How climate change intensified storm Boris’ extreme rainfall, revealed by near-real-time storylines

The paper addresses how much of the observed conditions in extreme weather events can be attributed to human-induced climate change and how such events may unfold in a warmer future. In mid-September 2024, storm Boris produced unprecedented rainfall across Central and Eastern Europe, causing casualties and extensive damage. Traditional rapid attribution relies on probabilistic methods using observations and large ensembles to quantify changes in likelihood or intensity, as exemplified by the World Weather Attribution group and approaches using circulation analogs. However, record-shattering events challenge analog-based methods due to their exceptional nature, and different climate models can yield divergent answers. Moreover, probabilistic concepts can be difficult to communicate to nonspecialists. Event-based storylines have emerged as a complementary approach. By nudging a climate model with the observed large-scale free-tropospheric winds to reproduce an actual event, and then imposing the same circulation in different background climates (preindustrial, present-day, future warmed), the method isolates thermodynamic effects of warming on daily weather while setting aside circulation frequency changes. The goal is to provide high signal-to-noise, relatable what-if scenarios in near-real time for both scientific and public communication.

The study situates itself within rapid event attribution frameworks that use probabilistic analyses based on observations and large-ensemble modeling to estimate changes in likelihood or intensity of extremes (e.g., World Weather Attribution, national weather services). Circulation-analog approaches have also been used for rapid contextualization. Challenges include the scarcity of suitable analogs for unprecedented extremes and inter-model spread, complicating clear communication of probabilities. Storyline approaches have been developed to complement probabilistic methods, representing uncertainty in a physically grounded way by conditioning on circulation patterns while varying thermodynamic backgrounds. Prior applications include heatwaves on land and in the ocean, and translating historical extremes into future scenarios. This work extends storyline methods toward an automated, near-real-time global system.

The authors employ the Alfred Wegener Institute’s coupled climate model AWI-CM-1-1-MR, which participated in CMIP and is documented in the IPCC AR6. Storyline simulations are generated by nudging the model’s large-scale free-tropospheric circulation to the observed evolution of horizontal winds at synoptic and planetary scales (horizontal scales larger than about 1000 km). Nudging is applied between 700 hPa and 100 hPa, allowing the atmospheric boundary layer to evolve freely; thus some dynamic adjustments remain possible. Three background climates are used: preindustrial, present-day, and a future scenario with global mean warming of about 4 °C relative to preindustrial (approximately 2.7 °C warmer than today). For each climate, a 5-member ensemble is run, and analyses are based on ensemble means. The system is automated and updated daily with a 3-day latency. An associated web tool (https://climate-storylines.awi.de) disseminates results from 1 January 2024 onward and allows users to select parameters (e.g., daily mean, minimum, and maximum 2-m air temperature, sea surface temperature, total precipitation), background climate, region, and visualization type. Calculations: anomalies for SST are relative to 1951–2014; ERA5 anomalies use ERA5 climatology. Present-day climatology is from the 5-member ensemble mean of free-running AWI-CM-1-1-MR historical simulations used to branch the storylines. Changes in precipitation, air and sea surface temperature, and vertical wind velocity are computed as differences between ensemble-mean storylines across climates; relative changes are given with respect to present-day values. Precipitation changes are additionally scaled by local 2-m air temperature changes to compare with Clausius–Clapeyron expectations. A broad event region (11–25°E, 46–54°N) is defined to quantify large-scale changes in event intensity. Areas with total precipitation exceeding 100 mm over the event are summed to quantify severe rainfall extent in each storyline; relative area changes are referenced to present-day simulations. Code and data: the ocean component FESOM v1.4 source code is available via Zenodo; ECHAM6 requires registration on the MPI-ESM user page. Free runs are available on ESGF; nudged storyline simulations are archived on Zenodo. ERA5 reanalysis data are from ECMWF.

• The near-real-time storyline system reproduces storm Boris’ spatial extent and exceptional rainfall. Simulated peak accumulated precipitation over 12–16 September 2024 reaches near 200 mm, close to the observed approximately 225 mm.
• Attribution: In present-day conditions relative to a preindustrial background, storm Boris deposited about 9% more rain due to human-induced warming. The area experiencing extreme rainfall above 100 mm was 18% larger than it would have been in a preindustrial climate.
• Projection: In a globally 4 °C warmer world (about 2.7 °C warmer than today), an analog event would likely shift slightly to the northeast, with a 14% larger area exceeding 100 mm compared to present-day. The overall increase relative to preindustrial is modest (about +2%) and not statistically significant for some metrics.
• Mechanisms: Increased water vapor in source regions under warming contributes to intensification; changes in vertical velocity and increased convective available potential energy, especially over the Eastern Mediterranean and Black Sea moisture source regions, are implicated in the enhanced upstream rainfall deposition. A dipole pattern in vertical ascent changes mirrors precipitation changes, reflecting interplay between thermodynamic and dynamic responses.

The results directly address the attribution question by isolating the thermodynamic contribution of anthropogenic warming to storm Boris’ rainfall and indicate how a similar circulation pattern could manifest in a warmer future. The storyline approach offers high signal-to-noise, relatable scenarios that can be communicated during or shortly after impactful events, complementing probabilistic attribution rather than replacing it. While nudging constrains large-scale winds, some dynamic adjustments occur, and thermodynamic changes can influence dynamical characteristics of low-pressure systems. The approach does not assess changes in the frequency of circulation patterns but provides physically consistent outcomes for a given circulation. The system’s accessibility via a web tool supports transparent, near-real-time dissemination to scientific and non-scientific audiences, potentially enhancing public understanding and decision-making.

An automated, near-real-time global storyline system can rapidly provide robust, evidence-based information on the role of climate change in ongoing extreme events. For storm Boris, human-induced warming increased total rainfall and expanded the area of extreme precipitation, with further expansion possible under stronger warming. The approach is a valuable addition to the climate information toolkit and communication portfolio. Future work aims to expand to multi-model storylines, increase resolution, and integrate impact models (agricultural, hydrological, urban), leveraging improved data access and initiatives like Destination Earth to broaden applications, including AI-driven analyses.

• Results are derived from a single climate model; some findings may be model-specific.
• The atmospheric resolution is coarse (about 100 km), limiting representation of orographic effects and fine-scale precipitation structures.
• The method constrains only large-scale winds above 700 hPa; dynamic adjustments are still possible, and disentangling dynamics versus thermodynamics can be region-specific.
• The approach is less effective in the tropics where convection dominates over large-scale advection.
• Storylines do not evaluate changes in the frequency of circulation patterns and thus cannot replace probabilistic attribution; findings may not be generalizable across regions or event types.