1. Introduction
The Athabasca oil sands development is a large industrial oil mining and upgrading operation located in the Canadian boreal forest near Fort McMurray in the northeastern part of the province of Alberta. One of the oil extraction methods utilized involves excavating a bitumen–sand mixture from large open pit mines. The bitumen is separated from the sand and then upgraded into synthetic crude oil. The entire process causes a substantial land disturbance and uses an enormous amount of energy (Kelly et al. 2009). The geographic location and details of the surface disturbance of the Athabasca oil sands are depicted in Figure 1 (left and center), using the most recent available data (2007) when the surface disturbance was about 530 km2. The Government of Alberta (2016) reported that the extent of the surface disturbance in 2013 had increased to 895 km2, approximately 88 km2 of which consisted of tailings ponds used to store waste water effluents.
(left) The location of the oil sands and Fort McMurray in Alberta (modified from Brown et al. 2011). (middle) The oil sands land cover in 2007 (modified from Brown et al. 2011). (right) Our model modifications. We modeled the oil sands development as an approximately circular 650-km2 disturbance of barren ground (pink area). We added waste heat to the atmosphere in a smaller area within the center of the disturbance (blue area). ©Earth Interactions. Used with permission.
Citation: Earth Interactions 22, 3; 10.1175/EI-D-17-0013.1
Research investigating the meteorological effects of land surface disturbances has demonstrated that industrial operations, such as oil refineries, can influence thunderstorm development. For example, Guan and Reuter (1995) found that oil refinery cooling towers could enhance cumulus rain shower precipitation amounts. Further research by Guan and Reuter (1996) suggested sensible heat emissions from the cooling towers were the dominant factor in such enhancements but that the combined effect of the cloud condensation nuclei and sensible heat emissions was also significant. In contrast, Steiger and Orville (2003) suggested that enhanced lightning effects observed in proximity to oil refineries near Lake Charles, Louisiana, were mainly due to cloud condensation nuclei emissions, concluding that heat emissions were not a significant factor. Adding more complexity, Brown et al. (2011) found that climatological cloud-to-ground lightning was not affected by the Athabasca oil sands development. These contradictory results suggest that more research is needed to fully understand how industrial developments can affect convection and thunderstorms.
In this study, we investigated the connection between the Athabasca oil sands surface disturbance and thunderstorms. Alberta is prone to severe thunderstorms, with many regions frequently experiencing damage because of hail, wind, and tornadoes (Smith et al. 1998). Although the majority of thunderstorm research in Alberta is focused on the heavily populated Edmonton–Calgary urban corridor and the severe thunderstorm–prone lee side of the Rocky Mountains (Taylor et al. 2011), thunderstorms certainly occur in other areas of the province. For example, 7-cm-diameter hail from a severe thunderstorm caused $15 million CAD in damage to Fort McMurray in 2007 (Crewe 2008). Thunderstorms require warm, humid air at lower levels and cooler air aloft. Often a warmer layer of air (the capping inversion) forms near 800 mb and blocks the humid air at the surface from interacting with the cooler air aloft (Strong 1986). A capping inversion causes thunderstorm initiation to be very sensitive to disturbances in the boundary layer known as triggers (Strong 1986). Some possible triggers could include land-cover variations, frontal surfaces, or topographical features.
The extensive literature on inadvertent weather modification suggests that land surface disturbances associated with urban areas can enhance convection and increase rainfall, although there have been some contradictory findings. For example, Changnon et al. (1976) documented enhanced rainfall, thunderstorms, and hail downwind of St. Louis, Missouri. Westcott (1995) found enhanced cloud-to-ground lightning in the urban cores of cities, and Ashley et al. (2012) documented higher radar reflectivity in urban cores, to confirm increased precipitation as the cause. Niyogi et al. (2011) showed that thunderstorms exhibited modified radar signatures as they passed over the city of Indianapolis, Indiana. Using a numerical model, Niyogi et al. (2006) simulated the structure of a mesoscale convective system both in the presence and absence of surface modifications associated with Oklahoma City, Oklahoma, finding the city affected the structure of the convective system. However, not all research shows that urban areas cause precipitation or thunderstorm enhancement. For example, Nkemdirim (1981) could not find evidence that the city of Calgary, Alberta, modified precipitation. Numerical simulations by Schmid and Niyogi (2013) have even indicated that some urban surface disturbances may result in lower precipitation. In situations where artificial surface disturbances appear to result in higher precipitation or thunderstorm enhancement, researchers propose three possible causes: the urban heat island, increased surface roughness, and cloud condensation nuclei emissions (Changnon et al. 1976; Steiger et al. 2002).
Heat islands as a manifestation of inadvertent weather modification have been studied extensively for many years. Oke (1973) documented heat islands in many small and large midlatitude cities in North America. Along with waste heat emissions, heat islands are suspected to enhance thunderstorms. For instance, Bornstein and Lin (2000) found that an urban heat island was responsible for enhanced precipitation in Atlanta, Georgia. Similarly, Guan and Reuter (1996) found that waste heat emissions from an oil refinery in Illinois were the dominant cause of convective shower enhancement. Baik et al. (2007) suggested that urban heat islands affect thunderstorms through enhanced surface convergence and that the urban heat island circulation is strongest in the midafternoon, when the rural–urban surface temperature difference is the lowest. Brown et al. (2011) found that the Athabasca oil sands heat island is strengthening as the development increases in size. Additionally, lower vegetation coverage in urban areas has been demonstrated to produce an “urban dry island” (Schmid and Niyogi 2013). Rozoff et al. (2003) found that lower evapotranspiration causes lower humidity and lower instability in urban cores. Brown et al. (2011) found that the Athabasca oil sands development is creating an increasingly strong dry island with the growing extent of the development.
Much less research has specifically discussed the effect of surface roughness on inadvertent weather modification. Rozoff et al. (2003) found that a higher surface roughness in urban core areas slightly increased moisture convergence and marginally reduced the urban heat island effect, with a small overall contribution to thunderstorm enhancement. The surface roughness of the oil sands is not known. It might be less than the surrounding forests, which have been cleared to yield barren industrial terrain and smooth tailings ponds. Conversely, surface roughness might be higher in some areas because of the density of upgrading facilities and open pit mines. The exact effect of surface roughness modifications on thunderstorms needs to be researched more thoroughly.
Cloud condensation nuclei have been suggested to cause precipitation enhancement in inadvertent weather modification (Changnon et al. 1976; Steiger and Orville 2003). Many researchers have investigated the effect of cloud condensation nuclei; however, many results have been inconclusive or disagree with those of other studies. For example, Steiger and Orville (2003) suggested that cloud condensation nuclei could be the dominant factor in lightning enhancement, despite Reuter and Guan (1995) having concluded that its overall effect was small. More likely, the effect of cloud condensation nuclei is highly complex (Dixon and Mote 2003). Zhong et al. (2015) found that cloud condensation nuclei reduced precipitation at the urban core center and enhanced precipitation downwind. Hazewinkel et al. (2008) showed that the oil sands development emits large amounts of pollutants into the atmosphere, and Howell et al. (2014) discussed how these emissions can act as cloud condensation nuclei. In their estimation, the pollution emissions from the oil sands development do not function as cloud condensation nuclei at first, but a substantial amount of cloud condensation nuclei form as the plume ages (Howell et al. 2014).
In this study, we have investigated how the Athabasca oil sands development might affect the intensity, initiation time, and duration of thunderstorms, given the development has produced a massive land surface disturbance and emits large quantities of waste heat (Kelly et al. 2009). Since the complex process of cloud condensation nuclei is poorly understood, and modeling procedures for this process are not well established, we considered this factor outside the scope of our research. Given that Brown et al. (2011) did not find any climatological lightning enhancement of the oil sands development in previous research, we did not expect to find any drastic inadvertent weather modification of thunderstorms. However, it is certainly plausible that in certain cases the oil sands heat island could serve as a trigger for thunderstorm initiation if the atmosphere was unstable with a weak capping inversion.
2. Experimental design and case study days
2.1. WRF Model
We used the Weather Research and Forecasting (WRF) Model, a nonhydrostatic regional numerical weather prediction model (Skamarock et al. 2008), to carry out sensitivity experiments for areas in the vicinity of the oil sands development. Pennelly and Reuter (2017) found that the WRF Model is suitable for forecasting daily weather conditions for Alberta and could be used to simulate heavy precipitation events (Pennelly et al. 2014). In our simulations, three nested domains were centered on the Athabasca oil sands development (Figure 2), and the North American Regional Reanalysis (NARR; NCEP/NWS/NOAA 2005) data were used for our initial and boundary conditions. On each day, the simulation started at 0600 UTC (midnight local), permitting approximately 12 h of model spinup, and ran for 36 h to capture the evolution of each thunderstorm until it dissipated.
A map depicting our three nested model domains in relation to the oil sands development in northeastern Alberta.
Citation: Earth Interactions 22, 3; 10.1175/EI-D-17-0013.1
Various subgrid-scale physics processes were parameterized in the model simulations, which are summarized in Table 1. The intent of this study is not to evaluate the merits of different physics schemes. Thus, some of the schemes were chosen simply because they were the defaults or because they operated well with other chosen schemes. The most important scheme for these simulations was the convective scheme. We used the Grell–Freitas convection scheme because it has been designed to adjust its parameterization strategy for different scales, and it has demonstrated validity at some of the smaller scales employed in our simulations (Grell and Freitas 2014). We did not employ a convective scheme in our smallest domain because the grid spacing of 2 km should be fine enough to handle explicit convection. We selected the Noah land surface model because it is the most advanced land surface model available in WRF to date and includes sophisticated representations of soil–vegetation–atmosphere interactions as well as urban or barren ground physics (Chen and Dudhia 2001). At high resolutions, the convective simulations require a precipitation microphysics scheme that includes snow and graupel; thus, the Lin et al. (1983) microphysics scheme was used. Finally, we used the default rapid radiative transfer model to parameterize the shortwave and longwave radiation, as well as the default MM5 surface layer scheme (Jiménez et al. 2012), and the default Yonsei University planetary boundary layer scheme (Hong et al. 2006).
The WRF configuration for our model runs. The cumulus scheme was developed by Grell and Freitas (2014). The microphysics scheme was developed by Lin et al. (1983). The default Rapid Radiative Transfer Model (RRTM) was used for the longwave and shortwave schemes.
2.2. WRF land-cover and waste heat emissions
The WRF Model does not adequately classify the land cover of the oil sands development in its database; moreover, the database is outdated and does not include the rapid land-cover changes occurring in the oil sands development over the past 10 years. Thus, we simulated this surface disturbance in the WRF Model framework by creating a new “oil sands” land-cover category through modifying the existing properties of a similar type using a simple text file. The oil sands land-cover category was given the same surface parameters as the “barren” category included in the WRF database. To add waste heat to the model atmosphere, we wrote a new module for the WRF Model that allowed us to specify our waste heat amount as watts per square meter in a configuration file and then add it to the lowest eta level of the atmosphere (about 50 m thick). A constant of 100 W m−2 was added to 10% of the oil sands area (the blue area in Figure 1, right), equivalent to an average of 10 W m−2 over the entire area (as discussed further below). Thus, our simulations included a combination of the waste heat emissions and the land-cover modifications. Even though the oil sands development was rapidly increasing in size during our case study days, the waste heat value was kept constant because our objective was to compare different case study days influenced by the current extent of the oil sands development rather than temporal changes associated with changing extent.
Since data on waste heat emissions from the oil sands facilities were unavailable, we approximated this parameter with the following method: Brown et al. (2011) calculated the total waste heat flux from the oil sands development as approximately 3 W m−2, assuming the waste heat released into the atmosphere was 10% of the production energy spread over the entire area. Because new developments have caused the disturbed area and the total energy to increase significantly in recent years, we used an average of 10 W m−2 in our simulations. Notably, these estimates only approximate the total energy and could potentially represent a source of error.
It does not appear realistic to distribute the emitted energy over the entire oil sands development area. Satellite images and map data suggest that only a small percentage of the disturbed area is covered with upgrading and refining equipment, justifying our application of 100 W m−2 of waste heat to 10% of the area in these simulations. This sensible heating area is much larger and much less concentrated than assumed in the work of Guan and Reuter (1995), but we believe it more accurately represents the actual conditions. In reality, the upgrading facilities are clustered at multiple locations in the oil sands development. For the sensitivity experiments, we concluded that the clearest results could be obtained by idealizing this distribution as a single cluster. To simulate the total Athabasca oil sands in the WRF Model, we converted a total area of 650 km2 to the oil sands land-cover category using a large circular area over its geographic location rather than replicating the exact oil sands footprint. The additional waste heat was added into a smaller area of 65 km2 in the center of the large circular surface disturbance area.
2.3. Case study days and method of analysis
Given that Brown et al. (2011) found little impact of the Athabasca oil sands development on climatological lightning, we anticipated difficulties finding suitable case study days to demonstrate thunderstorm enhancement. Thus, we used a combination of weather radar (from Environment and Climate Change Canada 2016), weather stations, and upper-air data to identify days when thunderstorms occurred in the vicinity of the oil sands development. Radar and surface data were used to exclude days when a synoptic-scale trigger, like a cold front, triggered thunderstorms, a restriction that should help ensure that the oil sands development was the primary trigger. We restricted our days to when one of the two nearest soundings indicated moderate convective available potential energy (CAPE; greater than 500 J kg−1) and relatively lower convective inhibition (CIN; between −10 and −50 J kg−1); however, the nearest available upper-air soundings were still approximately 300 km from the site. Weather radar was an invaluable tool for confirming convection in these cases; thus, we further limited our analysis to days when historical radar data were available from the Environment and Climate Change Canada agency’s external website (2007–14). Furthermore, we sought out case study days where the storm radar reflectivities were at least 40 dBZ and referenced rain gauge measurements from nearby weather stations for evidence of precipitation. A lack of recorded precipitation at a weather station did not necessarily exclude a thunderstorm day because the storm may have passed in between the sparsely spaced weather stations. Using a combination of soundings, weather radar, and precipitation measurements, we were able to settle on 10 case study days suitable for thunderstorm enhancement: 14 July 2007, 29 July 2007, 8 August 2009, 29 July 2010, 30 July 2010, 14 August 2011, 30 June 2013, 23 July 2014, 29 July 2014, and 6 August 2014.
In this study, we used the Stein and Alpert (1993) method of factor separation to quantify differences in the pure and combined effects between our numerical simulations. Factor separation requires that all factors, and all factor combinations, be identified and considered. Thus, for each day we conducted four numerical experiments: one with no heat or land-cover changes (case 0), one with heat added to the atmosphere (case H), one with the land cover changed to barren (case B), and one with both (case HB). The land area that was modified is shown in Figure 1 (right). It was necessary to produce simulations with neither factors activated as well as with both factors activated because the combined effect of both factors would not necessarily equal the sum of the individual factors. Importantly, the simulation with both factors activated (the control) should accurately represent the actual geographical and meteorological conditions at the oil sands development during each case.
2.4. Difficulties modeling thunderstorms
Thunderstorms and convection are highly nonlinear processes, presenting some of the most difficult meteorological phenomena to simulate using even the most sophisticated numerical models. Typically, the nonlinear synergistic interactions of weather systems increase with a smaller grid spacing, so that small perturbations quickly amplify in moist convection (Hohenegger and Schär 2007). However, Elmore et al. (2002) found a number of cases where large perturbations seemed to have little effect on thunderstorm development. Numerical model simulations of convective initiation continue to improve, but even the most sophisticated nowcasting systems in data-rich areas continue to struggle with convective initiation (Sun et al. 2014). Although numerical simulations in the 1–4-km range can forecast convection explicitly (without parameterization), forecasting the convective precipitation amount and distribution remains a challenge (Bryan et al. 2003). Weisman et al. (2008) found both success and failure using WRF to explicitly simulate convection on a 4-km grid in the U.S. Midwest.
A great deal of the recent research on numerical simulations of convection incorporates dense observation, upper-air, and radar networks to create accurate initialization analyses (Sun et al. 2014). In fact, Xue et al. (2013) showed that ingesting radar data were required to simulate convective precipitation accurately. Even with a dense observation network, the storms investigated by Rozoff et al. (2003) still occurred 2 h earlier than the storms they simulated. In contrast, data are scarce near Fort McMurray and the Athabasca oil sands development. Because of the difficulties modeling thunderstorms, this study was not concerned with exact agreement between the modeled thunderstorms and reality. Instead, we considered the model adequate if the initiation time was close (within a couple of hours), the storm motion was close (within 45°), and the storm intensities were close (within about 10 dBZ). Our simulations produced thunderstorms with radar reflectivities greater than 40 dBZ, as observed in reality; however, the exact locations and timings were not necessarily the same. Before presenting our results in the next section, we will indicate how each of our “control” simulations somewhat matched the observed convection from the radar and precipitation measurements.
3. Numerical simulation results
3.1. Agreement with observations
To examine the simulated release of waste heat and convection, we analyzed vertical cross sections of temperature and water vapor mixing ratio for all case study days on a latitude approximately through the center of the oil sands development. Although the cross-section latitude was slightly varied for each case, depending on the wind direction, it was most frequently between 57.0° and 57.1°N. The cross section for 29 July 2010 is shown in Figure 3. The two panels on the far left show the temperature (top) and water vapor mixing ratio (bottom) for the boreal forest case (case 0). The next panels to the right of these show the difference between each of the three modified cases (cases H, B, and HB) and the boreal forest case (case 0) at 1815 UTC (temperature on top and mixing ratio on the bottom). Changing the land-cover categorization from boreal forest to barren (middle-left panels) causes additional heat to be inputted into the atmosphere, as barren ground has a much lower Bowen ratio. This process converts more solar radiation into sensible versus latent heat and lowers the amount of transpiration and the water vapor mixing ratio. When waste heat was directly added to the atmosphere (middle-right panels), a concentrated heat plume formed, but the mixing ratio was essentially the same. When both factors were activated (far right panels), the resulting temperature and mixing ratio fields were approximately the sum of the fields produced by each individual factor. All case study days produced similar results, with minor variations.
(top left) A cross section of the model simulated temperature (°C) near the oil sands development from the case with no barren modifications or waste heat. (next three top row) The difference between the three modified simulations and the first one. (bottom) As in (top), but presenting the water vapor mixing ratio (g kg−1) instead of the temperature. All images are at 1815 UTC 29 Jul 2010, with the cross section at latitude 57.1°N.
Citation: Earth Interactions 22, 3; 10.1175/EI-D-17-0013.1
In all simulations, the warming effect of the land-cover change was substantially larger than that of the added waste heat. While sensible heat emissions did not alter the water vapor mixing ratio, land-cover modification did reduce the mixing ratio values. The average heat island effect was about 1°C, while the average dry island effect was about 2 g kg −1. During maximum daytime heating in the afternoon, the plume of warm dry air mixed well above the 800-mb level in most cases. The top of the plume was consistently composed of relatively cooler and moister air caused by the higher-reaching updraft and possibly cloud formation. Thus, we are confident that enhanced lift was produced by the simulated oil sands development. The modeled heat and dry island strength appear to be consistent with the results and predictions of Brown et al. (2011). Adding about 5 times more waste heat resulted in the effect of the waste heat emissions being about equal to the effect of the land-cover modification.
Nevertheless, the mechanisms by which the Athabasca oil sands could influence convection are complex. The sum of land-cover modification and waste heat emissions creates a significant heat island with the potential to add energy or trigger thunderstorms. Additionally, the existence of a dry island effect with the removal of vegetation causes significant drying to potentially reduce energy for thunderstorms. Thus, the oil sands effect on thunderstorms critically depends on whether the thunderstorm enhancement mechanism can be characterized as increasing instability caused by warmer conditions, decreasing instability caused by drier conditions, or a trigger caused by the heat island. Brimelow et al. (2011) determined that the size of a drought-induced vegetative disturbance should be greater than 18 000 km2 in order to affect instability enough to modify thunderstorms. However, Knowles (1993) found that forest fire burn areas (which are significantly higher impact than drought) at 400 km2 sometimes created convective circulations to trigger thunderstorms. Pielke and Uliasz (1993) showed that larger disturbance sizes had a greater impact on vertical velocity. Thus, at just under 1000 km2, the oil sands development may be more suited to triggering thunderstorms as opposed to inducing widespread instability modifications.
We further examined the convection produced by the model by comparing model-simulated radar images with both actual radar images and actual precipitation measurements, noting that our analysis remained hampered by data sparsity. Few rainfall measurements were available in the Fort McMurray area, and the nearest weather radar was just over 200 km away. This distance causes the radar to detect only the top half of storms in the region, and it is not able to detect storms north of the oil sands development. The data sparsity can affect our model comparisons with reality as well as the accuracy of the model initialization.
The initiation time of the simulation with no oil sands development (case 0) is compared to that of the actual radar data in Table 2. We defined the initiation time from the Cold Lake radar (CWHK) as when convective cells in the general vicinity of the oil sands development had a reflectivity greater than about 40 dBZ, comparing the actual thunderstorm initiation time with the modeled initiation time. We also compared the storm motion between the Cold Lake radar and the model-simulated storms. Data from the innermost domain and the middle domain were used. We stress again that our focus was on general convective characteristics rather than the exact location of the storms, as it is very difficult, if not impossible, to accurately simulate exact observed storm locations and intensities using a numerical model many hours in advance.
A comparison of the storm motion and initiation time for the simulated thunderstorms in the case with no oil sands development (case 0), with the real thunderstorms as detected by radar. The cases where the simulated and real thunderstorms differed by more than 4 h are bolded. The simulated initiation times are from a combination of the innermost and middle domain, giving a larger areal sample of convective initiation.
The results of Table 2 show that on six case study days, the difference in time between model initiation and reality (from radar) was less than an hour. In most of these cases, the model forecasted the convection slightly earlier than it had occurred. However, on the other four case study days, the model initiated convection four or more hours later than observed in reality, necessitating further investigation. As indicated in the next section, the initiation time results from 3 of these 4 days were improved with the addition of fluxes from the oil sands development. On all case study days, the motion of the real storms was very similar to the modeled storms (Table 2); minor variations are likely due to the model’s misinterpretations of right-moving multicell or supercell storms. However, according to our previously discussed criteria, the model performed adequately on all of these case study days and should not be expected to reproduce the specific storm tracks. That the WRF Model initiated reasonable convection, and generally forecasted the track well, is nevertheless impressive, given the sparse initialization data.
3.2. Summary of all case study days
The results of the effect of the oil sands development on thunderstorm intensity are shown in Table 3. Three variables were examined: the maximum simulated radar reflectivity, the average total condensate in the atmosphere, and the average total surface rainfall. The maximum simulated radar reflectivity was used as a proxy for the maximum storm strength. With respect to the condensate and rainfall, however, average values provide a better summary of the total impact of the oil sands development. Thus, we used the average total condensate and average surface-accumulated rainfall. All of the columns in Table 3 provide the results of the factor separation analysis. For example, on 29 July 2010, the maximum reflectivity in case 0 was 49.5 dBZ. The value for case B was 2.3 dBZ less than the value for case 0, and the value for Case H was 0.5 dBZ less than the value for case 0. As previously discussed, the value for case HB is not necessarily the sum of cases B and H. Case HB was 0.5 dBZ greater than if cases B and H were simply summed together.
The results of the factor separation for each case study day. The factor separation method was applied for three variables: maximum reflectivity (dBZ), average total condensate (g kg−1), and the total average rainfall (mm). The value for column 0 is the actual value for case 0. The values for columns B and H are the difference between each case and case 0. The value for column HB is the values for columns 0, B, and H subtracted from the actual value for case HB.
The maximum simulated radar reflectivity was the maximum value at any time of the day at any grid cell in domain 3, the results for which were largely inconclusive. When the oil sands development was activated, some days had slightly higher values (29 July 2014), while other days had slightly lower values (29 July 2010). Most frequently, the HB simulation was less than the sum of both the H and B simulations. The largest reduction was 2.3 dBZ, and the largest increase was 1.8 dBZ. These changes were very small, almost inconsequential, without any trend in either direction. Thus, we cannot conclude that there was any discernable impact of the oil sands development on the maximum simulated radar reflectivity.
The average condensate was calculated as the average of the total condensate in domain 3 (rain, snow, hail, cloud water, and cloud ice) over the entire three-dimensional domain and full duration of the simulation. Again, the results were inconclusive. When the oil sands development was activated, some days had slightly higher values of total condensate (30 July 2014), and other days had lower values (14 July 2007). The HB simulation was sometimes positive and sometimes negative. Once again, all changes were small and the oil sands development did not appear to have any discernable impact on the total condensate.
The total average rainfall was calculated as the total accumulated rainfall for the day averaged over all grid cells in domain 3. Large variability existed because some days produced short-lived, fast-moving storms over a small area (resulting in very low average rainfall), whereas other days produced long-lived, slower-moving storms and multiple rounds of back-building storms (with a larger average rainfall). These results too were inconclusive. On some case study days slightly higher values (30 July 2007) occurred in the oil sands simulations, while on other days there were slightly lower values (14 July 2007); however, the differences were only a few millimeters. Overall, the oil sands development did not appear to cause any discernable impact on the average total rainfall measurements.
In summary, these results would appear to indicate that the oil sands development does not affect the intensity of thunderstorms. We could not find any impact on the maximum radar reflectivity, the average total condensate, or the average accumulated rainfall in the model simulations, leading to the conclusion that the oil sands development is likely too small to cause widespread changes to the convective instability.
In contrast to the thunderstorm intensity results, we noted in the simulations some large differences in the thunderstorm initiation time and duration, which are related phenomena. Thunderstorm initiation is mostly affected by triggering, while the duration indicates the total impact of the thunderstorm (e.g., a storm that lasts twice as long could produce double the precipitation). Table 4 shows the initiation time and storm duration for the four sensitivity simulations on each case study day. Here, we considered the initiation time of the storm to be when the maximum reflectivity in the innermost domain was greater than 20 dBZ. In eight of the case study days, thunderstorm initiation and duration were not affected by the oil sands development in the model simulations. However, two of the case study days experienced large shifts in the initiation time and duration. When the oil sands were activated in the simulations, thunderstorms on 29 July 2010 were initiated 2 h earlier than the simulation where the oil sands were not activated. On 29 July 2014 they were initiated 1 h earlier. These results demonstrate that the oil sands development would appear to have some influence on thunderstorm triggering. Further examination of these 2 days in greater detail will help to suggest why in these two cases the oil sands appear to have produced such a greater impact on thunderstorms.
The initiation time (UTC) and storm duration (h) for each model run for each case study day. The initiation time and duration difference columns are the difference between the 0 (zero factors activated; the boreal forest case) and the HB cases (all factors activated; the full oil sands case). The 2 days with the largest initiation time and duration differences are bolded. The initiation times are from the innermost domain only and thus might not equal the values in Table 2.
3.3. Further analysis of 29 July 2010 and 29 July 2014 case study days
The model simulations suggest that conditions on 29 July 2010 and 29 July 2014 seemed particularly sensitive to land-cover modifications and industrial waste heat emissions. Exploring these 2 days further, including a detailed analysis of the initiation time, duration, and storm motion will provide further insight into the reasons for these two cases of thunderstorm sensitivity to the oil sands development.
On 29 July 2010, the Cold Lake weather radar (Figure 4, left panels) detected a thunderstorm developing directly over the oil sands development at about 1830 UTC. This thunderstorm passed immediately over the automated weather station at the Fort McMurray International Airport about an hour later. The radar echoes were 50–55 dBZ, and the storm dropped about 20 mm of heavy rain, with a visibility as low as 400 m at the point when it passed over the airport weather station (between 1949 and 2024 UTC). As the storm continued to the southeast, it slowly weakened and eventually dissipated. Other isolated thunderstorms were observed on the radar images after the first storm, but none were triggered over the oil sands development.
Observed radar images of the thunderstorm that developed near the oil sands development on 29 Jul 2010 and 29 Jul 2014 (Environment and Climate Change Canada 2016). Arrows indicate the storm motion. Some of the radar data has been manually optimized for this visualization, without any changes affecting data interpretation.
Citation: Earth Interactions 22, 3; 10.1175/EI-D-17-0013.1
The development of the storm simulated by the WRF Model is shown in Figures 5 and 6. Figure 5 (left) shows the simulated column maximum radar reflectivity for the four simulations at 2145 UTC 29 July 2010, about the time when the simulated storms initiated. Figure 6 (top) shows a time series of the maximum simulated radar reflectivity from all grid cells within the innermost domain. We used Stein and Alpert’s (1993) method of factor separation to further quantify the results (Table 5), which we discuss next. When the land cover is the natural boreal forest and no waste heat is added, thunderstorms were initiated at 2315 UTC. However, when we added the waste heat, a small, short-lived storm was initiated 1 h and 30 min earlier than the boreal forest case. When we changed the land use to barren, the storm was also initiated 1 h and 30 min earlier than the boreal forest case. This storm was a little stronger and lasted longer, indicating that the land-cover modification had a greater effect than the addition of waste heat, in this case. The simulated thunderstorms were initiated earliest with the activation of both factors, a full 2 h earlier than the case with no factors activated. In this case, including both factors simultaneously produced a weaker effect than the sum of each factor alone (Table 5). These simulation results were an improvement on the no oil sands case, but the initiation time was still later than the actual radar observations. The earliest simulated storm developed at 2115 UTC, a full 2 h and 45 min later than the observed storm. In other experiments (not shown), the simulated storm developed much closer to the observed time if much more waste heat was added (beyond what we calculated as reasonable), which suggests that this particular case was very sensitive to the amount of waste heat emissions.
Simulated model radar images for (right) 2145 UTC 29 Jul 2010 and for (left) 2315 UTC 29 Jul 2014. Simulation 0 is the simulation with no factors activated. Simulation B is the simulation with the land cover changed to barren. Simulation H is the simulation with the waste heat emissions. Simulation HB is the simulation with both factors activated.
Citation: Earth Interactions 22, 3; 10.1175/EI-D-17-0013.1
The maximum simulated radar reflectivity on (top) 29 Jul 2010 and (bottom) 29 Jul 2014 for the four simulations. Simulation 0 is the simulation with no factors activated. Simulation B is the case with the land surface changed to barren. Simulation H is the case with the waste heat added. Simulation HB is the case where both factors were activated. Also shown is the observed maximum radar reflectivity from the Jimmy Lake Radar Station (WHN).
Citation: Earth Interactions 22, 3; 10.1175/EI-D-17-0013.1
The results of the factor separation analysis (bolded) on the storm initiation time and total storm duration for 29 Jul 2010 and 29 Jul 2014 for each of the model runs. The separated factors for case 0 are the actual value for that case. The separated factors for cases H and B are the difference between each case and case 0. The separated factors for case HB is the difference between case HB and a simple addition of factors H and B.
Results from the simulations on 29 July 2014 also suggested that the initiation time was very sensitive to the presence of the oil sands development. Selected data from the Cold Lake weather radar for this day are shown in Figure 4 (right side). Three separate thunderstorms developed in proximity to, but not immediately over, the oil sands development (although it is difficult to pinpoint the exact location of initiation, as the initial thunderstorm development occurred below the lowest-elevation angle of the radar). The first storm formed at 1910 UTC, about 40 km south of Fort McMurray. At about 2030 UTC, some weak thunderstorms formed about 90 km east of Fort McMurray. At 0030 UTC, a very weak shower formed about 10 km east of Fort McMurray, and at about 0100 UTC, some thunderstorms formed about 40 km to the east. All of the storms moved slowly to the southeast, and based on radar data, none appeared to form directly over the oil sands development.
Results from the simulations of 29 July 2014 are shown alongside the simulations for 29 July 2010 in Figures 5 and 6. Figure 5 (right) shows the simulated column maximum radar reflectivity for the four simulations at 2315 UTC 29 July 2014, and Figure 6 (bottom) shows a time series of the maximum simulated radar reflectivity from all grid cells within the innermost domain. Again, the factor separation method was used to interpret the results (Figure 5). The simulations with the barren land cover initiated storms about 1 h earlier than the boreal forest simulation. Interestingly, there was very little difference between waste heat simulation and the boreal forest simulation; adding waste heat had very little effect on the simulated thunderstorms on this day. The simulation with both factors activated was very similar to the simulation with the land-cover modifications. The factor separation of the initiation time and duration did not show any difference between the sum of the individual factor simulations and the simulation with both factors activated simultaneously (Figure 5). Notably, the effect of the oil sands development on this case study day was less than that on 29 July 2010. The earliest simulated storm was at 2145 UTC, which was still 2 h and 35 min later than the observed 1910 UTC initiation on the radar. Similar to 29 July 2010, other experiments (not shown) where we emitted additional waste heat to the simulation caused the initiation time to be much closer to the observed radar initiation time. However, adding that amount of heat could not be justified based on the information we were able to obtain about the actual oil sands development.
Another case for discussion is 6 August 2014. The analysis did not indicate anything particularly unusual about this day. Nevertheless, some of the simulations in which a much higher than realistic heat flux was added (not shown) caused a thunderstorm to initiate over the oil sands over an hour earlier. In using the realistic heat flux, we did see an area with a reflectivity of less than 20 dBZ form over an hour earlier only when we activated the oil sands; this appears to have been a small cumulus cloud. These results suggest that 6 August 2014 was partially susceptible to thunderstorm modification, as thunderstorms were unaffected by the unrealistic increased heat flux on the other seven case study days in question.
On four case study days, the initiation time from the numerical model did not agree with what was observed on radar, and the model initiated the thunderstorm more than 4 h later than observed. Three of those four days appear to have been sensitive to boundary layer triggering, such that the addition of heat fluxes associated with the oil sands development triggered convective clouds or thunderstorms earlier and much closer to the observed initiation time; 29 July 2007 was the only day that was simulated poorly. We do not have an immediate explanation for its divergence from the pattern, although notably this was the same thunderstorm that resulted in $15 million CAD in insured property damages in Fort McMurray (Crewe 2008).
3.4. Analysis of aircraft measurements
The natural question arising from the results of this study is why some case study days appear to be affected by the oil sands development, while the remainder does not. To address this issue, we computed the difference in thunderstorm duration between the simulation with zero factors activated and the simulation with both factors activated, which was used to represent the magnitude of the oil sands’ effect on thunderstorm duration. Thunderstorm duration was used instead of initiation time to better capture the total impact of the thunderstorm, although results would tend to be similar in either case. As a proxy for midlevel instability, we calculated the temperature difference between 850 and 500 mb (also known as the vertical totals index) from commercial aircraft measurements taken at the Fort McMurray Airport. Our investigation of the 850–500-mb temperature difference drew on partial insights from the research literature. For instance, Dixon and Mote (2003) investigated many days with urban-enhanced precipitation in Atlanta, Georgia, finding that days with more urban enhancement tended to have higher lapse rates below 600 mb and higher humidity between 900 and 600 mb.
Aircraft meteorological data relay (AMDAR) soundings (ESRL/GSD 2016) were obtained from the National Oceanic and Atmospheric Administration website (at http://amdar.noaa.gov/). Benjamin et al. (1999) found that AMDAR data can be as accurate as radiosonde sounding data, and Schwartz et al. (2000) used AMDAR data to evaluate forecast model accuracy. Although radiosonde balloons are not regularly launched near Fort McMurray, between two and eight aircraft submitted AMDAR soundings from the Fort McMurray airport on each of our case study days. AMDAR data do not necessarily contain measurements at exactly 850 and 500 mb, so these levels required interpolation. We computed the difference between the 850 and 500 mb temperatures for all AMDAR soundings after 1800 UTC on each day, using the median for the model comparison such that outlier values would not skew our results.
The relationship between the 850–500-mb temperature difference and the extent that the oil sands development affected thunderstorm duration for each case study day is shown in Figure 7. The highest temperature difference (largest midlevel instability) was on 29 July 2010 (30.6°C), with the second highest temperature difference on 29 July 2014 (30.5°C), and on both of these days the oil sands development simulations caused the greatest increase in thunderstorm duration. Thus, we can confidently state that for days with an 850–500-mb temperature difference greater than 30°C, the oil sands development affected thunderstorm initiation time and duration to a certain degree. In fact, the only 2 days that appear to have been strongly affected by the oil sands development were days with an 850–500-mb temperature difference greater than 30°C. However, there were two other days with a temperature difference slightly less than 30°C for which the oil sands development did not appear to affect thunderstorms (30 July 2010 and 14 July 2007). Additionally, the day on which the oil sands affected thunderstorms with the addition of a great deal more waste heat (6 August 2014) had an 850–500-mb temperature difference of less than 30°C. Thus, there seemed to be some association between days with very high lapse rates and strong modification of thunderstorm initiation time by the oil sands; however, it is difficult to make strong conclusions within the scope of this study based on only 10 case study days.
The storm duration difference between the case with zero factors activated (case 0) and the case with both factors activated (case HB) vs the 850–500-mb temperature difference.
Citation: Earth Interactions 22, 3; 10.1175/EI-D-17-0013.1
4. Discussion and conclusions
Our research presents the following findings:
Waste heat created by the oil sands development does not appear to cause significant changes to thunderstorm intensity by directly adding sensible heat to the atmosphere and increasing buoyancy.
Land-cover modification by the oil sands development does not appear to cause significant changes to thunderstorm intensity by adding sensible heat to the atmosphere and increasing the Bowen ratio.
In rare cases, both waste heat and land-cover modifications resulting from the oil sands development appears to cause thunderstorms to initiate substantially earlier and to last longer.
Stein and Alpert’s (1993) method of factor separation helped determine the relative contributions of the waste heat and land-cover modification to thunderstorm initiation time and intensity.
Days on which the oil sands development appears to have significantly modified thunderstorms had an 850–500-mb temperature difference greater than 30°C.
The WRF Model was able to adequately simulate the heat and dry islands produced by the oil sands development, even with sparse initialization data. The simulated heat and dry islands were about the expected magnitude and extended fully throughout the boundary layer by the afternoon. The WRF Model also simulated convection and convection initiation reasonably well for most of the case study days. On 6 of the 10 case study days, convection was initiated at nearly the same time as observed in reality. On two of the four case study days when convection was not initiated at the expected time, the time was then significantly closer to that observed in reality when the oil sands development was added into the model. On a third day, if much more heat was added to the model, the timing of convective initiation was improved as well. However, one case study day with a massive hailstorm was not simulated well with the model regardless of the amount of heat added in the simulations.
In all cases, regardless of which factors were activated in the simulations, there was minimal modification to convection intensity. Although a factor may have slightly increased or decreased the intensity of the convection, there were no discernable patterns across the cases. Thus, the oil sands do not seem to be causing more thunderstorms or stronger thunderstorms.
On two case study days, adding the oil sands development into the model caused the convective initiation time to be earlier than without. These results seem to have been produced mainly by the change in land cover and only partially by the waste heat emissions; however, these cases were rare. The timing of thunderstorm modification by the oil sands seems most likely when large midlevel lapse rates exist. We found that days with significant simulated thunderstorm modification had an 850–500-mb temperature difference greater than 30°C, and we are confident that the oil sands land-cover and waste heat modifications to the simulated atmosphere are what caused this inadvertent weather modification to occur.
In this study, we did not account for cloud condensation nuclei because research has given conflicting results, and Dixon and Mote (2003) have suggested that this is because the effect of cloud condensation nuclei might be more complex than previously thought. Zhong et al. (2015) argued that cloud condensation nuclei might increase precipitation downwind of an emission site, and Howell et al. (2014) found that pollution from the oil sands development could cause cloud condensation nuclei to form downwind. Thus, it is possible that an increased amount of cloud condensation nuclei emitted from the oil sands could help to initiate convection even earlier. Moreover, this mechanism could partly account for discrepancies between the observed and modeled convection on the days when convection was initiated too late in the simulation.
We urge some caution regarding our estimates of waste heat emissions from the oil sands development. We found that three case study days were potentially sensitive to additional waste heat emissions over and above the 10% waste that we calculated as input. Given the large uncertainty of the actual amount of waste heat emitted, these results may suggest that the actual amount of waste heat emitted from the oil sands could be significantly higher than our estimates. We added what we thought was reasonable, but a combination of more waste heat and the inclusion of cloud condensation nuclei might help explain why the model results differed from reality for some of the case study days.
In our study, the Athabasca oil sands development appears to act as a thunderstorm trigger and appears to cause convective initiation by eroding the capping inversion in very unstable environments. The oil sands do not appear to be large enough to modify the amount of instability available to thunderstorms. Few studies have specifically implicated triggering as a means of thunderstorm modification. We also note that while most studies investigated large cities, our study investigated an industrial surface mine site. Industrial sites differ from cities in that they have few people, few buildings, and usually much less vegetation. We specifically did not use an urban model to represent the conditions at the oil sands development owing to the lack of tall buildings retaining heat by reemission. Additionally, there is a lack of studies that have explored inadvertent weather modification in the boreal forest. Most studies that we were able to find investigated agricultural areas with a much warmer, more humid climate, such as St. Louis, Missouri (Changnon et al. 1976; Rozoff et al. 2003); Atlanta, Georgia (Bornstein and Lin 2000); and Oklahoma City, Oklahoma (Niyogi et al. 2006). Often cities like this have been chosen because they are in areas with minimal synoptic forcing (Ashley et al. 2012).
Further research might investigate how often the conditions for thunderstorm duration enhancement exist for industrial operations in the Athabasca oil sands versus in Houston or Louisiana. Such research could potentially aid our understanding of why the oil refineries cause such strong climatological lightning enhancement in Houston and Louisiana, but not in northeastern Alberta. It is conceivable that there is little climatological lightning enhancement at the oil sands development because there appears to be only 1 or 2 days yr−1 that are conducive to modifying thunderstorms. It is possible that the conditions for thunderstorm enhancement noted in this research (an 850–500-mb temperature difference greater than 30°C) are more common near the Gulf of Mexico, resulting in the sum of the thunderstorm durations to be higher and thus impacting climatological lightning.
Acknowledgments
We acknowledge that some paid time was provided by the Meteorological Service of Environment and Climate Change Canada to complete this project. We also acknowledge the Department of Atmospheric Sciences at the University of Utah for providing insight into modifying the WRF Model code. We acknowledge the anonymous reviewers, whose comments substantially improved the manuscript.
References
Ashley, W. S., M. L. Bentley, and T. Stallins, 2012: Urban induced thunderstorm modification in the southeast United States. Climatic Change, 113, 481–498, https://doi.org/10.1007/s10584-011-0324-1.
Baik, J. J., Y. H. Kim, J. J. Kim, and J. Y. Han, 2007: Effects of boundary-layer stability on urban heat island-induced circulation. Theor. Appl. Climatol., 89, 73–81, https://doi.org/10.1007/s00704-006-0254-4.
Benjamin, S. B., B. E. Schwartz, and R. E. Cole, 1999: Accuracy of ACARS wind and temperature observations determined by collocation. Wea. Forecasting, 14, 1032–1038, https://doi.org/10.1175/1520-0434(1999)014<1032:AOAWAT>2.0.CO;2.
Bornstein, R., and Q. Lin, 2000: Urban heat islands and summertime convective thunderstorms in Atlanta: Three case studies. Atmos. Environ., 34, 507–516, https://doi.org/10.1016/S1352-2310(99)00374-X.
Brimelow, J. C., J. M. Hanesiak, and W. R. Burrows, 2011: On the surface convection-feedback during drought periods on the Canadian Prairies. Earth Interact., 15, https://doi.org/10.1175/2010EI381.1.
Brown, D. M., G. W. Reuter, and T. K. Flesch, 2011: Temperature, precipitation, and lightning modification in the vicinity of the Athabasca oil sands. Earth Interact., 15, https://doi.org/10.1175/2011EI412.1.
Bryan, G. H., J. C. Wyngaard, and J. M. Fritsch, 2003: Resolution requirements for deep moist convection. Mon. Wea. Rev., 131, 2394–2416, https://doi.org/10.1175/1520-0493(2003)131<2394:RRFTSO>2.0.CO;2.
Changnon, S. A., Jr., R. G. Semonin, and F. A. Huff, 1976: A hypothesis for urban rainfall anomalies. J. Appl. Meteor., 15, 544–560, https://doi.org/10.1175/1520-0450(1976)015<0544:AHFURA>2.0.CO;2.
Chen, F., and J. Dudhia, 2001: Coupling an advanced land surface–hydrology model with the Penn State–NCAR MM5 modelling system. Part I: Model implementation and sensitivity. Mon. Wea. Rev., 129, 569–604, https://doi.org/10.1175/1520-0493(2001)129<0569:CAALSH>2.0.CO;2.
Crewe, A., 2008: Hailstorm a year later: It can happen again. Fort McMurray Today, 29 July 2008, http://www.fortmcmurraytoday.com/2008/07/29/hailstorm-a-year-later-it-can-happen-again-2.
Dixon, P. G., and T. L. Mote, 2003: Patterns and causes of Atlanta’s urban heat island–initiated precipitation. J. Appl. Meteor., 42, 1273–1284, https://doi.org/10.1175/1520-0450(2003)042<1273:PACOAU>2.0.CO;2.
Elmore, K. L., D. J. Stensrud, and K. C. Crawford, 2002: Explicit cloud-scale models for operational forecasts: A note of caution. Wea. Forecasting, 17, 873–884, https://doi.org/10.1175/1520-0434(2002)017<0873:ECSMFO>2.0.CO;2.
Environment and Climate Change Canada, 2016: Canadian historical weather radar. Accessed 15 March 2016, http://climate.weather.gc.ca/radar/index_e.html.
ESRL/GSD, 2016: Aircraft data web. NOAA, Earth System Research Laboratory, Global Systems Division, accessed 25 March 2016, http://amdar.noaa.gov/.
Government of Alberta, 2016: Alberta energy: Facts and statistics. Accessed 31 January 2016, http://www.energy.alberta.ca/OilSands/791.asp.
Grell, G. A., and S. R. Freitas, 2014: A scale and aerosol aware stochastic convective parameterization for weather and air quality modeling. Atmos. Chem. Phys., 14, 5233–5250, https://doi.org/10.5194/acp-14-5233-2014.
Guan, S., and G. W. Reuter, 1995: Numerical simulation of a rain shower affected by waste energy released from a cooling tower complex in a calm environment. J. Appl. Meteor., 34, 131–142, https://doi.org/10.1175/1520-0450-34.1.131.
Guan, S., and G. W. Reuter, 1996: Numerical simulation of an industrial cumulus affected by heat, moisture, and CCN released from an oil refinery. J. Appl. Meteor., 35, 1257–1264, https://doi.org/10.1175/1520-0450(1996)035<1257:NSOAIC>2.0.CO;2.
Hazewinkel, R. R. O., A. P. Wolfe, S. Pla, C. Curtis, and K. Hadley, 2008: Have atmospheric emissions from the Athabasca oil sands impacted lakes in northeastern Alberta, Canada? Can. J. Fish. Aquat. Sci., 65, 1554–1567, https://doi.org/10.1139/F08-074.
Hohenegger, C., and C. Schär, 2007: Predictability and error growth dynamics in cloud-resolving models. J. Atmos. Sci., 64, 4467–4478, https://doi.org/10.1175/2007JAS2143.1.
Hong, S.-Y., Y. Noh, and J. Dudhia, 2006: A new vertical diffusion package with an explicit treatment of entrainment processes. Mon. Wea. Rev., 134, 2318–2341, https://doi.org/10.1175/MWR3199.1.
Howell, S. G., A. D. Clarke, S. Freitag, C. S. McNaughton, C. Kapustin, V. Brekovskikh, J.-L. Jimenez, and M. J. Cubison, 2014: An airborne assessment of atmospheric particulate emissions from the processing of the Athabasca oil sands. Atmos. Chem. Phys., 14, 5073–5087, https://doi.org/10.5194/acp-14-5073-2014.
Jiménez, P. A., J. Dudhia, J. F. González-Rouco, J. Navarro, J. P. Montávez, and E. García-Bustamante, 2012: A revised scheme for the WRF surface layer formulation. Mon. Wea. Rev., 140, 898–918, https://doi.org/10.1175/MWR-D-11-00056.1.
Kelly, E. N., J. W. Short, D. W. Schindler, P. V. Hodson, M. Ma, A. K. Kwan, and B. L. Fortin, 2009: Oil sands development contributes polycyclic aromatic compounds to the Athabasca River and its tributaries. Proc. Natl. Acad. Sci. USA, 106, 22 346–22 351, https://doi.org/10.1073/pnas.0912050106.
Knowles, J. B., 1993: The influence of forest fire induced albedo differences on the generation of mesoscale circulations. M.S. thesis, Department of Atmospheric Sciences, Colorado State University, 94 pp.
Lin, Y., R. D. Farley, and H. D. Orville, 1983: Bulk parameterization of the snow field in a cloud model. J. Climate Appl. Meteor., 22, 1065–1092, https://doi.org/10.1175/1520-0450(1983)022<1065:BPOTSF>2.0.CO;2.
NCEP/NWS/NOAA, 2005: NCEP North American Regional Reanalysis (NARR). NCAR/UCAR, Research Data Archive, Computational and Information Systems Lab, accessed 15 December 2016, http://rda.ucar.edu/datasets/ds608.0/.
Niyogi, D., T. Holt, S. Zhong, P. C. Pyle, and J. Basara, 2006: Urban and land surface effects on the 30 July 2003 mesoscale convective system event observed in the Southern Great Plains. J. Geophys. Res., 111, D19107, https://doi.org/10.1029/2005JD006746.
Niyogi, D., P. Pyle, M. Lei, S. P. Arya, C. M. Kishtawal, M. Shepherd, F. Chen, and B. Wolfe, 2011: Urban modification of thunderstorms: An observational storm climatology and model case study for the Indianapolis urban region. J. Appl. Meteor. Climatol., 50, 1129–1144, https://doi.org/10.1175/2010JAMC1836.1.
Nkemdirim, L. C., 1981: Extra urban and intra urban rainfall enhancement by a medium sized city. J. Amer. Water Resour. Assoc., 17, 753–759, https://doi.org/10.1111/j.1752-1688.1981.tb01294.x.
Oke, T. R., 1973: City size and the urban heat island. Atmos. Environ., 7, 769–779, https://doi.org/10.1016/0004-6981(73)90140-6.
Pennelly, C., and G. W. Reuter, 2017: Verification of the Weather Research and Forecasting Model when forecasting daily surface conditions in southern Alberta. Atmos.–Ocean, 55, 31–41, https://doi.org/10.1080/07055900.2017.1282345.
Pennelly, C., G. W. Reuter, and T. Flesch, 2014: Verification of the WRF Model for simulating heavy precipitation in Alberta. Atmos. Res., 135–136, 172–192, https://doi.org/10.1016/j.atmosres.2013.09.004.
Pielke, R. A., and M. Uliasz, 1993: Influence of landscape variability on atmospheric dispersion. J. Air Waste Manage. Assoc., 43, 989–994, https://doi.org/10.1080/1073161X.1993.10467181.
Reuter, G. W., and S. Guan, 1995: Effects of industrial pollution on cumulus convection and rain showers: A numerical study. Atmos. Environ., 29, 2467–2474, https://doi.org/10.1016/1352-2310(95)00169-Y.
Rozoff, C. M., W. R. Cotton, and J. O. Adegoke, 2003: Simulation of St. Louis, Missouri, land use impacts on thunderstorms. J. Appl. Meteor., 42, 716–738, https://doi.org/10.1175/1520-0450(2003)042<0716:SOSLML>2.0.CO;2.
Schmid, P. E., and D. Niyogi, 2013: Impact of city size on precipitation-modifying potential. Geophys. Res. Lett., 40, 5263–5267, https://doi.org/10.1002/grl.50656.
Schwartz, B. E., S. G. Benjamin, S. M. Green, and M. R. Jardin, 2000: Accuracy of RUC-1 and RUC-2 wind and aircraft trajectory forecasts by comparison with ACARS observations. Wea. Forecasting, 15, 313–326, https://doi.org/10.1175/1520-0434(2000)015<0313:AORARW>2.0.CO;2.
Skamarock, W. C., and Coauthors, 2008: A description of the Advanced Research WRF version 3. NCAR Tech. Note NCAR/TN-475+STR, 113 pp., https://doi.org/10.5065/D68S4MVH.
Smith, S. B., G. W. Reuter, and M. K. Yau, 1998: The episodic occurrence of hail in central Alberta and the Highveld of South Africa: Research note. Atmos.–Ocean, 36, 169–178, https://doi.org/10.1080/07055900.1998.9649610.
Steiger, S. M., and R. E. Orville, 2003: Cloud-to-ground lightning enhancement over southern Louisiana. Geophys. Res. Lett., 30, 1975, https://doi.org/10.1029/2003GL017923.
Steiger, S. M., R. E. Orville, and G. Huffines, 2002: Cloud-to-ground lightning characteristics over Houston, Texas: 1989–2000. J. Geophys. Res., 107, https://doi.org/10.1029/2001JD001142.
Stein, U., and P. Alpert, 1993: Factor separation in numerical simulations. J. Atmos. Sci., 50, 2107–2115, https://doi.org/10.1175/1520-0469(1993)050<2107:FSINS>2.0.CO;2.
Strong, G. S., 1986: Synoptic to mesoscale dynamics of severe thunderstorm environments: A diagnostic study with forecasting applications. Ph.D. thesis, University of Alberta, 345 pp.
Sun, J., and Coauthors, 2014: Use of NWP for nowcasting convective precipitation: Recent progress and challenges. Bull. Amer. Meteor. Soc., 95, 409–426, https://doi.org/10.1175/BAMS-D-11-00263.1.
Taylor, N. M., and Coauthors, 2011: The Understanding Severe Thunderstorms and Alberta Boundary Layers Experiment (UNSTABLE) 2008. Bull. Amer. Meteor. Soc., 92, 739–763, https://doi.org/10.1175/2011BAMS2994.1.
Weisman, M. L., C. Davis, W. Wang, K. W. Manning, and J. B. Klemp, 2008: Experiences with 0–36-h explicit convective forecasts with the WRF-ARW model. Wea. Forecasting, 23, 407–437, https://doi.org/10.1175/2007WAF2007005.1.
Westcott, N. E., 1995: Summertime cloud-to-ground lightning activity around major midwestern urban areas. J. Appl. Meteor., 34, 1633–1642, https://doi.org/10.1175/1520-0450-34.7.1633.
Xue, M., F. Kong, K. W. Thomas, J. Gao, Y. Wang, K. Brewster, and K. K. Droegemeier, 2013: Prediction of convective storms over continental United States with radar data assimilation: An example case of 26 May 2008 and precipitation forecasts from spring 2009. Adv. Meteor., 2013, 259052, https://doi.org/10.1155/2013/259052.
Zhong, S., Y. Qian, C. Zhao, R. Leung, and X.-Q. Yang, 2015: A case study of urbanization impact on summer precipitation in the greater Beijing metropolitan area: Urban heat island versus aerosol effects. J. Geophys. Res. Atmos., 120, 10 903–10 914, https://doi.org/10.1002/2015JD023753.
