Introduction

On 3-4 January 2024, Denmark experienced a severe weather event marked by a storm system that brought heavy snowfall, rainfall, strong winds, coastal storm surge, and high sea waves in much of the country1. The northern Jutland region saw up to 30 cm of snow in blizzard conditions2. The Danish Meteorological Institute (DMI) issued multiple warnings, including high-risk warnings for blizzards and heavy snowfall in northern Jutland and heavy rainfall warnings for southeastern Denmark3. This storm event is an example of a compound flood event in which multiple climatic drivers interact, causing significant socio-economic impacts4,5. The issue of compound events is not unique to Denmark, but is increasingly recognised as a global challenge6,7,8. For example, Hurricane Florence in 2018 revealed significant compound effects when storm surge and overland runoff converged9, while Hurricane Harvey in 2017 demonstrated how urbanisation can intensify the impacts of extreme precipitation and flood risks10 in multiple regions in the United States. This global perspective on compound events reinforces the importance of studying these phenomena to underscore the importance of integrating compound event forecasting into the value chain of weather services to better prepare for and mitigate the impacts of extreme weather events.

The value chain of the weather service can be conceptualised as a pathway linking observations to weather forecasts, hazard forecasts, socio-economic impact forecasts. Finally, this chain leads to warning messages that inform protective decision-making, with communication and interpretation bridges that facilitate the flow of information between domains of expertise11,12,13. Weather services primarily focus on short-term atmospheric conditions, providing real-time data, forecasts (ranging from hours to weeks), and warnings for severe hazards such as storms, floods, or heatwaves. In contrast, climate services deal with longer time scales, providing information on the degree of change, absolute or relative14,15. Forecasts provide probabilistic information about future atmospheric conditions, while warnings are issued when specific thresholds of hazardous conditions are expected to be exceeded, necessitating immediate action16. Other services include, e.g., datasets for more experienced end-users to utilise15, dedicated advice or consulting services17, etc. Despite advances in data integration and special forecasting services across the weather and climate domains18,19,20,21, it is important to recognise that the communication “valley” within the value chain of the weather service hampers the ability to take timely and protective actions. This communication “valley” refers to the gap between scientific forecasts and public understanding, where detailed meteorological information cannot be effectively communicated to decision-makers or the public22. The challenge of conveying compound events, due to their multivariate nature, further emphasises the need for improved communication strategies that bridge this “valley”, enabling more actionable, context-relevant warnings and responses23,24.

Effective communication has become an essential topic in flood early warning systems (FEWS), especially in the face of climate change, which intensifies the frequency and severity of extreme events25,26,27,28. The study by Wachinger et al.29 highlighted that personal factors, such as previous flood experience and perceived vulnerability, significantly influence how individuals interpret and respond to flood warnings. Furthermore, flood warnings are most effective when they incorporate clear, actionable advice and use accessible formats that avoid technical jargon30. As demonstrated by Kreibich et al.31, warnings substantially reduce flood damage only when people are provided with clear guidance on appropriate actions to take, underscoring the importance of actionable messaging in FEWS. The integration of action-oriented icons and clear, time-sensitive texts within multi-hazard warning messages can significantly boost public understanding and encourage timely protective behaviours, especially in extreme weather situations32. The devastating flooding event in Valencia, Spain, in 2024 serves as a stark reminder of the importance of such measures. With a death toll of 205, this tragedy highlights how inadequate communication and pre-conditions, such as the two-year drought that hardened the soil, exacerbated the impact of flash floods33. In addition, research emphasises that community-tailored messages that reflect local hazards and vulnerabilities can enhance the relevance and impact of warnings, helping residents better understand the immediate actions required34. As extreme weather events become more frequent and complex, refining warning messages to increase awareness and reduce ambiguity is critical in promoting proactive responses across diverse populations35.

The concept of compound events, and this specific terminology, has garnered increasing attention in the scientific literature during the last ten years5,6,8,36,37,38,39,40, although multiple drivers leading to extreme and coinciding events and conditions have seen attention much further back in time in, e.g., multivariate approaches such as in Loganathan et al.41. In the study of compound events, researchers, among others, attempt to define and classify these events based on various criteria in the need for shared characteristics and typologies for processes of a very complex nature40. Different articles have proposed definitions and typologies for compound events, reflecting the diverse range of phenomena encompassed by this concept. In this study, we adopt the framework proposed by Zscheischler et al.39, which is widely considered a seminal reference to define and categorise different types of compound events. These are: (a) preconditioned events: events where the impacts of a hazard are amplified by antecedent conditions. For example, heavy rainfall on already saturated soil may exacerbate flooding; (b) multivariate events: events that involve the simultaneous occurrence of multiple hazards, such as a coastal storm surge coinciding with heavy rainfall, leading to enhanced damage or risk; (c) temporally compounding events: events where hazards occur in close succession, with one event increasing the vulnerability or exposure to subsequent events; (d) spatially compounding events: events where hazards occur in different locations but affect the same larger system, such as widespread flooding occurring in different regions of a country that overloads the national response capacity. While compound events encompass a complex nature of multi-hazard scenarios, in this study we specifically focus on compound floods, characterised by the interaction of multiple flood-related drivers. We then assess the challenges and develop targeted communication strategies within the weather service value chain.

Focussing on compound flood events, recent research underscores the amplified risks and complex impacts these events bring40,42,43,44. For example, a study by Wahl et al.36 highlights how storm surges, combined with heavy rainfall, can significantly exacerbate coastal flooding risks, particularly in low-lying regions. Similarly, the 2013 winter storm Xaver highlighted the significant amplifying effect of coastal storm surge and fluvial flood hazards in northwestern Europe, with the study by Ganguli et al.45 showing that during this event, river discharge linked to a 50-year compound flood was up to 70 % higher than peak discharge levels typically recorded at individual sites. Another example is an analysis of Lake Como (Italy) by Banfi et al.46, which illustrates how the temporal clustering of rainfall can be a predominant driver of lake floods, as seen in 70 % of events over the past four decades. These examples of compound flood events demonstrate how different climate drivers can combine to create more severe and often unexpected impacts and highlight the need for integrated forecasting and warning systems to address such events effectively.

Building on these insights, national meteorological services, such as DMI, are working to enhance warning systems by integrating compound events into their product portfolio47. For DMI and the Danish society, the main outlet is constituted of the basic service that provides forecasts and warnings, such as DMI’s website48, a dedicated weather app, and email alerts to emergency service actors. Important dissemination channels downstream of DMI are local emergency service communication through e.g. websites and SMS alerts as well as broadcasts on television and radio. Additionally, tailored warnings are sent to individual municipalities and organisations based on service agreements (see warning areas in Fig. 1). Denmark faces significant challenges related to sea level rise and an increase in storm surges as a result of climate change, threatening its low-lying coastal areas and infrastructure49,50,51. Moreover, the combined impact of storm surges and wave action poses an additional layer of risk, intensifying flood hazards and shoreline erosion52,53,54. Urbanisation further compounds these issues, as the expansion of cities with impervious surfaces exacerbates runoff and reduces natural drainage, increasing vulnerability to pluvial and urban flooding55,56,57. Addressing these interconnected challenges requires the integration of advanced hydrometeorological modelling and the development of comprehensive communication strategies to ensure that warnings effectively convey the risks associated with compound events.

Fig. 1: The figure shows a geographical distribution of warning areas (marking municipalities and coastal stretches) overlaid on a base map of Denmark.
figure 1

The inset in the top-left panel indicates Denmark’s location within Europe, while the inset in the top-right panel highlights the island of Bornholm, a municipality geographically separate from mainland Denmark. The names of municipalities, regions, and seas specifically mentioned in the paper are annotated.

Full size image

Therefore, in the context of a national warning service, our definition of compound events is tailored to focus on the simultaneous issuance of multiple warnings relevant to specific municipalities, emergency response teams, and sector-specific stakeholders. This operational definition emphasises the need to recognise and address overlapping hazards, such as heavy rain and storm surge warnings issued concurrently, which can lead to exacerbated risks36,58. In practice, these multi-warning scenarios pose significant challenges for local authorities and the public, such as transportation, water management, and urban planning, as they require coordinated responses to multiple hazards occurring within the same area at the same time.

In this study, we examine the DMI warning service for Denmark over the past decade as a case study to provide general guidance for how to: (1) Evaluate the relevance and performance of an existing warning service in relation to compound floods. This is done by analysing historical warnings over a long period of time as well as a detailed case description of significant single compound events. (2) Develop understandable and locally relevant flood types that can compound with each other, based on a standardised global framework for general hazard classification. (3) Propose modifications to the current forecast and warning conceptualisation, and discuss perspectives in a context relevant to other meteorological service providers. While our primary focus is on compound flood events, given their significant impact on Danish society, we also consider other elements, such as fog, thunderstorms, and prolonged ice conditions, in cases where these occurred in relation to floods.

Methods

Operational forecasting systems at DMI

This section provides an overview of the operational forecasting systems at DMI that are relevant for predicting compound flooding events. These systems span weather prediction, ocean and storm surge modelling, hydrological forecasting, and flood/inundation mapping, providing a comprehensive framework to support Denmark’s flood warning and management efforts.

  1. 1.

    Weather/atmosphere/rainfall: DMI operates a range of Numerical Weather Prediction (NWP) models to forecast rainfall and other weather-related phenomena that contribute to flooding. These models include DMI’s weather model HARMONIE, version DINI (Denmark, Iceland, the Netherlands and Ireland), and its ensemble version, DINI EPS, as well as the ECMWF model59. These systems are supported by rain gauge networks and weather radar observations, which provide real-time data on precipitation intensity and distribution. This combination of forecast models and observation networks ensures that DMI can predict and monitor rainfall events, including the heavy rainfall that can trigger pluvial flooding and contribute to riverine flooding.

  2. 2.

    Ocean/storm surge: To forecast storm surges and coastal flooding, DMI utilises the DKSS (Danish Storm Surge) model code and setup, which is based on the HBM (HIROMB-BOOS Model)60. This system predicts water levels and storm surges along Denmark’s coastlines. In addition, WAM (wave model)61 is used to forecast wave conditions that can further contribute to coastal flooding during storm surges. Sea level observations from a network of tidal gauges provide critical validation data for these models, ensuring that the forecasts accurately reflect real-time conditions along Denmark’s vulnerable coastlines.

  3. 3.

    Hydrology/river runoff: At the time of writing, DMI does not have a hydrological forecasting system in operation. Instead, hydrological warnings are based on observations of current river water levels and predefined rainfall thresholds. However, DMI is currently in the process of setting up a national HYPE model for Denmark called “DK-HYPE”. HYPE is a hydrological model developed by SMHI for lumped conceptual modelling of catchment hydrology62 and is used in the national flood warning systems of Sweden and various other locations. DK-HYPE will provide predictions of river discharge at > 3000 locations in Denmark, and these discharge predictions will then be converted to estimates of water level through empirical rating curves. DMI is Denmark’s partner to EFAS (European Flood Awareness System), which provides pan-European predictions of river discharge. As a partner, DMI receives flood notification from EFAS, but operational experience at DMI has shown that the hydrological LISFLOOD model at EFAS struggle with the slow baseflow dynamics that characterises the hydrology of Danish surface water. Flood notifications are therefore not that reliable.

  4. 4.

    Flood/inundation mapping: DMI publishes maps of flood extent caused by storm surges and local pluvial events, which have been developed with the company SCALGO. The storm surge flood mapping uses predictions of sea levels from the DKSS ocean model as boundary conditions. The mapping is then based on pre-computed flow pathways from the coast to all inland pixels in a digital elevation model. If the elevation in a pixel is lower than the predicted sea water level at the end of the flow path, then the pixel will be inundated. Inundation maps for pluvial events are static blue spot maps computed for rainfall scenarios ranging from 5-150 mm, which are uniformly distributed throughout the country on each scenario map. Emergency services therefore have to choose the map that corresponds to the amount of rainfall in the current forecasts.

Existing warning categories for flooding

Here we describe the flood-relevant weather warning categories that we already have at DMI (see Table 1).

  1. 1.

    Rainfall (Indirect pluvial flooding): These warnings are issued when rainfall intensity exceeds the capacity of natural or man-made drainage systems, leading to localised flooding. The category includes heavy rain warnings that account for pluvial flood risks, particularly in urban and low-lying areas where drainage systems can become overwhelmed.

  2. 2.

    Coastal flooding: These warnings are issued when elevated water levels in coastal regions, caused by storm surges, pose a flood risk. This category includes warnings for both coastal flooding and storm surges in areas exposed to high sea levels, such as the western coast of Jutland or the Baltic Sea regions (Fig. 1). Storm surges are typically driven by meteorological factors such as strong winds and low atmospheric pressure, which push seawater onto land.

  3. 3.

    Hydrology (Indirect fluvial flooding): These warnings pertain to inland flooding caused by rivers exceeding their normal banks due to prolonged heavy rainfall or snowmelt. River flood warnings are generally issued when water levels are expected to rise significantly, leading to potential overflows, especially in regions with large river systems or flood-prone river basins.

Table 1 Current weather warning categories issued by DMI
Full size table

In addition to these individual warning categories, DMI also issues warnings for compound events, where multiple flood hazards overlap in time or space. Examples include warnings for freezing rain followed by heavy rainfall or storm surges coinciding with strong winds (Last rows in Table 1). These multi-hazard warnings are essential for addressing complex flood risks that result from interacting weather phenomena.

Historical weather charts to validate compound event dynamics

The DMI Weather Archive, accessible at https://www.dmi.dk/vejrarkiv, serves as a critical resource to analyse historical weather systems during compound events. This archive offers detailed weather charts and visualisations that allow users to explore past weather patterns and atmospheric conditions. By providing a comprehensive visual account of the weather developments, the archive enables the validation of weather systems linked to the compound events analysed in this study.

Specifically, these data were instrumental in justifying the temporal scope of each event by tracing the evolution of key atmospheric drivers over time. For example, through interactive and dynamic charts, we cross-verified the progression of storm systems and precipitation events to ensure accurate event categorisation and duration assessment. This rigorous use of historical weather data underpins the robustness of our event-based analysis and enhances our understanding of the meteorological contexts of compound events.

Archiving and analysing historical warnings using the NinJo system

DMI uses the NinJo system to manage, archive and analyse historical warnings related to weather and hydrological hazards63. NinJo is a collaborative meteorological workstation system developed by a consortium of several national meteorological services, including Germany, Switzerland, Denmark, Canada, and Iceland, and integrates meteorological observations, forecasting, and warning systems into a unified platform. Although primarily designed for operational meteorology, NinJo also plays a critical role in managing historical data for post-event analysis and research.

At DMI, the NinJo system archives detailed records of all weather-related warnings issued over time. This includes information on the type of warning (for example, storm surge, heavy rain, strong winds), the geographical area covered, the severity (category) of the event, and the timeframe in which the warning was active. Importantly, NinJo retains metadata on how warning categories have evolved over time. For example, historical records reflect changes in classification systems, the introduction of new warning types (such as compound event warnings), and refinements in criteria for issuing specific alerts. The availability of such detailed archives enables DMI to perform long-term analyses of weather patterns, the effectiveness of warnings, and trends in weather-related hazards. Analysis of the archived data can reveal trends in how often specific warning categories were issued, how frequently compound events occurred, and whether the public and emergency services responded effectively to multi-hazard scenarios.

Selection of recent, representative compound events

In addition to the analysis on previously issued warnings, this study also includes an analysis of five recent specific cases of compounding nature from the last five years, but with distinct features that distinguish them.

These events cover all individual categories of flood hazards that were identified (see the section Flood Typology For Effective Hazard Communication) in various combinations of compounding hazards. These events are all regarded as having a rather significant impact on the local-to-national society and reflect just a recent subset whereby other combinations could potentially occur. Due to their recent occurrence, all the data relevant to a thorough study of the events and their impacts are available.

The cases are addressed in a systematic approach by qualitatively and quantitatively (with regard to timing, especially) describing and analysing the events in terms of: (1) underlying large-scale drivers of the event, (2) the warnings that DMI sent out for the event, (3) retrospective evaluation of the impacts and consequences of the event, and (4) what were the lessons learnt about our warning service from the event?

Results

Current state of the flood warning service at DMI

In this section, we present an analysis of the DMI’s weather warnings issued between July 2014 and February 2024. DMI currently operates a comprehensive warning system that issues alerts for a variety of weather-related hazards (see the NinJo warning system in the section Archiving and Analysing Historical Warnings Using the Ninjo System, and individual warning categories in the section Existing Warning Categories for Flooding and Table 1). The system is designed to monitor and forecast individual hazards, such as coastal storm surges, heavy rainfall, and strong winds, in Denmark’s municipalities and coastal stretches (Fig. 1).

The categories of dangerous weather phenomena that DMI’s warning service uses are mainly designed for individual hazards, such as strong winds, fog, storm surge, etc (Fig. 2a). There are seven multi-warning categories that describe multivariate and/or temporal compounding phenomena (Fig. 2b, detailed definition see Table 1). However, the current multi-warning categories only cover precipitation-related hazards, e.g. multivariate compounding of heavy rainfall and thunderstorms, or temporal compounding of heavy rainfall and freezing rain phenomena. The system has not yet taken into account the full spectrum of compound events, such as those involving combinations of storm surges, fluvial flooding, and wind. The only other way to communicate about compound events is by issuing multiple single-hazard warnings to the same local emergency response units, e.g. both a storm surge warning and a strong wind warning, each of which has its own individual timelines and spatial extent. It is then up to the local emergency response units to evaluate whether multiple single hazards can cause compounding phenomena and if there is a risk that the compounding of hazards will increase impacts within their local area of responsibility.

Fig. 2: The multi-panel figure illustrates a comprehensive overview of Denmark’s weather warning patterns over the past decade.
figure 2

The left panel a presents the 10-year distribution of different weather warnings by group name, highlighting “Multi-warning'', “Wind'', and “Fog” as the most frequently issued warnings. The right panel b displays a detailed distribution of multi-warning group by cause name, indicating that “Heavy rain and flash flood” are the most common combinations.

Full size image

In order to develop a better understanding of how and where compounding phenomena of different types can occur, we analysed the temporal and spatial overlaps of the issued single hazards. In Fig. 3, chord diagrams represent the frequency and interconnectivity of different types of warning, such as coastal storm surge, wind, rain, and snow, in five representative municipalities in Denmark. Esbjerg and Copenhagen, for example, show significant overlaps between coastal storm surges and wind, suggesting that these areas are particularly vulnerable to these simultaneous hazards. However, inland municipalities such as Viborg experience different combinations of hazards, such as thunderstorms and rain. By mapping these connections, the figure highlights how each region faces unique challenges, emphasising the need for localised communication strategies to ensure timely and relevant warnings.

Fig. 3: Chord diagrams display the connections between various weather warnings issued by DMI from 2014-2024 across five municipalities in Denmark: Viborg, Aarhus, Esbjerg, Svendborg, and Copenhagen.
figure 3

Each chord diagram highlights the combination of warnings, such as coastal storm surge, wind, rain, snow, and thunderstorm, represented by colour-coded sectors. The connecting bands illustrate periods (more than 10 hours) where multiple warnings were issued concurrently, emphasising the most frequent combinations in each municipality. The base map show the location of the municipalities, indicating the variation in hazard combinations across Denmark.

Full size image

Analysis of multi-warnings in recent compound flooding events

To complement the findings of the previous sections on regional variations in hazard combinations, this section will focus on a detailed examination of individual severe weather events that involved one or more aspects of compound floods. We present five recent and representative compound flood events in Denmark. The events are described in terms of (1) the large-scale drivers causing the event, (2) issued warnings, (3) impacts and consequences, (4) challenges from weather service provider’s point of view (see the selection of events and criteria used to describe them in the section Selection of Recent, Representative Compound Events). A detailed evaluation of the accuracy of warnings with respect to event type, timeliness, and spatial-temporal coverage is beyond the scope of this analysis. Instead, a key focus includes the main learning points with respect to the warning system’s ability to predict and adequately communicate about the event.

11–14 October 2019: preconditioned

The weather event from 11-14 October 2019 was characterised by preconditioning factors that led to widespread weather warnings across Denmark. A period of consistent westerly winds had been affecting the country, preloading the region with unstable weather patterns. The country had been experiencing a prolonged period of consistent westerly winds, which had pre-loaded the region with unstable weather patterns, though this was not an extreme situation. The low-pressure system that followed the preconditioning led to heavy rainfall, thunderstorms, and strong winds, resulting in a flooding event that was challenging to predict.

  • Large-scale drivers: The event was driven by a low-pressure system moving across the North Sea, intensifying as it approached Denmark. Westerly winds were predominant before the event, which, combined with the low-pressure system, contributed to increased wind speeds, heavy rain, and scattered thunderstorms across the country.

  • Warnings: DMI issued multiple warnings during this period, but without the period overlap. As shown in Fig. 4a, there are no overlapping warnings, with heavy rain (dark blue) on 11 October first, and no warnings on 12 October, and then wind (red) and thunderstorms (brown) were active on 13 and 14 October.

  • Impacts and consequences: Figure 4b shows that the storm system had a significant spatial impact, particularly in the southern and eastern parts of Denmark, like Vejle. Thunderstorm warnings were primarily concentrated along the eastern coast, while rain and wind warnings were more widespread, leading to concerns over localised flooding and damage to infrastructure.

  • Challenge: This event highlighted the need to recognise preconditioned scenarios where existing extreme or non-extreme weather patterns over previous days created a vulnerability to compounding hazards in certain regions. The gap of more than 24 hours between the event on 11 and 13 October makes it unclear whether the two weather systems had a coupled effect. Therefore, this event emphasised the need for seamless integration between meteorological and hydrological models, allowing real-time updates and adjustments to forecasts, thus providing more precise and actionable warnings.

Fig. 4: Weather warnings issued by DMI during 11–14 October 2019.
figure 4

a Timeline showing the duration and overlap of weather warnings categorised by group names: wind (red), thunderstorm (brown) and storm surge (blue). b Geographical distribution of weather warnings for the same period. Single warnings use the same colours as the top panel, overlayed with hatching. Overlapping warnings include Thunderstorm & Rain (dark brown), Wind & Rain (plum) and Thunderstorm & Wind & Rain (sand).

Full size image

February/March 2020: multivariate, temporal, spatial

  • Large-scale drivers: A series of deep low-pressure systems from the North Atlantic ocean hit Denmark. Each system brought heavy westerly winds and large amounts of frontal precipitation.

  • Warnings: As shown in the timeline plot in Fig. 5a, DMI issued numerous storm surge warnings (blue colour) that were consistently active from February 9 to February 26, reflecting the prolonged nature of high sea levels along Denmark’s coastlines. Wind warnings (red colour) were also issued intermittently throughout this period, with concentrated warnings around February 15–17 and February 25–26. Despite the persistent heavy precipitation leading to record-breaking monthly rainfall, no direct fluvial flood warnings were issued as this event occured before hydrological warnings were implemented at DMI.

  • Impacts and consequences: Fluvial floods occured along rivers in Jutland and Funen that happened due to temporal compounding of successive large-scale frontal precipitation events. With a national average of 136 mm, February 2020 broke the record for largest amount of precipitation ever recorded in a February month (archives go back to 1860’s). There were large-scale spatial compounding of fluvial floods, which hit many river catchments and administrative regions simultaneously across Jutland and Funen. This meant that DEMA (the Danish Emergency Management Agency) were forced to prioritise where to allocate resources, including both personnel and equipment. Along the Wadden Sea coastline, there were also severe compounding of fluvial and coastal floods (Fig. 5b), and therefore also multivariate drivers (precipitation and wind). Heavy westerly winds with resulting high sea water levels forced the local sluice operators to close the gates at the mouth of Ribe River for an extended period of time. This meant a gradual build up of inundated areas on the inland side of the sluice as the river could not discharge into the ocean. At the peak of the event, extensive floods occurred in the city of Ribe.

  • Challenges: At the time, DMI did not have a fluvial flood warning category, nor a hydrological prediction system, and this event showed a clear need for this. It also showed that such a hydrological prediction systems needs to be able to account for gradual build up of water levels / flows over several weeks due to successive frontal precipitation systems. The buildup of water in Wadden sea streams and the associated inundation due to prolonged and severe westerly winds and the closed sluice gates was also somewhat unprecedentend in recent times. Such sluice-systems are present in many locations around Denmark and a flood warning system must be able to account for (1) extended storm surge periods, (2) river runoff, and (3) sluice gate management to adequately handle this phenomenon.

Fig. 5: Weather warnings issued by DMI during February 2020, highlighting a temporally compound event.
figure 5

a Timeline showing the duration and overlap of weather warnings categorised by group names: wind (red) and storm surge (blue). Start and end times of each warning are shown to highlight the persistent and overlapping nature of warnings characteristic for a temporally compound event. b Geographical distribution of weather warnings for the same period. Single warnings use the same colours as the top panel, overlayed with hatching. There were no overlapping warnings. This map emphasises the extensive impact of this temporally compound event across Denmark, showing how different regions were simultaneously affected by multiple types of weather hazards over an extended period.

Full size image

13 March, 2023: Preconditioned

  • Large-scale drivers: Over a six-day period, a cold spell with several days of snowfall had left 20-50 cm of snow on the ground prior to the flood event. The floods were triggered by a warm front that swept across Denmark on 13 March delivering 15–30 mm of rainfall. Due to the warm front, air temperatures suddenly increased from ≈ 0 ∘C (freezing) to 9 ∘C. This change produced large amounts of snowmelt. The duty forecasters were therefore forced to utilise experience and a more qualitative approach to estimate what might happen.

  • Warnings: At the time, DMI had no prediction system that could account for snow melting, leaving the event without a clear and specific warning category. While standard forecasts indicated a warming trend and rainfall, there was no explicit mechanism to communicate the risk of compounding effects from snowmelt and rain, nor their potential for exacerbating flood hazards.

  • Impacts and consequences: There were pluvial and fluvial floods across Northern Jutland, and local emergency services had to work hard to avoid the collapse of a river embankment protecting the town of Elling.

  • Challenges: The duty forecasters investigated that snow depth parameters of numerical weather prediction models were very different from the observed manual gaugings of snow depth. On top of this, DMI’s snow depth measurements are not performed in real-time, instead they are manual measurements once a day at 8am in the morning. It was therefore not possible to follow the progress of snowmelt as the event happened. At this point in time, DMI was already obliged to issue fluvial flood warnings, but due to its early stage only EFAS flood notifications was available, which did not show any flood issues. This event therefore further underpinned the immediate need for a dedicated hydrological modelling system with a good representation of snowmelt processes and the challenges in slow “updates” of snow depth measurements during such events.

19–22 October, 2023: preconditioned, multivariate, spatial

  • Large-scale drivers: Several days of severe westerly winds forced large amounts of ocean water from the North Sea through the Danish straits into the Baltic Sea, where sea levels rose to approximately 40 cm above normal conditions (Fig. 6b). A low-pressure system then moved from the North Atlantic over the southern part of Great Britain, while a high-pressure system developed over Northern Sweden leading to a large pressure gradient across Denmark. This caused the winds over Denmark to shift into severe gale force easterlies (Beaufort scale 9) with gust of violent storm strength (Beaufort scale 11); see the section Existing Warning Categories for Flooding. Rainfall in the southern half of Denmark amounts to approximately 15–35 mm over a 24–h period.

  • Warnings: Due to the confidence in the forecasted atmospheric patterns and the severity of predictions, so-called early “pre-warnings” of storm surges were issued already five days ahead of the event. The updated warnings of the most severe category were then sent 36 hours in advance (Fig. 6a).

  • Impacts and consequences: The preconditioning of the Baltic Sea water levels from westerly winds followed by the easterly storm notably increased flood impacts, and led us to classify the compounding nature of the event as both preconditioned and spatial. In most locations, the event was a univariate storm surge, but there were also cases of compound floods at the mouth of small rivers (see also ref. 64), occurring on the coasts of the Baltic Sea, the eastern coasts of Southern Jutland as well as in Køge Bay and Præstø Bay. The spatial extent of the storm surge was significant in all municipalities with coastlines facing the Baltic Sea, which meant that all local municipal response teams in the southern parts of Denmark were active and DEMA (the Danish Emergency Management Agency) had to prioritise its resources in the most critical locations. Peak seawater levels were estimated to exceed a 100-year return period in several places. The total flood damage is still not known at the time of writing, but damages eligible for compensation from the emergency fund of the Danish Natural Hazard Council exceeded 150 million euros65.

  • Challenges: This was one of the best-warned severe weather events in the history of DMI. The early warning (5-6 days of notice) was incredibly important for national and local emergency services to organise the response. As a result, there were no casualties, despite the immense severity, and protective measures could be taken well in advance. However, the lack of a dedicated hydrological prediction system made it very difficult to predict where river-coastal compound floods could occur.

Fig. 6: Weather warnings issued by DMI during 19–22 October 2023, highlighting a spatial compound event.
figure 6

a Timeline showing the duration and overlap of weather warnings categorised by group names: wind (red), rain (dark blue), foggy (grey), and storm surge (blue). b Geographical distribution of weather warnings for the same period. Single warnings use the same colours as the top panel, overlayed with hatching. Overlapping warnings include Foggy & Rain (dark grey), Rain & Wind (plum) and Foggy & Rain & Wind (sand). This map emphasises the extensive impact of this spatial compound event across Denmark, showing how different regions were simultaneously affected by multiple types of weather hazards.

Full size image

2–7 January 2024: preconditioned, multivariate, spatial

  • Large-scale drivers: A low-pressure system moved directly over Denmark from the southwest, while a high-pressure system was situated over Finland. This gave strong easterly winds over most of Denmark with a minor storm surge and high waves along the Baltic sea coastline (Fig. 7). There was a strong north-south temperature gradient across Denmark with 5 ∘C in the south and −2 ∘C in the north. The low pressure system brought extreme amounts of frontal precipitation. The days following the event saw easterly winds and an extended cold spell with many days of freezing temperatures over most of the country.

  • Warnings: The operational numerical weather prediction (NWP) models were extremely sensitive to the exact path of the low pressure centre as it passed over Denmark, and consecutive and ensemble-based NWP simulations were very inconsistent in predicting the rain-snow boundary, amounts of precipitation, and the wind directions, which also had substantial effects on the accuracy of the ocean modelling system for storm surge and waves. Therefore, the warnings were issued with a significant overlapping area with snow and rain warnings in central Jutland. The erratic predictions of wind directions also meant that the warning for storm surge in Southeastern Denmark were issued with very short lead time before the impacts (Fig. 7b).

  • Impacts and consequences: Large areas of the southern islands of Denmark (Funen, Lolland, Falster, Møn) broke the previous national record for 24-hour accumulated precipitation on a January day with > 50 mm. Following the wettest year on record in 2023, the soils were completed saturated, which preconditioned extreme pluvial floods where large urban and rural areas were water logged for several weeks following the event. Several of the water logged areas experienced severe disruptions to drainage and sewerage infrastructure for an extended period of time. The cold spell in the days after the event froze the water logged areas deeply, causing severe disruption of transportation infrastructure. As an example, the highway towards the Rødbyhavn ferry harbour was covered by a deep layer of ice, and it took several days for the emergency services to restore access to the highway. There were fluvial floods across large areas of southern Denmark, which were also preconditioned by the saturated soils. The easterly winds caused a storm surge along the Baltic Sea coasts, which was exacerbated by high waves. There were multivariate fluvial and coastal floods in the cities of Køge, Præstø and in areas of the eastern parts of Southern Jutland. Due to the temperature gradient, the precipitation fell as heavy snowfall in the central and northern parts of the country causing large transportation disruptions of both roads and railways. People were stuck in their cars in freezing conditions for up to 24 hours on the E45 highway north of Aarhus. From the national perspective of DEMA, the scale of the floods and snow problems was extreme and very complex. Even at local, municipal scale the situation was very complex. An example of this being Hedensted municipality on the eastern coast of Central Jutland, which had temperatures close to zero degrees and was therefore at the rainfall-snow boundary. Within its small 30 × 40 km2 municipality area, the response had to be prioritised between a snow storm in the northwest, fluvial and pluvial floods in the south and east, and high waves prompting concerns about a potential breach of the sea dikes at Bjørnsknude. At the point/location scale, the floods in the town of Præstø were multivariate fluvial and coastal floods along the mouth of Tubæk river, but also preconditioned by the fact that the 100-year storm surge event three months prior on 20-21 October, 2023, had broken the town’s sluice-gate system, which still had not been repaired by the time of this event.

  • Challenges: The spatial scale of end-user responsibilities plays a critical role in determining and effectively communicating the specific compounding hazards that are relevant to different stakeholders. We saw that different kinds of compound events occur at different spatial scales: national, municipal, location/point. The inability of a national weather service to keep track of the operational status of local flood mitigating structures, such as the broken sluice-gate system in Præstø, can seriously complicate flood hazard and impact predictions. The inconsistency and uncertainty of NWP runs can be very large during certain weather events, especially in terms of precipitation amounts and wind directions. This seriously complicates the communication of warnings to end users and the public. While DMI operates NWP ensembles for internal use, only a single, deterministic run is made visible on maps for the public. DMI therefore has very limited options for communicating uncertainties, and interested end users (including emergency services) have to call a telephone hotline, or read a short free text section on DMI’s website where the duty forecasters can write about uncertainties if they find it relevant. This gap underscores the need for integrating uncertainty information directly into warnings.

Fig. 7: Weather warnings issued by DMI during 2–7 January 2024, highlighting a multi-hazard event.
figure 7

a Timeline showing the duration and overlap of weather warnings categorised by group names: wind (red), rain (dark blue), snow (yellow), and storm surge (blue). b Geographical distribution of weather warnings for the same period. Single warnings use the same colours as the top panel, overlayed with hatching. Overlapping warnings include Snow & Rain (green), Snow & Wind (orange), Rain & Wind (plum) and Snow & Rain & Wind (sand).

Full size image

Flood typology for effective hazard communication

Notably, none of DMI’s current warning categories consider floods explicitly (with the exception being some of the existing multi-warning categories for rainfall that mention the risk of flash floods in case of severe rainfall). Instead, the only indirectly relevant warning categories focus on two variables that can lead to floods: rainfall and sea level. In order to better warn against floods, it is therefore highly relevant to establish a categorisation of local flood types to include in forecasts, warnings, and communication outlets. Such classification does not exist for Denmark, and we therefore propose the following approach for using a global hazard framework to establish local flood hazard categories.

To support global efforts in implementing the Sendai Framework for Disaster Risk Reduction, the United Nations Office for Disaster Risk Reduction (UNDRR) developed the comprehensive report “Hazard Definition and Classification Review”66, which offers a unified approach to defining hazards and lists 302 distinct types. This classification spans categories such as meteorological, hydrological, extraterrestrial, geohazards, environmental, and technological hazards. A supplementary report, Hazard Information Profiles (HIP), provides detailed definitions of these hazards, forming a global standardised lexicon for the identification of hazards67.

Of these 302 hazards, 12 types of flood related hazards were identified, two of which were discarded as irrelevant for Denmark (ice jam floods and glacial lake outbursts). The remaining ten hazards were aggregated into five categories based on flood occurrence patterns and their relevance to emergency response. These are coastal floods, fluvial floods, pluvial floods, groundwater floods, and technological floods (see Fig. 8 and the section Existing Warning Categories for Flooding). Most of the original ten fit directly into one of the five aggregated groups, but some are split into two of the aggregated groups (estuarine = coastal + fluvial; snowmelt = pluvial and fluvial). This simplified categorisation improves communication and aligns with the responsibilities of emergency services.

Fig. 8: Doughnut chart illustrating the classification of flood hazards relevant for Denmark.
figure 8

The inner circle depicts five overarching flood categories: Coastal, Fluvial, Pluvial, Groundwater, and Technological floods, while the outer circle the original flood hazard types from the UNDRR's HIP-profiles. The two shaded flood types (groundwater and technological) are not currently represented in DMI's warning system.

Full size image

Following the categorisation, we assess how the current warning categories at DMI could be modified to better fit the five targeted flood categories. The current warning categories that are used to warn against weather variables that might cause flooding are storm surge, hydrology, rainfall, and the multiwarning categories that include rainfall.

  1. (1)

    Of the five categories, coastal flooding is likely the hazard that is handled best by the current DMI warning system. This is the case since existing warning systems predict sea levels along the entire Danish coastline and assess the exceedance of specific water level criteria defined through continuous conversations between DMI and the local municipal emergency response units.

  2. (2)

    Fluvial flooding has until recently only been indirectly warned for by DMI through the rainfall warning categories that are based on the exceedance of at least one of several rainfall thresholds (15 mm in 30 minutes, 24 mm in 6 hours or 30 mm over 24 hours). These thresholds are, at present, constant for the whole country and do not vary seasonally or with antecedent catchment conditions (such as soil saturation). However, a new national hydrological model is in development at the time of writing, which will provide predictions of river discharge at certain locations, and local warning thresholds are intended to be defined through conversations between DMI and the municipal emergency response units. The hydrological model will be able to account for antecedent catchment conditions, seasonal influences, and snowmelt, which will remedy many of the current shortcomings and result in predictions that can be used to warn for the fluvial flooding category.

  3. (3)

    Pluvial flooding, as we define it, encompasses floods that are generated by rainfall running off surfaces into terrain depressions, but also floods that occur when local drainage system capacities are overwhelmed. These floods have until now also only been indirectly warned through the same rainfall warning categories described under “fluvial flooding”. Providing a warning of pluvial floods will require several changes to the current warning system. In terms of fluvial floods, the risk of pluvial floods is influenced by antecedent surface conditions, as well as soil type characteristics and local drainage capacity. A surface water balance model (or a complete hydrological model) would be able to quantify local infiltration capacities to adequately assess whether predicted rainfall amounts pose a risk of surface floods. Sewers and urban drainage systems are designed based on engineering guidelines that specify how much rainwater they should be able to handle. In Denmark, the main guideline is that combined and separate sewer systems should be able to handle a 10- and 5-year return period rainfall event without discharging water to urban surfaces. Information on sewer system type and local return period estimates for rainfall is available on the national scale and could be accounted for to move from rainfall warnings to pluvial flood warnings.

  4. (4)

    Groundwater flooding occurs when the water table rises to ground level, inundating areas from below. This phenomenon develops at slower rates compared to pluvial, fluvial, and coastal floods. It is therefore the least relevant individual category in a flood warning system that considers timescales of a few days ahead at most. None of the current DMI warning categories are directly related to groundwater floods. However, its impacts can be persistent and challenging to predict without an extensive network of groundwater monitoring stations68. Currently, Denmark lacks a well-established dataset and real-time monitoring network for groundwater gauges, which hampers the ability to effectively forecast such events69. Establishing a more comprehensive network of groundwater gauges, combined with advanced hydrological models that can integrate these data, such as the national scale, MIKE SHE - MIKE HYDRO model complex of the Geological Survey of Denmark and Greenland70, is necessary to enhance early warning capabilities71.

  5. (5)

    Technological flooding, including dam and dike failures, sluice gate failures, and similar infrastructure breakdowns, remains a poorly addressed hazard in the current warning framework. This became evident during, for example, the January 2024 storm surge, where sluice gates in the town of Præstø had not been repaired following the October 2023 storm surge event. This resulted in floods within the city along the downstream parts of the local river. So far, it has been the responsibility of local stakeholders to evaluate whether the predicted hazards pose a risk to their technical structures. Incorporating technological flood hazards into the warning service would require not only technical modelling of, e.g., dike breach or sluice failure scenarios, but also active coordination with local authorities and private sector stakeholders responsible for critical infrastructure. This would be a huge national task that would require extensive coordination with local public and private stakeholders on the operational status of critical infrastructure.

Discussion

By examining past warnings and case events, we have identified key gaps in the existing forecasting and warning framework. Building on these findings, we propose improvements to enhance the integration of compound events into Denmark’s warning services, with broader implications for other weather service providers.

For preconditioned events, one of the key challenges is the identification and prediction of variables that significantly influence local flood risks across complex spatial and temporal trajectories, including, e.g., soil moisture levels, snow cover, and the degree of filling of the Baltic Sea72. For example, high soil moisture reduces the infiltration capacity of the ground, leading to increased surface runoff during heavy rainfall. This issue was a critical factor in the unsatisfactory forecast prior to and during the October 2019 flood event (Section: 11–14 October 2019: Preconditioned), where the saturated soil could not absorb additional rainfall, resulting in severe pluvial and fluvial flooding. Similarly, snowpack accumulation and its subsequent meltwater can amplify pluvial and fluvial flooding, especially when combined with rainfall, as observed in the March 2023 event (Section: 13 March, 2023: Preconditioned). During this event, the lack of a meltwater forecasting system led to an unanticipated magnitude of floods. Some review literature, such as that by Zscheischler et al.5 and Bevacqua et al.40, highlights the importance of soil moisture as a predictor for pluvial and fluvial floods, pointing to a wealth of studies (see references therein) that underscore its critical role. However, obtaining a well-validated soil moisture dataset for forecast purposes remains a substantial challenge73. Limited soil moisture observations, despite the availability of satellite products (such as ESA CCI and NASA SMAP), along with the coarse spatial resolution and inconsistencies of models like ERA5-Land, make it difficult to accurately incorporate observations of this variable into real-time forecasts. Similarly, snow cover, snowpack, and temperature are critical variables for predicting snowmelt floods in regions with significant winter snowfall74. Tuttle et al.75 emphasised that snow water equivalent (SWE) and soil moisture are key drivers of snowmelt floods, particularly in mountain areas. However, accurately capturing snow dynamics in forecasting systems is equally challenging due to the complexity of snowmelt processes and the limited availability of real-time data76. Many hydrological forecasting systems implicitly or explicitly account for antecedent soil moisture and snowpack conditions as part of their model predictions. However, without reliable observations of these variables for assimilation into models or real-time assessment of forecast performance, there is still significant uncertainty about their influence leading up to a given event. Being able to quantify, visualise, and communicate about the current state of important preconditioning variables, independent of the forecasts and warnings for the main impact-related variable of interest (e.g., river water level) would be very relevant for end users of warnings.

The analysis of multivariate compound events in Denmark reveals several key combinations of flood hazards that are particularly problematic due to their simultaneous (e.g., the event in January 2024, Section: 2–7 January 2024: Preconditioned, Multivariate, Spatial) or sequential nature (e.g., the event in February 2020, Section: February/March 2020: Multivariate, Temporal, Spatial). The complexity of forecasting and preparing for compound events, particularly in the timing and overlap of multiple hazard drivers, highlights the critical need for integrated models that can accurately simulate these interactions and their combined impact. DMI’s current operational models mainly focus on a single flood hazard (numerical weather predictions for rainfall and snow, ocean models for sea levels, and rainfall-runoff models for river discharge) without accounting for the interactions between them. This results in underestimations of risk and incomplete hazard assessments. Integrating cross-disciplinary data and models can provide a more comprehensive understanding of these events (see review papers77,78,79, and references therein). One solution involves the coupling of models80,81 or fully integrated flood models, but these are often still extremely computationally heavy, making them difficult to use for operational services, especially when considering ensemble predictions and forecast uncertainty quantification. Another option is the development of seamless and on-demand modular models82 or the “tales of future weather” approach83 that can be adapted to local conditions and include various hazard interactions, thus offering municipalities more accurate and actionable warnings. Our study of these events shows how spatial overlaps and the varying durations of multiple flood drivers underscore the importance of understanding where and when these interactions occur. Although we provide several examples of different multivariate compound floods and use historical warnings to analyse which hazards can be compounded where (Fig. 3), there is still a great need for more scientific work that systematically studies this with observations and models84,85.

For single-hazard-type temporally compounding events, where multiple flood drivers occur in a sequence over a short time span, the existing suite of modelling systems at DMI are generally skilful in their reproduction. Both the hydrological (HYPE) and oceanic (HBM) models at DMI readily account for the gradual accumulation of river discharge and sea levels, respectively. Hence, these two models, supplemented by the DMI NWP model (HARMONIE) individually, provide sufficient temporal resolution and predictive capabilities to support warnings for individual flood drivers and sequential hazard events86. In contrast, challenges are seen in model performance where multivariate interactions are present across model domains. An example includes the February 2020 event (Section: February/March 2020: Multivariate, Temporal, Spatial), where the reproduction of sea levels and river discharge at the Ribe river sluice gate presented challenges that current models were not equipped to fully address. Another critical gap in the current multi-warning category is the lack of a framework to handle overlapping hazards across both temporal and spatial scales87, particularly with wind and storm surge events, e.g., the February 2020 event. The coastal municipalities often face coastal flooding and high winds during major events (Fig. 5). Municipalities are required to manage simultaneous risks: strong winds impacting land-based infrastructure and storm surges threatening coastal defences. This mismatch complicates resource allocation and response prioritisation, as the absence of integrated multi-warning categories obscures the interconnected nature of these hazards. A refinement of the warning framework to incorporate spatially overlapping wind and storm surge events as part of the multi-warning system could enhance clarity for municipalities and ensure more cohesive response strategies.

Spatially compounding events, where multiple hazards occur simultaneously in different locations or interact across various scales (e.g., the event in January 2024, Section: 2–7 January 2024: Preconditioned, Multivariate, Spatial), present unique challenges for warning systems, particularly when considering the spatio-temporal evolution of hazards and their interactions87. Our experience is that different end-users, such as national agencies, local authorities, and private individuals, operate on vastly different spatial scales. During the January 2024 event, at the local town level, many people were dealing with a single flood hazard (e.g., pluvial, fluvial, or coastal), whereas at larger municipal and national scales, the event was extremely complex with floods and other weather hazards compounding in preconditioned and multivariate ways. This makes it crucial to tailor warning services according to these spatial considerations88. The importance of spatial scale becomes evident when considering how these different users react to warnings. For example, a spatially compounding coastal storm surge that affects both the western Danish coasts and Kattegat might require a coordinated national response, but within each affected region, local municipalities would prioritise resources based on their specific vulnerabilities, such as protecting key urban areas or maintaining critical lifelines like road networks and water supply systems. Similarly, at the household level, private residents might only be concerned if their specific location is directly impacted, requiring highly localised point-based warnings.

In previous studies, considerable emphasis has been placed on developing frameworks and theoretical models for effective FEWS to improve public understanding and response during extreme events25,26,28,89. However, these studies largely focus on conceptual guidelines rather than offering concrete, actionable methods for implementation. At national and regional levels, warnings should focus on providing a clear overview of broad flood risk areas, ensuring that major stakeholders and emergency response teams are aware of potential multi-hazard compound situations90. Our work bridges this gap by introducing a practical approach with ’informational text advisories,’ which offer detailed, context-specific guidance that combines the clarity of straightforward text with the immediacy of actionable advisories.

Such a “special warning service” approach to service communication is essential for different types of compound events91. The lack of well-validated soil moisture data and the difficulty in incorporating snow dynamics into current models limit the effectiveness of automated warnings. For example, a shift toward incorporating more detailed context-specific information into warning communications, such as text message alerts, could significantly improve the clarity and usability of these warnings. Moreover, spatially compounding events require a tiered approach to warnings that differentiates between broad, national-scale alerts and finely resolved, site-specific information. To make the benefits of informational advisories explicit, we compare the existing approach with an enhanced version incorporating such advisories in Table 2.

Table 2 Flood warning examples to compare current warning information and enhanced informational advisories
Full size table

The transition to dedicated and impact-based predictions is widely recognised in the scientific community as a critical step to enhance national warning services92,93,94,95. However, Denmark has a long-standing tradition of decentralised emergency response management services. The traditional responsibility of DMI has been to focus only on hazards. Danish law on public emergency services says that impact mitigation responsibilities are with the local emergency authorities and further split into local “sector responsibilities” (i.e., everyone has to protect themselves; e.g., local water utility company must prevent damage to water infrastructure). This relates to the severe challenges for a national weather service to maintain an updated overview of all local-scale knowledge on, e.g. impact-mitigating strategies such as climate change adaptation projects, the use of mobile flood-mitigating services such as water tubes, and sluice operations. If DMI were to produce impact maps, based on high-resolution inundation maps, these would quickly become outdated as local emergency services continuously adapt and refine their methods and strategies for impact mitigation. Furthermore, workshops with local Danish emergency services have revealed that these agencies do not seek real-time impact assessments from DMI. Instead, they prefer to receive precise hazard predictions that they can incorporate into their own decision-making frameworks. This feedback underscores a key point: the role of a national warning service should focus on delivering reliable hazard forecasts, which local responders can then use in conjunction with their in-depth understanding of community-specific vulnerabilities and assets91.

However, implementing national-level impact models has merit in large-scale emergencies where local responders may lack the expertise or resources to develop, run, and maintain detailed impact models. In such cases, a national impact model, while less detailed than local models, could provide a consistent and overarching picture of large-scale impacts. This would be especially beneficial for ensuring coordination between multiple municipalities and regions during large spatial emergencies. To strike a balance, DMI should explore creating a two-tiered system: (1) high-quality hazard predictions to serve as the core product for regional and national decision-makers and (2) supplementary impact scenarios that provide a macrolevel view of large-scale emergencies. By acknowledging and leveraging both the strengths of localised knowledge and the consistency of national-level models, this approach could improve preparedness while respecting the decentralised emergency management framework in Denmark. Importantly, the success of such a system depends on fostering strong collaborations between national agencies and local responders to ensure that all stakeholders can benefit from integrated hazard and impact forecasts.

By providing a systematic evaluation of warning services, developing tailored hazard categories, and proposing actionable modifications, our study serves as a framework for improving national warning systems. The insights gained here not only improve Denmark’s ability to address compound flood hazards but also offer valuable perspectives for other meteorological service providers worldwide.