Building a multisectoral weather impact database and piloting machine-learning based impact forecasts

The direct, indirect and cascading impacts of extreme weather events and climate change have created a growing need for developing new methods for forecasting the scale of impacts. However, the progress has been hindered by fragmented and poorly available information on the realized impacts. This information is a prerequisite for developing impact forecasting tools that are based on machine learning or other novel technologies.In this work, we created a weather impact database, where impact information from various sectors such as rescue services, electricity distribution, insurance and transport was successfully merged. Most of the impact data time series are years or even decades long and cover the whole country of Finland. As far as we know this may be a unique database from a global perspective, where detailed weather impact data have been able to be collected exceptionally comprehensively.After successfully establishing the database, we trained a machine learning model to demonstrate one of its potential applications. The model, trained with impact data, was designed to predict rescue service workloads and anticipate traffic accidents, particularly during adverse weather conditions. To study its effectiveness, a group of seasoned weather data users tested the novel forecasts in challenging real-world weather conditions. Their initial feedback was encouraging, highlighting significant potential for further advancements in impact forecasting and its practical use in enhancing public safety. We expect that the use of the impact database and forecasts can improve preparedness for hazardous weather, informed decision-making, and refine the assessment of necessary response measures.