GEMCLIM is a modelling platform developed for operational numerical weather prediction at Environment and Climate Change Canada (Côté et al., 1998; Yeh et al., 2002). The dynamical core of the model uses a semi-Lagrangian and semi-implicit two-time-level scheme (Côté et al., 1998). The LBCs are provided every 300 s. The prognostic variables are gradually relaxed to the LBCs through a sponge zone with a width of ten grid points. An additional lateral boundary zone with a width of ten grid points is used for semi-Lagrangian interpolation. The physical parametrizations used in the model include the boundary layer (Benoit et al., 1989; Delage, 1997), radiation (Li and Barker, 2005), sub-grid orographic gravity wave drag (McFarlane, 1987), deep convection (Kain and Fritsch, 1990), shallow convection (Bélair et al., 2005), cloud and large-scale condensation (Sundqvist et al., 1989). The land surface component applies the Canadian Land Surface Scheme, version 3.5 (Verseghy, 2000; Langlois et al., 2014). The model also uses a one-dimensional column lake model (Martynov et al., 2013) to simulate the interaction between the atmosphere and lakes. GEMCLIM also applies the Town Energy Balance model (Masson et al., 2002; Roberge and Sushama, 2018) to simulate the energy and water balance between urban areas and the atmosphere. The GEMCLIM simulations for the Metro Vancouver region are performed in two nested domains. As shown in Figure 1, the external domain is at a resolution of 0.11° × 0.11° (∼10 km for simulating the large-scale urban ambient environment) and the internal domain uses a typical convection-permitting horizontal grid spacing of ∼3 km (0.03° × 0.03°, at urban-scale). Both the external and internal domains excluding the sponge zones consist of 101 × 101 grid points.

The geophysical fields required for the GEMCLIM simulations are provided by the Climate Change Initiative Land Cover (Poulter et al., 2015) database for land-mask and vegetation, NASA Shuttle Radar Topography Mission (SRTM) v3.0 datasets for topography (Farr et al., 2007), Global Soil Dataset for Earth System Modeling for soil characteristics (Shangguan et al., 2014) and Canadian Vector Data v.9.0 dataset for urban properties.

To physically estimate PMP in Vancouver, the GEMCLIM experiments driven by spatially adjusted LBCs from the ERA-Interim reanalysis (Dee et al., 2011) are used to simulate selected extreme precipitation events in Metro Vancouver. PMP values (defined by the greatest regional-average 72-h total precipitation) are analyzed among the GEMCLIM simulations for each extreme precipitation event. The LBCs from ERA-Interim are spatially shifting in both longitude and latitude to drive GEMCLIM in order to capture the PMP with maximized AR intensities (in terms of IVT) hitting Vancouver with the optimal direction. The LBCs are meridionally shifted from 4.5° southward to 7.5° northward with a shifting interval of 0.75°. In the zonal direction, the LBCs are shifted from 6° westward to 10° eastward with a shifting interval of 1.0°. Thus, each event is simulated by 17 × 17 model runs corresponding with different spatial shifting of LBCs. We limit the range of LBC shifting because a large spatial shifting can mistakenly introduce non-AR systems outside the storm track regions to the simulation domain (Ishida et al., 2015b). The simulation for each event is initialized by a spin-up simulation of one year prior to the event. The simulations of the events driven by the spatially adjusted LBCs start seven days prior to the peak of regionally averaged 72-h total precipitation of the event and last for fourteen days.

Using the Daily Surface Weather and Climatological Summaries (Daymet, Thornton et al., 2017) at 1-km resolution, we select 30 most intense (in terms of 72-h total precipitation) extreme precipitation events associated with ARs for the period 1980∼2017 in Metro Vancouver to be simulated. Daymet is based on a local regression interpolation method to produce gridded estimates of daily meteorological variables using weather station data (Thornton et al., 1997). This dataset has been widely used for analyzing regional precipitation (Newman et al., 2020) and validating precipitation simulation (Diaconescu et al., 2016; Krebs et al., 2018; Huang et al., 2020) over Canada. Information of the selected events using Daymet, including the date of occurrence and intensity in terms of the regionally averaged 72-h total precipitation, is shown in Table 1. Daymet is also used to validate the GEMCLIM simulation of precipitation for the selected historical extreme precipitation events. To show how the physical PMP estimation performs differently from the statistical estimation, the GEV analysis based on the Fisher-Tippett distribution (Hershfield and Kohler, 1960; Nobilis et al., 1991; Kotowski and Kaźmierczak, 2013) fitted to the precipitation data by the Gumbel method is used to statistically estimate the PMP using Daymet and compared with the result of the physical method.

To investigate the mechanisms involved in the occurrence of PMP, the environmental variables associated with precipitation are analyzed. As suggested by Guan et al. (2010), Ralph and Dettinger (2012) and Liang and Sushama (2019), the large-scale environments, featured by vertically integrated water vapor transport (IVT), vertically integrated water vapor (IWV) and low-level air temperature are important characteristics of ARs responsible for different types of precipitation, hence these features are displayed for the simulated historical events and PMP estimations using the GEMCLIM simulations at 10-km resolution. To locally examine the dynamical precipitation processes in Vancouver, the changes in the dynamical factors, including the low-level vertical velocity and moisture transport convergence are analyzed using the 3-km simulations. We also calculate the large-scale precipitation efficiency, defined by the ratio of total precipitation to IWV during a certain period of time, to indicate how effective are the atmospheric dynamical processes in converting the available total column moisture to precipitation (Tuller, 1973; Ye et al., 2014).

To understand how the LBC shifting generates PMP in Metro Vancouver, we first analyze the GEMCLIM simulations for the most intense extreme precipitation event (occurring in October 2003) over Metro Vancouver for the period 1980∼2017 (event ID 1 in Table 1). Figure 2A shows that the PMP value (corresponding with the greatest precipitation in the matrix) is identified with LBCs shifted northward by 3.75° and eastward by 4°. During the historical event, Vancouver was affected by a landfalling AR with northeastward IVT of up to above 1100 kg m^–1s^–1 offshore Washington (state) of the United States (Figure 2B). The GEMCLIM simulation shows that the maximum 72-h precipitation reached up to 600 mm across the north of Metro Vancouver (Figure 2C). For the PMP simulation of the event, the maximum IVT within the AR plume increases by about 100 kg m^–1s^–1 and becomes more concentrated as affected by the spatially adjusted LBCs (Figure 2D). The PMP simulation shows that the maximum 72-h precipitation reached above 1000 mm in North Vancouver, implying an increase by about 67% relative to the historical simulation (Figure 2E).

In view of all the GEMCLIM simulations for the 30 selected historical extreme precipitation events, the PMP values (corresponding with the greatest precipitation increase relative to the historical precipitation values) are identified with LBCs shifted northward by 3.75° and eastward by 7° on average (Figure 3). In general, the regional average of 72-h precipitation over Vancouver is increased by northeastward shiftings of LBCs. Decreased precipitation is found when the LBCs are shifted southward and westward. If the LBCs are shifted only in the meridional (zonal) direction, the greatest precipitation increase is found when LBCs are shifted 0.75° northward (1° eastward), while the magnitudes of increase are much smaller than those generated by LBCs shifted in both directions. It is noticed that the direction of spatial shift corresponding with the PMP values is different from those presented by Ohara et al. (2011) and Ishida et al. (2015b), which suggested a southward shift (approximately 5∼6° southward) of the LBCs for capturing PMP. This is mainly due to the difference in the targeted area as these studies focused on the south of the main active region of ARs over the western North America.

Figure 4 shows the PMP values, in terms of the maximum regional average of 72-h total precipitation, over Metro Vancouver for the 30 selected extreme precipitation events with respect to the observed and simulated historical precipitation values. Compared to the observations, GEMCLIM successfully represents the most intense precipitation event that occurred on October 16th, 2003 (Event ID 1) and the second most intense event on January 18th, 2005 (Event ID 2). Although GEMCLIM shows overestimation (by 37%) of precipitation in general and the model overestimates the precipitation of the two most intense events by 34% and 45%, respectively, compared to the observations, the historical simulations are well correlated with the observation with a correlation coefficient of 0.6, which is statistically significant (p-value < 0.001). The greatest PMP value (783.35 mm) is found in the simulation of the third most intense precipitation event in history (November 9th, 1999, Event ID 3), which indicates precipitation increase by a factor of 2.7 relative to the corresponding historical precipitation. Also, the greatest PMP value is 1.3 times larger than the precipitation maximum during the most intense historical event (October 16th, 2003). On average, PMP values for all the events show a precipitation increase by 81% relative to the mean precipitation of historical simulations.

The spatial distributions of different types of precipitation for the historical extreme precipitation events and PMP estimates are presented in Figures 5, 6. The PMP estimated by the statistical method [i.e., 50-year return period of precipitation, following Clavet-Gaumont et al. (2017) and Afzali-Gorouh et al. (2018)] indicates up to 2 fold increases in precipitation (Figures 5B,C), particularly in the northern part of Vancouver’s downtown (i.e., the West Vancouver and North Vancouver District). The GEMCLIM simulations of historical events (Figure 5D) show relatively intense precipitation over the north of Metro Vancouver, which is generally in agreement with the precipitation pattern from Daymet (Figure 5A). The PMP simulations from GEMCLIM driven by the spatially adjusted LBCs (Figures 5E,F) generally resembles the pattern of PMP estimated by the statistical method, though a smaller magnitude of precipitation increase is found. Also, the PMP simulated by GEMCLIM shows larger precipitation increases (approximately 300 mm) in the southern part of Vancouver.

For the changes in the solid type precipitation, although the PMP simulations suggest an overall decrease in snowfall (by 12% on average) relative to the historical events (especially over most of the mid-south of Vancouver), a significant increase (>22.5 mm) in snowfall is displayed over the pre-slope downtown districts of Vancouver (Figure 6C). Moreover, the PMP simulations suggest an obvious increase (by up to 3 mm) and southward extension of freezing rain (Figure 6F), posing a greater hazardous impact on the transportation and electricity infrastructure of the city (Odon et al., 2017, 2019). The more widespread freezing rain implies a more pronounced thermal stratification in the lower troposphere. These can be explained by the maximized effects of ARs, including the concentrated warm advection and positive differential vorticity advection near the leading edges of AR plumes as discussed by the study of Liang and Sushama (2019).

To further understand the local hydrological response to the occurrence of PMP, the changes in the 72-h cumulative surface runoff for PMP relative to the historical events are analyzed (Figure 7). Over the northwest of Metro Vancouver, the surface runoff is approximately tripled by the occurrences of PMP (Figure 7B compared to A). Moreover, a pronounced extension of the overland flow is displayed over the east of the region during the occurrences of PMP (Figure 7B). These changes possibly indicate a dramatic increase in the intensity of runoff-induced hydrological extremes such as flood (Costa, 1987; Segond et al., 2007; Konrad and Dettinger, 2017), soil erosion (Nearing et al., 2005; Zuazo and Pleguezuelo, 2009) and debris flows (Coe et al., 2008; Kean et al., 2013).

Figure 8 displays the changes to the large-scale environments related to precipitation with maximized AR impacts. The simulated AR intensity (in terms of IVT) for the selected extreme precipitation events displayed more intense and concentrated IVT affecting Vancouver for the PMP simulations (Figures 8B,C) compared with the historical simulations (Figure 8A). The AR axes (shown by red lines in Figures 8A,B) also penetrate deeper toward inland when PMP occurs, which can greatly enhance heavy rainfall and flooding (Mahoney et al., 2016). Meanwhile, the IWV across the passage of ARs is approximately doubled by the maximized AR intensity, which partially explained the increase of precipitation as shown in Figures 5E,F. For the changes in the thermal environment, a warmer lower-troposphere (by about 2.1°C in terms of the vertically averaged temperature between 850 and 1000 hPa) is found near Vancouver (Figure 8I). This is possible due to the overlapped warm sectors of AR-related storms (Lavers et al., 2011) when the AR-related IVTs are shifted toward the vicinity of Vancouver. Although warmer air temperatures are favorable for the increase of IWV and precipitation, increases in the low-level temperature may result in shifts from snowfall to rainfall during AR events (Guan et al., 2010). This partly explains the overall reduction of snowfall over the mid-south of Vancouver (Figures 6B,C).

Simulations of the urban-scale dynamical factors for the historical extreme precipitation events and their PMP estimations are shown in Figure 9. When PMP occurred, we found approximately doubled magnitudes of vertical ascent between 850 and 1000-hPa levels over Vancouver (Figures 9B,C), especially in the pre-slope districts in the north of the city. Moreover, the low-level IVT associated with PMP (Figures 9E,F) tends to exhibit more pre-slope convergence (by 2∼6 mm/h). These imply a more favorable dynamical environment for generating precipitation. To further analyze how can the more favorable dynamical environments lead to increases to precipitation, the precipitation efficiency is calculated and presented in Figures 9G–I. Compared to the simulated historical events, the PMP simulations show increases in precipitation efficiency (by around 39%) in the pre-slope districts of Vancouver. This indicates that the occurrences of PMP are associated with the maximized IWV conditions from ARs as well as the stronger ability of the dynamical factors to deplete atmospheric moistures.