1. Introduction
Over the past decades computational resources have increased significantly in operational weather centers around the world and with this has come an increased capacity to run complex and computationally demanding numerical weather prediction (NWP) systems. More and more resources are now devoted to larger ensemble prediction systems, complex data assimilation systems, and more sophisticated physical parameterizations, including radiative transfer schemes, land surface schemes, and moist process schemes. For regional systems this also includes larger coverage of limited-area model (LAM) domains and increased horizontal and vertical resolution. Operational centers are quickly moving toward the convective scale, that is, kilometer-scale regional NWP systems, with horizontal grid spacings (Δx) on the order of 1–3 km. For example, the National Oceanic and Atmospheric Administration (NOAA) has an operational 3-km prediction system, the High-Resolution Rapid Refresh (HRRR; Pinto et al. 2015); Météo-France runs the Applications de la Recherche à l’Opérationnel à Méso-Echelle (AROME) model at 2.5 km (Seity et al. 2010); the Deutscher Wetterdienst (DWD) runs a version of the Consortium for Small-Scale Modeling model (COSMO-DE) at 2.8 km (Baldauf et al. 2011); and the Met Office has a 1.5-km configuration of the Unified Model (Lean et al. 2008).
High-resolution modeling is computationally costly. For each reduction of horizontal grid spacing by a factor of 2 there is an increase1 in the total number of gridpoint calculations and hence the total run time by a factor of 8. With kilometer-scale models, there is also the added cost of using more complex and realistic physics parameterizations. While such costs may have relatively less impact in research mode, for operational NWP the costs must be justified in terms of added forecast value. There are some distinct benefits of using kilometer-scale configurations over coarser-resolution systems. For example, the effective model resolution (i.e., the resolution of the smallest scales of motion) is higher (e.g., Skamarock and Klemp 2008); there is better representation of orography and other important sources of topographic forcing; there is a reduced need to parameterize deep convection, which can be problematic for forecasting summertime convective storms; and it becomes valid and beneficial to use more detailed parameterizations for cloud and precipitation processes.
Over the past two decades, the Meteorological Service of Canada (MSC), which provides official operational weather prediction from Environment and Climate Change Canada [ECCC; formerly called Environment Canada (EC) prior to 2016], and the Meteorological Research Division (MRD), which develops and maintains the Global Environmental Multiscale (GEM) model (Côté et al. 1998; Girard et al. 2014) and the NWP systems, have been experimenting with and moving toward kilometer-scale NWP. In 1997, MSC ran various operational systems including what is now referred to as the Regional Deterministic Prediction System (RDPS), which at the time provided short-term numerical guidance from 48-h forecast runs on a global grid with variable horizontal resolution and a grid spacing of 24 km over the region covering a large portion of North America (Bélair et al. 2000). To begin examining the effects of higher horizontal resolution, two real-time experimental forecast domains with grid spacings of 15 km covering most of Canada were run; the domains were referred to as the High-Resolution Meteorological Application (HiMAP; Côté et al. 2010). It was soon realized that this was insufficient to resolve the complex terrain of western Canada. In 1999, the performance of GEM with Δx of approximately 4 km was explored, expanding the focus to include the ability to forecast deep convection over the Canadian prairies. The 4-km GEM configuration, still with a global grid but with a high-resolution region over the area of interest, showed some skill over the regional 24-km prediction system (RDPS) in forecasting large-scale convective storms, such as the storm that produced a devastating (Fujita scale) F3 tornado over Pine Lake, Alberta, during the summer of 2000 (Erfani et al. 2003). The 4-km model also demonstrated improved skill, in comparison to RDPS and HiMAP, in forecasting weather phenomena related to orography. In the summer of 2001 an experimental 2.5-km system was set up over the Great Lakes in support of the Effects of Lake Breezes on Weather (ELBOW) project studying the interaction of synoptic-scale weather systems with lake-breeze fronts (Sills et al. 2002). Subjective evaluation during ELBOW showed improvement in the prediction of lake-breeze fronts for the 2.5-km model compared to the 24-km RDPS. These and other experiments provided a consistent general conclusion that there is potential for significant added forecast value in using high-resolution NWP systems, particularly with regard to high-impact weather.
In the last few years EC has been involved in several projects that involve real-time kilometer-scale NWP using GEM. Besides serving the research and/or operational needs of the projects directly, involvement in these projects has helped to test GEM and to advance its model development, as well as furthering MSC’s approach to high-resolution NWP. In 2008 the World Meteorological Organization (WMO) sponsored a forecast demonstration project (FDP) called Mesoscale Alpine Programme–Demonstration of Probabilistic Hydrological and Atmospheric Simulation of Flood Events (MAP-DPHASE), wherein weather centers from various countries ran real-time high-resolution NWP models over the region of the European Alps (Rotach et al. 2009). EC contributed to this FDP by running the GEM model, which was nested into a 2.5-km LAM grid over the FDP domain. During the summer of 2008, researchers from EC and several Canadian universities collaborated on the Understanding Alberta Boundary Layers Experiment (UNSTABLE) to characterize planetary boundary layer environments that are conducive to the initiation of severe convection in Alberta (Taylor et al. 2011). For this project EC ran an experimental real-time 1-km LAM domain over southern Alberta, whose simulations provided an opportunity to examine the potential benefits of downscaling from 2.5-km runs. A WMO-sponsored research demonstration project (RDP) referred to as the Forecasting and Research for Olympics Sochi Testbed 2014 (FROST-2014) took place in the region of Sochi, Russia, during the Winter Olympic and Paralympic Games in 2014. In addition to a suite of specialized instrumentation, this RDP involved various weather centers running high-resolution deterministic and ensemble predictions. EC provided real-time high-resolution GEM model output, from nested 2.5-km, 1-km, and 250-m grids. In the summer of 2015, the Pan American Games took place in the regions around Lake Ontario near Toronto, Ontario, Canada, for which EC ran a high-resolution GEM forecast system with a configuration similar to that for FROST-2014, where the 250-m grid also used the detailed Town Energy Balance model, which accounts for the complex surface fluxes in urban areas (Masson 2000).
As a specific move toward an operational kilometer-scale NWP system at MSC, two experimental, real-time, 2.5-km domains were implemented in 2004: one in western Canada and the other over the Great Lakes regions (see Fig. 1). Two more domains were later implemented: one in the Arctic and the other over the eastern Canadian maritime region. An additional experimental 2.5-km domain over the high Arctic was implemented in 2009 on a seasonal basis (July–October) in support of the shipping activities within the region. This set of 2.5-km domains eventually became known as EC’s experimental High Resolution Deterministic Prediction System (HRDPS). In support of forecasting for the Vancouver 2010 Winter Olympic and Paralympic Games, EC also ran a special high-resolution forecast system with nested 10-, 2.5-, and 1-km domains (not shown) centered over the Vancouver–Whistler region of British Columbia (Mailhot et al. 2012). This model configuration was later implemented into the experimental version of HRDPS in 2011. In the fall of 2012, the system was upgraded again, with the west 2.5-km domain having two daily 42-h integrations, and in January 2013 this domain was granted formal operational status.
Locations of the computational model domains for the 10-km RDPS (blue) and the 2.5-km HRDPS (red). The Gulf of St. Lawrence region is indicated by GSL. The dashed rectangles denote other experimental 2.5-km domains (see main text).
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
By this time, there were clear benefits of kilometer-scale NWP to MSC in terms of the added forecast value over the operational RDPS, which by that time had a Δx of approximately 10 km, based on forecaster feedback during the Vancouver 2010 Olympic Games and subjective evaluation of the experimental 2.5-km HRDPS windows. It was clear, however, that the continuation of the multidomain approach would be impractical and insufficient for numerical guidance of a Canada-wide operational system. Therefore, in November 2014 the experimental 2.5-km pan-Canadian HRDPS was introduced, with 48-h integrations run four times per day, thereby replacing the multigrid system (including the operational west domain, but excluding the two Arctic domains that remain). It is expected that this system will soon become operational and will eventually be the official source of short-term (days 1 and 2) numerical guidance by MSC.
The purpose of this article is to provide a detailed overview of the pan-Canadian HRDPS and to illustrate the added forecast value over the existing operational 10-km system through comparison to observations of common weather elements and objective scores. The remainder of the article is organized as follows. Section 2 provides an overview of the technical aspects of the pan-Canadian HRDPS. Section 3 summarizes the objective verification that was performed against surface and upper-air observations, with comparison to verification scores from the operational RDPS. Section 4 presents a short comparison of the two systems for a case of deep convection. Aspects of recent and current development are outlined in section 5, with concluding comments given in section 6.
2. Description of the pan-Canadian 2.5-km system
a. The GEM model
All operational weather prediction systems run at Environment Canada, deterministic and ensemble, use the GEM forecast model. A description of the dynamical core can be found in Côté et al. (1998) with details on recent developments in Girard et al. (2014). GEM is a nonhydrostatic atmospheric model that solves the fully compressible Euler equations. The model uses a combination of semi-implicit time differencing and semi-Lagrangian advection, which allows for stable model integrations with Courant numbers larger than 1, hence with much larger time steps than are required with Eulerian advection schemes.
Several horizontal grid options are available, including a global uniform grid (latitude–longitude projection), a variable-resolution global grid, and a LAM grid with a latitude–longitude projection. The option has been recently added for a global yin-and-yang configuration, which consists of two LAM grids “wrapped” around the globe similar to a baseball (Qaddouri 2011). The recent versions of GEM use a Charney–Phillips vertical coordinate with staggered thermodynamic and momentum levels. GEM is also capable of one-way self-nesting.
b. Main components of the HRDPS
1) Overview
This section describes the configuration of the pan-Canadian HRDPS that was implemented in November 2014, with results presented in the following two sections. Recent changes to the system are summarized in section 5. The HRDPS provides a deterministic forecast on a regional domain with Δx of 0.0225° latitude (approximately 2.5 km) with 48-h integrations four times daily, initialized at 0000, 0600, 1200, and 1800 UTC. The domain (Fig. 1) covers a large portion of Canada, extending from west to east from the Pacific to the Atlantic Oceans, covering much of the territories northward and extending southward into the northern United States.
The HRDPS is driven by the RDPS (Fig. 1; i.e., the RDPS provides initial and lateral boundary conditions) and consists of three main components: a land data assimilation system, a coupled 2.5-km GEM 6-h cycle system, and a full 48-h 2.5-km GEM forecast run. For a given integration of the 6-hourly atmospheric cycle, which is run on the same grid as the full forecast run, the upper-air initial conditions (ICs) are provided by the RDPS analysis, valid at the same time, with lateral boundary conditions (LBCs) from the RDPS forecast, updated every 1 h. Also, hourly fields from the RDPS are used for levels above 10 hPa through the application of an upper-boundary nesting technique (McTaggart-Cowan et al. 2011). Initial hydrometeor fields come from the final integration time of the previous 6-h cycle run. The overall configuration of the HRDPS is depicted in Fig. 2.
Schematic of sequencing for the HRDPS, showing RDPS (R-10), HRDPS (HR-2.5), GSL (coupled atmosphere–ocean system), atmospheric (ATM), surface (SFC), clouds (CLD; hydrometeor), ICs, and BCs.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
Some configuration details for the GEM components of the HRDPS are as follows, with other details of dynamics/numerics and physics configurations used in the HRDPS summarized in Table 1. The model is integrated using 62 vertical levels. Clouds and precipitation are computed using a combination of the Kuo transient shallow convection scheme (Bélair et al. 2005) and the two-moment version of the Milbrandt and Yau (2005) bulk microphysics scheme (MY2), with contributions of boundary layer clouds from the MoisTKE boundary layer scheme (Bélair et al. 2005). Radiative transfer calculations are performed using a correlated-k distribution scheme (Li and Barker 2005). Surface processes are treated using the Interactions between Soil, Biosphere, and Atmosphere (ISBA) land surface scheme (Noilhan and Planton 1989; Bélair et al. 2003).
Summary of GEM configurations details.
2) Surface initial conditions
(i) CaLDAS
The ICs for the mean surface temperature and root zone soil moisture for the atmospheric components of the HRDPS are provided by a coupled 2.5-km configuration of the Canadian Land Data Analysis System (CaLDAS; Carrera et al. 2015). CaLDAS uses the ensemble Kalman filtering (EnKF; Reichle et al. 2002) assimilation methodology with 24 members to generate the surface conditions necessary to drive the HRDPS. The land surface analyses are produced on the same computational grid as the GEM forecast model. For each 6-h period the 24 ensemble members are forced with 0–6-h atmospheric fields from the 2.5-km HRDPS forecast run. Perturbations are added to the radiation, air temperature forcing, root-zone soil moisture, and mean surface temperature to ensure a sufficient ensemble spread. The precipitation forcing used to drive the individual members within CaLDAS is derived from the Canadian Precipitation Analysis (Mahfouf et al. 2007; Carrera et al. 2015) optimum interpolation (OI) methodology, where an ensemble of precipitation analyses is generated for each 6-h period by combining spatially perturbed 0–6-h HRDPS first-guess precipitation with perturbed observations from the surface synoptic observation (SYNOP) and METAR networks.
The data assimilated into CaLDAS are screen-level observations of temperature and dewpoint temperature along with surface observations of snow depth. The screen-level and dewpoint temperatures are used to analyze the mean surface temperature and root-zone soil moisture every 3 h with the EnKF. Every 6-h snow depth observations are combined with the individual first-guess snow depth fields using the OI technique to produce 24 snow depth analyses, where the mean of these analyses is provided to the HRDPS. Along with the mean surface temperature, root-zone soil moisture, and snow depth, CaLDAS also provides the following initial conditions to the HRDPS: superficial soil moisture, surface temperature, frozen soil, water retained in the vegetation, water retained in the snowpack, snow albedo, and snow density. These additional variables are not analyzed but represent the arithmetic mean of the 24 first-guess ensemble members.
Figure 3 shows an example of the near-surface (0–10 cm) soil moisture provided by the 2.5-km CaLDAS. Given the additional data that are assimilated and the higher spatial resolution compared with the RDPS analysis, use of CaLDAS for the surface ICs is an important component of the HRDPS. Among other reasons, the resolution of initial soil moisture fields can be important in kilometer-scale NWP because of the increased ability to resolve moisture gradients, which can be an initiation mechanism for summertime deep convection (e.g., Smith and Yau 1993). Future work on the HRDPS will involve closer examination on the skill of predicting convective initiation and other impacts of CaLDAS.
Example of near-surface (0–10 cm) soil moisture from 2.5-km CaLDAS (valid 1200 UTC 25 Jun 2011).
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
(ii) Gulf of St. Lawrence coupled ocean–atmosphere analysis
For most of the ocean and lake regions, the initial surface conditions come from the RDPS analysis. However, for the particular region of the Gulf of St. Lawrence (GSL; see Fig. 1), the sea surface temperature, ice surface temperature, ice fraction, and ice thickness are provided by an ocean-ice analysis for the GSL (Smith et al. 2012). An example of the information provided from this system is shown in Fig. 4, which presents the ice fraction from a forecast of this system interpolated onto the HRDPS grid (Fig. 4b) along with the equivalent field from the RDPS analysis (Fig. 4a). Note the considerable increase in detail of this field, from which comes increased resolution of surface fluxes in that region. In the future, a newly implemented (experimental) coupled atmosphere–ocean modeling system run over the Great Lakes will be considered as a source for initial surface conditions for the HRDPS, similar to the way the forecast fields from the coupled Gulf of St. Lawrence system are currently used.
Example of sea ice fraction interpolated onto the HRDPS grid from (a) RDPS analysis and (b) forecast from the coupled atmosphere (GEM 15 km)–ocean (GSL) modeling system (valid 0000 UTC 25 Mar 2009).
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
3) 2.5-km GEM cycle
(i) Description
The cycle of consecutive 6-h GEM integrations on the full 2.5-km computational domain is another component of the HRDPS. Prognostic cloud and precipitation fields (hydrometeor mass and number mixing ratios from the microphysics scheme) at the initial time come from the 6-h forecast of the previous run in the cycle. This procedure of “recycling” the microphysics tracers constitutes a “hot start,” whereby cloud fields are present at the initial time, as opposed to a “cold start,” where clouds and precipitation must form gradually over a spinup period from initially cloud-free conditions. Although this procedure can result in some inconsistencies in the cloud fields recycled from the previous 2.5-km run and the relative humidity from the RDPS, tests were conducted (not shown) and indicated that these effects are minor and that inconsistencies are removed after the first model time step.
Also, the GEM cycle and forecast components both use a filtered orographic height that is initialized with the orography field from the 10-km driving model (RDPS) and evolves gradually over 1 h to the full resolution used in the 2.5-km grid. This reduces instabilities caused by rapid acceleration of air due to abrupt increases in terrain height and problems associated with valleys that are deeper in the high-resolution grid where field values would otherwise be extrapolated. It was demonstrated for the 1-km GEM system used during Vancouver 2010 that the procedure of gradually evolving the orography to full resolution resulted in better predictions of nocturnal temperature inversions at valley stations that were otherwise missed (Mailhot et al. 2012).
Except for the shorter integration time, a given integration from the 6-h cycle is otherwise identical to a full forecast run that starts at the same time. The cycle component provides the flexibility for the possible reduction of frequency of forecast runs, with no disruption to the cycle, or an increase in frequency of the cycle and also provides the infrastructure for the development of an upper-air data assimilation system.
(ii) Effects of recycling of hydrometeor fields
Without the hot start, clouds would form gradually over a period of several hours as the model generates high-resolution motion fields and grid-scale saturation, and the formation of precipitation, which comes almost entirely from the grid-scale condensation (microphysics) scheme, would be delayed. The effect of hydrometeor recycling is illustrated with a comparison of the set of HRDPS control simulations with the 80 benchmark cases (see section 3) to a sensitivity experiment set with the hydrometeor recycling shut off.
Figure 5 shows an example of the first 3 h of the column-maximum equivalent reflectivity from the model [see Milbrandt et al. (2008) for computation details] for a single case, comparing a simulation with a cold start (recycling off) and a run with a hot start (recycling on). Over the first few hours, the cloud fields gradually appear during the spinup period, taking approximately 3 h until the cold-start simulation has a comparable quantity of clouds compared with the hot-start simulation. In contrast, the initial reflectivity field for the hot-start run already has the same degree of overall coverage as it does after 3 h.
Column-maximum model equivalent reflectivity at each hour for HRDPS simulations with hydrometeor recycling (left) off and (right) on.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
Since clouds form gradually for a cold-start run, there is also a greater spinup period for precipitation at the surface. To examine this more quantitatively, the spinup time for precipitation was determined for each set of runs, which is the time it takes for the precipitation field to reach a statistical equilibrium. The 1-h accumulated precipitation was averaged over the entire domain and for all 80 cases. The average hourly precipitation versus forecast hour for each set is plotted in Fig. 6. The accumulation after the first hour is very small but increases quickly with time over the first 7 h before reaching a quasi-equilibrium state, with a distinct diurnal cycle and a constant gradual background increase. This constant increase is related to a gradual precipitation increase in the global modeling system, from which the model lateral boundary conditions ultimately originate (indirectly, through the RDPS). The hourly precipitation for the run set with hydrometeor recycling on eventually reaches a very similar equilibrium state. The spinup time of the hot-start runs appears to be about approximately 5 h, 2 h less than for the cold-start runs. However, during the spinup period itself, the recycling-on runs have hourly precipitation amounts that are much closer to the equilibrium values (the dashed line in Fig. 6b, extrapolated from the equilibrium period to earlier hours). For example, the average precipitation after only 1 h is approximately 65% of the equilibrium value for the recycling-on runs, compared with 12% for the recycling-off runs.
Domain-averaged 1-h accumulated precipitation over all 80 benchmark cases, for run sets with hydrometeor recycling off (blue) and on (red) for (a) the full 48 h and (b) the first 8 h only. The near-horizontal dashed lines denote the peak equilibrium values (throughout the diurnal cycle). The percentages indicate the ratio of the model precipitation values to the equilibrium values (extrapolated back to 0 h) at the given times.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
4) 2.5-km GEM forecast run
Finally, the system includes the full GEM forecast run itself. As described above, the forecast component is identical to the cycle component except that it is integrated for the full 48-h forecast period.
3. Objective evaluation and comparison to the 10-km operational system
a. Evaluation methods
To evaluate objectively the forecast skill of the HRDPS and to compare directly with that of the 10-km operational RDPS [whose configuration details are similar to those in Mailhot et al. (2006)], a set of 80 benchmark cases (40 winter, 40 summer, during 2011) were run using each system, with runs staring at 0000 and 1200 UTC but otherwise separated by 3 days to provide meteorologically independent cases. At the time of the implementation of the HRDPS in 2014 the RDPS had not been modified since 2011, so this benchmark test set was perfectly valid. Standard skill scores for model forecasts against surface station observations for common weather elements were computed. These included screen-level temperature T, humidity (dewpoint temperature Td), wind speed Vspd, and wind direction Vdir as well as 6-h accumulated precipitation. Model precipitation values at grid points nearest to the stations were used to compute skill scores for each prediction system. It is recognized that this type of point comparison of precipitation forecast scores for models with different horizontal resolutions generally means that the high-resolution model suffers more from the “double penalty” problem of having both a false alarm at the model point and a missed forecast at the observation point for forecasts that are correct but displaced. More rigorous precipitation evaluations will, in the future, include methods that are more appropriate for high-resolution models, such as neighborhood verification methods (Ebert 2009). Standard evaluation metrics for each system were computed using an in-house verification package, described in appendix A.
Model cloud cover was evaluated by comparison against human observations of fractional cloudiness over the celestial dome, the region seen by a human observer. To approximate the model field of this observed weather element, the 2D model fractional cloud cover over a column, as diagnosed by the radiation scheme based on maximum random overlap of the explicit and implicit clouds at each level, was spatially averaged over an area of approximately 70 km × 70 km (27 × 27 points for the HRDPS and 7 × 7 points for the RDPS), assuming a human observer’s range to be approximately 35 km. The reason for the spatial averaging is illustrated in Fig. 7, which shows an example of the unsmoothed 2D cloud fraction from the HRDPS along with model grid points (small red dots for the HRDPS, large green dots for the RDPS), with the large red dot denoting the center point (near the observation location) of the square area over which the model fields are averaged. In this example the station is located near a sharp gradient in model cloudiness; the spatial averaging greatly reduces changes in results as a result of slight displacements of model clouds and effectively mimics the observations. This, of course, is not a strict simulated observation but it serves as a convenient means of estimating the general quality of the model cloudiness and of contrasting the cloud covers from the two prediction systems.
Example of model 2D fractional cloud cover field from the HRDPS. Pink dots denote grid points from the HRDPS (Δx = 2.5 km) grid; solid green circles denote the approximate gridpoint locations from the RDPS (Δx = 10 km). The solid red circle denotes the center point over which the spatial averaging is performed over a 70 km × 70 km area for comparisons with surface observations from the nearest station.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
The evaluation of upper-air forecasts of temperature T, dewpoint depression T − Td, geopotential height Φ, and zonal and meridional wind speeds was also done through comparison with radiosonde observations and the computation of biases and the standard deviation of error.
b. Results
The results from the objective verification of the HRDPS benchmark cases against observations, and with corresponding verification of the RDPS for comparison, are summarized below.
1) Surface fields
To avoid smoothing of the diurnal signals from runs with different forecast lead times, the runs initialized at 0000 and 1200 UTC were scored separately. Only the 0000 UTC runs are shown, but the results for the 1200 UTC runs are qualitatively similar, just shifted in time. In Figs. 8–14 the bias (BIAS) and standard error (STDE) for model versus observation results, for both the HRDPS (in red) and the RDPS (in blue), are indicated at each forecast hour, with scores from the winter and summer cases shown separately. The differences in the scores between the two systems, Δ = (|BIASRDPS| − |BIASHRDPS|) and Δ = (|STDERDPS| − |STDEHRDPS|), are indicated at the bottom of each panel, with the shading denoting the 5%–95% confidence interval from the bootstrapping (see appendix A). A positive (negative) value of Δ indicates a better score for the HRDPS (RDPS) for each metric. The 2-m T BIAS and STDE computed over the entire domain are shown in Fig. 8. Since the full computational grid has regions with different types of geography and weather, scores are also computed over six different subdomains: the mountainous region of British Columbia (BC), the Canadian prairie provinces (Prairies), Ontario and the surrounding region (QC-ON), the eastern Canadian maritime region (Maritimes), the northern region (North), and the southern part of the domain that extends south of the U.S.–Canada border (USA). For the scores computed over these subdomains, the BIAS and STDE for the 2-m T are shown in Figs. 9 and 10, respectively; scores for 2-m Td are shown in Figs. 11 and 12; and scores for Vspd are shown in Figs. 13 and 14.
(a),(b) BIAS and (c),(d) STDE for 2-m T (°C) for (left) winter and (right) summer cases initialized at 0000 UTC for the HRDPS (red) and RDPS (blue) in the region of the full HRDPS domain. The top portion of each panel shows the scores; the bottom portion shows Δ = (|BIASRDPS| − |BIASHRDPS|) in (a) and (b), and Δ = (|STDERDPS| − |STDEHRDPS|) in (c) and (d), with the gray shading denoting the 5%–95% confidence interval.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
As in Fig. 8, but for the BIAS of 2-m T (°C) over the specific region indicated in each panel for the (a)–(f) winter and (g)–(l) summer cases. The scores in the USA subdomain at 24 and 48 h have been removed because of contaminated values.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
As in Fig. 9, but for STDE of 2-m T (°C).
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
As in Fig. 9, but for BIAS of 2-m Td (°C).
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
As in Fig. 9, but for STDE of 2-m Td (°C).
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
As in Fig. 9, but for BIAS of 10-m wind speed (m s−1).
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
As in Fig. 9 but for STDE of 10-m wind speed (m s−1).
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
The HRDPS near-surface temperature is consistently warmer than the 10-km model in winter when considering the entire domain (Fig. 8), being an improvement in some regions (e.g., Figs. 9e,f) and causing deterioration in others (Fig. 9d). The STDE is neutral on average when considering the entire domain as a whole (Fig. 8) but is generally reduced in most of the specific regions (Figs. 10a,b,c,e,f), indicating improvement for the HRDPS. In the summer, over the entire HRDPS domain there is a cold bias for the HRDPS compared with a warm bias for the RDPS (smaller in magnitude; Fig. 8b) and with a net-neutral STDE (Fig. 8d). Regionally, there are mixed changes in the T for both the bias (Fig. 9) and the STDE (Fig. 10).
For humidity, there is a general increase in the moist bias in the winter (Figs. 11a–f), which is consistent with upper-air scores for much of the troposphere (see below); however, the STDE is generally reduced, except at later forecast times (after 42 h) in certain regions, such as the Prairies (Figs. 11a–f). For the summer cases, there is also an increased moist bias (Figs. 11g–l) but now this change represents a general improvement over the RDPS, as does the general reduction in STDE (Figs. 12g–l). Recent tests (not shown) have indicated that the screen-level temperature and humidity HRDPS forecasts can be improved considerably by modifying the interface between the radiative transfer scheme and the microphysics through improved consistency among the schemes for the computation of cloud optical properties. Specifically, this involves using the values of the effective radii of cloud droplets and ice crystals that are predicted from the two-moment microphysics scheme for the computation of cloud optical depth, single-scattering albedo, and asymmetry factor, rather than using only the hydrometeor mass contents and imposing assumptions on the particle sizes, as is done by necessity for larger-scale models, which generally use simpler condensation schemes.
For the screen-level wind speed in both the winter and summer, there is a considerable regional dependency on the changes in biases, with a bias increase in the BC region (improvement; Figs. 13a,g), and a decrease in the Prairies (deterioration; Figs. 13d,j), Maritimes (deterioration; Figs. 13c,i), and QC-ON (improvement; Figs. 13f–l). The same algorithm was used throughout the domain to compute the surface roughness length, which heavily influences the parameterized surface friction and hence the wind speed. However, different algorithms could be applied to different regions in order to reduce the wind speed bias in specific regions. The STDE is reduced by up to 0.2 m s−1 in some regions (Fig. 14). The scores for the screen-level wind direction (not shown) are changed only very slightly.
For 6-h precipitation amounts exceeding 0.5 mm, the bias and percent correct for the entire domain are shown in Fig. 15. There is a distinct improvement in the HRDPS in reducing the overprediction from the RDPS for both winter and summer (Figs. 15a,b). The percent correct is also improved for the winter (Fig. 15c), though for the summer there is a net-neutral change (Fig. 15d). The evaluation of model precipitation and comparisons of models with different spatial resolutions is subtle for a variety of reasons (Ebert 2009) and requires in-depth analysis, which is beyond the scope of this study, but which is the topic of ongoing work. The scores presented here are shown to illustrate that the precipitation forecasts from the HRDPS are at least as good as that from the RDPS.
As in Fig. 9, but for (a),(b) BIAS and (c),(d) % CORRECT for 6-h accumulated precipitation amounts greater than or equal to 0.5 mm for (left) winter and (right) summer cases over the region of the entire HRDPS.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
2) Cloud cover
For cloudiness, there is a distinct improvement in the HRDPS for partial cloudiness and overcast conditions, particularly in the winter, based on this evaluation. Unlike the RDPS, the frequency bias scores for the HRDPS in the winter are nearly perfect (Figs. 16b–d), where perfect is a frequency bias of 1, and there is approximately a 10% improvement in the percent correct (Fig. 17a). For the summer, the change in frequency bias is not as impressive, with the HRDPS now having too little fractional cloudiness (Figs. 16f,g), but there is still improvement for summer overcast conditions (Fig. 16h) and a slight improvement is shown in the percent correct (Fig. 17b). Despite the overall improvement, the comparisons to the observations illustrate that the HRDPS still needs a spinup period of several hours to obtain overcast conditions (Figs. 16d,h). We remark that the MY2 microphysics scheme has a known tendency to overpredict upper-level ice leading to excessively large anvils for cases of deep convection (Cintineo et al. 2014), which may possibly deteriorate the HRDPS cloudiness in the summer. This bias in MY2 has been addressed, and the modifications will be considered for a future version of the system.
As in Fig. 9, but for frequency bias (FREQ_BIAS) in fractional cloudiness over the region of the entire HRDPS domain for values ranging from (a) 0 to 0.25, (b) 0.25 to 0.50, (c) 0.50 to 0.75, and (d) 0.75 to 1.00.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
As in Fig. 9, but for % CORRECT for fractional cloudiness over the entire domain for (a) winter and (b) summer cases.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
As a caveat, we reiterate that this evaluation of model cloudiness is not based on a strict comparison of matching observed and model fields. Furthermore, the representation of clouds in an NWP model is a complex combination of different schemes parameterizing different types of clouds; thus, there is no unique way to compute the model cloud fraction. Current development of the HRDPS involves evaluation of model clouds based on simulated GOES brightness temperatures and comparison to satellite observations, as in Cintineo et al. (2014). Nevertheless, we argue that the above evaluation indicates that the cloudiness in the HRDPS is quite reasonable and is an improvement over that from the RDPS.
3) Summary of evaluation of surface fields and cloudiness
The objective scores presented in Figs. 9–17 for the near-surface variables and cloud cover in the various regions of the HRDPS domains are summarized in Tables 2 and 3 in terms of net improvement or deterioration compared to the RDPS. For example, the wintertime 2-m T bias in the maritime region (Fig. 9c) is deemed to be slightly improved, indicated by the single plus sign (+) in Table 2. There is some geographical and seasonal dependence on the change in results. There are some notable weaknesses in the HRDPS in some cases; for example, the temperature and humidity in the Prairies and the northern regions are distinctly deteriorated in the winter. This evaluation provides a guide for targeting future improvements to the system. Overall, however, the scores indicated that the forecast skill of the HRPDS is generally improved over that of the RDPS. The high-resolution system therefore exhibits the distinct potential to provide improved numerical guidance for the prediction of important weather elements.
Summary of objective scores for screen-level fields, precipitation, and cloud cover for the winter cases for the indicated fields, metrics, and geographic regions. The plus (+) and minus (−) signs denote a net improvement or deterioration, respectively, for the HRDPS compared with the RDPS, with a single sign indicating a slight overall change, two signs indicating a pronounced change, and a solidus (/) denoting a net-neutral impact.
4) Upper-air scores
The verification of model forecasts of upper-air fields by comparison against radiosonde observations is a common practice in weather centers when evaluating large-scale models. Given the relatively large areal coverage of the HRDPS compared to kilometer-scale systems in use at other centers and the availability of sounding stations throughout much of the domain, it was feasible to compute standard skill scores for upper-air temperature, humidity, geopotential heights, and winds for the HRDPS benchmarks cases and compare the results with those of the RDPS. Note that while the main goal of a kilometer-scale model is to improve the forecasts of surface weather elements, this verification exercise is performed as an important “sanity” check to ensure that there is no obvious deterioration of upper-air fields and to identify possible systematic problems. Later we summarize the bias and STDE of the aforementioned fields for the 48-h forecasts of the HRDPS and the RDPS for all 80 test cases for the winter and summer (Figs. 19 and 20, respectively. Note that the in-house tool for computing upper-air scores evaluates the dewpoint depression T − Td, not Td.
The scores are computed at every synoptic hour and grow in time for both systems. The differences in errors between the HRDPS and the RDPS, where they exist, are greatest at the maximum lead time (48 h), shown in Figs. 18 and 19. Similar differences but with smaller magnitudes exist at earlier times (not shown). Overall, the mass (geopotential height) and wind fields for the two models are nearly identical. For the winter cases, there is some increase in error in air temperature for the HRDPS, and the cold bias below 400 hPa in both systems changes to a slight warm bias at the lowest level for the HRPDS (Fig. 18d); in the summer the error is essentially unchanged (except slightly at the lowest levels), but the bias throughout most of the troposphere is notably improved (Fig. 19d). One notable problem in the HRDPS that is not present in the RDPS is a distinct moist bias in the mid- to lower troposphere, particularly in the winter, as indicated by the negative bias in T − Td (Figs. 18e and 19e) but with very little bias in T (Figs. 18d and 19d). Recent tests indicate that this moist bias may originate from the microphysics scheme and could be corrected by modifying certain process rates (e.g., reducing the sublimation of snow); work to correct this is ongoing. This result illustrates the utility of examining upper-air scores, without which the depth of this moist bias in the HRPDS may not have been detected.
Upper-air scores, based on comparison to radiosondes over the HRDPS domain, vs air pressure p at 48 h for winter cases for (a) zonal wind speed (m s−1), (b) meridional wind speed (m s−1), (c) geopotential height (dam), (d) temperature (°C), and (e) dewpoint depression (°C). The dashed (solid) lines denote BIAS (STDE) with red (blue) for the HRDPS (RDPS). The colored boxes denote statistical significance at the value indicated with red (blue) indicating that the HRDPS (RDPS) has a better score. Scores at the 1000-hPa level have been removed.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
As in Fig. 18, but for summer cases.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
4. Example of a case of deep convection
In an operational weather center it is crucial to demonstrate that a new prediction system aimed at replacing an existing operational one does not degrade (and hopefully improves) the forecast guidance for common weather elements evaluated by conventional metrics, as discussed in detail in the previous section. Nevertheless, one of the main reasons for moving toward a computationally expensive kilometer-scale NWP system is for the improved guidance for high-impact weather. Because of temporal and spatial scale issues and other considerations, this is more difficult to evaluate objectively and systematically. Thus, to illustrate briefly and subjectively the potential added value of the HRDPS over the operational 10-km system for one type of high-impact weather, a case of deep convection is presented.
Figure 20 compares a summertime case from the benchmark set (see section 3) of a small-scale convective outbreak over the northern U.S. plains on 8 July 2011. Model precipitation rates from the RDPS (Figs. 20a,b) and HRDPS (Figs. 20c,d) at two forecast lead times (14 and 18 h) are shown along with reflectivity from the HRDPS (Figs. 20e,f) and from the observed NEXRAD mosaic (Figs. 20g,h). The HRDPS better captures the strong convective regions with high precipitation rates (>30 mm h−1)—in terms of location, structure, and magnitude—in southern North Dakota at 0200 UTC (left column) and in South Dakota at 0600 UTC (right column). The precipitation rates are not observed directly, but the values from the HRDPS (Figs. 20c,d) are consistent with the observed radar reflectivity values of >45 dBZ [e.g., based on reflectivity–rain rate (Z–R) relations commonly used by the U.S. National Weather Service]. In contrast, the precipitation rates from the RDPS in those regions (Figs. 20a,b) are diffuse and weak. The HRDPS also discriminates well between the convective regions, as indicated by higher reflectivity values (e.g., >40 dBZ; Figs. 20e–h), and stratiform regions, as indicated by lower reflectivity values (<40 dBZ; Figs. 20e–h) by comparison of the observed reflectivities.2 Since the RDPS uses a diagnostic precipitation scheme (with very simple microphysics and no in-cloud precipitation) and a convective parameterization scheme, it is meaningless to compute model reflectivity for this system.
Model precipitation rates valid at (left) 0200 UTC (14-h forecast time) and (right) 0600 UTC (18-h forecast time) 8 Jul 2011 for the (a),(b) RDPS and (c),(d) HRDPS; (e),(f) equivalent reflectivity (column maximum) for the HRDPS; and (g),(h) observed reflectivity from NEXRAD 1-km mosaic.
Citation: Weather and Forecasting 31, 6; 10.1175/WAF-D-16-0035.1
Even with downscaling from the 10-km model, the 2.5-km system clearly has potential added forecast value for high-impact weather associated with deep convection. This may be partly attributed to the better surface initial conditions, but it is certainly strongly related to the higher capacity to resolve the storm dynamics as a result of the finer grid spacing and the use of a detailed microphysics scheme that can better represent the important diabatic processes of latent heat release due to condensation and cooling due to evaporation and melting, which are important for cold pool dynamics in deep convective systems.
It is worth noting that the differences between the RDPS and the HRDPS exhibited in Fig. 20 will likely have little or no impact on standard QPF scores, such as those discussed in section 3; thus, the added value from the high-resolution model tends to get obscured or overlooked when using conventional metrics only. Eventually, systematic and objective methods to evaluate model storm structures and reflectivity patterns from the HRDPS will be explored, as well as approaches to evaluating other model weather elements related to various types of high-impact weather.
5. Recent updates
After the official initial implementation of the HRDPS in November 2014, further development of the system was undertaken, which led to a set of modifications in December 2015. Their impacts are not documented in detail here, but in order to provide a complete description of the current state of the HRDPS (at the time of writing), the main changes included in the 2015 update are described briefly below.
a. Use of a deep convective parameterization scheme
The most recent upgrade included the introduction of the use of the Kain and Fritsch (1990; KF) convective parameterization scheme (CPS) to treat subgrid-scale deep convection. Atmospheric models with Δx of 1–4 km (or less) have in the past generally not used a CPS since this grid spacing was commonly considered outside of the “gray zone” for convection, where deep convection is considered to be sufficiently resolved (e.g., Zhang et al. 1998). More recently, it has been noted that with any grid spacing greater than 1 km the model is still in the gray zone and that there may be a practical benefit or even a need for using a CPS at this scale (e.g., NOAA Center for Weather and Climate Prediction 2015). The KF scheme was therefore tested in the HRDPS with parameter settings adjusted appropriately, scaled from those used in the RDPS based on the horizontal grid spacing. While there are conceptual arguments against the validity of the closure assumptions in a mass-flux convective scheme at this scale, the KF scheme was readily available in GEM for testing.
The results in the HRDPS were mixed, but there was a distinct reduction in the bias in the diurnal cycle of hourly summertime precipitation for high precipitation amounts compared with runs without the use of a CPS (not shown). Also, for strong convective cases such as squall lines, the precipitation patterns often appeared to be more realistic, being more continuous and less “spotty.” Given the apparent positive impact on precipitation in the HRPDS and the flexibility that is permitted by the current experimental status of the system, the KF scheme was included in the configuration of the 2015 upgrade to the system. Detailed examination of using the KF scheme at this scale is being conducted and will be reported upon in an upcoming study.
b. Modification to the microphysics to improve freezing rain
Another change that was introduced was a modification to the MY2 microphysics scheme to address a problem that was discovered after the initial HRDPS implementation that involved missed forecasts of freezing rain. Upon investigation, one problem that emerged was an inconsistency in the energy budget routine of ISBA that produced excess surface temperatures under conditions of precipitating rain onto previously accumulated snow, which incorrectly resulted in rapid melting and ultimately boundary layer temperature profiles that did not support freezing rain. This problem was corrected immediately in the real-time system. It was also discovered that under correctly forecast temperature inversion layers at the leading edge of warm fronts there was systematic reglaciation of rain in the microphysics, which resulted in the explicit precipitation at the surface being incorrectly forecast as solid frozen precipitation (graupel) rather than freezing rain. A modification to the MY2 scheme was made (see appendix B) and is included in the 2015 HRDPS upgrade, which successfully reduced the incorrect reglaciation of rain and solved the problem of missed freezing rain forecasts to which this was attributed.
6. Discussion and conclusions
A detailed overview of ECCC’s real-time experimental 2.5-km deterministic NWP system, the HRDPS, has been provided. Standard objective skill scores for the prediction of common meteorological surface fields compared with station observations, as well as fractional cloud cover and comparison to radiosonde observations, have been presented along with comparisons of the performance of the operational 10-km deterministic system. While there is some geographical and seasonal dependence on the results, the 2.5-km system was shown to provide an overall improvement for surface fields and cloud cover. With the exception of a moist bias in the troposphere in the winter, the effects on the upper-air scores were near neutral or improved. The HRDPS is currently (at the time of writing) run in experimental mode by the MSC operations section and direct model output from the system is available in gridded binary (GRIB2) format online through Datamart (http://dd.meteo.gc.ca/model_hrdps/). Preparation is under way (for postprocessed fields, model output statistics, etc.) in anticipation of the system becoming officially operational.
Although some increased forecast value from the HRDPS has been demonstrated, it is arguable that the improvements shown alone may not merit the added computational cost over the 10-km system. While the focus here has been on standard metrics of common weather elements, there is considerably more potential for added value to short-term guidance from a kilometer-scale NWP system, much of which may be obscured or overlooked with evaluation using standard bulk verification statistics. In addition to the benefit of higher resolution of topographic forcing, the fact that a kilometer-scale model resolution dictates the need for a prognostic precipitation scheme and makes valid the use of a detailed parameterization of cloud microphysics invites the exploitation of improved simulation capabilities that pertain to high-impact weather. For example, even at 2.5-km grid spacing the GEM model using MY2 has been shown to have an impressive ability to simulate storm structure and reflectivity patterns of deep convective storms (Taylor et al. 2011); with the two-moment treatment of cloud water, visibility through fog can be parameterized (Mailhot et al. 2012); the representation of the snow size distribution allows for the explicit prediction of the instantaneous solid-to-liquid ratio of falling snow (Milbrandt et al. 2012); the ability to predict aircraft icing potential exists due to a two-moment scheme’s capacity to simulate supercooled cloud water (Reisner et al. 1998); and the explicit prediction of hail and hail sizes is possible in principle (Milbrandt and Yau 2006).
Most of these aspects are challenging to observe and quantify systematically; hence, objective model verification is difficult. Nevertheless, there has been an increased use of kilometer-scale deterministic NWP systems in weather centers worldwide, enhancements to detailed meteorological observational systems, developments in verification techniques appropriate for high-resolution models (e.g., Ebert 2009), and a general increase in concern around high-impact weather. Thus, it is reasonable to expect that more attention will be paid to these unconventional applications of kilometer-scale modeling systems and that there will be more application of the guidance of these types of fields into forecast systems.
Further development of the HRDPS and the testing of new components and configurations is under way in an effort to improve the system’s performance in anticipation of the switch to operational status. One major component being tested and considered is the switch from MY2 to the Predicted Particle Properties (P3) microphysics scheme recently developed by Morrison and Milbrandt (2015). This scheme has been shown to be a conceptual improvement, is less computationally costly, and produces competitive results compared to standard state-of-the-art microphysics schemes (Morrison et al. 2015). The potential benefits to the modification to the cloud droplet nucleation in the microphysics based on the “aerosol aware” approach of Thompson and Eidhammer (2014) are also being examined. Another item to be tested is the use of the Town Energy Balance model in order to improve the prediction of surface fluxes in urban areas. Increased vertical resolution in the lower troposphere, in particular with the lowest model level much closer to the surface, has been shown to improve the prediction of low-level winds and temperature in GEM (Bernier and Bélair 2012) and will, therefore, also be tested.
Eventually, it is likely that the 2.5-km system will also have its own upper-air data assimilation (DA) system, most likely a type of EnKF. Research and development toward this end is ongoing (Caron et al. 2015). With its own upper-air DA system, the system would no longer be downscaled from the 10-km model, but would be a fully independent convective-scale prediction system. Regardless, the 2.5-km deterministic system will likely become MSC’s primary source of numerical guidance for short-term NWP in the foreseeable future.
Acknowledgments
We express our thanks to the many individuals from various sections in Environment Canada, including the Analysis and Prediction section of MSC, the MRD national laboratories, and regional forecast offices who have provided useful feedback on the HRDPS and earlier development versions over the past several years. We would also like to acknowledge the contributions of Amin Erfani and Jocelyn Mailhot, who led the development of the original GEM-LAM-2.5 project, which set the stage for operational kilometer-scale NWP in Canada.
APPENDIX A
Description of the Verification Package
The objective evaluation of model surface fields and fractional cloud cover against point observations was performed using an in-house verification package, referred to as USTAT, which computes several standard skill scores. In general, when the verification period is a valid representation of the verification population and each event is independent and part of the same distribution, then bootstrapping the sampling period is equivalent to estimating the verification population. This procedure is based on drawing observations with replacement. Meteorological data are spatially and temporally correlated, which forces the use of bootstrapping in blocks to capture the dependence structure of neighboring observations. The width of the confidence interval provides an estimate of the uncertainty inherent in the process of population sampling. USTAT performs this statistical procedure. The stations are precisely located to the degrees, minutes, and seconds (resolution of about 30 m) to verify very high-resolution models. A strict quality control dataset is grafted into the package. This current version can verify the temperature at 2 m, the dewpoint temperature at 2 m, wind speed and direction at 10 m, and 6-h accumulated precipitation amounts, as well as the total cloud cover.
APPENDIX B
Modification to the Microphysics
In the MY2 scheme, frozen hydrometeors can be initiated from the so-called three-component freezing mechanism whereby collection occurs between rain and cloud ice, snow, or graupel to form graupel or hail (Milbrandt and Yau 2005). Investigation into the freezing rain problem revealed that the excessive reglaciation of supercooled rain occurred primarily as a result of trace amounts of graupel acting sufficiently as an embryo for rain to refreeze via this mechanism. In the MY2 version originally used in the HRDPS, three-component freezing could occur at all temperatures below 0°C for collection between rain and either ice or graupel or below −10°C for rain and snow. A modification was implemented whereby all three-component freezing was restricted to temperatures colder than −5°C. This change is somewhat ad hoc but has physical justification since drop freezing rates are temperature dependent, increasing with the amount of supercooling (Pruppacher and Klett 1997). Furthermore, MY2 assumes the drop temperature to be equal to the air temperature, which may imply an overestimation of the degree of supercooling for drops falling from (or being advected upward from) relatively warm air that require time for their temperatures to adjust to the environment.
REFERENCES
Baldauf, M., Seifert A. , Förstner J. , Majewski D. , Raschendorfer M. , and Reinhardt T. , 2011: Operational convective-scale numerical weather prediction with the COSMO model: Description and sensitivities. Mon. Wea. Rev., 139, 3887–3905, doi:10.1175/MWR-D-10-05013.1.
Bélair, S., Méthot A. , Mailhot J. , Bilodeau B. , Patoine A. , Pellerin G. , and Côté J. , 2000: Operational implementation of the Fritsch–Chappel convective scheme in the 24-km Canadian regional model. Wea. Forecasting, 15, 257–274, doi:10.1175/1520-0434(2000)015<0257:OIOTFC>2.0.CO;2.
Bélair, S., Crevier L.-P. , Mailhot J. , Bilodeau B. , and Delage Y. , 2003: Operational implementation of the ISBA land surface scheme in the Canadian regional weather forecast model. Part I: Warm season results. J. Hydrometeor., 4, 352–370, doi:10.1175/1525-7541(2003)4<352:OIOTIL>2.0.CO;2.
Bélair, S., Mailhot J. , Girard C. , and Vaillancourt P. , 2005: Boundary layer and shallow cumulus clouds in a medium-range forecast of a large-scale weather system. Mon. Wea. Rev., 133, 1938–1960, doi:10.1175/MWR2958.1.
Bernier, N. B., and Bélair S. , 2012: High horizontal and vertical resolution limited-area model: Near-surface and wind energy forecast applications. J. Appl. Meteor. Climatol., 51, 1061–1078, doi:10.1175/JAMC-D-11-0197.1.
Caron, J.-F., Milewski T. , Buehner M. , Fillion L. , Reszka M. , Macpherson S. , and St-James J. , 2015: Implementation of deterministic weather forecasting systems based on ensemble-variational data assimilation at Environment Canada. Part II: The regional system. Mon. Wea. Rev., 143, 2560–2580, doi:10.1175/MWR-D-14-00353.1.
Carrera, M. L., Bélair S. , and Bilodeau B. , 2015: The Canadian Land Data Assimilation System (CaLDAS): Description and synthetic evaluation study. J. Hydrometeor., 16, 1293–1314, doi:10.1175/JHM-D-14-0089.1.
Cintineo, R., Otkin J. A. , Xue M. , and Kong F. , 2014: Evaluating the performance of planetary boundary layer and cloud microphysical parameterization schemes in convection-permitting ensemble forecasts using synthetic GOES-13 satellite observations. Mon. Wea. Rev., 142, 163–182, doi:10.1175/MWR-D-13-00143.1.
Côté, J., Gravel S. , Méthot A. , Patoine A. , Roch M. , and Staniforth A. , 1998: The operational CMC–MRD Global Environmental Multiscale (GEM) model. Part I: Design considerations and formulation. Mon. Wea. Rev., 126, 1373–1395, doi:10.1175/1520-0493(1998)126<1373:TOCMGE>2.0.CO;2.
Côté, J., Girard C. , Plante A. , McTaggart-Cowan R. , Milbrandt J. , and Qaddouri A. , 2010: Non-hydrostatic modelling with the GEM model. Preprints, Workshop on Non-hydrostatic Modelling, Reading, United Kingdom, ECMWF. [Available online at http://www.ecmwf.int/sites/default/files/elibrary/2010/15129-workshop-non-hydrostatic-modelling-global-environmental-multiscale-gem-model.pdf.]
Ebert, E. E., 2009: Neighborhood verification: A strategy for rewarding close forecasts. Wea. Forecasting, 24, 1498–1510, doi:10.1175/2009WAF2222251.1.
Erfani, A., Méthot A. , Goodson R. , Bélair S. , Yeh K. S. , Côté J. , and Moffet R. , 2003: Synoptic and mesoscale study of a severe convective outbreak with a nonhydrostatic Global Environmental Multiscale (GEM) model. Meteor. Atmos. Phys., 82, 31–53, doi:10.1007/s00703-001-0585-8.
Girard, C., and Coauthors, 2014: Staggered vertical discretization of the Canadian Environmental Multiscale (GEM) model using a coordinate of the log-hydrostatic-pressure type. Mon. Wea. Rev., 142, 1183–1196, doi:10.1175/MWR-D-13-00255.1.
Kain, J. S., and Fritsch J. M. , 1990: A one-dimensional entraining/detraining plume model and its application in convective parameterization. J. Atmos. Sci., 47, 2784–2802, doi:10.1175/1520-0469(1990)047<2784:AODEPM>2.0.CO;2.
Lean, H. W., Clark P. A. , Dixon M. , Roberts N. M. , Fitch A. , Forbes R. , and Halliwell C. , 2008: Characteristics of high-resolution versions of the Met Office Unified Model for forecasting convection over the United Kingdom. Mon. Wea. Rev., 136, 3408–3424, doi:10.1175/2008MWR2332.1.
Li, J., and Barker H. W. , 2005: A radiation algorithm with correlated-k distribution. Part I: Local thermal equilibrium. J. Atmos. Sci., 62, 286–309, doi:10.1175/JAS-3396.1.
Mahfouf, J.-F., Brasnett B. , and Gagnon S. , 2007: A Canadian Precipitation Analysis (CaPA) project: Description and preliminary results. Atmos.–Ocean, 45, 1–17, doi:10.3137/ao.v450101.
Mailhot, J., and Coauthors, 2006: The 15-km version of the Canadian regional forecast system. Atmos.–Ocean, 44, 133–149, doi:10.3137/ao.440202.
Mailhot, J., Milbrandt J. A. , Giguère A. , McTaggart-Cowan R. , Erfani A. , Denis B. , Glazer A. , and Vallée M. , 2012: An experimental 1-km resolution forecast model during the Vancouver 2010 Winter Olympic and Paralympic Games. Pure Appl. Geophys., 171, 209–229, doi:10.1007/s00024-012-0520-6.
Masson, V., 2000: A physically-based scheme for the urban energy budget in atmospheric models. Bound.-Layer Meteor., 94, 357–397, doi:10.1023/A:1002463829265.
McTaggart-Cowan, R., Girard C. , Plante A. , and Desgagné M. , 2011: The utility of upper-boundary nesting in NWP. Mon. Wea. Rev., 139, 2117–2144, doi:10.1175/2010MWR3633.1.
Milbrandt, J. A., and Yau M. K. , 2005: A multimoment bulk microphysics parameterization scheme. Part II: A proposed three-moment closure and scheme description. J. Atmos. Sci., 62, 3065–3081, doi:10.1175/JAS3535.1.
Milbrandt, J. A., and Yau M. K. , 2006: A multimoment bulk microphysics parameterization. Part IV: Sensitivity experiments. J. Atmos. Sci., 63, 3137–3159, doi:10.1175/JAS3817.1.
Milbrandt, J. A., Yau M. K. , Mailhot J. , and Bélair S. , 2008: Simulation of an orographic precipitation event during IMPROVE-2. Part I: Evaluation of the triple-moment control run. Mon. Wea. Rev., 136, 3873–3893, doi:10.1175/2008MWR2197.1.
Milbrandt, J. A., Glazer A. , and Jacob D. , 2012: Predicting the snow-to-liquid ratio of surface precipitation using a bulk microphysics scheme. Mon. Wea. Rev., 140, 2461–2476, doi:10.1175/MWR-D-11-00286.1.
Morrison, H., and Milbrandt J. A. , 2015: Parameterization of ice microphysics based on the prediction of bulk particle properties. Part I: Scheme description and idealized tests. J. Atmos. Sci., 72, 287–311, doi:10.1175/JAS-D-14-0065.1.
Morrison, H., Milbrandt J. A. , Bryan G. , Ikeda K. , Tessendorf S. A. , and Thompson G. , 2015: A new approach for parameterizing microphysics based on prediction of multiple ice particle properties. Part II: Case study comparison with observations and other schemes. J. Atmos. Sci., 72, 311–339, doi:10.1175/JAS-D-14-0066.1.
NOAA Center for Weather and Climate Prediction, 2015: Workshop on Parameterization of Moist Processes for Next-Generation Weather Prediction. NCWCP Workshop Summary Rep., College Park, MD, 13 pp. [Available online at http://www.dtcenter.org/events/workshops15/mm_phys_15/MoistProcessesWorkshopSummary_27-29Jan2015.pdf.]
Noilhan, J., and Planton S. , 1989: A simple parameterization of land surface processes for meteorological models. Mon. Wea. Rev., 117, 536–549, doi:10.1175/1520-0493(1989)117<0536:ASPOLS>2.0.CO;2.
Pinto, J. O., Grim J. A. , and Steinter M. , 2015: Assessment of the High-Resolution Rapid Refresh model’s ability to predict mesoscale convective systems using object-based evaluation. Wea. Forecasting, 30, 892–913, doi:10.1175/WAF-D-14-00118.1.
Pruppacher, H. R., and Klett J. D. , 1997: Microphysics of Clouds and Precipitation. 2nd ed. Kluwer Academic, 954 pp.
Qaddouri, A., 2011: Nonlinear shallow-water equations on the Yin-Yang grid. Quart. J. Roy. Meteor. Soc., 137, 810–818, doi:10.1002/qj.792.
Reichle, R. H., Walker J. P. , Koster R. D. , and Houser P. R. , 2002: Extended versus ensemble Kalman filtering for land data assimilation. J. Hydrometeor., 3, 728–740, doi:10.1175/1525-7541(2002)003<0728:EVEKFF>2.0.CO;2.
Reisner, J., Rasmussen R. M. , and Bruintjes R. , 1998: Explicit forecasting of supercooled liquid water in winter storms using the MM5 model. Quart. J. Roy. Meteor. Soc., 124, 1071–1107, doi:10.1002/qj.49712454804.
Rotach, M. W., and Coauthors, 2009: MAP D-PHASE: Real-time demonstration of weather forecast quality in the Alpine region. Bull. Amer. Meteor. Soc., 90, 1321–1336, doi:10.1175/2009BAMS2776.1.
Seity, Y., Brousseau P. , Malardel S. , Hello G. , Bénard P. , Bouttier F. , Lac C. , and Masson V. , 2010: The AROME-France convective-scale operational model. Mon. Wea. Rev., 139, 876–913, doi:10.1175/2010MWR3425.1.
Sills, D., Taylor P. , King P. , Hocking W. , and Nichols I. , 2002: ELBOW 2001—Studying the relationship between lake breezes and severe weather: Project overview and preliminary results. Preprints, 21st Conf. on Severe Local Storms, San Antonio, TX, Amer. Meteor. Soc., P12.7. [Available online at http://www.yorku.ca/pat/research/dsills/papers/SLS21/sls21_p127.pdf.]
Skamarock, W. C., and Klemp J. B. , 2008: A time-split nonhydrostatic atmospheric model for weather and forecasting applications. J. Comput. Phys., 227, 3465–3485, doi:10.1016/j.jcp.2007.01.037.
Smith, G. C., Roy F. , and Brasnett B. , 2012: Evaluation of an operational ice–ocean analysis and forecasting system for the Gulf of St. Lawrence. Quart. J. Roy. Meteor. Soc., 139, 419–433, doi:10.1002/qj.1982.
Smith, S. B., and Yau M. K. , 1993: The causes of severe convective outbreaks in Alberta. Part II: Conceptual model and statistical analysis. Mon. Wea. Rev., 121, 1126–1133, doi:10.1175/1520-0493(1993)121<1126:TCOSCO>2.0.CO;2.
Taylor, N. M., and Coauthors, 2011: The Understanding Severe Thunderstorms and Alberta Boundary Layers Experiment (UNSTABLE) 2008. Bull. Amer. Meteor. Soc., 92, 739–763, doi:10.1175/2011BAMS2994.1.
Thompson, G., and Eidhammer T. , 2014: A study of aerosol impacts on clouds and precipitation development in a large winter cyclone. J. Atmos. Sci., 71, 3636–3658, doi:10.1175/JAS-D-13-0305.1.
Zhang, D.-L., Hsie E.-Y. , and Moncrieff M. W. , 1998: A comparison of explicit and implicit predictions of convective and stratiform precipitating weather systems with a meso-β-scale numerical model. Quart. J. Roy. Meteor. Soc., 114, 31–60, doi:10.1002/qj.49711447903.
For a uniform horizontal grid spacing Δx with a reduction factor fΔx = Δx1/Δx2 and a corresponding model time step reduction factor fΔt = fΔx (for a linear reduction in time step with grid spacing), and with no change in the number of vertical levels, the increase in number of gridpoint calculations per model integration is given by (fΔx)3.
Note, while the model reflectivity shown for the HRDPS is the column-maximum (which was available at the time for the benchmark cases), except for lower values (e.g., <20 dBZ) the general patterns and magnitudes tend to be similar to 1-km model reflectivities for the HRDPS. Thus, qualitative comparison to the 1-km NEXRAD radar observations is reasonable.
