Mesoscale Convective Systems in Northeastern North America: identification and evaluation with the convection-permitting version of the Canadian Regional Climate Model

Abstract

Extreme precipitation events are often associated with mesoscale meteorological phenomena, such as mesoscale convective systems (MCS). Convection-permitting models (CPMs), which operate at high spatial resolution, have enhanced our ability to represent atmospheric processes associated with mesoscale phenomena. This study aims to identify and evaluate MCSs and associated precipitation events in northeastern North America over the 2015–2022 period using various observational and model-based products. A tracking algorithm is used to identify and characterize MCSs in the ERA5 reanalysis, satellite-based (IMERG) data, radar observations (STAGE-IV and MRMS), and two simulations performed with the sixth version of the Canadian Regional Climate Model (CRCM6). These simulations are performed at horizontal grid spacings of 12 km and 2.5 km (CRCM6-12 and CRCM6-2.5, respectively), with the higher-resolution (2.5 km) simulation operating in CPM mode. Radar observations indicate that MCSs occur most frequently from May to September and typically initiate in the early afternoon. The lower horizontal resolution products (IMERG, CRCM6-12, and ERA5) underestimate both the mean occurrence and interannual variability of MCSs compared to the reference dataset STAGE IV-MERGIR, with mean biases of − 17%, − 50%, and − 88% and standard deviations of 23, 17, and 5.8 MCSs per year, respectively (reference standard deviation = 32 systems per year). The CRCM6-2.5 CPM model configuration accurately reproduces key MCS characteristics, including their intra-annual occurrence, size, duration, intensity, diurnal cycle, and their contribution to total and extreme precipitation. Notably, MCSs contribute more to extreme precipitation events than to the total precipitation. The CRCM6-2.5 model significantly improves the representation of convective processes at finer scales compared to lower-resolution products, although it slightly overestimates precipitation in comparison to radar observations.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00382-026-08102-6.

Introduction

Mesoscale Convective Systems (MCS) are generally defined as large clusters of cumulonimbus clouds that produce a continuous area of precipitation extending at least 100 kilometers in one direction (Houze 2004). They are considered the largest of the convective storms and can take on a variety of structures. The most common examples are squall lines, bow echoes, and mesoscale convective complexes. According to Markowski and Richardson (2010), the formation and development of MCSs depend on key atmospheric ingredients: sufficient moisture, convective instability (often quantified using the Convective Available Potential Energy (CAPE)), and a lifting mechanism such as synoptic fronts, outflow boundaries, or orographic effects. However, for convective storms to develop into larger, organized systems such as MCSs, an additional ingredient is required: vertical wind shear (Markowski and Richardson 2010). The authors emphasize that vertical shear, particularly in the 0–6 km layer, plays a crucial role in organizing and maintaining these systems. They also note that low-level shear enhances uplift along gust fronts, promoting the formation of new cells and influencing storm organization, particularly in multicellular convection. These systems are notorious for causing extreme weather events in most mid-latitude and tropical regions, often bringing heavy rainfall, strong winds, and landslides (Houze 2004; Schumacher and Rasmussen 2020). While MCS precipitation is essential to the hydrological cycle, its accumulation over short periods can lead to adverse impacts, including flooding, disruption of hydropower and wind energy operations, and significant risks to communities and human life.

In many regions, MCSs contribute significantly to total and extreme precipitation, particularly during summer. They account for 30-70% of the total warm-season precipitation in parts of the central and eastern United States (U.S.) (Feng et al. 2018, 2019; Prein et al. 2020), approximately for 70% of the summer precipitation in the semi-arid Sahel (Goyens et al. 2012; Laurent et al. 1998), and for 20-60% of total precipitation over Meiyu Frontal Zone in Central and Southern China (Cui et al. 2020). More recently, Silva et al. (2023) found that MCSs contribute more than a third of total precipitation in different regions of Europe and can be responsible for more than 60% of extreme precipitation (events exceeding the 98 percentile) in some areas. Using hourly variables from observation datasets such as the Integrated Multi-satellite Retrievals for Global Precipitation Measurement (IMERG) and the gridded merged multi-satellite geostationary infrared (MERGIR) over the period 2000-2020, Prein et al. (2023) analyzed the contribution of MCS to the total precipitation across different regions of the globe. They found that they are more abundant in the Inter-Tropical Convergence Zone and in some mid-latitudes regions such as central eastern South America and the central and eastern U.S. Their frequency is higher during spring and summer over the eastern and central U.S., where they contribute to more than 60% of the hourly extreme precipitation (99 percentile), in agreement with previous findings (Prein et al. 2017).

Understanding the environmental drivers of MCS - especially in high-impact regions such as the U.S. - is critical given its larger role in precipitation. Feng et al. (2019) analyzed the characteristics and large-scale environments of MCSs over the U.S. MCSs were found to be most frequent during spring and summer, especially over the northern Great Plains region, where the monthly mean number of events typically exceeds 10 during summer. In contrast, they found weak MCS activity over the Appalachian Mountains during summer. Their results indicate that in spring, MCSs typically develop ahead of a mid-level trough associated with baroclinic waves, with strong low-level convergence and upper-level divergence. They also highlight the important role of the Great Plains low-level jet in transporting humidity from the Gulf of Mexico and enhancing instability. During summer, Feng et al. (2019) found that MCSs tend to form under or ahead of a high-pressure ridge, where baroclinic forcing is weak, and both low-level convergence and upper-level divergence are less pronounced. They also suggested that MCSs’ environments are characterized by high low-level humidity, strong diurnal heating, particularly east of the Rocky Mountains, or sea-breeze convergence along the eastern coastal regions. In western Canada, Li et al. (2020) analyzed precipitation systems using multiple precipitation datasets and convection-permitting WRF simulations using a 4-km horizontal grid. Their study found that MCSs are more frequent in the Canadian Prairies during summer and tend to propagate faster than those in the western coastal regions. Subsequently, Hwang and Li (2022) evaluated key environmental parameters associated with MCS characteristics, including vertical wind shear, most unstable convective available potential energy, convective inhibition, storm-relative helicity, relative humidity, and isentropic potential vorticity. They categorized storms by day/night occurrence and short-/long-lived duration to isolate distinct formation regimes. Their study established that the most unstable convective available potential energy (MUCAPE) and the 0-3 km vertical wind shear were the dominant factors influencing MCS duration. Across Europe, Silva et al. (2023) observed peak summer frequency over continental areas, particularly mountainous regions. They found that the summer precipitation diurnal cycle shows a peak during the late afternoon and early evening, mainly driven by convective instability. In contrast, winter MCS activity is more frequently associated with frontal systems.

Several algorithms have been developed to identify and track MCSs, often employing different methods and criteria based on precipitation, brightness temperature, or a combination of both. One of the earliest algorithms, proposed by Maddox (1980), used satellite infrared imagery to identify MCS over the central and eastern U.S., relying on cold cloud tops with a brightness temperature lower than − 32 C and an area greater than . Similarly, Goyens et al. (2012) used satellite infrared imagery to track cold cloud pixels with a brightness temperature lower than − 40 C and a contiguous area larger than over the Sahel region, combining these data with satellite-based precipitation measurements. Feng et al. (2019) combined satellite infrared data (brightness temperature lower than − 32 C, area larger than , persisting at least 6 h) and hourly precipitation information to define and track MCSs. Other works (Prein et al. 2020; Li et al. 2020; Hwang and Li 2022) have used the MODE Time Domain (MTD) algorithm (see (Clark et al. 2014)), which considers only contiguous precipitation regions in time and space by using specific thresholds for the area, intensity, and lifetime.

Unlike isolated storms, MCSs can extend over hundreds of kilometers (Houze 2004). Their formation and development can also be affected by various geographical and environmental factors, such as topography and large-scale atmospheric circulation. Accurately representing the characteristics of MCSs is essential for understanding their impact on larger circulation patterns and their contribution to the total and extreme precipitation events. Several studies have demonstrated the advantages of high-resolution simulations in capturing these complex systems (Prein et al. 2021; Weisman et al. 1997; Lucas-Picher et al. 2021). For example, Weisman et al. (1997) suggested that 4-km, non-hydrostatic simulations could capture the mesoscale organization and net transports of midlatitude MCS, whereas 8–12 km simulations exhibit slower convective evolution and overestimated rainfall and circulation strength compared to higher resolutions. Lucas-Picher et al. (2021) also indicated that horizontal grid spacings between 1 and 4 km are adequate to explicitly simulate most aspects of deep convection.

Global climate models (GCMs) typically have horizontal grid spacing greater than 50 km, while regional climate models (RCMs) between 10 and 50 km (as those used in the recent North American CORDEX Program; see (Giorgi et al. 2008)). However, at these coarse resolutions, fine-scale meteorological processes can still be poorly represented, such as deep convection. Convection-Permitting Models (CPMs) address this limitation by employing RCMs at higher resolution (4 km horizontal grid spacing) with deep convection parameterizations deactivated, enabling the explicit simulation of mesoscale phenomena including MCSs (Prein et al. 2015). CPM configurations have been implemented in several modeling systems, such as the Weather Research and Forecasting developed in U.S. (Skamarock et al. 2019), or with the Met Office Unified Model developed in UK (Clark et al. 2016). While this approach resolves deep convection explicitly, other parameterizations remain necessary, including those for microphysics, shallow convection, or turbulence (Prein et al. 2015; Lucas-Picher et al. 2021). As a result, they provide a more realistic representation of atmospheric processes at finer scales, for instance improving the simulation of the convective precipitation diurnal cycle (Prein et al. 2015; Ban et al. 2014; Gao et al. 2017). For example,  Prein et al. (2021) investigated how varying horizontal resolutions, ranging from 250 m to 12 km, influence the characteristics of MCSs. They found that using a grid spacing of 4 km significantly improved the representation of main MCS features, including maximum precipitation, cold pool intensity, and cloud-top temperatures. However, despite these improvements, some characteristics such as mass fluxes near cloud tops and draft velocities remained overestimated at resolutions between 1 and 4 km in comparison with 250 m simulations. Likewise, Prein et al. (2020) also showed the added value of high-resolution numerical simulations in accurately representing MCS characteristics, such as propagation speed, size, and total precipitation.

While extensive research has been conducted in various mid-latitude and tropical regions to understand MCS characteristics (Houze 2018), the literature review has revealed a lack of studies in regions like northeastern North America. In addition, there has not been any study looking at MCSs as represented using the latest version of the Canadian Regional Climate Model (CRCM6), including a CPM configuration. The main aim of the present study is to evaluate the CRCM6’s ability to simulate MCSs over a northeastern North American region and to assess the added value of higher horizontal grid spacing (2.5 km vs. 12 km). To do so, an identification and tracking algorithm based on hourly brightness temperature and precipitation is used (Prein et al. 2021), including data from multiple observations, reanalysis, and simulations. In addition, we use two model configurations: one in CPM mode at 2.5-km grid spacing and another at 12-km grid spacing, to assess the added value of the higher resolution.

The remainder of this study is structured as follows: Sect. 2 describes the datasets used, including reanalysis, simulations, and observational datasets. Section 3 discusses the methods, including data pre-processing, the MCS identification and tracking algorithm, and the evaluation metrics. Section 4 presents the results, Sect. 5 the discussion, and Sect. 6 the conclusions.

Data

ERA5 reanalysis data

ERA5 is the fifth-generation reanalysis produced by the European Centre for Medium-Range Weather Forecasts (ECMWF, (Hersbach et al. 2020)) which offers significant improvements over earlier generations, such as ERA-Interim reanalysis. It provides hourly meteorological data from 1940 to the present, covering a global domain at a horizontal resolution of about 31 km (0.25 ) with 137 levels extending from the surface to 0.01 hPa. ERA5 includes information for several atmospheric, oceanic, and land surface variables by assimilating a variety of reprocessed datasets, including in situ observations, satellite measurements, and ground-based radar-gauge composites such as the STAGE IV quantitative precipitation estimates (Hersbach et al. 2020). It is based on the Integrated Forecasting System (IFS), which uses 12-hour windows (0900–2100 UTC and 2100–0900 UTC) to collect observations. A short forecast is initiated 9 h into each window (at 18:00 UTC and 06:00 UTC), providing the starting point for the next assimilation cycle  (Hersbach et al. 2020).

In this study, different ERA5 variables are used for multiple purposes. Hourly temperature, geopotential height, horizontal winds, and specific humidity at all 37 pressure levels were used to drive CRCM6-12 simulations (Roberge et al. 2024). Hourly precipitation and top net thermal radiation were used to identify and track MCS over the study area (see Table 1). Specifically, top net thermal radiation, which is equivalent to the negative of outgoing longwave radiation (OLR), was used to estimate brightness temperature (BT) following Yang and Slingo (2001):

1

where is the Stefan–Boltzmann constant.

Table 1.

Input data sources information used in the study (reanalysis, model simulations, satellite, and radar observations) from 2015 to 2022, including native spatial () and temporal resolution (), and variables (precipitation (PR), brightness temperature (BT), and outgoing longwave radiation (OLR))

Data source	Abb.	Δx	Δt	Variables	References
CRCM6-GEM5	CRCM6-12	0.11 ( 12 km)	1-h	Total PR (kg m s) OLR (W m)	Roberge et al. (2024)
CRCM6-GEM5	CRCM6-2.5	0.0225 ( 2.5 km)	1-h	Total PR (kg m s) OLR (W m)	Roberge et al. (2024)
ERA5	ERA5	0.25 ( 31 km)	1-h	Total PR (m) Top net thermal radiation (J m)	Hersbach et al. (2020)
IMERG v07	IMERG	0.1 ( 11 km)	30-min	PR (mm h)	Huffman et al. (2015)
MERGIR	MERGIR	4 km	30-min	BT (K)	Janowiak et al. (2017)
STAGE IV	STAGE IV	4 km	1-h	PR (mm h)	Fulton et al. (1998) Lin and Mitchell (2005)
MRMS radar	MRMS	1 km	1-h	PR (mm h)	Zhang et al. (2016)

It should be noted that we do not expect the ERA5 reanalysis to realistically represent MCSs, as the relatively high precipitation-intensity thresholds used to identify MCSs are unlikely to be met at such coarse spatial resolution (0.25 ). However, given the widespread use of ERA5 as a large-scale reference dataset in the climate community, including ERA5-derived MCSs remains useful for providing contextual background rather than for a direct comparison of mesoscale convective features.

CRCM6-GEM5 simulated data

The newly developed version 6 of the Canadian Regional Climate Model (CRCM6-GEM5; see (Roberge et al. 2024; Llerena et al. 2023)) is used based on two horizontal grid spacings (12 and 2.5 km, see Table 1). The atmospheric component of the CRCM6-GEM5 model is based on version 5.1.1 of the Global Environmental Multiscale (GEM5) model  (McTaggart-Cowan et al. 2019a). In addition, CRCM6-GEM5 uses version 3.6 of the Canadian Land Surface Scheme (CLASS3.6, see (Verseghy 2012)) and the FLake lake model to represent lake temperatures (see (Martynov et al. 2012; Mironov et al. 2010)).

A simulation with a horizontal grid spacing of 12 km, referred to as CRCM6-12, is driven by atmospheric conditions and sea surface temperatures from the ERA5 reanalysis. It is integrated over the CORDEX North American domain  (Giorgi and Gutowski 2015) with a total of 655 x 655 in latitudes and longitudes, covering most of North America and adjacent oceans (see blue dashed line in Fig. 1). Large-scale spectral nudging is implemented for levels above the 0.85 hybrid level and horizontal scales greater than 200 km, with a relaxation time scale of 8 h. Subgrid-scale physical processes in the CRCM6-12 model are represented using the Predicted Particle Properties (P3) scheme for clouds and precipitation (Morrison and Milbrandt 2015; Milbrandt and Morrison 2016) with a subgrid cloud and precipitation scheme to improve its sensitivity to the model resolution (Jouan et al. 2020; Chosson et al. 2014), the Kain-Fritsch scheme for deep convection (McTaggart-Cowan et al. 2019b), the Bechtold scheme for shallow convection (Bechtold et al. 2001), a Monin-Obukhov surface layer scheme (McTaggart-Cowan et al. 2019a) while using a minimum Obukhov length to prevent decoupling by limiting the minimum wind speed at the lowest prognostic level (Foken 2006), and the scheme developed by  Bélair et al. (2005) for the atmospheric boundary layer clouds.

Fig. 1.

Simulation domains with CRCM6 over North America (blue dashed line) with a resolution of 0.11 ( 12 km), and over northeastern North America with a resolution of 2.5 km (red dashed line). The Stage IV dataset (presented in Table 1) is used alongside MRMS to evaluate MCS within the Stage IV domain (dark green area) and for MCS identification

The second configuration, CRCM6-2.5, is run over a domain centered on southern Quebec, with 1132 x 1060 grid points (Fig. 1, red dashed line region). Parameterization schemes being used are similar to CRCM6-12 with the exception of an explicit representation of deep convection (i.e., the Kain-Fritsch deep convection scheme is turned off) and the use of the Kuo Transient scheme for shallow convection (Bélair et al. 2005). CRCM6-2.5 still uses a Monin-Obukhov surface layer scheme, but uses a constant value of 2.5 m s for the minimal wind speed at the lowest prognostic level to prevent decoupling.

The CRCM6-2.5 configuration is driven only at the lateral boundaries using the CRCM6-12 simulation presented above. CRCM6-2.5 is driven using horizontal wind components, temperature, specific humidity, liquid cloud hydrometeors, ice hydrometeors, and rain hydrometeors from the 71 model levels of the CRCM6-12 simulation. The use of an intermediate simulation between the coarse driving data (i.e., ERA5) and the convection-permitting CRCM6-2.5 simulation allows for a higher physical coherence  (Roberge et al. 2024; Scinocca et al. 2016). Yet, it is important to note that, as shown by Roberge et al. (2024), even when driving the model with hydrometeors from the parent simulation, small-scale features close to the borders are partially developed (i.e., spatial spin-up issue). This can lead to a lack of precipitation close to the borders and can impact the identification of MCSs, particularly for those entering from the western or southwestern boundaries. This spin-up effect could lead to the underestimation of MCS activity near these edges of the domain for the CRCM6-2.5 simulation.

As for the CRCM6-12 simulation, we use hourly precipitation and hourly OLR from both simulations. Details of the model configuration at 12 km and 2.5 km resolutions are summarized in Table S1.

Observed data

IMERG-satellite precipitation

The Integrated Multi-satellite Retrievals for Global Precipitation Measurement (IMERG) version 07 is a satellite-based dataset that provides global precipitation estimates at half-hourly timescales and 0.1 of horizontal resolution from 2000 to the present ((Li et al. 2023; Huffman et al. 2015); see Table 1). IMERG combines data from multiple sensors, such as passive microwave radiometers and infrared sensors, to accurately capture precipitation patterns. Its global coverage allows better analysis of precipitation patterns and variability in areas over land and sea or in areas with sparse or no reliable surface observations. Picart et al. (2024) compared multiple precipitation products over northeastern Canada and the U.S. and found that STAGE IV and IMERG showed consistent agreement with the other dataset products included in the analysis, indicating no notable deviations. The IMERG data have been used in several studies to identify and track MCSs (Prein et al. 2023; Silva et al. 2023; Feng et al. 2021a).

MERGIR-satellite brightness temperature

The gridded merged multi-satellite geostationary infrared (MERGIR) product is used as the brightness temperature data source. MERGIR is produced by the Climate Prediction Center (CPC) of the National Oceanic and Atmospheric Administration (NOAA), the National Centers for Environmental Prediction (NCEP), and the National Weather Service (NWS). It provides data merged from the European, Japanese, and U.S. geostationary satellites  (Janowiak et al. 2017). The data is provided every 30 min on a 4-km equivalent latitude/longitude grid spanning the latitude band from 60N to 60S ((Huffman et al. 2023); see Table 1).

STAGE IV radar precipitation

The National Center for Atmospheric Research (NCAR) archives the Stage IV data (see Table 1). It creates mosaics for hourly and 6-hourly covering the Continental U.S. (CONUS), gridded in their original Hydrologic Rainfall Analysis Projection (HRAP). The NWS River Forecast Centers (RFCs) precipitation estimates are combined and bias-adjusted in near real-time with automated gauge measurements (Prat and Nelson 2015). In addition, this product includes a manual quality control step applied to the hourly maps produced by the RFCs (Nelson et al. 2016). The data is provided at hourly intervals on a 4 km polar stereographic grid.

MRMS radar precipitation

The Multi-Radar Multi-Sensor (MRMS) system has been implemented at the NCEP (see Table 1). This radar-based data integrates data from other sources, such as environmental data, satellite, lightning, and rain gauge observations (Zhang et al. 2016). It also includes information from WSR-88D radar sites over the CONUS domain and radars over southern Canada operated by Environment and Climate Change Canada (ECCC). MRMS provides four quantitative precipitation estimation (QPE) products: radar-based, gauge-based, local gauge bias-corrected radar, and gauge-and-climatology-merged QPE, with a spatial resolution of 1 km and 2 min update cycle. The dataset has been continuously updated since late 2014 and, at the time of writing, is available through February 2025. MRMS has demonstrated improvements in precipitation estimation over the CONUS when using radar-based products compared to other radar products (Zhang et al. 2016). In this study, we have selected the QPE radar-only 1-h precipitation accumulation product.

Radar data were considered as the reference dataset for this analysis, as they provide more detailed and reliable information (Moazami and Najafi 2021). Table 1 provides a brief overview of the products and variables used as input for the MCS algorithm (described in Sect. 3), as well as the precipitation fields used to compare observed and simulated data.

Methods

Data preprocessing: regridding and missing values

Our analysis focuses on the region that is common to all datasets and is denoted as the STAGE IV domain, represented by the dark green area in Fig. 1. All data have been interpolated using the conservative method to the CRCM6-12 grid (see its use using different precipitation products in (Diaconescu et al. 2015)). This regridding approach ensures a consistent comparison across datasets that originally represent different spatial scales (Luca et al. 2021). Although extreme values in high-resolution data could be locally smoothed out as indicated by Luca et al. (2021), regional-scale patterns remain unchanged. Specifically, the STAGE IV data were first interpolated from their native 4-km polar stereographic projection to a regular 0.036 lat/lon grid using a nearest-neighbor approach and then interpolated using the conservative approach to the CRCM6-12 grid. When needed, precipitation and brightness temperature variables were converted to mm h and Kelvin, respectively.

Three of the four observational products (MRMS, STAGE IV, and MERGIR) contain missing data over the period (2015–2022) and the region of analysis. Figure S1 in the Supplementary Information (SI) shows that MRMS, STAGE IV, and MERGIR have a total of 3 411 256, 731 854, and 3 139 013 missing values, respectively. As the presence of missing data in observations can affect the identification and tracking of MCSs, we account for missing data in various ways. First, for each dataset, we identify instances where missing data occurred in isolation (i.e., no consecutive missing values are present) in specific grid points. In such cases, a linear interpolation is applied to estimate the missing hourly precipitation or brightness temperature based on the surrounding data points. Figure S1 shows that this approach replaces about 2 085 440 values for the MRMS product and 0 values for the STAGE IV and MERGIR products. Second, for the MRMS and STAGE IV products, missing values in one dataset are replaced using the corresponding hourly values from the other dataset. This leads to 1 262 240 values being replaced by STAGE IV in the MRMS dataset and 668 278 values by MRMS in the STAGE IV.

Where neither interpolation nor replacement was possible, missing values were retained in precipitation and brightness temperature datasets. In this case, to make the comparison as fair as possible, these missing values were systematically propagated to the exact spatio-temporal locations across all datasets by using binary masks. Consequently, MCS identification and tracking were performed using hourly data containing missing values. Figure S1 shows that the final percentage of retained missing data is 0.0066% and 0.33% for precipitation and BT, respectively.

MCS identification and tracking algorithm

In this study, the Python-based MCS identification and tracking algorithm developed by Prein et al. (2021), and modified by Argüeso (2025) is used. Unlike other storm identification methods (e.g., (Clark et al. 2014)), this algorithm incorporates two input variables: precipitation (PR) and BT.

The algorithm begins using threshold values for the combined hourly fields of PR ( 5 mm h; see (Prein et al. 2017, 2020)) and BT ( 241 K; see (Feng et al. 2019, 2016; Prein et al. 2023, 2024)). For each hour, this step creates two binary 2D masks, one for each variable, that contain ones for grid points satisfying the threshold and zeros otherwise. For each variable, 3D objects are then created by connecting masked elements that are adjacent in time. Each of these 3D objects is then assigned a unique identifier. Short-lived or small objects are filtered out by applying a minimum duration ( 4 h; see (Haberlie and Ashley 2019; Prein et al. 2020)) and a minimum area (PR area 1000 km and BT area 5000 km) thresholds. These thresholds for precipitation and BT area were applied to exclude the smallest MCSs that are not resolved at the common spatial grid used, thus filtering very localized convective cells. After identifying 3D objects for both PR and BT variables, a 3D object is defined by the overlap between PR and BT objects. For each PR object, the algorithm checks whether it temporally overlaps with at least one BT object at any grid point. Consequently, precipitation associated with warmer, non-deep clouds is excluded. Overlapped PR-BT 3D objects are classified and labeled as an MCS if they persist for at least four consecutive hours and meet the additional following criteria at least once during their lifetime:

• MCS(PR) area 1000 km

• MCS(BT) area 5000 km

• Min(BT) 225 K (Prein et al. 2024, 2023)

• Max(PR) 15 mm h

In addition, the major axis length (L) of an MCS is calculated based on the PR object, and MCS objects are discarded unless 100 km 1 000 km. This criteria allows for filtering out MCSs that are too small (Houze 2004; Haberlie and Ashley 2019; Feng et al. 2019) or too large (i.e., those approaching synoptic scales).

To evaluate the robustness of the MCS definition, a series of sensitivity experiments were conducted in which some of the thresholds used in the identification and tracking procedure were modified (see Table S2 for details). Specifically, the effects of changing the maximum major axis length, the BT thresholds, and the minimum area of PR objects were tested (see Table S2 for a short description of experiments). The mean monthly distribution of MCS counts (see Fig. S2) showed that experiments with warmer BT thresholds detected more MCS. Notably, the differences between the various configurations were most evident in the high-resolution datasets (e.g. STAGE IV-MERGIR, MRMS-MERGIR and CRCM6-2.5), suggesting that these datasets are more sensitive to variations in the BT thresholds. In contrast, low-resolution products (e.g., IMERG-MERGIR, CRCM6-12, and ERA5) showed smaller variations, suggesting that BT and PR threshold values are less accurate or robust in these datasets. Despite these differences, the monthly pattern remained consistent across all experiments, with the highest MCS frequencies occurring in July. The statistical significance of the monthly distribution differences relative to the STAGE IV-MERGIR reference product was evaluated using the Mann–Whitney test (see Fig. S3). The results indicate that none of the tested modifications caused statistically significant changes to the monthly distribution of MCS frequencies. Seasonal relative mean differences in the fraction of precipitation associated with MCSs were also calculated for various precipitation thresholds (see Fig. S4). Differences were calculated as the percentage difference, where . The largest differences () were observed in experiments that modified BT thresholds or the PR minimum area, particularly during summer and autumn. In contrast, the smallest differences () occurred in experiments where the maximum major axis length was set to 1200 km, followed by those where this constraint was not applied. These percentage differences may be considerably affected by variations in the spatial extent of MCS influence, as changes in their occurrence and coverage can considerably alter their contribution to precipitation across the domain.

For each identified MCS, multiple characteristics are calculated for each hour of its lifetime. The center of mass, which is used as the location of the MCS, represents the central coordinates of the object at any given time step. The translation speed of the object is determined by calculating the Euclidean distance between the location at consecutive time steps and converting these distances from degrees to kilometers. The areal-mean and areal-maximum precipitation rates are also calculated at each time step. The area (or size) of the MCS object is also calculated at each specific hour and expressed in square kilometers.

It is also important to note, as highlighted by Prein et al. (2024), that the frequency and several key characteristics of MCSs depend strongly on the specific algorithm used for their detection and tracking. This methodological dependency should be considered when interpreting our results and when comparing them to other studies employing different tracking approaches. Although this aspect is not explicitly assessed in this work, the sensitivity tests also show that the results are dependent on the thresholds applied in the identification procedure.

Evaluation metrics

The Pearson correlation coefficient (r) and the associated p-value were computed using the SciPy Python library (Virtanen et al. 2020).

The contribution of MCSs to summer precipitation is quantified by calculating two fractions at each grid point over the study area. The calculation is performed globally over all summer hours (June–July–August) across all available years. Defining the hourly precipitation rate as during summer (June-July-August), the fraction of the total summer precipitation larger than 0.1 mm h associated with MCSs is defined as follows:

2

where denotes hourly precipitation values associated with an MCS. The total summer extreme precipitation associated with MCSs is defined as follows:

3

where q(x) denotes the percentile of hourly precipitation, calculated using hourly values larger than 0.1 mm h.

Results

Temporal distribution of the frequency of MCSs

Figure 2 presents the monthly-mean (a) and annual total (b) number of MCSs from 2015 to 2022 over the STAGE IV region (dark green area in Fig. 1). The monthly-mean number of MCSs was calculated by counting each unique MCS identifier for specific months. In cases where the MCS spanned two months, the MCS was associated with the month in which it spent most of its lifetime. Despite biases in absolute counts, all datasets consistently indicate that the majority of MCSs occur between May and September, with the largest number in July (Fig. 2a). Using the STAGE IV-MERGIR and MRMS-MERGIR as reference datasets, we found an average of 263 and 276 MCSs per year, respectively, over the study period (Fig. 2b). STAGE IV-MERGIR shows an interannual variability of 32 systems per year, as indicated by the standard deviation of annual counts (see Table 2). We emphasize that these values are sensitive to the specific MCS detection and tracking thresholds (e.g., (Prein et al. 2024; Feng et al. 2019)), and care must be taken when comparing absolute counts across datasets or studies. The spatial distribution of MCS occurrences during summer is further illustrated in Fig. S5, highlighting regions with the highest MCS initiation activity. The maximum influence occurs over the Midwest, with a secondary peak east of the Appalachians in STAGE IV-MERGIR observations (see Fig. S5).

Fig. 2.

Monthly (a) and annual (b) distribution of total number of MCSs in 2015–2022 over the STAGE IV region (see Fig. 1). a shows the monthly-mean total number of MCSs from 2015–2022, and b shows the total annual counts. The top values indicate the approximate mean annual MCS count averaged over 2015–2022, while bottom values represent the Pearson correlation coefficients (r) and p-value between each dataset and the reference STAGE IV-MERGIR observations

Table 2.

Annual variability of MCS occurrences across different datasets

Dataset	σ	CV	Bias (%)
STAGE IV-MERGIR	32.0	0.12	0.0
MRMS-MERGIR	24.9	0.09	4.9
IMERG-MERGIR	23.2	0.10	− 16.5
CRCM6-12	17.3	0.13	− 49.7
CRCM6−2.5	34.4	0.10	28.3
ERA5	5.8	0.18	− 87.9

Standard deviation (), coefficient of variation (CV), and percentage bias (in %) of annual MCS counts for each dataset relative to the STAGE IV product. The coefficient of variation was calculated as the ratio of the standard deviation () to the mean ():

In contrast, the IMERG-MERGIR, CRCM6-12, and ERA5 datasets underestimate both the mean number of MCSs and their interannual variability, with approximate averages of 219, 132, and 32 per year and mean biases of − 17%, − 50%, and − 88%, respectively. Their standard deviations are 23, 17, and 6 MCSs per year, respectively, compared to the STAGE IV-MERGIR reference dataset (Table 2). This underestimation is also evident in the spatial distribution, with lower frequencies of MCS initiation observed in the regions of primary formation in the STAGE IV-MERGIR (Fig. S5). The CRCM6-2.5 simulation overestimates the annual MCS count, averaging 337 systems per year (28% more than STAGE IV-MERGIR) with a standard deviation of 34 systems per year. In terms of spatial distribution (Fig. S5), CRCM6-2.5 shows a local minimum along the southwestern edge of the domain and a significant peak in MCS initiation to the east of the Appalachians and along the North shore of the Cape Hatteras which is affected by not only MCS activities, but also by intense storm systems that moved or formed over the western (tropical and sub-tropical) North Atlantic along the eastern USA shorelines (i.e., the nor’easter storm events, see (Chen et al. 2025)). These patterns are not evident in the STAGE IV-MERGIR dataset.

The annual distribution of MCSs (Fig. 2b) is highly correlated between the two radar products (r(STAGE IV-MERGIR,MRMS-MERGIR) = 0.93, p-value = 0.0008) and between STAGE IV-MERGIR and IMERG-MERGIR (r = 0.83, p-value = 0.01). In contrast, the correlations between STAGE IV-MERGIR and the CRCM6 simulations are weaker and not statistically significant at the 95% confidence level, with correlation coefficients of 0.50 and 0.45 for the CRCM6-2.5 and the CRCM6-12 simulations, respectively. The interannual variability obtained from the ERA5 reanalysis product shows an even lower correlation, with a value of 0.26, when compared with the STAGE IV-MERGIR product. Most products agree that the highest MCS activity was observed in 2018, except ERA5. It is important to note that these correlations are subject to large errors given the small sample size (only 8 years). Figure S6 in the SI provides a joint monthly and yearly distribution of the number of MCSs for all six datasets considered.

Figure 3 shows the number of MCSs as a function of the initiation hour (the hour when an MCS was first identified), for all datasets, in UTC (a) and local (b) times. It considers all MCS events throughout the year, without differentiating between seasons or months. Local time is determined based on the longitude of the mass center of the system (, (Stensrud 2007)). STAGE IV-MERGIR and MRMS-MERGIR references show a clear diurnal cycle with a peak in MCS initiation between 1700 and 2100 UTC, with relative amplitudes of and 0.13, respectively. IMERG-MERGIR () and CRCM6-2.5 () align well with this pattern, though the latter overestimates the number of MCS during this peak period, particularly at 1700 and 1800 UTC. In contrast, ERA5 exhibits two dominant peaks: one at 0700 UTC and another at 1900 UTC. These peaks are likely influenced by the 12-hour cycle, at 0600 and 1800 UTC, that is used in ERA5 to assimilate observed data (Hersbach et al. 2020; Lavers et al. 2022), which also contributes to its relatively high amplitude values . Additionally, CRCM6-12 shows no clear diurnal cycle in MCS initiation times with . Local time analysis reveals a peak in MCS initiation between 1200 and 1600 h, with strong agreement among STAGE IV-MERGIR, MRMS-MERGIR, and CRCM6-2.5. IMERG-MERGIR, though detecting fewer MCSs, also captures a peak within this period. In contrast, CRCM6-12 exhibits no distinct diurnal cycle (), while ERA5 shows two peaks at 1300 and 1500 local time. In addition, the diurnal cycle of MCS initiation to the west and east of the Appalachians is presented in Fig. S7. The results show that the timing of MCS initiation is broadly consistent in the two regions in the STAGE IV-MERGIR reference dataset, although more MCSs initiate in the late evening and early morning hours in the Midwest region than in the area east of the Appalachian Mountains, where they are more common in the afternoon. Furthermore, while the CRCM6-2.5 model reproduces well the peak at 1300 local time in the east of the domain, it tends to show a peak too early in the west of the domain. However, there is also noticeable differences between MCS initiation diurnal cycle west and east of 85W in the two radar products, with higher peak values of MCS activities in early afternoon in MRMS than in STAGE IV east of the Appalachian Mountains.

Fig. 3.

Histogram of MCS initiation time per hour, considering total MCS events over the whole years during the 2015–2022 period over the STAGE IV domain from different data sources. Bars are shown in 2-hour intervals. a shows initiation times in UTC, and b in local time, computed using the longitude of the mass center of the system (local time , (Stensrud 2007)). This distinction helps capture regional variations in timing relative to local solar time. The metric corresponds to the relative amplitude of each dataset, where C is the hourly MCS count and is the total number of occurrences over the diurnal cycle. Note that all MCS events are considered throughout the year, but the initiation time per hour is strongly affected by peak MCS activity from May to September, as shown in Fig. 2

MCS characteristics

Figure 4 presents histograms and probability density functions (using kernel density estimation) of lifetime-integrated MCS characteristics, including the MCS median size (Fig. 4a), the median areal-mean precipitation rate (Fig. 4b), the lifetime (Fig. 4c), and the track length, calculated from the hourly movements of MCS centroids from initiation to decay (Fig. 4d). The median size distribution (Fig. 4a) shows that STAGE IV-MERGIR and MRMS-MERGIR references capture smaller MCSs (approximately 3500 - 5000 km). CRCM6-2.5 closely follows the radar-based distributions, while the CRCM6-12 detects slightly larger MCSs. In contrast, IMERG-MERGIR and ERA5-MERGIR detect a higher frequency of larger MCSs (exceeding 5000 km and 10000 km, respectively), than radar-based reference datasets. The MCS areal-mean precipitation rate (Fig. 4b) shows the highest frequency between 8–9 mm h in STAGE IV–MERGIR and MRMS–MERGIR. Compared with radar products, IMERG–MERGIR, ERA5–MERGIR, and CRCM6-12 strongly underestimate the occurrence of areal-mean precipitation rates higher than 9 mm h. In contrast, the CRCM6-2.5 simulation reproduces well the distribution observed in the radar observation products, although it overestimates the frequency of high precipitation rates. The MCS lifetime distribution (Fig. 4c) shows that most systems last less than 24 h, with a peak at about 10 h across most datasets. ERA5-MERGIR overestimates the MCS lifetime relative to radar-based reference datasets, with a higher frequency of MCS lasting more than 10 h, peaking between 15 and 25 h. CRCM6-12 also shows slightly longer-lived MCSs with too high frequencies at around 20 h. The total track length (Fig. 4d) indicates that MCSs typically travel about 500 km in radar-based datasets. CRCM6-2.5 is consistent with these values, while CRCM6-12, IMERG–MERGIR, and ERA5-MERGIR tend to overestimate the occurrence of long track distances. Complementary empirical cumulative distribution functions () are provided in Fig. S8 in the SI to further illustrate the ability of each dataset to represent extreme values of MCS characteristics.

Fig. 4.

Histogram and probability density functions of lifetime-integrated MCS characteristics over the STAGE IV domain, including the MCS median size (a), median of the areal-mean precipitation rate (b), lifetime (c), and track length calculated from the hourly movements of MCS centroids from initiation to decay (d). All MCS characteristics are computed using each dataset (STAGE IV-MERGIR, MRMS-MERGIR, IMERG-MERGIR, CRCM6-12, CRCM6-2.5, and ERA5) during 2015–2022. The numbers in panel (a) indicate the total number of samples used for the histograms. Kernel density estimation bandwidth is calculated using Silverman’s rule (Weglarczyk 2018). Light lines represent the histogram, while fully opaque lines correspond to the probability density function

Figure 5 presents histograms and the probability density function of MCS hourly occurrences for their translation speed (Fig. 5a), size (Fig. 5b), areal-maximum precipitation rate (Fig. 5c), and areal-mean precipitation rate (Fig. 5d). Translation speeds (Fig. 5a) are generally consistent across most datasets, peaking at around 40–50 km h. However, the CRCM6-2.5 and CRCM6-12 simulations suggest slightly slower-moving systems, while the IMERG-MERGIR and ERA5 products exhibit faster-moving MCSs, compared to radar products. The hourly size distribution (Fig. 5b) shows a high frequency of small MCSs ( 5000 km) in all products, especially in the radar-based datasets (STAGE IV-MERGIR, MRMS-MERGIR), IMERG-MERGIR and both CRCM6 simulations. Consistent with Fig. 4a, lower resolution datasets (IMERG-MERGIR, CRCM6-12, and ERA5) overestimate the frequency of large MCSs ( 10 000 km) compared to radar-based datasets. In contrast, CRCM6-2.5 is more consistent with radar observations and thus accurately captures the distribution of MCS sizes. The distributions of the areal maximum (Fig. 5c) and mean (Fig. 5d) precipitation rates reveal peaks near 18 mm h and 9 mm h, respectively, in STAGE IV-MERGIR and MRMS-MERGIR. The CRCM6-2.5 simulation represents the distribution of precipitation rates very well, although it shows a greater number of hours with high precipitation intensities (exceeding 30 mm h and 12 mm h for the areal-maximum and areal-mean values, respectively) (Fig. 5c and d). In contrast, the coarser-resolution products such as IMERG–MERGIR, CRCM6-12, and ERA5 strongly underestimate the peak precipitation intensity relative to radar-based observations with peak values of around 11 mm h and 7 mm h for maximum and average rates, respectively. Figure S9 shows Complementary empirical cumulative distribution functions () to further illustrate the ability of each dataset to represent extreme values of MCS characteristics.

Fig. 5.

Histogram and probability density functions of MCS hourly characteristics over the STAGE IV domain, including translation speed (a), size (b), areal maximum precipitation (c), and areal mean precipitation (d). All MCS characteristics are computed using each dataset (STAGE IV-MERGIR, MRMS-MERGIR, IMERG-MERGIR, CRCM6-12, CRCM6-2.5, and ERA5) during 2015–2022. The numbers in panel (a) indicate the total number of samples used for the histograms. Kernel density estimation bandwidth is calculated using Silverman’s rule (Weglarczyk 2018). Light lines represent the histogram, while fully opaque lines correspond to the probability density function

Figure 6 shows the evolution of key MCS characteristics - translation speed, size, maximum areal precipitation, and mean areal precipitation- over the lifetime of MCS lasting less than 30 h. In this analysis, each MCS only contributes to the composite during its actual lifetime, with no temporal interpolation. As a result, the number of samples decreases at later stages, particularly beyond 20 h. This may lead to increased variability in the composite values. However, all MCSs are included in the composites, capturing the full range of observed behaviours. Translation speed (Fig. 6a) remains relatively consistent throughout the MCS lifetime across most datasets, with minimal variation. However, CRCM6-2.5 systematically underestimates translation speed compared to the radar-based reference datasets (MRMS/STAGE IV-MERGIR). IMERG-MERGIR overestimates MCS track speed, particularly during the early stage of MCS development, relative to radar-based reference datasets. In contrast, ERA5 seems to have a more erratic bias or non-homogeneous moving speed of MCSs during their lifetimes, especially beyond 25 h (see Fig. 6a). MCS size increases rapidly in the first few hours following initiation (Fig. 6b), reaching a maximum size between 3 and 10 h after the initiation for STAGE IV-MERGIR and MRMS-MERGIR. As noted previously in Fig. 4, lower-resolution datasets such as ERA5 and IMERG-MERGIR tend to overestimate the size values relative to the radar-based STAGE IV/MRMS-MERGIR observations, while CRCM6-12 and 2.5 simulations agree well with these data. Maximum and mean areal precipitation (Figs. 6c–d) peak concurrently with MCS size, typically between 3 and 5 h after initiation. The radar-based products report the highest maximum precipitation during the early hours. CRCM6-2.5 consistently overestimates maximum and mean hourly precipitation, with peaks exceeding 20 mm h and 10 mm h, respectively (as shown in Figs. 6c, d). Overall, MCSs exhibit peak intensity during their growth phase (3–5 h), characterized by rapid increases in size and precipitation intensity, followed by a gradual decline in activity toward the end of their lifetime across all datasets.

Fig. 6.

Evolution of MCS characteristics for duration less than 30 h, namely: translation speed (a), size (b), areal maximum precipitation (c), and areal mean precipitation (d). The x-axis represents each hour of the MCS lifetime, while the y-axis shows the median values of each MCS characteristic. All MCS characteristics are computed using each respective data (STAGE IV-MERGIR, MRMS-MERGIR, IMERG-MERGIR, CRCM6-12, CRCM6-2.5, and ERA5) during the 2015–2022 period over the STAGE IV domain (see Fig. 1)

Figure 7 shows the diurnal cycle of MCS characteristics, derived by analyzing all MCS at each local time hour. According to STAGE IV-MERGIR and MRMS-MERGIR, MCS sizes reach their maximum during the late afternoon and early morning (Fig. 7a). IMERG-MERGIR and ERA5 display a more pronounced diurnal cycle than radar-based references, while CRCM6-2.5 closely reproduces the observed pattern. In contrast, CRCM6-12 simulates mainly no diurnal variation in MCS size (Fig. 7a). Maximum and mean areal precipitation (Figs. 7b–c) in the reference datasets peak between 15:00 and 19:00 local time. CRCM6-2.5 significantly overestimates both maximum and mean precipitation rates, particularly during the afternoon and early evening hours (14:00–21:00). Meanwhile, CRCM6-12, ERA5, and IMERG–MERGIR show weaker diurnal signals, with ERA5 showing an underestimation in afternoon precipitation, with minimum values between 13:00 and 15:00 when peak activity would be expected.

Fig. 7.

Diurnal cycle of MCS characteristics over the STAGE IV domain: size (a), areal maximum precipitation (b), and areal mean precipitation (c). The plot is generated by calculating the median of all hourly MCS values corresponding to each hour of the day (expressed in local time) computed over the 2015–2022 period from each dataset (STAGE IV-MERGIR, MRMS-MERGIR, IMERG-MERGIR, CRCM6-12, CRCM6-2.5, and ERA5)

MCSs contribution to precipitation

Figure 8 shows the fraction of the summer (JJA) precipitation that can be attributed to MCSs for all six datasets (see Eq. 2). STAGE IV-MERGIR and MRMS-MERGIR indicate that, averaged over the analysis region, MCSs account for 24% of the summer precipitation. Both radar reference datasets agree that some areas southwest of the Great Lakes and on the east coast receive as much as 50% of summer precipitation from MCSs. They also agree that the lowest contribution from MCSs to summer precipitation is found over the Appalachian Mountains and over the northern part of the Great Lakes. According to the IMERG-MERGIR product, the contribution is on average 22%, slightly underestimating the precipitation fraction in most of the domain. CRCM6-12 and ERA5 considerably underestimate the contribution of MCSs to the summer precipitation across the study area, with an average of 15% and 7%, respectively. In contrast, CRCM6-2.5 tends to overestimate MCS-related precipitation (average of 30%), particularly identifying a much larger proportion of MCS summer precipitation over the Appalachian Mountains and the southeastern coastlines compared to STAGE IV-MERGIR and MRMS-MERGIR reference datasets. However, CRCM6-2.5 reproduces similar patterns of the contribution of precipitation from MCSs when compared to radar products, particularly the maxima in regions southwest of the Great Lakes and along the east coast.

Fig. 8.

Fraction of total summer precipitation associated with MCSs over the STAGE IV domain. MCS-associated precipitation is calculated by summing precipitation at each grid point labeled as MCS, while total precipitation is determined by considering only precipitation values 0.1 mm h, as defined by Eq. 2. Red numbers indicate the domain-mean values computed excluding the Appalachian Mountain areas, shown as white contours. Values are computed for each dataset over the 2015–2022 period (STAGE IV-MERGIR, MRMS-MERGIR, IMERG-MERGIR, CRCM6-12, CRCM6-2.5, and ERA5)

Figure 9 shows the fraction of summer hourly precipitation values above the 99 percentile attributed to MCSs, as defined by Eq. 3. As a reference, Fig. S10 in the SI shows the 99 percentile of hourly summer precipitation rates for all datasets during 2015–2022 over the STAGE IV domain. Based on MRMS–MERGIR and STAGE IV–MERGIR reference observations, MCSs account for an average of approximately 60% of extreme summer precipitation over the study area. As revealed in Fig. 8, the highest contributions occur southwest of the Great Lakes and along the east coast, where MCS-related extremes cover larger areas and have higher fractions. In contrast, smaller contributions are observed over the Appalachian Mountains and the northern Great Lakes. IMERG–MERGIR shows an average contribution of 71%, with a general overestimation compared with radar products in certain regions, particularly over the Great Lakes, the Appalachian Mountains, and the east coast. CRCM6-12 shows a mean contribution of 69%, overestimating the values compared to radar products over the Great Lakes and the western part of the domain. CRCM6-2.5 reproduces very well the MCS contribution to extreme precipitation, with a domain-wide mean of 62%. However, while the domain-mean value is very close to the radar-based estimations, the CRCM6-2.5 simulation shows higher fractions in specific regions such as the Appalachians, the east coast, and regions west of the Great Lakes, with patterns that closely align with the spatial distribution of total summer precipitation (Fig. 8). In contrast, there is an area south of Lake Michigan where an underestimation is observed compared to the reference datasets. The higher fraction of MCS-related extreme precipitation in CRCM6-12 compared to CRCM6-2.5 is likely due to CRCM6-12 producing fewer MCSs, which leads to lower total MCS extreme precipitation. CRCM6-12 also has lower total extreme precipitation overall. As a result, the MCSs that do occur contribute a larger share of extreme precipitation, even though their absolute precipitation amounts remain lower relative to CRCM6-2.5. The ERA5 reanalysis considerably underestimates the MCS contribution to extreme precipitation, with a domain-wide mean of only 45%. The highest values in ERA5 are found southwest of the Great Lakes, while little or no contribution is captured along the coast, where extreme MCS-related events are observed in the reference datasets.

Fig. 9.

Fraction of the total MCS extreme precipitation relative to the total extreme precipitation in summer over the STAGE IV domain. Extreme precipitation is defined as those hourly precipitation events above the 99 percentile. MCS extreme precipitation refers to the accumulated precipitation during hours in which MCSs are present and the precipitation exceeds the 99 percentile threshold (Eq. 3). Red numbers indicate the domain-mean values computed excluding the Appalachian Mountain areas, shown as red contours. Values are computed for each dataset over the 2015–2022 period (STAGE IV-MERGIR, MRMS-MERGIR, IMERG-MERGIR, CRCM6-12, CRCM6-2.5, and ERA5)

Figure 10 shows the average proportion of MCS-related precipitation to various extreme precipitation thresholds for each season within the STAGE IV region. These thresholds are calculated using high percentiles (98, 99, 99.5, and 99.9) while only including hourly precipitation events above 0.1 mm h . The right panels of Fig. 10 further illustrate the biases relative to the reference dataset STAGE IV-MERGIR, summarizing the mean absolute error (MAE) values for each data source and season. As previously noted, the contribution of MCSs to total precipitation is relatively weak across all seasons in this region, except in summer, where this can reach 20-30 % (i.e., as suggested in radar and CRCM6-2.5 data). In winter (DJF), MCS contributions are generally the lowest. While IMERG-MERGIR and CRCM6-2.5 tend to overestimate this contribution (MAE values of 0.10 and 0.11, respectively), CRCM6-12 closely aligns (MAE = 0.02) with STAGE IV-MERGIR reference dataset. At the highest threshold (q99.9), MCSs account for up to 43% of extreme precipitation in STAGE IV-MERGIR radar observations dataset. However, it is important to note that the sample size of MCSs is low in winter (i.e. on average, there are less than 10 MCS per winter season) and the uncertainties in MCS contribution to precipitation extremes are then high. In fact, the differences in MCS contribution to precipitation extremes among all data products are higher than in any other season (Fig. 10). During spring (MAM), there is better agreement between STAGE IV-MERGIR, MRMS-MERGIR (MAE=0.01), and CRCM6-12 (MAE=0.04), showing contributions of around 60% at the most extreme percentiles. However, CRCM6-2.5 (MAE = 0.11) and IMERG-MERGIR (MAE=0.11) continue to overestimate MCS contributions relative to STAGE IV-MERGIR, as ERA5 (MAE=0.32) reveals the strongest underestimation of all the products and seasons. The overestimation by CRCM6-2.5 likely reflects the fact that MCSs in spring are more strongly influenced by synoptic-scale systems, which are already captured reasonably well by CRCM6-12. In summer (JJA), MCSs become the dominant drivers or contributors of extreme precipitation. Contributions exceed 60% in radar observations and in most datasets with the highest MCS influences for all percentiles, except ERA5. Here, CRCM6-2.5 aligns well with radar-based datasets (MAE = 0.03), demonstrating the clear advantage of convection-permitting resolution in representing convectively driven extremes during summer months. CRCM6-12 and IMERG-MERGIR show slight overestimations, with MAE values of 0.08 and 0.09, respectively. In autumn (SON), MCSs continue to make a substantial contribution to extreme precipitation, with values similar to those observed in spring. There is a good agreement among the radar datasets and IMERG-MERGIR (MAE=0.02) and CRCM6-2.5 (MAE = 0.02), all of which show contributions between 50% and 60%. In contrast, ERA5 consistently underestimates the contribution of MCSs across all seasons, particularly during the cold and temperate months (DJF, MAM, and SON) with MAE values of 0.21, 0.32 and 0.20 respectively. Overall, the results suggest that MCSs can account for over 60% of the most extreme hourly precipitation events.

Fig. 10.

Seasonal mean fraction of hourly MCSs precipitation relative to the precipitation over the STAGE IV domain, from winter to fall seasons (top left to the bottom right panels). The left and middle columns show the fraction of precipitation exceeding multiple thresholds: the total precipitation (all hourly data great than 0.1 mm h values), and extreme thresholds corresponding to the 98, 99, 99.5, and 99.9 percentiles (denoted as q98, q99, q99.5, and q99.9, respectively). Note that the 0.1 mm h threshold is also used to define precipitating hours when calculating percentiles. MCS precipitation refers to the accumulated precipitation during hours in which MCSs are present and the precipitation exceeds the total or the respective extreme threshold (Eqs. 2 and 3). MAE (right-hand column) was calculated using STAGE IV as the reference dataset, considering all precipitation thresholds for each season. Values are computed for each dataset over the 2015–2022 period (STAGE IV-MERGIR, MRMS-MERGIR, IMERG-MERGIR, CRCM6-12, CRCM6-2.5, and ERA5)

Discussion

This study evaluates the performance of the sixth version of the Canadian Regional Climate Model (CRCM6), including a convection-permitting configuration run at 2.5-km grid spacing, at representing MCSs from 2015 to 2022 over northeastern North America. An MCS tracking algorithm is applied to identify and track MCSs based on hourly precipitation and brightness temperature data. The performance of the CRCM6 model is analyzed by comparing MCS characteristics derived from CRCM6 simulations at 12 km and 2.5 km resolution with observational datasets, including radar (STAGE IV and MRMS), satellite (IMERG and MERGIR), and reanalysis (ERA5).

Over northeastern North America, radar observations (STAGE IV-MERGIR and MRMS-MERGIR) show a seasonal cycle that is qualitatively consistent with earlier findings, with peak MCS activity from May to August (Haberlie and Ashley 2019; Prein et al. 2020). A direct comparison with other studies is not straightforward due to differences in methodological choices, including the tracking algorithm, the specific definition of MCSs, and the datasets used. Recent multi-tracker analysis (Prein et al. 2024) suggests that the absolute values of MCS metrics, such as frequency, size, duration, and precipitation contribution, are sensitive to the tracking methodology. Nevertheless, our findings are consistent with previous studies in terms of certain MCS characteristics (Prein et al. 2020; Feng et al. 2021a; Haberlie and Ashley 2019). Key characteristics such as size, mean precipitation, lifetime, translation speed, and track length align well with the results reported by Prein et al. (2020), who studied MCSs east of the Continental Divide in the U.S. using STAGE IV and the Weather Research and Forecasting (WRF) model. Prein et al. (2020) found that MCSs generally had sizes smaller than 20 000 km, mean precipitation rates between 5 and 10 mm h, lifetimes of 10 h on average, translation speeds typically between 20 and 60 km h, and track lengths of approximately 500 km or less on average. The present results show some similarities with these findings, particularly in terms of mean and maximum precipitation rates, duration, and translation speed. Our findings are also consistent with previous studies in terms of MCS contribution to total precipitation (Feng et al. 2021a; Haberlie and Ashley 2019). Using IMERG and MERGIR data over the 2001-2019 period, Feng et al. (2021a) found that MCSs contribute approximately 20-30% of total precipitation in southeastern Canada and 30–40% in parts of the U.S. Midwest, and over the eastern U.S. coast. Similarly, Haberlie and Ashley (2019) reported that MCSs contribute most significantly during summer in the U.S. Corn Belt region (40-50%) and the Northeast U.S. (20–30%), as defined in their study.

Our results show that the CRCM6-2.5 simulation overestimates summer MCS activity over northeastern North America by about 20%, with the largest overestimation over the Appalachian Mountains (see Figs. 8 and 9). It is unclear to what extent this finding reflects limitations of the model itself or of the radar-based reference datasets (STAGE IV-MERGIR and MRMS-MERGIR). For instance, Prein et al. (2020) noted that STAGE IV could underestimate precipitation systems over mountainous regions, particularly along the southeastern coast of the U.S. and in the Appalachians, a limitation also mentioned by Fairman et al. (2016) and Zhang et al. (2016). In these areas, the WRF model overestimated MCS track density by more than 70% compared to STAGE IV (Prein et al. 2020). Additionally, the spatial spin-up near the lateral boundaries of CRCM6-2.5 implies that MCSs entering from the western or southwestern edges may not fully develop in their early stages (see (Roberge et al. 2024)). This limitation likely contributes to the underestimation of MCS occurrence near the domain edges, as can be seen in Fig. 8 and in Fig. S5. Conversely, the total number of MCSs at the time of initiation in summer is overestimated in CRCM6-2.5 and systematically lower in STAGE IV-MERGIR radar-based reference datasets over the Appalachian Mountains. This is consistent with the 70% overestimation of MCS track density reported in Prein et al. (2020) when using a convection-permitting configuration of the WRF model. Therefore, radar uncertainties in these regions should be considered to prevent apparent model biases from being overinterpreted.

The performance of ERA5 in representing MCS characteristics reveals several limitations, including a substantial underestimation of the number of summer systems, an underestimation of the precipitation intensity, and large discrepancies in the spatial extent, the duration, and the initiation time of MCSs. These limitations are largely expected, as ERA5 is a global reanalysis with relatively coarse horizontal resolution (grid spacing of about 0.25 ) and relies on parameterized convection, thereby precluding the explicit representation of MCSs. The underestimation of MCS frequency by the ERA5 reanalysis has been documented by Chan et al. (2022) across the Indo-Pacific region for the 2001-2019 period. Consistent with our findings, Chan et al. (2022) also found that ERA5 detected a higher frequency of large MCSs with lower areal mean precipitation rates compared to their analysis based on the IMERG-MERGIR product. The lack of MCSs in ERA5 is likely related to its inability to represent high-intensity hourly precipitation during summer, already reported by Chen et al. (2024) over eastern North America, Llerena et al. (2023) in southern Quebec, Gomis-Cebolla et al. (2023) in Spain, and Lavers et al. (2022) globally. In addition, MCS’s initiation times in ERA5 exhibit anomalous and artificial peaks at 0700 and 1900 UTC, likely associated with its 12-hour assimilation cycle. Hersbach et al. (2020) reported artificial discontinuities in wind speed in the boundary layer, which are likely caused by transitions between assimilation periods, leading to systematic jumps. Such behavior may also influence the diurnal cycle of precipitation associated with MCSs, as suggested by the minimum/maximum amount observed in Fig.  3a, b.

Using the CRCM6-2.5 CPM configuration improves the identification of MCSs events compared to the CRCM6-12, particularly identifying the multiannual and seasonal frequency, MCS size, lifetime, and precipitation intensity. This improvement is likely related to the CPM’s higher capacity to capture more detailed precipitation patterns compared to coarse simulations, especially precipitation extremes, as suggested by  Lucas-Picher et al. (2021),  Prein et al. (2021),  Ban et al. (2021), and  Prein et al. (2015). These findings are consistent with those of Ban et al. (2021), who evaluated a multi-model ensemble of 23 simulations at approximately 3 km resolution over the greater Alpine region between 2000 and 2009, comparing them to high-resolution observational datasets and coarser RCMs. Their study showed that high-resolution simulations outperformed coarser-resolution models in reproducing hourly precipitation, the extreme summer precipitation events (defined as the 99.9 of hourly precipitation), and the diurnal cycle of mean and extreme precipitation. CRCM6-2.5 also closely aligns with reference data in capturing the initiation of convective activity associated with MCSs. This is consistent with Feng et al. (2019), who reported that most summer MCSs initiate in the afternoon, suggesting that they may be driven by strong diurnal diabatic heating. Multiple studies further support the ability of CPMs to realistically simulate MCS characteristics, as well as the timing and intensity of precipitation (Prein et al. 2017, 2020; Feng et al. 2021b; Ban et al. 2021; Lucas-Picher et al. 2017; Prein et al. 2015). However, compared to radar observational datasets STAGE IV–MERGIR and MRMS–MERGIR, the CRCM6-2.5 simulation tends to overestimate both the maximum and mean precipitation rates. Similar biases were found by Prein et al. (2020) based on the WRF model across various U.S. regions, including the Midwest, Mid-Atlantic, and Northeast.

The overestimation of MCS frequency and rainfall intensity in the CRCM6-2.5 simulation likely stems from physical rather than methodological factors. Because the identification and tracking of MCSs are applied after conservative interpolation of BT and PR fields onto the CRCM6-12 grid, the observed differences cannot be attributed to the BT–PR matching or the tracking procedure. Instead, they appear to reflect limitations in model physics, particularly the treatment of convection, boundary-layer dynamics, and microphysical processes at convection-permitting scales. Internal sensitivity experiments (Lahaie and Luca 2025) conducted with the CRCM6-2.5 model show that when deep convection is explicitly treated (i.e., without the Kain–Fritsch parameterization), the overestimation of precipitation extremes is more pronounced than when the deep convection Kain–Fritsch scheme is activated. This suggests that the interactions between boundary-layer processes, convective organization, and microphysics strongly influence the bias structure in the 2.5 km simulation, an aspect that will be investigated in more detail in future work.

Since CRCM6-2.5 is a relatively new configuration, few studies have evaluated its performance for extreme precipitation, which makes it challenging to compare results with similar work. This study helps fill that gap by offering a useful reference for future research using CRCM6-2.5, and supporting its continued development and improvement in representing extreme convective events.

While synoptic and mesoscale environmental conditions such as CAPE and vertical wind shear are known to strongly influence MCSs development and organization, a detailed analysis of these factors is beyond the scope of this already extensive study. Future studies will build on these results by investigating the large-scale and mesoscale environments associated with MCSs and examining how these conditions may evolve under climate change over the full CRCM6-2.5 domain (i.e., including the Canadian part of the domain). This is of particularly importance, as prior studies have shown that changes in atmospheric circulation or dynamics can increase precipitation extremes compared to moderate precipitation (see (Pendergrass 2018; Pendergrass and Hartmann 2014)), while the sensitivity of such extremes to warming remains uncertain in convective regimes (O’Gorman 2015).

Conclusions

We evaluated the characteristics of mesoscale convective systems (MCSs) over northeastern North America using multiple observational datasets (radar- and satellite-based), a reanalysis product, and two configurations of the CRCM6-GEM5 regional climate model. In this study, we assumed that the radar-derived MCSs-identified by combining radar precipitation data with satellite-based brightness temperature-provide the most reliable estimates of MCS activity in the region. The main conclusions from this work are as follows:

• In northeastern North America, MCSs are most frequent from May to September and account for a substantial portion of hourly extreme summer precipitation, highlighting their role as important drivers of severe hydrometeorological events in the region.

• The convection-permitting version of the model (CRCM6-2.5) successfully reproduces most key characteristics of MCSs in the region, including their monthly frequency, interannual variability, size, duration, precipitation intensity, typical initiation times, and their contribution to hourly extreme precipitation.

• ERA5 fails to capture most MCSs and their key characteristics, especially those linked to precipitation intensity and the diurnal precipitation cycle. As such, ERA5 is included here as a large-scale reference dataset commonly used in the climate community, providing a general context rather than a direct comparison of mesoscale convective features.

• The 12-km version of the model (CRCM6-12) offers improvements in the representation of MCSs compared to the driving ERA5 reanalysis. However, it still fails to capture several key aspects, including a pronounced underestimation of MCS frequency, and a tendency to produce systems that are too large, too long-lived, and too weak in terms of precipitation intensity.

Given the ability of MCSs to generate intense precipitation events in short periods of time, their adequate representation is crucial for assessing current and future hydrometeorological risks. Although certain limitations remain, such as overestimation of rainfall intensity and some uncertainty over mountainous areas, where the capacity to capture precipitation occurrence and intensity from observational products is limited. Such observational limitations can have a significant impact on the evaluation of high-resolution simulations such as CRCM6-2.5, as they can increase the apparent overestimation of precipitation intensity and make it difficult to accurately attribute model biases.

Assessing the current characteristics of MCS is critical for understanding how these systems behave under present conditions. This understanding will provide a baseline for comparison under future warming trends, helping us anticipate the impacts of climate change on MCS dynamics and the associated risks in order to increase our collective management’s ability to adapt to these changes. Further works are underway using the CRCM6-2.5 CPM model to also identify and evaluate the changes in precipitation extremes from the potential evolvement of MCS characteristics under global warming, using different GCMs available from the Coupled Model Intercomparison Project Phase 6 (CMIP6) available simulations as boundary conditions for the CRCM6-12 model. These runs will then be used to drive the CRCM6−2.5 CPM model to evaluate the changes in MCS characteristics over northeastern North America, including the southern Quebec area.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Milena Alpizar: Conceptualization; investigation; writing—original draft; writing—review and editing; methodology. Alejandro Di Luca: methodology, writing - review and editing; supervision. Philippe Gachon: Funding acquisition; writing—review and editing; methodology; supervision. François Roberge: numerical simulations.

Funding

Alejandro Di Luca was funded by the Natural Sciences and Engineering Research Council of Canada (NSERC) grant (RGPIN–2020–05631). Authors were also supported by the Discovery Grant program and the Alliance program grant of the Natural Sciences and Engineering Research Council of Canada (NSERC; i.e. 2022–05032 and 576492–2022, respectively), obtained by Pr. Philippe Gachon; and by the scholarships 1653 “Programme de bourses de soutien universel au doctorat" from UQAM.

Data availability

The data generated by the MCS tracking algorithm applied to the various datasets will be made publicly available in an open-access repository prior to the final publication of this article. These data will include key MCS characteristics derived from the algorithm, such as location (coordinates), time of occurrence, areal extent, areal-mean and maximum precipitation, total accumulated precipitation, and translation speed for each dataset analyzed in this study.

Declarations

Conflict of interest

The authors have no relevant financial or non-financial interests to disclose.