1. Introduction
Extreme near-surface wind speeds are a significant natural hazard, affecting crops, power lines, property, and other infrastructure. They are a major driver of natural catastrophes and insurance losses in Europe (Schwierz et al. 2010), the United States (Changnon 2009), and Canada (Sandink et al. 2019; IBC 2017), and are of consequence for the wind power (Pryor et al. 2020) and the construction industries (Schuldt et al. 2021). In urban regions, extreme winds are an important consideration for the design of tall structures, with the National Building Code of Canada (NBCC) prescribing the “design wind speed” as a design parameter representing the maximum wind load a structure must be able to withstand (Teran et al. 2022; NRC 2015). Clearly, there is societal value in better characterizing extreme wind events in urban regions and how they are affected by climate change, for the purpose of engineering design and climate change adaptation.
Modeling of extratropical wind extremes and projected changes under climate change are highly uncertain (Pryor et al. 2020). One reason for this is that the coarse resolution of global climate models (GCMs) renders them unable to directly simulate microscale and mesoscale turbulence in the atmospheric boundary layer, which are important for representing extreme wind speeds. This is reflected in an energy deficit in the simulated wind spectrum in the mesoscale range (10–100 km), and thus an underestimation of extreme wind speeds (Skamarock 2004; Larsén et al. 2012; Pryor and Hahmann 2019). Kumar et al. (2015) evaluated annual maximum wind speeds in 15 CMIP5 GCMs against the ERA-Interim dataset and found that the models performed poorly at simulating historical trends and spatial variability in mountainous regions. Jeong and Sushama (2019) produced several dynamically downscaled projections of 50-yr return-period wind speed (U50) in Canada and the contiguous United States using a limited area regional climate model (RCM) of 0.44° resolution driven by two different GCMs, and found disagreement in the sign of the projected changes for several regions. This is consistent with Pryor et al. (2012), who found strong model dependence on projections of U50 based on 13 different GCM–RCM pairings. Evidently, both GCMs and RCMs struggle to represent extreme near-surface wind speeds and their trends, and this has been a barrier for studying how they may be affected by climate change.
Such results motivate the need to more precisely evaluate how models represent the mesoscale and larger-scale dynamical processes associated with extratropical extreme near-surface winds. Cold-season extreme winds in Canada and the northern United States are typically associated with extratropical cyclones (ETCs) originating in the lee of the Rocky Mountains and tracking to the east (Alberta clippers) or northeast (Colorado lows), or originating in the Gulf of Mexico or Atlantic Ocean and tracking to the north (nor’easters) (Cheng et al. 2014; Booth et al. 2015; Letson et al. 2021). In the United States, high gust factors associated with extreme winds tend to occur in regions of high ETC activity (Letson et al. 2018), and associated ETCs are an order of magnitude more intense in both relative vorticity and anomalous sea level pressure (SLP) than the mean of cyclones which follow similar tracks (Letson et al. 2021). In mountainous regions such as the North American Cordillera, extreme winds are heavily influenced by topographically controlled winds like chinooks and foehns (Sherry and Rival 2015). ETC and topographically controlled surface wind extremes are linked to large-scale circulation features at the surface and upper troposphere. For example, Goyette (2011) studied the synoptic conditions during extreme winds in Switzerland using both reanalysis and GCMs and found characteristic patterns of SLP and 250 hPa winds, including a deep surface low and strong divergence aloft. Upper-level circulations feature synoptic structures such as Rossby waves (Wirth and Eichhorn 2014) and jet streaks (Clark et al. 2009) that are linked to high-impact surface weather. Jet streaks are local maxima of wind speed in the jet stream. They provide favorable conditions for severe weather by inducing upward vertical motion through strong upper-level divergence (Uccellini and Kocin 1987). This rising motion contributes to cyclogenesis by reducing surface pressure below. The divergence is enhanced by the curvature of Rossby waves in which jet streaks are often embedded, and is typically located in the right entrance (upstream) region and left exit (downstream) region (Clark et al. 2009). These regions are where jet streak–induced severe weather typically occurs (Uccellini and Kocin 1987). Jet streaks also contribute to the ageostrophic component of extreme winds through downward transport of high-momentum air (Durkee et al. 2012). These studies demonstrate the interest in characterizing large-scale drivers of extreme winds, and that numerical models are a potentially useful tool for doing this, despite not reliably simulating near-surface wind speeds.
The purpose of this work is to identify the large-scale atmospheric conditions associated with extreme winds in Canadian cities, to assess how well climate models used for climate change projection can represent these patterns, and how this is affected by model resolution. High-resolution modeling is computationally expensive, so our results will show where it can add value and where lower resolution suffices. We focus on urban extreme events because the population density and infrastructure in cities exposes more people to the effects of damaging impacts. We choose Canada as the study domain because relatively few recent works have investigated drivers of extreme winds here compared to other midlatitude regions like the contiguous United States and Europe.
We use weather station observations to identify extreme wind events in the historical record, and study how reanalysis and operational analysis data capture the occurrence and scale of these events. To study the effects of model resolution on these results, we first use two products from the ECMWF: the ERA5 dataset (Hersbach et al. 2020), and the high-resolution ECMWF Operational Analysis (ECMWF 2016), which we refer to as ECOA. Using composite analysis, we then identify the circulation patterns both near the surface (ETCs) and in the free troposphere (jet streaks) that are characteristic of extreme winds. This analysis provides indicators that can be used to assess how climate models that do not assimilate observations represents these circulation patterns.
Our model focus is the NCAR Community Earth System Model (CESM) (Hurrell et al. 2013; Kay et al. 2015; Zarzycki and Jablonowski 2014), including global and variable-resolution (VR-CESM; Zarzycki and Jablonowski 2014) versions of the model. The different versions permit a comparison of how model resolution affects the representation of ETC and jet streak structures associated with extreme winds. Our VR-CESM grid has spatial resolution refined over southern Ontario, chosen to cover the Greater Toronto and Hamilton Area. This region has the largest population and population density in Canada and is therefore especially sensitive to extreme weather impacts. To the knowledge of the authors, this work is the first Canada-wide survey of synoptic drivers of extreme wind events in urban regions, and the first to use VR-CESM to study extreme winds in Canada, expanding on Wang et al. (2018, 2020), which used VR-CESM to study winds in California.
2. Data and methods
a. Study domain
We study extreme wind events in seven major Canadian cities: Vancouver, Calgary, Winnipeg, Toronto, Ottawa, Montreal, and Halifax. The location of each city is shown with the ERA5 topography in Fig. 1 in the online supplemental material. These cities span the country from the west to the east coast, and their metropolitan areas contain ∼47% of Canada’s population (Government of Canada 2022). Including cities from different regions of the country probes extreme events influenced by different local factors. These include complex topography in the mountains near Vancouver and Calgary, open terrain in the Canadian prairies (Winnipeg), the Great Lakes (Toronto) and Canadian Shield (Ottawa, Montreal), and land–sea contrast on the Atlantic coast (Halifax). Most studies on extreme weather in these regions, including high-resolution modeling studies, focus on temperature and precipitation (Gula and Peltier 2012; Roy et al. 2014; Erler et al. 2015; Curry et al. 2016; Wazneh et al. 2020). We expand on these works by considering extreme winds.
b. Data sources
1) Observational data
Extreme wind events in the observed record are identified using observations of 10 m wind speed and direction at Environment and Climate Change Canada (ECCC) weather stations in the seven cities. The averaging period for measurements is 1, 2, or 10 min depending on the station, but differences between stations are inconsequential because we identify extremes at each station independently [section 2c(1)]. When possible, data from multiple stations are used in each urban region, to capture wind variability on a similar spatial scale to a reanalysis or GCM grid cell. Identifying information for each station is presented in Table 1, along with the Ruggedness Index (RIX), computed using Shuttle Radar Topography Mission elevation data at 30 m resolution (Farr et al. 2007). RIX is the percentage of locations along 3.5 km transects at 30° intervals that have a slope ≥ 30% (Letson et al. 2018). RIX well above 0% indicates complex terrain, and poor expected coupling between the station and synoptic winds. Each station selected has RIX near 0%, therefore they are expected to be representative of the regional circulation. To ensure a long enough record to characterize extreme events, only stations with at least 20 years of hourly data are selected. We subsample the observations from hourly to 6-hourly frequency to match the ECOA and CESM output.
Metadata of ECCC weather stations used for identifying historical extreme wind events. The station ID is an internal numbering system in the ECCC station database, and is provided only for reference purposes. The station ID of a weather station may change over time. Data from multiple station IDs are combined together when they share a common WMO ID or if the station coordinates are unchanged with a change of station ID. The WMO ID is a permanent number assigned to each station by the World Meteorological Organization.
2) Analysis and reanalysis data
To analyze how a numerical model represents observed extreme wind events, we use analysis and reanalysis from ECOA and ERA5. Analysis/reanalysis is produced by constraining a weather forecast model with observations through data assimilation, making it consistent with the observed record. ERA5 is produced using the ECMWF’s IFS model version CY41r2. ECOA uses newer versions of the IFS over time (starting with version CY41r2), but there have been no documented changes in the calculation of boundary layer processes. Each uses a time step of 720 s. We compare ECOA to ERA5 to investigate whether refined resolution improves the representation of extreme wind events, analogous to our study of refined resolution in CESM [section 2b(3)]. ERA5 has a spatial resolution of 0.25° latitude × 0.25° longitude (∼30 km). We subselect its hourly data to 6-hourly intervals from 1979 to 2020. ECOA has a spatial resolution of ∼9 km and provides 6-hourly data starting in 2016. We compare ECOA to observations during only the ECOA data period (2016–20). We mask the reanalysis and analysis data when observations are missing. We use the ERA5 and ECOA 10 m horizontal wind speed and components (u and υ) as near-surface winds. Mean sea level pressure (SLP) and horizontal wind speed and components on the 300 hPa pressure level are used for identifying the large-scale circulation patterns associated with extreme wind events.
3) Climate model output
We use data from three configurations of CESM. This section details each model configuration, including the CESM1 Large Ensemble and two idealized simulations designed to study the impact of model resolution. Figure 1 shows maps of the spatial resolution for each configuration. As with ERA5, we use SLP and 300 hPa winds for studying the simulated large-scale circulation patterns.
(a)–(c) Grid resolutions of the three CESM configurations used. Resolution is calculated by taking the square root of the area of the grid cell. While VR-CESM and CAM-SE-UNIF both have ne30np4 resolution away from the refinement region, they are not identical because the cubed sphere of the VR-CESM grid is rotated to place the entire refinement region on one cube face, which reduces numerical artifacts. The CAM-SE-UNIF grid uses the default orientation of the cubed-sphere grid. Note that grid line-like features in (b) and (c) are cells bordering adjacent Gauss–Lobatto–Legendre elements on the cubed-sphere grid.
Citation: Journal of Climate 36, 13; 10.1175/JCLI-D-22-0719.1
(i) CESM large ensemble
CESM-LE is a set of 40 ocean–atmosphere–land–sea ice coupled simulations of the twentieth- and twenty-first-century climates, each with slightly perturbed initial conditions (Kay et al. 2015). Its spatial resolution is 0.9° latitude × 1.25° longitude (∼110 km), and its physics time step is 1800 s (Neale et al. 2012). We use ensemble members 001–035, which were all run on the NCAR machines. Six-hourly instantaneous output for simulation years 1990–2005 is publicly available for download on the NCAR Climate Data Gateway, giving 560 total simulation years. Instantaneous 10 m winds are not archived, so we use winds on the lowest model hybrid-sigma level for identifying extreme events. We derive 300 hPa winds by linearly interpolating the wind components from the hybrid-sigma levels to pressure levels with the NCL vinth2p_ecmwf function.
(ii) VR-CESM
Spatial resolution is known to impact the representation of extreme wind speeds in climate models (Larsén et al. 2012; Pryor and Hahmann 2019). We investigate the effects of refined resolution on the synoptic conditions during extreme winds using VR-CESM with high resolution over southern Ontario. VR-CESM offers the ability to simulate regional climate at high resolution with reduced computational cost relative to global high resolution, but still more costly than uniform coarse resolution (1 year of simulation requires ∼18 000 core-hours on the Scinet Niagara cluster (Loken et al. 2010; Ponce et al. 2019), versus ∼1200 core-hours for uniform 1° resolution). VR-CESM is an atmosphere–land general circulation model (AGCM) “time-slice” simulation using the CESM2.1.0 F2000C5 component set (year 2000 boundary conditions, including greenhouse gas and aerosol concentrations) with an active atmosphere [CAM5; Neale et al. (2012), the same as CESM-LE] and land model (CLM5; Lawrence et al. 2019). CLM5 was chosen instead of CLM4 (used in CESM-LE) because its updated lake model has reduced temperature biases (Subin et al. 2012), and this simulation was first used by Zhang et al. (2022) for studying temperature in the Great Lakes region. It has been extended from 20 to 30 years for this paper. Sea surface temperatures and sea ice are prescribed using the 1990–2010 monthly climatology of CESM-LE instead of observations, to improve consistency across the simulations. Because the boundary conditions are identical for each simulation year, each year represents a sample of atmosphere–land internal variability under the prescribed forcing, analogous to a different ensemble member. The first simulation year is discarded to allow the model climate and adjust to the forcing and initial conditions (spinup).
VR-CESM employs the spectral element dynamical core of the Community Atmosphere Model (CAM-SE; Dennis et al. 2012). CAM-SE uses a cubed-sphere grid for spatial discretization, in contrast with the finite volume dynamical core (CAM-FV) and latitude–longitude grid of CESM-LE. CAM-SE uses the same horizontal diffusion and Lagrangian vertical discretization as CAM-FV, and solves the same moist hydrostatic equations. It improves upon CAM-FV by locally conserving mass and moist energy without fixers (Dennis et al. 2012). Resolution on the cubed-sphere grid is specified using numbers Ne, the number of elements on one side of a cube face, and Np, the number of points on the side of the Gauss–Lobatto–Legendre grid that makes up each element (Lauritzen et al. 2018). The grid cell size decreases from ne30np4 (∼110 km) over most of the globe to ne480np4 (∼7 km) over southern Ontario. Accordingly, the physics time step is reduced to 450 s to satisfy the Courant–Friedrichs–Lewy constraint. Regions of intermediate resolution (required for numerical stability) cover most of North America. Further details about this simulation are found in the supporting information of Zhang et al. (2022). Unlike CESM-LE, we archived 10 m wind speed output for VR-CESM and for CESM-SE-UNIF to identify extreme near-surface wind speeds. The lowest model level wind components are used for calculating the wind direction.
(iii) Uniform resolution CESM with spectral element dynamical core
To isolate the effects of refined resolution, we have performed a 30-yr time-slice simulation using identical models and boundary conditions to VR-CESM, but globally uniform ne30np4 resolution (and an 1800 s time step). We refer to this simulation as CESM-SE-UNIF. This method follows the protocol of Zarzycki et al. (2015), which studied the effects of regional refinement in VR-CESM on various aspects of the climate using comparable AGCM simulations.
c. Methods
1) Extreme event identification
We identify extreme wind speeds at each station using the 98th-percentile wind speed, excluding the boreal summer (JJA). This season is excluded because synoptic storms are most common in the extended cold season (Cheng et al. 2014), so comparably few JJA events occur in any city (supplemental Fig. 2), and the composite patterns for JJA events, particularly in the free troposphere, show different characteristics than other seasons (supplemental Figs. 3 and 4). The 98th-percentile wind speed has been used as the threshold for identifying extreme winds in numerous previous studies (Brönnimann et al. 2012; Hanley and Caballero 2012; Welker and Martius 2015; Lukens et al. 2018), and balances focus on events with the greatest impacts with having a sufficient sample size to characterize them. Although not all top-2-percentile events cause damaging impacts, less intense events can be used to determine large-scale drivers of extremes and to evaluate how models represent them (Sillmann et al. 2017). Using a local threshold is consistent with our intent to study events relevant for engineering design because NBCC design wind speeds, which determine the capacity of infrastructure to sustain wind loads, are prescribed locally. Therefore, a constant wind speed threshold would not correspond to the same impact risk in different cities. ERA5 (Betts et al. 2019; Gualtieri 2021) and CESM (Wang et al. 2018) wind speeds are negatively biased relative to observations, so using a threshold unique to each data product accounts for this bias and ensures consistency across data products. To ensure events contributing to the composites are independent, we discard events which occur within 24 h of an event with higher wind speed at the same station. Extreme events in the CESM output are identified using the same procedure with the simulated near-surface wind speed in the grid cells nearest to each station. Because the simulations do not assimilate observations, they reflect only a plausible sequence of weather, not the historical record. The 98th-percentile wind speed in CESM-LE is calculated using the distribution across all 35 ensemble members. The independence criterion only applies for events from the same ensemble member. Events from all ensemble members contribute to the composites.
Captured Extremes in Analysis and Reanalysis
To ensure the synoptic patterns in ERA5/ECOA are reflective of conditions associated with extreme winds, we identify observed extreme events whose occurrence is “captured,” i.e., coincides with occurrences in the (re)analysis data. We call an observed event “captured” when it occurs within 24 h of an extreme event in the ECMWF product, using a synoptic time scale as a coincidence criterion. We chose 24 h because a smaller window gives results sensitive to both increases and decreases of the window size. Increasing the window by 12 h increases the overall proportion of events captured by <1%, but reducing it to 12 h decreases the proportion by 5%. We use only the captured events for analyzing large-scale circulations and compare the percentage of captured events between ERA5 and ECOA to check whether finer resolution improves the ability of the IFS to capture extreme wind events.
2) Design wind speed
3) Scale of spatial variability
To characterize the typical spatial scale of the extreme wind events, we use the decorrelation length scale (DCLS) of the ERA5 and ECOA 10 m wind speed. This metric was developed for assessing the impact of model resolution on precipitation variability (De Benedetti and Moore 2017) and wind speed variability (De Benedetti and Moore 2020). The DCLS for a location is the average distance over which the variability in a given field is explained by the variability at that location. It is computed by first calculating the Pearson correlation r between the wind speed at one grid cell and all surrounding grid cells, and then calculating the average distance from the chosen grid cell to the nearest closed contour for which r equals some critical value rcrit. Following De Benedetti and Moore (2020), we use rcrit = 0.9. As noted in De Benedetti and Moore (2020), the DCLS is best interpreted as a relative measure; the distances themselves are not necessarily physically meaningful, but comparing DCLS values can determine higher and lower spatial variability. We compare the DCLS for extreme events in ERA5 to ECOA, and from the captured extremes to the events whose occurrence is not captured, to investigate differences between the captured and not-captured extreme events.
4) Composite analysis
To identify large-scale circulation patterns associated with extreme wind events, we produce composite averages of the SLP and 300 hPa wind fields for the captured events, grouping together events at stations within a single city. Composite analysis is a classical technique for studying meteorological patterns associated with extreme weather, including extreme temperatures (Grotjahn et al. 2016), precipitation (Barlow et al. 2019), cyclones (Wirth and Eichhorn 2014), and windstorms (Goyette 2011). We produce separate composites for events with observed near-surface wind direction in different quadrants of the wind rose: northeast (NE), northwest (NW), southeast (SE), and southwest (SW). This is done because extreme winds at nearby stations occasionally have different typical wind directions, and indeed, we find that the synoptic patterns vary with the local wind direction (discussed with the results, section 3b). These direction quadrants were chosen for consistency with previous literature, which found a preference for southwesterly winds among extremes in the Great Lakes region (Booth et al. 2015), and that the northwest and southwest quadrants made up the majority of high-wind events in the eastern United States (Gilliland et al. 2020). We confirm that these quadrants are appropriate for all selected cities by applying k-means clustering to the SLP and 300 hPa wind speeds for the captured events with k = 4. The typical wind direction in the most populated cluster corresponds to the most frequent quadrant in six of the seven cities (supplemental Fig. 5). The RMSE (across all stations/cities) between the average wind direction for the most populated cluster and the most populated quadrant is 7.8°, which is less than the 10° increment with which ECCC observations record the wind direction. This supports our decision to stratify the composites based on the four chosen wind direction quadrants, which has the benefit of being more physically interpretable than k-means clustering.
3. Results
a. ERA5 and ECOA wind speeds
Agreement between the ERA5 wind speeds and observations is visualized in the scatterplots in Fig. 2. In each city ERA5 winds are generally weaker than station observations [consistent with Gualtieri (2021) and Betts et al. (2019)], and the bias is larger for the extreme events than for the dataset as a whole. Other metrics such as Pearson correlation (r) and root-mean-square error (RMSE) also consistently show worse agreement between ERA5 and the station wind speeds for extremes than overall. Because the captured events [section 2c(1)], are the ones to be studied further, the metric of most interest for this part of the analysis is the percentage of extreme events captured. Overall, approximately half (48% across all cities) of the observed extremes are captured, but the captured events tend to be those with highest observed wind speed.
ERA5 10 m wind speeds vs observed station wind speeds in each selected city. Colored markers indicate extreme events observed at a station in the indicated city. Blue stars represent extreme events captured by the reanalysis (i.e., coincident occurrence in reanalysis and station data). Orange triangles represent observed extremes that did not meet the criteria to be “captured.” Gray dots represent either events not identified as extremes, or extreme events that occurred within 24 h of a stronger wind speed. Percentages in the panel titles indicate the fraction of observed extremes captured by the reanalysis, and integers indicate the total number of events in that city. Black contours show Gaussian kernel density estimates of the joint distribution of the observed and reanalysis wind speed, logarithmically spaced in base 10 with exponents −0.35 to −0.5 with increments of 1/3.
Citation: Journal of Climate 36, 13; 10.1175/JCLI-D-22-0719.1
To confirm that using only the captured events retains focus on extreme wind events relative to engineering design, we calculate U50 using all annual maxima, and compare to values computed using only annual maxima that are captured. This yields only minor differences for most stations (Table 2). The median difference across all stations is a decrease of 0.08 m s−1 for captured-only. For COP UPPER station (Calgary), using only captured events results in a large decrease of 7.7 m s−1. This may be related to the complex topography in the vicinity of the station. Similarly, Burlington (Toronto) and McNab’s Island (Halifax) see moderate decreases (1.8 and 2.2 m s−1, respectively) when excluding noncaptured annual maxima. Both of these stations are along shorelines (Lake Ontario and Halifax Harbour) and their extremes may be influenced by wind fetch. These results are sensitive to the method of fitting the Gumbel distribution to the data. Using Maximum Likelihood Estimation (MLE), instead of MoM, gives large decreases for Burlington and McNab’s Island, but only a moderate decrease for COP UPPER. Evidently, it is difficult to get robust estimates of the 50-yr return-period wind speed using samples of fewer than 50 years, but the general result is that the captured extremes are those most relevant for engineering design, because they include the majority (79%) of annual maxima used for estimating design wind speeds. Questions regarding sensitivity to the estimation method are left for future research.
Design wind speeds at each weather station used for identifying extreme wind events, calculated using all annual maximum wind speeds, and only annual maxima that are captured as extreme events by ERA5. Values are provided for estimates calculated using the method of moments (MoM) and maximum likelihood estimation (MLE). Units are m s−1. Period of study is 1979–2020.
Similar scatterplots for ECOA wind speeds are shown in Fig. 3. While ECOA has improved bias, RMSE, and correlation, the percentage of events captured does not systematically increase. Across all cities, ECOA captures 52.0% of events. Vancouver and Winnipeg do see large increases of 41% and 51% in ERA5 to 60% and 58% in ECOA, respectively. These increases are largely attributable to the change in sampling period. Restricting ERA5 to the ECOA period (2016–20), 53% of extremes in Vancouver and 58% in Winnipeg are captured in ERA5. For other cities, changes are small, and none decrease by more than 5%. The overall percentage captured for ERA5 during this period is 51.6%. ECOA does systematically capture smaller-scale spatial variability than ERA5, measured by the DCLS described in section 2c(3) (Fig. 4). We calculate the DCLS separately for captured and noncaptured extremes at each station, and average across stations to get a representative value for each city. While some cities (Toronto, Montreal, Calgary) show evidence that the noncaptured extremes are characterized by shorter DCLS than the captured extremes, the opposite is true for others (Winnipeg, Ottawa, Halifax), so no systematic relationship is identified. This is consistent with the result that ECOA does not consistently capture a greater proportion of extreme events, and hence the ability of the IFS to capture the timing of observed extreme wind events is likely not strongly controlled by model resolution. We do not calculate design wind speeds for the ECOA period, as 5 years is too small of a sample. We conclude that ERA5 can reliably capture the timing of occurrence of the strongest observed wind events at the selected stations, and it has no fundamental difference with ECOA in this regard.
As in Fig. 2, but with ECOA 10 m wind speeds from 2016 to 2020.
Citation: Journal of Climate 36, 13; 10.1175/JCLI-D-22-0719.1
Decorrelation length scales for observed extreme wind events, calculated using both ERA5 and ECOA 10 m wind speeds. Warm colors indicate extremes not captured by the (re)analysis product, and cool colors indicate captured extremes.
Citation: Journal of Climate 36, 13; 10.1175/JCLI-D-22-0719.1
b. Large-scale circulation patterns in ERA5
1) Sea level pressure
Figure 5 shows composite maps of SLP anomalies for captured events with local wind direction in the most common quadrant for each city. The large scale nature of the features identified in the composites suggest that they should be able to be represented in climate models. Anomalies are calculated by subtracting the season average SLP from the field at the time of each extreme, and subtracting the zonal mean. We focus on the most common wind direction quadrant to characterize the most typical synoptic conditions associated with extreme winds. Consistent with Gilliland et al. (2020), the predominant near-surface wind directions are westerly (northwest and southwest quadrants), though southeast is most common for Vancouver and Halifax. All composites show a negative pressure anomaly located near the city, and the majority show a dipole or wave pattern, with a high-pressure anomaly on the opposite side of the city. Composites using ECOA are nearly identical to ERA5 (not shown). Composites for noncaptured events have weaker SLP anomalies than the captured events (supplemental Fig. 6), consistent with the result that captured events typically have higher wind speed (Fig. 2). The displacement between the SLP minimum (cyclone center) and the location of the extreme wind speeds is consistent with Welker and Martius (2015). We add that the location of the SLP minimum relative to the city partially determines the local wind direction because the direction of the geostrophic wind at the city diagnosed from the SLP (e.g., blowing with low pressure to the left) is in the quadrant of the extreme winds. An exception is extremes in the NE quadrant, where the geostrophic wind is southeasterly. Ageostrophic processes must contribute significantly for these events to have northeasterly wind. This is discussed further along with the results for composites in CESM [section 3c(1)]. We note that the spatial patterns seen in the SLP composites are not unique to extreme (versus moderately strong) winds, but the magnitude of the anomalies is. Composites for wind speeds between the 70th and 72nd percentiles show qualitatively similar spatial patterns but have SLP anomalies with much lower magnitude (supplemental Fig. 7). Thus, the composites do contain useful information about extreme wind events specifically.
Composites of ERA5 mean sea level pressure anomalies (with zonal mean removed) for captured extremes in each city for the most common near-surface wind direction among the extreme events at that city. Stippling indicates where the SLP deviates significantly from the seasonal climatology at the 5% level as computed using a t test with false discovery rate correction. Percentages in the panel titles indicate the frequency of the indicated wind direction for extreme events in the corresponding city. Red dots indicate the locations of the cities.
Citation: Journal of Climate 36, 13; 10.1175/JCLI-D-22-0719.1
2) Upper-level winds
Figure 6 shows composites of the 300 hPa wind field for the same events as Fig. 5. The majority of cases show a local maximum of wind speed located near the city of interest, indicating the presence of an upper-level jet streak. The composite jet streaks are typically located in regions of cyclonic curvature (troughs), especially for the eastern Canadian cities (Figs. 6d–g), but also for Vancouver (Fig. 6a), albeit with weaker curvature. In each of these cases except for Halifax (Fig. 6g) the city is located near the left exit region (e.g., poleward of the downstream end of the jet streak). The Halifax composite has a second jet streak north of the city for which Halifax lies in the right entrance (upstream) region. The upper-level divergence associated with the jet streak is evident in composites of 300 hPa velocity potential and divergent wind (Fig. 7). Strong divergence is present around the local minima in the velocity potential. These minima align closely with the negative SLP anomalies for the same events, supporting the notion that surface cyclones are induced or strengthened by jet streaks. While the composites for Calgary and Winnipeg do show significant upper-level wind speed enhancements (Figs. 6b,c) and moderate upper-level divergence (Figs. 7b,c), the jet streak winds are notably weaker than for the other cities. This is true for events in each quadrant (not shown). Therefore, some extreme events are related to jet streaks, but there may be different processes primarily responsible for extreme surface winds in these cities and the composites may not be characteristic of typical extremes.
As in Fig. 5, but showing the composite 300 hPa wind field for captured extremes in each city. Shading indicates the wind speed, and arrows indicate the wind direction. Cyan contours surround regions where the 300 hPa wind speed is significantly greater than the seasonal climatology at the 5% level as computed using a t test with false discovery rate correction. Wind direction in the panel titles indicate the quadrant of the observed near-surface wind direction for the events which contribute to the composite.
Citation: Journal of Climate 36, 13; 10.1175/JCLI-D-22-0719.1
As in Fig. 6, but showing the composite 300 hPa velocity potential (shading) and divergent wind vectors (arrows). Purple stars indicate the location of the minimum sea level pressure in the composites in Fig. 5.
Citation: Journal of Climate 36, 13; 10.1175/JCLI-D-22-0719.1
In summary, we have identified characteristic surface and upper-level circulation patterns associated with extreme wind events in major Canadian cities, by compositing ERA5 300 hPa winds and SLP during observed extreme wind events (except for Calgary and Winnipeg, which only show characteristic SLP patterns). In the following section, we will use these composites to evaluate the representation of urban extreme wind events in climate model simulations that do not assimilate meteorological observations.
c. Large-scale circulation patterns in CESM
1) Sea level pressure
To assess how a free-running climate model represents the identified circulation patterns, we produce the same composites in CESM-LE, VR-CESM, and CESM-SE-UNIF. Figure 8 shows the SLP composites for Toronto extreme events in both the SW (Figs. 8a–d) and NE (Figs. 8e–h) quadrants. While SW remains the most common direction for extreme winds in Toronto in CESM-LE, NW is the most common in VR-CESM with 46% of the extreme events, and in CESM-SE-UNIF with 44%. The SW composites for the models all strongly resemble the ERA5 composite in both magnitude and spatial pattern, except for the absence of a moderate high-pressure anomaly over western North America. For this case, the refined resolution does not appear to add value since the coarse-resolution models agree well with ERA5. In contrast, the composites for the less frequent NE-quadrant events shown in Figs. 8e–h display a case for which the refined resolution adds value. The ERA5 composite for NE extremes in Toronto (Fig. 8e) shows evidence of strong ageostrophic circulation, with the pressure dipole pattern oriented to the northeast, and southeasterly geostrophic wind. VR-CESM captures the tilted dipole pattern much better than the coarse-resolution models (Fig. 8f). The composites for CESM-SE-UNIF and CESM-LE (Figs. 8g,h) also imply strong ageostrophic flow, but the geostrophic wind has a larger easterly component than ERA5 because the pressure dipole is oriented meridionally. Composites for NE extreme winds in Montreal, Ottawa, and Halifax also show a tilted pressure dipole, and improvements in VR-CESM. Therefore, the higher resolution of VR-CESM over these cities can contribute to improved representation of ageostrophic circulations associated with extreme winds in Canada, but circulation systems for the most common wind directions, which do not have such strong ageostrophic flow, are adequately captured by coarse-resolution models.
Composites of sea level pressure during Toronto extreme wind events from the SW and NE quadrants, in (a),(e) ERA5 and (b)–(d),(f)–(h) the three configurations of CESM. Percentages in the panel titles indicate the proportion of the total extreme wind events in that data product with wind direction in that quadrant. Red dots indicate the location of Toronto. The plot in (a) is identical to Fig. 5d. Stippling shows statistical significance in the same manner as described in Fig. 5.
Citation: Journal of Climate 36, 13; 10.1175/JCLI-D-22-0719.1
To quantify how well CESM reproduces the ERA5 composites we calculate metrics of agreement (pattern correlation, RMSE, and standard deviation ratio) with ERA5 for each model configuration, and all cities and wind directions. The relevant features in the composites are localized near the city, so we calculate these statistics for a 40° × 40° domain centered at the city of interest to avoid inflating the measures of model skill. The domain size was chosen to include only the SLP and 300 hPa wind features near the city associated with extreme winds. We summarize these statistics using Taylor diagrams (Figs. 9a–c), plotted separately for each model for ease of inspection. The markers in the VR-CESM diagram (Fig. 9a) are more tightly clustered than for the other two models, with the exception of the composites for Vancouver and Calgary, which are poorly reproduced in all simulations. These cities are farthest away from the region of greatest refinement, so improvements are not expected. In support of the finding that VR-CESM improves representation of ageostrophic circulations, the NE composites for Toronto, Ottawa, Montreal, and Halifax all show better agreement with ERA5. VR-CESM has higher pattern correlations and lower RMSE than the coarse-resolution models. While the performance of CESM-LE and CESM-SE-UNIF is generally good for the eastern cities, VR-CESM shows clear improvement regarding the synoptic-scale surface variability associated with extreme winds in Canada.
Taylor diagrams for composites of (a)–(c) SLP and (d)–(f) 300 hPa wind speed in (top) VR-CESM, (middle) CESM-SE-UNIF, and (bottom) CESM-LE. On a Taylor diagram, the cosine of the azimuthal coordinate is the pattern correlation between the model and ERA5 composites, and the radial coordinate is the ratio between the spatial standard deviations of the model and ERA5 composites. The radii of the semicircles centered at the reference point are the centered root-mean-square error (RMSE) between the ERA5 and model composites, scaled by the ERA5 standard deviation. Arrows represent the wind direction quadrant for events contributing to each composite, while colors indicate the city. Note that certain results for Vancouver do not fall within the axis limits.
Citation: Journal of Climate 36, 13; 10.1175/JCLI-D-22-0719.1
2) Upper-level winds
Composites of 300 hPa winds for Toronto extreme events in the SW and NW quadrants (the two most frequent wind quadrants) are shown in Fig. 10. While the CESM composites do show enhanced wind speed and curvature somewhat indicative of a jet streak pattern like in ERA5, they underestimate both the magnitude of the wind speed within the jet streak and the strength of the curvature of the flow, especially for the SW events (Figs. 10a–d). In both cases, the VR-CESM and CESM-SE-UNIF composites show stronger upper-level wind speed and flow curvature than CESM-LE, closer to the ERA5 composites (Figs. 10a,e). An overall assessment of the agreement with ERA5 is depicted in the Taylor diagrams (Figs. 9d–f). As with the SLP anomalies, the CESM results for Vancouver and Calgary generally show poor agreement with ERA5, and VR-CESM shows improved results for Toronto, Montreal, Halifax, and Ottawa. For these cities, CESM-SE-UNIF also shows improvements relative to CESM-LE, consistent with the result that both the VR-CESM and CESM-SE-UNIF composites have jet streak winds speeds closer to those in the ERA5 composites than CESM-LE (Fig. 10). These results show how both model resolution and the numerics of the dynamical core affect the representation of upper-level jet streaks associated with extreme near-surface winds.
As in Fig. 8, but 300 hPa winds for extreme wind events in Toronto from the SW and NW quadrants. Shading shows the 300 hPa wind speed, and arrows indicate the wind direction. Cyan contours indicate regions where the wind speed is significantly greater than the season average, as in Fig. 6. Wind direction in the panel titles indicate the quadrant of the observed (ERA5) or simulated (CESM) near-surface wind direction for the events which contribute to the composite.
Citation: Journal of Climate 36, 13; 10.1175/JCLI-D-22-0719.1
To investigate why the composite jet streaks in CESM-LE are weaker than CESM-SE-UNIF and VR-CESM, we consider two possible explanations. The first is that CESM underestimates jet streak wind speeds, and this bias is less severe in VR-CESM. The second possibility is that CESM does simulate strong jet streaks, but the association between jet streaks and extreme surface winds is weaker. To investigate, we look for strong jet streaks in CESM over a city by calculating the pattern correlation between the CESM 300 hPa wind speed at all times and ERA5 composite 300 hPa wind speed for each direction quadrant. To focus on the jet streak, pattern correlations are calculated over the same 40° × 40° domain centered at the city of interest used for the Taylor diagrams. Timestamps of occurrence are sorted by pattern correlation, and times with the top N correlations are selected, where N is the number of events included in the CESM composite for that wind direction quadrant. For comparison, we do the same calculations for ERA5 300 hPa winds and the ERA5 composites. Composites of the 300 hPa winds during these times of maximum pattern correlation are shown in Fig. 11, and all have much higher upper-level wind speeds than their corresponding composites in Fig. 10. This provides evidence that there are strong upper-level jet streak events in each configuration of CESM, and thus the weak upper level winds in CESM are not simply due to an overall bias in 300 hPa wind speed.
Composites of 300 hPa winds for times when the 300 hPa wind speed has maximum pattern correlation with the ERA5 composite for SW extreme winds in Toronto. Shading indicates the 300 hPa wind speed, while the arrows indicate the 300 hPa wind direction. The red boxes show the domain over which the pattern correlations are calculated. Red dots indicate the location of Toronto.
Citation: Journal of Climate 36, 13; 10.1175/JCLI-D-22-0719.1
Having established that strong jet streaks are present in the CESM simulations, we investigate whether the strength of the relationship between upper-level jet streaks and extreme near-surface winds is stronger in CESM-SE-UNIF and VR-CESM than CESM-LE. We quantify this relationship using the probability that an extreme wind event occurs, conditioned on the presence of a strong jet streak (as defined above, via maximum pattern correlation with ERA5). To calculate this, we fit a Gaussian kernel density estimator to the distribution of surface winds near each city during the times identified as having strong jet streaks, and integrate it from the city’s 98th-percentile wind speed to infinity. Since jet streaks may cause severe weather in an extended region, not just a single grid cell, we use the maximum wind speed across the four grid cells nearest to each city for the CESM data. We do the same calculation for observations using the maximum wind speed across stations in each city, and the times of strong jet streaks identified in ERA5.
The conditional probabilities for each model and city are plotted in Fig. 12a. For most cities, one or both of VR-CESM and CESM-SE-UNIF have a stronger association between upper-level jet streaks and extreme near-surface winds than CESM-LE. The differences are largest for Toronto and Montreal, the two cities for which the observed conditional probability is greatest. These cities are within regions of very high resolution in VR-CESM, and show substantially higher conditional probabilities in VR-CESM than CESM-SE-UNIF or CESM-LE, suggesting that refined horizontal resolution plays a role in the vertical coherence of extreme winds. Ottawa, also in this region, does not have higher probability in VR-CESM, but for this city the other simulations already agree well with observations. While this metric is somewhat sensitive to the size of the domain from which model near-surface winds are selected, particularly for VR-CESM, where a larger domain means a much larger number of grid cells, we believe it to be robust. Our metric suggests that an upper-level jet streak over Winnipeg only weakly increases the chance of an extreme wind event, which is consistent with the weak jet streak in the Winnipeg composite (Fig. 6c). Further, the results are similar to the climatological correlations (i.e., over the complete dataset) between 300 hPa wind speed averaged over the 40° × 40° domain around the city, and the maximum wind speed across the 4 grid cells nearest to the city (Fig. 12b). All cities except Vancouver and Calgary have higher correlations in VR-CESM and CESM-SE-UNIF than CESM-LE, and Winnipeg remains the city with by far the weakest relationship between upper-level and near-surface winds.
(a) Conditional probabilities of the occurrence of an extreme wind event in each city in each model, given the presence of a strong upper-level jet streak. (b) Pearson correlations between 300 hPa wind, averaged over a 40° × 40° domain centered at the city, and maximum near-surface wind speed in the four grid cells nearest to the city (or across all stations nearest to each city, for ERA5/OBS). Results are unchanged when using the Spearman rank correlation instead of the Pearson correlation (not shown).
Citation: Journal of Climate 36, 13; 10.1175/JCLI-D-22-0719.1
This result is visualized through composite vertical profiles of the horizontal wind speed in Fig. 13. Like Fig. 10, ERA5 has higher peak wind speed than all three CESM simulations, but both VR-CESM and CESM-SE-UNIF show a tighter meridional structure and stronger vertical wind shear than CESM-LE, which most poorly resembles ERA5. Vectors indicating the meridional and vertical (pressure) velocities overlaid on the plot confirm the presence of upward vertical motion to the north of Toronto, coinciding with the region of upper-level divergence in Fig. 7. Our findings suggest that refined horizontal resolution and use of the CAM-SE dynamical core improves the vertical structure of wind speed for both extreme and nonextreme winds. This helps explain why CESM-LE shows the weakest jet streak winds in Fig. 10, but complete understanding of the mechanism by which resolution affects this vertical coupling is left for future study. Because of these issues regarding the vertical structure of winds in CESM-LE, it may not be a reliable model for studying jet streaks relating to extreme surface winds.
Composites of the vertical profile of sector zonal-average horizontal wind speed during Toronto extreme wind events from the SW and NW quadrants, in (a),(e) ERA5 and (b)–(d),(f)–(h) the three configurations of CESM. Sector averages are taken over a 40°-wide longitude band centered at the city’s longitude. Vectors indicate the direction and magnitude of the meridional and vertical velocities, with the vertical velocity expressed in pressure units (omega). 6-hourly Omega from CESM-LE is not available for download, so it was diagnosed from the instantaneous horizontal wind components and surface pressure output using the NCL function omega_ccm_driver. Wind direction in the panel titles indicate the quadrant of the near-surface wind direction for the events that contribute to the composite.
Citation: Journal of Climate 36, 13; 10.1175/JCLI-D-22-0719.1
4. Discussion and conclusions
Extreme wind speeds in urban regions pose hazards to human safety and infrastructure, therefore it is important to understand the processes that drive them, and how they may be affected by a changing climate. Confidence in future projections of extreme winds is low, because models and reanalysis struggle to reproduce the upper tail of the observed near-surface wind speed distribution and historical trends, and disagree regarding future trends. However, this does not necessarily mean that models do not contain useful information regarding extreme winds, especially those driven by synoptic-scale processes.
Our results show that the ERA5 captures approximately half (48%) of observed extreme wind events as extremes relative to its own biased wind speed distribution. With bias correction, ERA5 could be a useful tool for estimating design wind speeds, since it captures the majority (79%) of the annual-maximum wind speed events used for this calculation. We examined whether the higher resolution of ECOA would provide additional value for this study, and although it simulates finer-scale wind speed variability, it does not capture a substantially greater proportion of observed extreme wind events (52.0% and 70 of 78 annual maxima, versus 51.6% and 65 for ERA5 2016–20). We have identified circulation patterns in ERA5 in both the lower (all cities) and upper (except Calgary and Winnipeg) troposphere characteristic of extreme wind events in Canadian cities, which vary with the local wind direction. Near the surface, these events are associated with deep SLP minima that are part of synoptic-scale wave or dipole patterns (Fig. 5). The direction of the extreme winds is typically consistent with the direction of the geostrophic wind driven by the SLP anomalies, though events with northeasterly winds show strong ageostrophic flow. In the free troposphere, the extremes are associated with upper-level jet streaks (Fig. 6) that provide favorable conditions for cyclogenesis. It is possible that our focus on extreme events that occur as extremes in reanalysis winds may bias our analysis toward events with large scale structure. However, the captured events typically have stronger observed wind speeds than the noncaptured events (Figs. 2 and 3) and include most of the annual maxima used for estimating design wind speeds, and are thus most relevant regarding impacts on society and infrastructure.
We investigated how well GCM can represent the circulation patterns identified in ERA5 with three CESM simulations, one with regionally refined resolution. Each model configuration can skillfully reproduce the ERA5 SLP patterns for cities in central and eastern Canada, with pattern correlations close to or exceeding 0.9, standard deviation ratios near 1, and normalized RMSEs typically under 0.5 (Fig. 9). The models perform poorly for extremes in Vancouver and Calgary, possibly due to the influence of complex terrain, but further study and refined modeling of these areas is required. We find that regional refinement from CESM-SE-UNIF to VR-CESM improves the representation of near-surface ageostrophic circulations associated with extreme winds and vertical coupling between upper-level jet streaks and surface wind extremes. A limitation of our study of refined resolution is the focus on only southern Ontario, but because we see improvements for the NE composites in multiple cities, and also because jet streak circulations, which are themselves strongly ageostrophic, (Uccellini and Kocin 1987), are associated with midlatitude extreme weather more generally, we are confident that the result that refined resolution improves ageostrophic circulations is robust. We also find that CESM-SE-UNIF and VR-CESM, which use CAM-SE, show more realistic vertical coupling than CESM-LE, which uses CAM-FV, but differences cannot be attributed to the dynamical core because the SE and FV simulations use different boundary conditions, and the SE simulations do not include an active ocean model.
Agreement in the composite patterns from free-running model simulations and reanalysis builds confidence that models can credibly simulate the large-scale dynamical processes that drive extreme near-surface winds. This in turn supports the use of multivariate methods for statistical downscaling of extreme wind speeds, such as those used in Cheng et al. (2014) and Pryor and Barthelmie (2014). Where representation of the large-scale circulation patterns is poor, such as Vancouver or Calgary, statistically downscaled wind speeds will likely be contaminated with errors relating to errors in the large-scale circulation. We have shown that measures of consistency with ERA5 vary with both location and the local wind direction, so these factors should be considered when assessing confidence in model results regarding extreme winds. Mao and Monahan (2017) report that statistical downscaling skill for near-surface wind speed depends on wind direction (anisotropy), even away from complex topography. This may be because the skill in representing the large-scale circulation varies with the local wind direction. We find that northeasterly extreme winds in central and eastern Canada show strong ageostrophic circulation that is not well represented at coarse resolution, so the reported anisotropy could be related to the relative contribution of the ageostrophic wind to the total wind. Durkee et al. (2012) note that the ageostrophic wind can be decomposed into terms representing different physical processes. Of interest for our results is the isallobaric component of the ageostrophic wind, which is proportional to the local pressure tendency and points toward decreasing pressure, and Ekman friction in the planetary boundary layer, which rotates the wind toward lower pressure. This is consistent with the direction of the total wind for the NE composites. These processes, among others, may be responsible for the strong ageostrophic circulations in the NE composites, and may be better represented at higher resolution. Further investigation into this matter is planned. At this point, we have demonstrated that for cases such as the NE extremes, when ageostrophic winds are especially important, dynamical downscaling may be required to skillfully simulate both the local and synoptic-scale variability associated with extreme near-surface wind speeds. Additional simulations with refinement in regions with worse performance by coarse-resolution models such as western Canada (Erler et al. 2015; Wu et al. 2017) are planned as a part of upcoming work to more comprehensively quantify the effect of refined resolution on model representation of extreme winds in Canada.
Acknowledgments.
The authors acknowledge funding from the Natural Sciences and Engineering Research Council of Canada (NSERC) and the Centre for Climate Science and Engineering at the University of Toronto (Dean’s Strategic Fund DSF 18-30). Computations were performed on the Niagara supercomputer at the SciNet HPC Consortium. SciNet is funded by the Canada Foundation for Innovation; the Government of Ontario; Ontario Research Fund–Research Excellence; and the University of Toronto. This material is based in part on work supported by the National Center for Atmospheric Research (NCAR), which is a major facility sponsored by the National Science Foundation (NSF) and managed by the University Corporation for Atmospheric Research. The CESM project is supported primarily by the NSF.
Data availability statement.
ECCC weather station data are freely available for download at https://climate.weather.gc.ca/historical_data/search_historic_data_e.html. ERA5 data were retrieved from the Copernicus Climate Data Store (pressure levels: https://doi.org/10.24381/cds.bd0915c6, single levels: https://doi.org/10.24381/cds.adbb2d47). ECOA data were retrieved from the NCAR Research Data Archive, Dataset 113.1 (DOI: 10.5065/D68050ZV). CESM-LE data were retrieved from the NCAR Climate Data Gateway (https://www.cesm.ucar.edu/projects/community-projects/LENS/data-sets.html). SRTM 30 m resolution topography data were retrieved from https://portal.opentopography.org/raster?opentopoID=OTSRTM.082015.4326.1. Postprocessed output from the VR-CESM and CESM-SE-UNIF simulations, required to reproduce the figures, along with the SCRIP and EXODUS files for the variable-resolution grid, are archived at https://doi.org/10.5683/SP3/LFUYB3.
REFERENCES
Barlow, M., and Coauthors, 2019: North American extreme precipitation events and related large-scale meteorological patterns: A review of statistical methods, dynamics, modeling, and trends. Climate Dyn., 53, 6835–6875, https://doi.org/10.1007/s00382-019-04958-z.
Betts, A. K., D. Z. Chan, and R. L. Desjardins, 2019: Near-surface biases in ERA5 over the Canadian Prairies. Front. Environ. Sci., 7, 129, https://doi.org/10.3389/fenvs.2019.00129.
Booth, J. F., H. E. Rieder, D. E. Lee, and Y. Kushnir, 2015: The paths of extratropical cyclones associated with wintertime high-wind events in the northeastern United States. J. Appl. Meteor. Climatol., 54, 1871–1885, https://doi.org/10.1175/JAMC-D-14-0320.1.
Brönnimann, S., O. Martius, H. von Waldow, C. Welker, J. Luterbacher, G. P. Compo, P. D. Sardeshmukh, and T. Usbeck, 2012: Extreme winds at northern mid-latitudes since 1871. Meteor. Z., 21, 13–27, https://doi.org/10.1127/0941-2948/2012/0337.
Changnon, S. A., 2009: Temporal and spatial distributions of wind storm damages in the United States. Climatic Change, 94, 473–482, https://doi.org/10.1007/s10584-008-9518-6.
Cheng, C. S., E. Lopes, C. Fu, and Z. Huang, 2014: Possible impacts of climate change on wind gusts under downscaled future climate conditions: Updated for Canada. J. Climate, 27, 1255–1270, https://doi.org/10.1175/JCLI-D-13-00020.1.
Clark, A. J., C. J. Schaffer, W. A. Gallus Jr., and K. Johnson-O’Mara, 2009: Climatology of storm reports relative to upper-level jet streaks. Wea. Forecasting, 24, 1032–1051, https://doi.org/10.1175/2009WAF2222216.1.
Curry, C. L., B. Tencer, K. Whan, A. J. Weaver, M. Giguère, and E. Wiebe, 2016: Searching for added value in simulating climate extremes with a high-resolution regional climate model over western Canada. Atmos.–Ocean, 54, 364–384, https://doi.org/10.1080/07055900.2016.1158146.
Dawson, A., 2016: Windspharm: A high-level library for global wind field computations using spherical harmonics. J. Open Res. Software, 4, e31, https://doi.org/10.5334/jors.129.
De Benedetti, M., and G. W. K. Moore, 2017: Impact of resolution on the representation of precipitation variability associated with the ITCZ. Geophys. Res. Lett., 44, 12 519–12 526, https://doi.org/10.1002/2017GL075714.
De Benedetti, M., and G. W. K. Moore, 2020: Impact of model resolution on the representation of the wind speed field: An example from the United Kingdom. Quart. J. Roy. Meteor. Soc., 146, 367–379, https://doi.org/10.1002/qj.3680.
Dennis, J. M., and Coauthors, 2012: CAM-SE: A scalable spectral element dynamical core for the Community Atmosphere Model. Int. J. High Perform. Comput. Appl., 26, 74–89, https://doi.org/10.1177/1094342011428142.
Durkee, J. D., C. M. Fuhrmann, J. A. Knox, and J. D. Frye, 2012: Ageostrophic contributions to a non-convective high wind event in the great lakes region. Natl. Wea. Dig., 36 (1), 1–20.
ECMWF, 2016: ECMWF IFS CY41r2 high-resolution operational forecasts. National Center for Atmospheric Research, accessed 2 December 2021, https://doi.org/10.5065/D68050ZV.
Erler, A. R., W. R. Peltier, and M. D’Orgeville, 2015: Dynamically downscaled high-resolution hydroclimate projections for western Canada. J. Climate, 28, 423–450, https://doi.org/10.1175/JCLI-D-14-00174.1.
Farr, T. G., and Coauthors, 2007: The Shuttle Radar Topography Mission. Rev. Geophys., 45, RG2004, https://doi.org/10.1029/2005RG000183.
Gilliland, J. M., A. W. Black, J. D. Durkee, and V. A. Murley, 2020: A climatology of high-wind events for the eastern United States. Int. J. Climatol., 40, 723–738, https://doi.org/10.1002/joc.6233.
Government of Canada, 2022: Population estimates, July 1, by census metropolitan area and census agglomeration, 2016 boundaries. Accessed 13 January 2022, https://www150.statcan.gc.ca/t1/tbl1/en/tv.action?pid=1710013501.
Goyette, S., 2011: Synoptic conditions of extreme windstorms over Switzerland in a changing climate. Climate Dyn., 36, 845–866, https://doi.org/10.1007/s00382-010-0820-4.
Grotjahn, R., and Coauthors, 2016: North American extreme temperature events and related large scale meteorological patterns: A review of statistical methods, dynamics, modeling, and trends. Climate Dyn., 46, 1151–1184, https://doi.org/10.1007/s00382-015-2638-6.
Gualtieri, G., 2021: Reliability of ERA5 reanalysis data for wind resource assessment: A comparison against tall towers. Energies, 14, 4169, https://doi.org/10.3390/en14144169.
Gula, J., and W. R. Peltier, 2012: Dynamical downscaling over the great lakes basin of North America using the WRF regional climate model: The impact of the great lakes system on regional greenhouse warming. J. Climate, 25, 7723–7742, https://doi.org/10.1175/JCLI-D-11-00388.1.
Hanley, J., and R. Caballero, 2012: The role of large-scale atmospheric flow and Rossby wave breaking in the evolution of extreme windstorms over Europe. Geophys. Res. Lett., 39, L21708, https://doi.org/10.1029/2012GL053408.
Hersbach, H., and Coauthors, 2020: The ERA5 global reanalysis. Quart. J. Roy. Meteor. Soc., 146, 1999–2049, https://doi.org/10.1002/qj.3803.
Hurrell, J. W., and Coauthors, 2013: The Community Earth System Model: A framework for collaborative research. Bull. Amer. Meteor. Soc., 94, 1339–1360, https://doi.org/10.1175/BAMS-D-12-00121.1.
IBC, 2017: Facts of the property and casualty insurance industry in Canada. IBC Tech. Rep., 66 pp., http://assets.ibc.ca/Documents/Facts%20Book/Facts_Book/2017/Fact-Book-2017.pdf.
Jeong, D. I., and L. Sushama, 2019: Projected changes to mean and extreme surface wind speeds for North America based on regional climate model simulations. Atmosphere, 10, 497, https://doi.org/10.3390/atmos10090497.
Kay, J. E., and Coauthors, 2015: The Community Earth System Model (CESM) large ensemble project: A community resource for studying climate change in the presence of internal climate variability. Bull. Amer. Meteor. Soc., 96, 1333–1349, https://doi.org/10.1175/BAMS-D-13-00255.1.
Kumar, D., V. Mishra, and A. R. Ganguly, 2015: Evaluating wind extremes in CMIP5 climate models. Climate Dyn., 45, 441–453, https://doi.org/10.1007/s00382-014-2306-2.
Larsén, X. G., S. Ott, J. Badger, A. N. Hahmann, and J. Mann, 2012: Recipes for correcting the impact of effective mesoscale resolution on the estimation of extreme winds. J. Appl. Meteor. Climatol., 51, 521–533, https://doi.org/10.1175/JAMC-D-11-090.1.
Lauritzen, P. H., and Coauthors, 2018: NCAR release of CAM-SE in CESM2.0: A reformulation of the spectral element dynamical core in dry-mass vertical coordinates with comprehensive treatment of condensates and energy. J. Adv. Model. Earth Syst., 10, 1537–1570, https://doi.org/10.1029/2017MS001257.
Lawrence, D. M., and Coauthors, 2019: The Community Land Model version 5: Description of new features, benchmarking, and impact of forcing uncertainty. J. Adv. Model. Earth Syst., 11, 4245–4287, https://doi.org/10.1029/2018MS001583.
Letson, F., S. C. Pryor, R. J. Barthelmie, and W. Hu, 2018: Observed gust wind speeds in the coterminous United States, and their relationship to local and regional drivers. J. Wind Eng. Ind. Aerodyn., 173, 199–209, https://doi.org/10.1016/j.jweia.2017.12.008.
Letson, F. W., R. J. Barthelmie, K. I. Hodges, and S. C. Pryor, 2021: Intense windstorms in the northeastern United States. Nat. Hazards Earth Syst. Sci., 21, 2001–2020, https://doi.org/10.5194/nhess-21-2001-2021.
Loken, C., and Coauthors, 2010: SciNet: Lessons learned from building a power-efficient top-20 system and data centre. J. Phys. Conf. Ser., 256, 012026, https://doi.org/10.1088/1742-6596/256/1/012026.
Lowery, M. D., and J. E. Nash, 1970: A comparison of methods of fitting the double exponential distribution. J. Hydrol., 10, 259–275, https://doi.org/10.1016/0022-1694(70)90253-2.
Lukens, K. E., E. H. Berbery, and K. I. Hodges, 2018: The imprint of strong-storm tracks on winter weather in North America. J. Climate, 31, 2057–2074, https://doi.org/10.1175/JCLI-D-17-0420.1.
Mao, Y., and A. Monahan, 2017: Predictive anisotropy of surface winds by linear statistical prediction. J. Climate, 30, 6183–6201, https://doi.org/10.1175/JCLI-D-16-0507.1.
Moore, G. W. K., and J. L. Semple, 2006: Weather and death on Mount Everest: An analysis of the into thin air storm. Bull. Amer. Meteor. Soc., 87, 465–480, https://doi.org/10.1175/BAMS-87-4-465.
Neale, R. B., and Coauthors, 2012: Description of the NCAR Community Atmosphere Model (CAM 5.0). NCAR Tech. Note NCAR/TN-486+STR, 274 pp., https://www.cesm.ucar.edu/models/cesm1.0/cam/docs/description/cam5_desc.pdf.
NRC, 2015: National building code of Canada: 2015. National Research Council Canada, 714 pp., https://doi.org/10.4224/40002005.
Ponce, M., and Coauthors, 2019: Deploying a top-100 supercomputer for large parallel workloads: The Niagara supercomputer. Proc. Practice and Experience in Advanced Research Computing on Rise of the Machines (Learning) (PEARC ’19), Chicago, IL, Association for Computing Machinery, 1–8, https://doi.org/10.1145/3332186.3332195.
Pryor, S. C., and R. J. Barthelmie, 2014: Hybrid downscaling of wind climates over the eastern USA. Environ. Res. Lett., 9, 024013, https://doi.org/10.1088/1748-9326/9/2/024013.
Pryor, S. C., and A. N. Hahmann, 2019: Downscaling wind. Oxford Research Encyclopedia of Climate Science, Oxford University Press, https://doi.org/10.1093/acrefore/9780190228620.013.730.
Pryor, S. C., R. J. Barthelmie, and J. T. Schoof, 2012: Past and future wind climates over the contiguous USA based on the North American regional climate change assessment program model suite. J. Geophys. Res., 117, D19119, https://doi.org/10.1029/2012JD017449.
Pryor, S. C., R. J. Barthelmie, M. S. Bukovsky, L. R. Leung, and K. Sakaguchi, 2020: Climate change impacts on wind power generation. Nat. Rev. Earth Environ., 1, 627–643, https://doi.org/10.1038/s43017-020-0101-7.
Roy, P., P. Gachon, and R. Laprise, 2014: Sensitivity of seasonal precipitation extremes to model configuration of the Canadian regional climate model over eastern Canada using historical simulations. Climate Dyn., 43, 2431–2453, https://doi.org/10.1007/s00382-014-2066-z.
Sandink, D., G. Kopp, S. Stevenson, and N. Dale, 2019: Increasing high wind safety for Canadian homes: A foundational document for low-rise residential and small buildings. ICLR Research Paper Series 62, 128 pp., https://www.iclr.org/wp-content/uploads/2019/04/ICLR-Western-SCC-Increasing-High-Wind-Safety-2019_EN.pdf.
Schuldt, S. J., M. R. Nicholson, Y. A. Adams, and J. D. Delorit, 2021: Weather-related construction delays in a changing climate: A systematic state-of-the-art review. Sustainability, 13, 2861, https://doi.org/10.3390/su13052861.
Schwierz, C., P. Köllner-Heck, E. Zenklusen Mutter, D. N. Bresch, P.-L. Vidale, M. Wild, and C. Schär, 2010: Modelling European winter wind storm losses in current and future climate. Climatic Change, 101, 485–514, https://doi.org/10.1007/s10584-009-9712-1.
Sherry, M., and D. Rival, 2015: Meteorological phenomena associated with wind-power ramps downwind of mountainous terrain. J. Renewable Sustainable Energy, 7, 033101, https://doi.org/10.1063/1.4919021.
Sillmann, J., and Coauthors, 2017: Understanding, modeling and predicting weather and climate extremes: Challenges and opportunities. Wea. Climate Extremes, 18, 65–74, https://doi.org/10.1016/j.wace.2017.10.003.
Skamarock, W. C., 2004: Evaluating mesoscale NWP models using kinetic energy spectra. Mon. Wea. Rev., 132, 3019–3032, https://doi.org/10.1175/MWR2830.1.
Subin, Z. M., W. J. Riley, and D. Mironov, 2012: An improved lake model for climate simulations: Model structure, evaluation, and sensitivity analyses in CESM1. J. Adv. Model. Earth Syst., 4, M02001, https://doi.org/10.1029/2011MS000072.
Taylor, K. E., 2001: Summarizing multiple aspects of model performance in a single diagram. J. Geophys. Res., 106, 7183–7192, https://doi.org/10.1029/2000JD900719.
Teran, A. S., and Coauthors, 2022: Towards a computational workflow for studying the effects of climate change on wind loads on high-rise buildings in urban areas. Atmos.–Ocean, 60, 124–140, https://doi.org/10.1080/07055900.2022.2061412.
Uccellini, L. W., and P. J. Kocin, 1987: The interaction of jet streak circulations during heavy snow events along the East Coast of the United States. Wea. Forecasting, 2, 289–308, https://doi.org/10.1175/1520-0434(1987)002<0289:TIOJSC>2.0.CO;2.
Wang, M., P. Ullrich, and D. Millstein, 2018: The future of wind energy in California: Future projections with the variable-resolution CESM. Renewable Energy, 127, 242–257, https://doi.org/10.1016/j.renene.2018.04.031.
Wang, M., P. Ullrich, and D. Millstein, 2020: Future projections of wind patterns in California with the variable-resolution CESM: A clustering analysis approach. Climate Dyn., 54, 2511–2531, https://doi.org/10.1007/s00382-020-05125-5.
Wazneh, H., M. A. Arain, P. Coulibaly, and P. Gachon, 2020: Evaluating the dependence between temperature and precipitation to better estimate the risks of concurrent extreme weather events. Adv. Meteor., 2020, 8763631, https://doi.org/10.1155/2020/8763631.
Welker, C., and O. Martius, 2015: Large-scale atmospheric flow conditions and sea surface temperatures associated with hazardous winds in Switzerland. Climate Dyn., 44, 1857–1869, https://doi.org/10.1007/s00382-014-2404-1.
Wirth, V., and J. Eichhorn, 2014: Long-lived Rossby wave trains as precursors to strong winter cyclones over Europe. Quart. J. Roy. Meteor. Soc., 140, 729–737, https://doi.org/10.1002/qj.2191.
Wu, C., X. Liu, Z. Lin, A. M. Rhoades, P. A. Ullrich, C. M. Zarzycki, Z. Lu, and S. R. Rahimi-Esfarjani, 2017: Exploring a variable-resolution approach for simulating regional climate in the Rocky Mountain region using the VR-CESM. J. Geophys. Res. Atmos., 122, 10 939–10 965, https://doi.org/10.1002/2017JD027008.
Zarzycki, C. M., and C. Jablonowski, 2014: A multidecadal simulation of Atlantic tropical cyclones using a variable-resolution global atmospheric general circulation model. J. Adv. Model. Earth Syst., 6, 805–828, https://doi.org/10.1002/2014MS000352.
Zarzycki, C. M., C. Jablonowski, D. R. Thatcher, and M. A. Taylor, 2015: Effects of localized grid refinement on the general circulation and climatology in the community atmosphere model. J. Climate, 28, 2777–2803, https://doi.org/10.1175/JCLI-D-14-00599.1.
Zhang, X., P. J. Kushner, B. A. Saville, and I. D. Posen, 2022: Cold temperature limits to biodiesel use under present and future climates in North America. Environ. Sci. Technol., 56, 8640–8649, https://doi.org/10.1021/acs.est.2c01699.