Collaborative data-driven quantitative impact forecasting: applications in storm damage, human health and wildfires

Context. Extreme weather events are affecting society in unprecedented ways. Weather and climate services, including early warnings, play a fundamental role in preparing for and responding to these events, ultimately reducing loss and damage from the weather. In 2024, the EEA published the European Climate Risk Assessment, identifying different types of extreme weather as some of the most important climate change risks. To prepare for and respond to the rising weather and climate risks, the WMO has promoted the move towards impact-based forecasts (WMO, 2015, 2021). Knowing what to expect in terms of impacts is much more actionable for people and enables decision-makers to take targeted, cost-effective, and informed decisions (Harrowsmith et al., 2020). For some time, the Royal Netherlands Meteorological Institute (KNMI) and its Early Warning Centre (EWC) have been working towards impact-based warnings.Approach. Within the context of early warning services, KNMI and its EWC are experimenting with quantitative impact forecasting: "not what the weather will be, but what the weather will do." Through collaboration with decision-makers and end-users, we apply data-driven methods to make quantitative forecasts of impact. We provide these quantitative forecasts as an ensemble, so that the stakeholders can base their decision on the expected impact and the uncertainty of the forecast. We apply a general machine learning framework to train a model on the observed weather and impact data for the Netherlands, the latter being provided by the stakeholders. Then, we use this model to derive quantitative two-week impact forecasts.Results. We show results for three different fields: storm damage, human health, and wildfires. The storm damage results are aggregated to a 10 x 10 km spatial resolution and 24-hour time resolution, and we provide a two-week forecast of the expected number of daily storm damage emergency calls as received by the fire services at the combined emergency services dispatch centre. We are collaborating closely with the emergency service provider to communicate these quantitative impact forecasts in an efficient way. For human health (cardio-vascular and respiratory fatalities) as well as wildfires (nature fires and road-side fires), we are currently working on quantitative impact forecasts aggregated at the national and monthly levels. The spatial and temporal scale of forecasts will be further improved to the provincial and weekly levels. Lastly, we will highlight the current limitations and challenges of impact-based forecasts based on our practical experience, and further research needed in this area.