4 shows the percent change in frequency of January ice storms for a 1, 2, 3, 4, and 5 °C increase in vertical temperature profile for Scenario A. We find throughout the eastern United States, as temperatures increase, there is indeed a poleward trend: ice storms become less frequent in the south, and more frequent in the north. Figure 5 shows the percent change seasonally for a 3 °C rise for Scenario A. We find as temperatures increase, there is also a shift toward winter: southern locations experience fewer ice storms, while northern areas experience fewer in the spring/fall and more in the winter. We then tested Scenario B understand how a non-uniform temperature change simulating surface effects might affect results. Figure 6 shows the percent change in frequency of January ice storms for a 2, 3, 4, 5, 6, and 7 °C increase in vertical temperature profile for Scenario B with 100 % (50 %) of the temperature increase above (below) 850 hPa, and Fig. 7 shows the percent change in freezing rain events for a 3 °C temperature profile increase for a) November, b) December, c) January, d) February, e) March, and f) April for Scenario B with 100 % (50 %) of the temperature increase above (below) 850 hPa. Full values similar to Figs. 4 and 5 for

1 Please recall that we enacted a uniform temperature increase across all areas as opposed to differential heating in different places as might be expected from a globally averaged temperature change. For this reason the results should be interpreted only as changes in local temperature changes and not as a globally averaged warming. Author's personal copy

Climatic Change (2015) 133:209–222 215

Fig. 3 Fractional change in ice storm frequency for Scenario A. a Pittsburgh, PA (PIT), b Albany, NY (ALB), c Maniwaki, Quebec (WMW) Author's personal copy

216 Climatic Change (2015) 133:209–222

54o 54o N N a d 48o 48o N N

42o 42 o N N

36o 36o N N

30o 30o N N -100

o o 99o W 99o W W o o 63 W o o 63 90 W 81oW 72 W 90 W 81oW 72 W

54o o b N e 54 N

48o o N 48 N

4 o 42o 2 N N

36o 36o N N

30o 30o N N

o o o 99 W 63 W 99o W o o o W o o 63 90 W 81 W 72 W 90 W 81oW 72 W

150 o c 54 N 100 o 48 N

50 42 o 4 N 2 N

0 36 o N

-50 30o N

-100

9 o oW 9 W o 63 90oW o 72 W 81 W -150

Fig. 4 Percent changes in freezing rain events for a a 1, b 2, c 3, d 4, and e 5 °C temperature profile increase in January for Scenario A. Values shown are filtered with a 1 °C resolution. Stars indicate the temperature soundings used. The shaded area shows the range of zero lines for all Scenario A January runs. The spatial extent of the results is limited by the range of the radiosonde data warming below the 850 hPa of 50–90 % are given in the Supplementary Information.While Scenario A and Scenario B results are not strictly comparable,2 we find Scenario B agrees with

2 For example, it is unclear whether a Scenario A temperature increase of 2 °C would be equivalent to a Scenario B temperature increase of a) 4 °C above 850 hPa and 2 °C below 850 hPa (match the surface temperatures), b) 2 °C above 850 hPa and 1 °C below 850 hPa (match the temperatures above 850 hPa), or c) 2.66 °C above 850hPaand1.33°Cbelow850hPa(matchanaveragetemperature). Author's personal copy

Climatic Change (2015) 133:209–222 217 the general findings of Scenario A: a poleward and shift toward winter. In addition, we find that a temperature increase would result in an increased frequency of ice storm events

54o 54o a N d N

48o 48o N N

42o 42o N N

36o 36o N N

o 30o 30 N N

o o 99o 63 W 99o 63 W W o o W o o 90 W 81oW 72 W 90 W 81oW 72 W

54o 54 o b N e N

o 48o 48 N N

o 4 o 42 N 2 N

36 o 36o N N - 30 o 30o N N

o o o 99 W 63 W 99o 63 W o o oW W o o 90 W 81 W 72 90 W 81oW 72 W 150

54o 54o N c N f 100 48 o 48o N N 50

4 o 42o 2 N 0 N

0 36o -50 36 o N N

30 o -100 o N 30 N

-150 o o o 99 W 63 W 99o 3 W o o o W o o 6 90 W 81 W 72 W 90 W 81oW 72 W

Fig. 5 Percent changes in freezing rain events for a 3 °C temperature profile increase for a November, b December, c January, d February, e March, and f April for Scenario A. Values shown are filtered with a 1 °C resolution. Stars indicate the temperature soundings used. The shaded area shows the range of zero lines for all Scenario A 3 °C runs. The spatial extent of the results is limited by the range of the radiosonde data throughout much of the winter across Eastern Canada and in the U.S. west of the Appalachian Mountains, penetrating as far south as Tennessee. These areas coincide with higher altitudes, where in a warming climate we would still expect to find cold surface temperatures. Note, there are three factors contributing to uncertainty in these results: false positives (calculate ice storms but none actual reported), false negatives (no ice storms calculated when they were actually reported), and changes in events occurring at a finer time resolution than the available observational data. Regarding the first two (false positive and false negative), the Author's personal copy

218 Climatic Change (2015) 133:209–222

a 54o d 54o N N

48o 48o N N

42o 42o N N

36o 36o N N

30 o 30o N N

o o o 99 W 63 W 99o 63 W o o o W o o 90 W 81 W 72 W 90 W 81oW 72 W

5 o o b 4 N e 54 N

48o 48o N N

42o 42o N N

36o 3 o N 6 N

3 o o 0 N 30 N

o o o 150 99 W 63 W 99o W o o o W o o 63 90 W 81 W 72 W 90 W 81oW 72 W 100 c 54o 54o N f N 50

48o 48o N N 0

42o 42 o N N -50

36o 36o N N -100

o o 30 N 30 N -150

o o 99o W 99o 3 W W o o 63 W o o 6 90 W 81oW 72 W 90 W 81oW 72 W

Fig. 6 Percent changes in freezing rain events for a a 2, b 3, c 4, d 5°C,e 6 °C, and f 7 °C temperature profile increase in January for Scenario B with 100 % (50 %) of the temperature increase above (below) 850 hPa. Values shown are filtered with a 1 °C resolution. Stars indicate the temperature soundings used. The shaded area shows the range of zero lines for all Scenario B January runs. The spatial extent of the results is limited by the range of theradiosondedata

“gold standard” is of course the reported observations. The Kovacik 2014 M.S. thesis examines incidences of actual reported ice storms as a function of reported similar storms (FRZR, ice storms, etc.). We compared our results to this data set, and find good agreement. For the third type of uncertainty, radiosondes are only released twice daily. Given that ice storms and FRZR may be more likely at certain times of the day such as dawn and dusk, rather than during the two 12 h periods when the radiosondes are released, it is likely that the base Author's personal copy

Climatic Change (2015) 133:209–222 219

a 54o d 54o N N

48o 48o N N

42o 42o N N

36o 36o N N

30o 30o N N

o o 99o W 99o 3 W W o o 63 W o o 6 90 W 81oW 72 W 90 W 81oW 72 W

54o 54o b N e N

48o 48o N N

o 42o 42 N N

0 36o 36o N N

30o 30o N N

o o 99o 3 W 99o W 150 W o o 6 W o o 63 90 W 81oW 72 W 90 W 81oW 72 W

100 cf 54o 54o N N 50

48o o N 48 N 0

42o 4 o N 2 N -50 0 36o 36 o N N -100

o 30o 30 N N -150

99o oW 99o oW W o o 63 W o o 63 90 W 81oW 72 W 90 W 81oW 72 W

Fig. 7 Percent changes in freezing rain events for a 3 °C temperature profile increase for a November, b December, c January, d February, e March, and f April for Scenario B with 100 % (50 %) of the temperature increase above (below) 850 hPa. Values shown are filtered with a 1 °C resolution. Stars indicate the temperature soundings used. The shaded area shows the range of zero lines for all Scenario B 3 °C runs. The spatial extent of the results is limited by the range of the radiosonde data case number of ice storms is an underestimate. This is why we report changes from the base case: to alleviate uncertainties caused by the base value itself.

4 Discussion

This thought experiment investigated how ice storm frequency in the eastern United States might change as temperature profiles change. Prior studies examine ice storm Author's personal copy

220 Climatic Change (2015) 133:209–222 frequency changes from one GCM (Cheng et al. 2007, 2011); since future changes to vertical temperature profiles are highly uncertain, we have examined a wide range of changes to the vertical temperature profile that span both historic trends and trends suggested by GCMs, even though the latter are known to overpredict warming aloft. We find that, all else being equal, as temperatures increase, there is a poleward shift and a shift toward winter in ice storm occurrence; southern locations experience fewer ice storms, while northern areas experience fewer in the spring and fall and more in the winter. Second, surfaces effects can combine with the temperature effects to create even more winter ice storms in the United States than might be expected from a temperature shift alone. These results are robust to the vertical temperature profile changes assumed, encompassing previous GCM results. Not included in this simple analysis is the possibility that climate change might result in other phenomena such as oscillations in the location of the jet stream resulting in more frequent southerly incursion of the polar vortex, as happened in North America in the winter of 2013–14. Clearly, this thought experiment is just a first step in understanding what might happen in a future climate. While the analysis captures temperature and approximates differences in surfaces altering the probability of temperature inversions, it does not include precipitation effects. Thus, it has not captured possible changes to ice storm intensity, which could be a more important indicator than the ice storm frequency. One possibility is to examine historical temperature profiles during ice storms, binning them into “warmer” and “cooler” climates. Then, using historical data that includes complex temperature and precipitation interactions, one could try to characterize how frequencies and severities might change in a future climate. However, it is unclear that there are sufficient continuous radiosonde data at most weather stations to account for other signals causing variability, such as the El Nino-Southern Oscillation. A second possibility, informed by synoptic studies such as Ressler et al. (2011), would be to downscale General Circulation Model runs to a terrain-resolving, high resolution numerical weather model using techniques to capture the important terrain effects of cold- air damming along the eastern slopes of the Appalachians and in interior Washington State. Unfortunately, as noted above, a major limitation of these studies is that even state-of-the-art downscaling techniques cannot easily resolve ice storm occurrences. A model with much greater resolution is the North American Regional Reanalysis (NARR) dataset, but even this dataset cannot resolve ice storms, and therefore uses a mini ensemble of 3 ice storm parame- terizations to predict ice storms (Mesinger et al. 2006). Furthermore, this study does not consider instabilities in the jet steam which could sometimes bring ice storms further south, such as the 2013–2014 winter “polar vortex”.As future blocking patterns and other synoptic effects may change in a warmer climate, it is possible that increases in this variability may continue. Our results suggest that the future frequency, intensity and geographical distribution of ice storms is a topic that deserves greater attention, and that decision-makers in energy, transportation, health, and other critical infrastructures should begin considering the fact that they may need to adapt to a change in ice storms. Improved predictions would be critical to power companies, and electric grid consortia, emergency managers and local and regional governments as they develop risk mitigating strategies to provide effective and efficient essential services in a changing climate.

Acknowledgments We thank the Carnegie Mellon Electricity Industry Center and the Center for Climate and Energy Decision Making (created through a cooperative agreement between the NSF (SES-0345798) and Carnegie Mellon. Also thanks for significant support and input from the University of Wyoming for providing the temperature soundings, from Professor John Gyakum of McGill for suggestions on the design of the non- uniform temperature increase experiment, and from Carnegie Mellon’s Patti Steranchak for reviewing the manuscript. Author's personal copy

Climatic Change (2015) 133:209–222 221

References

Ahrens CD, Samson P (2011) Extreme weather & climate. Belmont, Brooks/Cole, Cengage Learning American Meteorological Society (2014) “Ice storm”. Glossary of meteorology. Last accessed July 7, 2014. http://glossary.ametsoc.org/wiki/Ice_storm Baldwin M, Treadon R, Contorno S (1994) Precipitation type prediction using a decision tree approach with NMC’s mesoscale Eta Model. Preprints, Tenth Conference on Numerical Weather Prediction, American Meteorological Society 30–31 Bourgouin P (2000) A method to determine precipitation type. Weather Forecast 15:583–592 Campbell RJ (2012) Weather-related power outages and electric system resiliency. Congressional Research Service. Report 7–5700, R42696, 18 pp CBSNews (2014) Four Deaths Blamed on Ice Storm. Last accessed July 7, 2014. http://www.cbsnews.com/ news/four-deaths-blamed-on-ice-storm/ Changnon SA, Changnon JM (2002) Major ice storms in the United States, 1949–2000. Environ Hazards 4(4): 105–111 Cheng CS, Auld H, Li G, Klaassen J, Li Q (2007) Possible impacts of climate change on freezing rain in south- central Canada using downscaled future climate scenarios. Nat Hazards Earth Syst Sci 7:71–87 Cheng CS, Li G, Auld H (2011) Possible impacts of climate change on freezing rain using downscaled future climate scenarios: updated for Eastern Canada. Atmosphere-Ocean 49(1):8–21 Cortinas JV, Bernstein BC, Robbins CC, Strapp JW (2004) An analysis of freezing rain, freezing drizzle, and ice pellets across the United States and Canada: 1976–90. Weather Forecast 19:377–390 Czys RR, Scott RW, Tang KC, Przybylinski RW, Sabones ME (1996) A physically based, nondimensional parameter for discriminating between locations of freezing rain and ice pellets. Weather Forecast 11:591–598 Druez J, McComber P, Farzaneh M (1999) Correlation between measurement of an ice detector and the mass of ice accreted on two different sized conductors. Can J Civ Eng 26(6):869–875 Fu Q, Manabe S, Johanson CM (2011) On the warming in the tropical upper troposphere: models versus observations. Geophys Res Lett 38, L15704 Hines P, Apt J, Talukdar S (2009) Large Blackouts in North America: historical trends and policy implications. Energy Policy 3:5249–5259 Holton JR (2004) An introduction to dynamic meteorology. Elsevier Academic Press, Burlington Houston TG, Changnon SA (2006) Freezing rain events: a major weather hazard in the conterminous US. Nat Hazards 40:485–494 IPCC D2013] In: Stocker TF et al Deds] Climate change 2013: The physical science basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate change. Cambridge University Press, Cambridge, 1535 pp Kovacik C A (2014) Preliminary climatology of U.S. ice storm frequency and a comparison between Northeast U.S. ice storm frequency and Teleconnections. M.S. Thesis, University of Oklahoma Lambert SJ, Hansen BK (2011) Simulated changes in the freezing rain climatology of North America under Global Warming using a coupled climate model. Atmosphere-Ocean 49(3):289–295 Lanzante JR, Free M (2008) Comparison of radiosonde and GCM vertical temperature trend profiles: effects of dataset choice and data homogenization. J Climate 21:5417–5435 McKitrick R, McIntyre S, Herman C (2010) Panel and multivariate methods for tests of trend equivalence in climate data series. Atmos Sci Lett 11:270–277 Melillo JM, Richmond TC, Yohe GW (eds) (2014) Climate change impacts in the United States: The Third National Climate Assessment. U.S. Global Change Research Program, 841 pp. doi:10.7930/J0Z31WJ2 Mesinger F, DiMego G, Kalnay E, Mitchell K, Shafran PC, Ebisuzaki W, Jović D, Woollen J, Rogers E, Berbery EH, Ek MB, Fan Y, Grumbine R, Higgins W, Li H, Lin Y, Manikin G, Parrish D, Shi W (2006) North American Regional reanalysis. Bull Am Meteorol Soc 87:343–360 Miller FP, Vandome AF, McBrewster J (eds) (2011) January 2007 North American ice storm. Online: Alphascript Publishing. 64 pp. ISBN-13: 9786134349062 Millward AA, Kraft CE (2004) Physical influences of landscape on a large-extent ecological disturbance: the Northeastern North American ice storm of 1998. Landsc Ecol 19:99–111 National Weather Service Forecast Office: Burlington, Vermont (2014) 10th Anniversary of the Devastating 1998 Ice Storm in the Northeast. Last accessed July 7, 2014. http://www.erh.noaa.gov/btv/events/IceStorm1998/ ice98.shtml North Carolina Public Utility Commission (NCPUC) (2003) Report of the North Carolina Utilities Commission and the Public Staff to the North Carolina Natural Disaster Preparedness Task Force, Response of the Electric Utilities to the December2002 Ice Storm. Raleigh, NC. 74 pp. Available: http://www.ncuc. commerce.state.nc.us/reports/part1ice.pdf Author's personal copy

222 Climatic Change (2015) 133:209–222

Personal communication. NCEP North American Regional Reanalysis modeling group. October 2013. Po-Chedley S, Fu Q (2012) Discrepancies in tropical upper tropospheric warming between atmospheric circulation models and satellites. Environ Res Lett 7, 044018. doi:10.1088/1748-9326/7/4/044018 Ramer J (1993) An empirical technique for diagnosing precipitation type from model output. In: Preprints 5th International Conference on Aviation Weather Systems. American Meteorological Society, Vienna, pp 227– 230 Ressler GM, Milrad SR, Atallah EH, Gyakum JR (2011) Synoptic-scale analysis of freezing rain events in Montreal, Quebec, Canada. Weather Forecast 27:362–378 Santer BD et al (2013) Identifying human influences on atmospheric temperature. Proc Natl Acad Sci U S A 110: 26–33 Shan L, Marr L, McCafferty RM (1998) Ice storm data base and ice severity maps. Atmos Res 46:159–168 Splawinski S, Gyakum JR, Atallah EH (2011) Atmospheric circulation structures associated with freezing rain in Quebec City, Quebec. McGill Sci Undergrad Res J 6(1):50 University of Wyoming (2014) Wyoming Weather Web. Last accessed: July 18, 2014. http://weather.uwyo.edu/ USAID (2014) A review of downscaling methods for climate change projections. U.S. Agency for International Development: Washington, DC. Available: http://www.ciesin.org/documents/Downscaling_CLEARED_ 000.pdf Wong CJ, Miller MD (2010) Guidelines for electrical transmission line structural loading ASCE manuals and reports on engineering practice No. 74. doi:10.1061/9780784410356