Mesoscale convective systems (MCSs) are a principal driver of high-impact weather across Europe, frequently leading to heavy precipitation, damaging winds, and large hail that pose substantial societal and economic risks . For instance, on 18 August 2022, a violent, long-lived convective system known as a derecho swept from the Balearic Islands across Corsica, Italy, and Austria, producing measured wind gusts up to 62.2 ms-1 (224 kmh-1) and extreme precipitation values, causing 12 fatalities, and leaving a trail of destruction over 1000 km long .

Given the devastating impacts of such extreme events, understanding how the characteristics of MCSs will evolve in a warming climate is critical. Climate models are our primary tools for projecting these changes, yet accurately simulating the complex spatiotemporal organization of MCSs remains a well-documented grand challenge . Consequently, there is a crucial need for more sophisticated, process-based model evaluation.

The scientific community is moving beyond simple bias maps toward assessing whether models can reproduce the physical processes that lead to high-impact weather, as this process-understanding is fundamental to building trust and generating robust climate information . Regional Climate Models (RCMs), such as those in the Coordinated Regional Downscaling experiment (EURO-CORDEX) ensemble (0.11°, ∼12.5 km horizontal grid spacing), rely on convective parameterizations and exhibit persistent biases, including an underestimation of hourly precipitation extremes and a premature triggering of the diurnal precipitation cycle . The recent advent of convection-permitting model (CPMs) in ensembles like EURO-CORDEX FPS on Convective phenomena at high resolution over Europe and the Mediterranean FPS-CONV; (∼3 km horizontal grid spacing) marks an important advancement by explicitly resolving deep convection, leading to marked improvements in hourly precipitation characteristics . Even CPMs can still produce overly intense and spatially confined convective cells and exhibit biases in storm location, underscoring the continued need for robust, process-oriented observational benchmarks for evaluation .

Recent high-resolution modeling studies investigating MCSs in a future climate have yielded mixed signals regarding their frequency, preventing simple generalizations across regions. While some convection-permitting simulations project widespread increases in MCS frequency over North America , others identify a stagnation or non-significant decrease in specific hotspots, such as the US Great Plains and the Amazon basin . Despite this regional heterogeneity in occurrence, there is a consensus that they might get more intense in a warming climate . However, the transferability of these findings is limited by the strong sensitivity of MCS simulations to region-specific model configurations , underscoring the necessity for a dedicated European evaluation. A prerequisite for such an analysis is the creation of a robust observational benchmark.

Developing a robust benchmark for Europe is difficult because synoptic-scale atmospheric fronts associated with mid-latitude cyclones frequently occur in the region . Past literature attributes European precipitation extremes to different phenomena depending on the timescale. While some studies suggest mesoscale convective systems (MCSs) primarily drive hourly precipitation extremes during the warm season , other research shows cold fronts also trigger a large fraction of these hourly events . Furthermore, when analyzing longer timescales (e.g., 6-hourly), atmospheric fronts directly cause the vast majority of mid-latitude precipitation extremes . As recent global quantifications demonstrate, these different attributions are not contradictory, because extreme precipitation can result from events where fronts and MCSs co-occur , and large-scale frontal ascent frequently forces organized convection .

Although fronts and MCSs coexist in nature, evaluating climate models requires researchers to cleanly separate their physical structures. If an observational reference dataset mixes these two systems, the evaluation process can become fundamentally flawed. Because Europe's precipitation climatology is heavily front-dominated, the environmental signal in any mixed dataset can become overwhelmingly dictated by large-scale baroclinic forcing. Even coarse-resolution RCMs represent these broad baroclinic dynamics reasonably well . However, accurately simulating the severe, short-duration hazards characteristic, such as the severe wind gusts of convective lines usually require convection-permitting scales . Consequently, if researchers generate storm-centered composites or analyze environmental conditions using a mixed dataset, the well-resolved synoptic features dominate the signal, making RCMs and CPMs appear structurally similar. This baroclinic dominance completely masks the intense, localized physics of organized convective storms, which often develop in entirely different environments such as the prefrontal warm sector . However, separating these systems is inherently difficult when their cloud and precipitation footprints become contiguous. Historically, tracking methods identified systems by their large, cold cloud shields, as demonstrated in the European climatology of . While early methods sometimes confused active convection with non-precipitating cirrus anvils , recent algorithms (e.g., TAMS, , and PyFLEXTRKR, ) have significantly improved detection by incorporating precipitation data alongside infrared brightness temperature . Yet, relying on infrared brightness temperature or outgoing longwave radiation (OLR) creates a fundamental barrier for evaluating climate models. Because regional climate model ensembles have historically lacked hourly OLR output, researchers cannot use these advanced trackers for a consistent, pan-European assessment of simulated MCSs. To diagnose climate model physics accurately, we need a tracking framework that structurally isolates organized mesoscale convection using only standard model output variables.

Regional climatologies have successfully characterized MCS archetypes over specific areas, such as Poland , the Mediterranean , and Spain . More recently studied convective Systems over Europe using satellite-based precipitation and lightning data for identification. Their work, which purposefully includes convectively active frontal systems to characterize all large organized precipitation, restricts the ability to diagnose process-level model biases. Moreover, these studies generally rely on variables that are not standard output in most climate model ensembles, preventing a consistent pan-European assessment.

To address these gaps, we introduce the EMMA-Tracker (Evolution-based MCS Model Assessment) and a 27-year (1998–2024) European reference climatology. This dataset provides a physically consistent benchmark of MCSs for evaluating regional and convection-permitting models. The EMMA-Tracker is a python-based algorithm designed to identify and track MCSs using hourly precipitation and atmospheric instability fields. The core novelty of the tracker lies in physics-based filters that identify an MCS as systems that maintain a clear shape and steady movement, remaining structurally distinct from any co-occurring synoptic triggers.

This paper is structured as follows: Section  details the input datasets and the multi-stage EMMA-Tracker algorithm. Section  presents a validation of these filters, demonstrating their effectiveness in separating MCSs from frontal systems and stationary thunderstorms. The resulting 27-year MCS climatology is presented in Sect. , focusing on spatial distribution, system characteristics, and contribution to heavy hourly precipitation. Section  concludes the implications and limitations of the methodology and dataset.

This section demonstrates the spatial effect of each filter and provides a quantitative validation of their role in separating organized mesoscale convection from other precipitation phenomena.

Figure  spatially illustrates the impact of this filtering cascade. The unfiltered warm-season climatology (Fig. a), which includes all tracked systems meeting the initial size, duration and convection criteria, shows high event frequencies in known convective hotspots, such as the Alps and the Balkan region . Fewer systems are tracked over North Atlantic, Great Britain, and the North Sea. This latter region coincides with the main North Atlantic storm track, which is dominated by synoptic-scale frontal precipitation . Figure e, which shows the relative effect of all combined filters, shows that the subsequent filters physically distinguish and remove these synoptic-scale systems.

The track straightness filter has the most significant quantitative impact, removing a large number of systems from both the Alpine region and the North Atlantic storm track (Fig. b). This filter effectively targets two distinct types of non-MCS phenomena. First, over complex orography, the inclusive precipitation threshold can artificially merge nearby, quasi-stationary orographic precipitation cells. These merged objects may meet the area and duration criteria but fail to propagate coherently, resulting in a low straightness index. Second, over the storm track, the precipitation-weighted centroid of large, frontal rainbands can jump erratically as different embedded precipitation cells intensify and wane, also leading to a low straightness index and subsequent removal. We acknowledge that this filter may also remove some MCSs embedded in frontal systems (e.g., squall lines). This is a deliberate methodological choice to ensure the final dataset is not contaminated by synoptic-scale systems, which is essential for our objective of evaluating model physics at the mesoscale.

The mean LI filter, which removes systems existing in, on average, stable environments (mean LI>1.5 K), has its strongest relative impact over the North Atlantic, further removing systems over the storm track region (Fig. c). Finally, the area volatility filter (Fig. d) acts as a clean-up step, removing remaining spurious tracks, particularly over the Norwegian coast and the North Sea. These systems exhibit physically implausible, rapid fluctuations in size, characteristic of the detection algorithm intermittently merging and splitting large, parts of frontal rainbands as well as orographic triggered local convection (Fig. S4 in the Supplement). The combined effect of all three filters (Fig. e) results in a final, refined MCS climatology (Fig. f) that is concentrated over continental Europe and key orographic regions, in strong agreement with the infrared-based climatology by .

To quantitatively validate the effectiveness of the postprocessing filters in separating propagating MCSs from frontally-associated systems, we collocated all identified system timesteps (accepted and rejected) with an independent hourly frontal database of both cold and warm fronts, based on IMERG . A system's timestep is marked as frontal if its precipitation mask (>1 mmh-1) overlaps with any frontal pixel from the database. Figure  presents the results of this collocation analysis over the 1998–2020 period covered by the frontal database.

The spatial distribution of the accepted MCS timesteps (Fig. a) shows the highest counts concentrated over continental Europe, particularly along the Alps, the Dinaric Alps, and parts of France and Germany, consistent with the final climatology (Fig. f). The corresponding frontal ratio for these accepted systems (Fig. b), calculated as the fraction of timesteps coinciding with a front, is generally low across these central European hotspots. Higher frontal ratios for accepted systems are observed over maritime regions like the North Sea and parts of the Mediterranean, as well as areas adjacent to the primary storm track.

The spatial distribution of the total count of rejected timesteps (Fig. c) exhibits prominent frontal contribution over the North Atlantic, the North Sea, and extending across the British Isles, directly coinciding with the typical North Atlantic storm track region. Primary maxima of rejected systems are found over the Alps, the Balkans, and parts of the Mediterranean. Examining the frontal ratio specifically for these rejected systems (Fig. d) reveals overwhelmingly high values over the North Atlantic and North Sea, frequently exceeding 80 % in most areas. High frontal ratios among rejected systems are also evident over the rest of continental Europe.

These contrasting patterns between accepted and rejected systems provide strong validation for the filtering methodology. The filters successfully retain systems primarily over continental Europe where the precipitation associated with fronts is lower in the warm-season (Fig. a and b). A large number of systems (Fig. c), particularly over the North Atlantic storm track, that are confirmed to be overwhelmingly associated with fronts (Fig. d) are effectively removed. This demonstrates the filters ability to isolate the intended population of predominantly non-frontal, propagating convective systems.

A cohesive region of frequently rejected systems extends from the eastern Alps southward along the Croatian coast and into the Balkans (Fig. c), highlights another important regional interaction. This area is known for significant thunderstorm activity, particularly in late summer and autumn . This activity is often linked to the interaction of synoptic systems with the Dinaric Alps and Adriatic regions . The interaction is enhanced by abundant low-level moisture and instability supplied by the warm Adriatic Sea during this period . While these interactions can produce large, long-lasting precipitation systems (as noted by regarding thunderstorm size and duration over the Adriatic) that trigger our initial detection, their formation mechanism is tied to synoptic and orographic forcing, not self-sustaining mesoscale dynamics. Therefore, their removal by our filters, confirmed by their frequent association with fronts in Fig. d, is physically appropriate for isolating distinct MCSs (see Fig. a for an example of a rejected quasi-stationary orographic system).

In this study, we presented the EMMA-Tracker, a novel algorithm specifically designed to identify and track mesoscale convective systems (MCSs) using only variables available in standard climate model output. A key innovation of this method is the application of physics-based filters that utilize the system's full spatiotemporal lifecycle, allowing for a robust separation of propagating MCSs from synoptically-driven or stationary precipitation features. We applied this tracker to IMERG precipitation and ERA5 data to generate a 27-year (1998–2024) warm-season climatology, representing the longest such dataset for Europe to date.

The final climatology shows that MCS activity is predominantly continental during summer, with the highest frequencies concentrated near major mountain ranges, particularly the Alps, Dinaric Alps. In September, a clear transition occurs as activity wanes over the continent and shifts toward maritime and coastal regions, especially the Mediterranean and Adriatic Seas. We could show that MCS are the most dominant source of heavy hourly precipitation (P99.9) over most of continental Europe.

This dataset is not only restricted for model evaluation but itself represents a robust observational record that enables in-depth climatological analysis of organized convection over Europe. Beyond climatological comparisons, the dataset is designed for process-based evaluation of European climate model ensembles, such as EURO-CORDEX and FPS-CONV. By providing the exact time and location of propagating systems, the EMMA-Tracker allows researchers to move beyond simple bias maps and investigate the physical drivers of convection – such as moisture flux, vertical wind shear, and cold pool strength – throughout the MCS lifecycle. This capability is essential for estimating the uncertainties in future climate projections and for understanding how these high-impact systems will evolve in a warming world. Ultimately, improving our ability to model these systems is a prerequisite for effective disaster preparedness and adaptation to future hydrological hazards.