France’s 63-Year Storm Simulation Puts Kilometer-Scale Climate Model to the Test

When a thunderstorm over the Massif Central dumps more than 120 millimeters of rain in a single hour, the difference between a wet street and a deadly flash flood comes down to details that most climate models simply cannot see. A new study published in Hydrology and Earth System Sciences has put one of the most advanced regional climate models in Europe through an unprecedented six-decade stress test over France, and the results reveal both a powerful new tool for understanding extreme rainfall and a sobering reminder of how hard convective storms remain to simulate.

The research, led by Nicolas Decoopman of the University of Grenoble Alpes together with Juliette Blanchet, Antoine Blanc, and Cécile Caillaud, evaluated the convection-permitting regional climate model AROME, developed at the French national meteorological research centre CNRM and operated by Météo-France. Unlike conventional climate models that approximate thunderstorms with statistical formulas, AROME resolves deep convection explicitly at a grid spacing of just 2.5 kilometers. The team assessed a remarkable 63-year simulation spanning 1959 to 2022, driven at its boundaries by the ERA5 global reanalysis, against thousands of rain gauge records from Météo-France stations across metropolitan France.

The scientific stakes are high because extreme precipitation behaves very differently from ordinary rain. The Clausius–Clapeyron relationship dictates that warmer air holds roughly 7 percent more moisture per degree Celsius of warming. Yet while global mean precipitation rises by only about 1 to 3 percent per degree, constrained by radiative and dynamic limits, intense convective events can condense nearly all of that surplus moisture, pushing short-duration extreme rainfall up by 5 to 8 percent per degree. In other words, the fiercest storms scale almost perfectly with the theoretical moisture ceiling, which is why a warming climate loads the dice for flash floods even when average rainfall barely changes.

France offers a compelling natural laboratory. The country has warmed by 1.7 degrees Celsius since the pre-industrial era, and 2 degrees in the Alps, and it has experienced a string of spectacular recent events: intense high-altitude rainfall that flooded the Écrins massif in June 2024, more than 600 millimeters of rain in 48 hours on the western slopes of the Massif Central that October, and violent thunderstorms in May 2025 with locally more than 120 millimeters per hour. Previous studies have shown that daily precipitation extremes in southeastern France have intensified by roughly 20 percent since the early 1990s, with estimates reaching up to 40 percent, and that in the southwestern Alps autumn 20-year return levels climbed by as much as 100 percent between 1958 and 2017.

To evaluate the model, the researchers applied the machinery of extreme value theory, fitting generalized extreme value distributions to annual maxima of daily and hourly precipitation. The 10-year return level, the rainfall amount exceeded on average once per decade, served as the primary metric. Crucially, the team went beyond a static climatology by employing non-stationary GEV models in which the location and scale parameters evolve linearly in time, including breakpoint variants that allow trends to begin in 1985, the period when French extreme-precipitation signals are thought to have emerged. The analysis drew on 1,583 daily stations with at least 50 years of quality-controlled data and 574 hourly stations with at least 25 years, using the hydrological year running from September to August so that autumn Mediterranean episodes are not artificially split across two calendar years.

At the daily scale, the verdict was largely favorable. AROME reproduced the large-scale spatial structure of French precipitation climatology with impressive fidelity: Pearson correlations of 0.95 for the annual frequency of wet days, 0.94 for annual totals, and 0.95 for the daily 10-year return level. The model correctly captured the aridity of the Rhône valley and Mediterranean coast, where fewer than 70 wet days per year are recorded, in contrast to the Alps, Pyrenees, and Massif Central, which receive more than 1,800 millimeters annually across 140 to 160 wet days. Mean biases were remarkably small, just 11 millimeters per year for annual totals and minus 2 millimeters for the daily return level, although structured regional errors appeared, including excesses of more than 300 millimeters per year along the Pyrenean ridge and deficits of up to 400 millimeters stretching from the northern Pre-Alps to the Vosges.

The trend analysis at the daily scale confirmed known patterns and revealed the model’s strengths and weaknesses. Observations show coherent increases of 5 to more than 30 percent in the 10-year return level along the Rhône Valley and in the southern Alps, a diagonal band of decrease from the Mediterranean to the Atlantic coast in winter, widespread positive spring trends locally exceeding 35 percent, and predominantly negative summer trends across the southern half of the country. AROME reproduced the broad spatial organization of these patterns, performing best in autumn with a correlation of 0.40, but it systematically underestimated trend amplitudes. Part of that dampening has a clear explanation: the model’s simulated warming over France is about one-third weaker than observed, which weakens the Clausius–Clapeyron-driven component of extreme rainfall trends, likely due to the model’s internal sensitivity to radiative forcing and its static land-use map.

The hourly results tell a more troubling story, and they constitute the first nationwide mapping of hourly extreme-precipitation trends in France. Station data over 1990 to 2022 show markedly noisier and more extreme signals than at the daily scale, with significant trends ranging from minus 50 percent to beyond plus 100 percent, concentrated in convective windows: monthly median trends reached 101.86 percent in June, 53.88 percent in October, and 43.28 percent in February. AROME captured the positive sign of these monthly peaks but drastically underestimated their magnitude, simulating a median June trend of just 17.70 percent, and its spatial correlation with observed hourly trends collapsed to nearly zero during convective months, 0.08 in June and 0.04 in March. The model also underestimated hourly 10-year return levels by 23.30 percent on average, with widespread deficits of 5 to 20 millimeters, and the authors note that gauge undercatch in windy conditions means the true underestimation is probably even larger.

Several factors conspire to make hourly extremes so difficult. A single convective core may occupy only part of a 2.5-kilometer grid cell, so a station can record a downpour far more intense than the grid-box average even when the storm is realistically simulated. The sparse hourly network, which is notably thin in the Alps and Pyrenees, further limits evaluation, and the short 33-year hourly record means the climate signal may still be emerging from year-to-year noise. Earlier work suggests that running the model at 1.3 kilometers improves the representation of peak intensities, and the authors found that correlations rise by about 30 percent for 6- and 9-hour accumulations, hinting that AROME may spread convective rain too widely in space and time, diluting local extremes.

The study’s conclusions carry real weight for how France prepares for a wetter, stormier future. The 63-year AROME simulation, one of the longest convection-permitting hindcasts ever produced, is demonstrably valuable for studying daily extremes and can support national-scale climate analyses, provided trend magnitudes are interpreted cautiously. But for the short-duration convective deluges that trigger flash floods, the model’s hourly trends are, in the authors’ words, almost nonexistent, and future evaluations should turn to high-resolution radar-gauge reanalyses such as Météo-France’s 1-kilometer COMEPHORE product. As kilometer-scale climate models edge toward operational use in projections, this work draws a clear line: they excel at the storms we can watch unfold over days, but the hour-long cloudbursts that matter most for human safety still test the limits of what even the finest simulation grids can resolve.

Subject of Research: Evaluation of a convection-permitting regional climate model's simulation of extreme precipitation trends in France

Article Title: Climatology and trends of extreme precipitation in France: evaluation of an explicit-convection regional climate model

Article References: Decoopman, N., Blanchet, J., Blanc, A., & Caillaud, C. (2026). Climatology and trends of extreme precipitation in France: evaluation of an explicit-convection regional climate model. Hydrology and Earth System Sciences, 30(18), 5833-5855. https://doi.org/10.5194/hess-30-5833-2026

Image Credits: AI Generated

DOI: 10.5194/hess-30-5833-2026

Keywords: extreme precipitation, convection-permitting model, AROME, France, flash floods, extreme value theory, climate change, Clausius–Clapeyron, hourly rainfall, regional climate modeling, ERA5 reanalysis, Météo-France