Conference Agenda

Background. Over 110,000 Europeans died as a result of the record-breaking temperatures of 2022 and 2023, emphasising the urgent need to strengthen existing emergency and resilience plans. Inherent to the adaptation strategy against climate change, governments need to develop a new generation of heat-cold-health early warning systems, using epidemiological models to transform the physical information of weather forecasts into health-related forecasts, and specifically targeting vulnerable groups. The forecast skill horizon of these impact-based systems however needs to be first demonstrated in order to generate trust among public health authorities and end-users.

Methods. Here we tested the forecast skill horizon of temperature related mortality forecasts in Europe, as a necessary step towards the release of the first operational continental heat-cold-health early warning system. We used a spatiotemporally-homogeneous daily mortality database, including almost 60 million counts of death in 147 contiguous European regions, representing their entire urban and rural population of 420 million people. We used state-of-the-art temperature-lag-mortality epidemiological models to transform bias-corrected ensemble weather forecasts into daily predictions of temperature related mortality. We compared the predictive skill of temperature forecasts and temperature related mortality predictions by using predictability assessment techniques widely used in operational weather forecasting.

Findings. We found that temperature forecasts can be used to issue skilful forecasts of temperature related mortality at lead times beyond 15 days in winter and beyond 11 days in summer, accounting for the real impacts of temperature on human health at very long lead times. Nonetheless, when compared with the original temperature forecasts, the forecast skill horizon of the forecasting system was differently reduced by season and location due to the epidemiological models. Overall, our study showed that the forecast skill is to a very large extent influenced by the forecast skill of the original weather forecasts, and to a much lesser extent by the epidemiological models. This means that further advancements in weather forecasting would automatically turn into an increase in the forecast skill horizon of heat-related forecasts.

Interpretation. The forecast skill assessment of operational weather forecast schemes, which are routinely done by meteorologist, cannot be used to directly characterise the forecast skill horizon of any derived impact-based health early warning system. Overall, our results indicate that a rigorous assessment of the forecast skill of health early warning systems is an unavoidable requisite to generate trust among public health authorities, and in this way, increase resilience and strengthen our early adaptation response to climate change.

Forecaster-Dot-Health (https://forecaster.health/), funded by the ERC Proof-of-Concept Grants HHS-EWS and FORECAST-AIR, is the first continental, impact-based early warning system issuing health warnings of heat- and cold-related mortality risks by sex and age (Ballester et al. 2024).

Every day, the system automatically downloads and processes the latest available (i) temperature observations and (ii) 51 ensemble member forecasts for the next 15 days. Then, it post-processes the ensemble of temperature forecasts to bias-correct them against the temperature observations used in the epidemiological models (see below). We chose a bias-correction method considering the most recent N = 30 pairs of observations and forecasts with respect to each forecast start date (BC-30).

To transform the bias-corrected weather forecasts into predictions of heat- and cold-related mortality risks, we used a time-series quasi-Poisson regression model to derive estimates of region-specific temperature-lag-mortality risks (Ballester et al. 2023, Gallo et al. 2024). For that purpose, we used the daily temperature and mortality dataset of the ERC Consolidator project EARLY-ADAPT (https://early-adapt.eu/), which includes spatiotemporally homogeneous data for the period 2000-2019 in 654 contiguous regions from 32 European countries. We used these epidemiological associations to transform the temperature forecast for a given location, forecast date and forecast lead time into 5 warning categories: a baseline warning state (“none”), when the risk of death is minimum, and 4 categories of heat and cold warnings (“low”, “moderate”, “high” and “extreme”), corresponding to increasing levels of risk of death.

A key aspect of the system is that, in each location, we estimated separate epidemiological associations for each sex and age group. These sex-specific and age-specific epidemiological associations were exclusively estimated with mortality records of the respective sex or age group, and therefore, they quantify the actual risk of death of the population subgroup at any given temperature and location based on real data. This means that we issue independent health warnings for each population subgroup based on the temperature forecasts and its corresponding sex-specific or age-specific epidemiological association.

In this presentation, I will give an overview of the methodology, including (i) the processing of the weather forecasts, (ii) the fitting of the epidemiological associations, (iii) the interdisciplinary integration of data and models in both areas, and (iv) the automatization of the system to deliver updated health warnings every day. I will also show (a) initial results for the health warnings by sex, age and geographic area, and (b) the gained experience and feedback from a range of European public health agencies after running the system for almost 8 months. Finally, I will sketch the new features that the system will include during the year 2025.

The projected increase in heatwave intensity and frequency will have far-reaching consequences for the human and natural environment of Switzerland. Two particularly important consequences are heat-related excess mortality in the low-lying areas and heat-related acceleration of climate-change-induced alpine permafrost thawing in high-elevation areas. The latter will potentially have far-reaching impacts on alpine hazards, ecosystems, infrastructure, and tourism. In this interdisciplinary project, we assess the potential of using subseasonal heatwave predictions as a basis for early warning systems for the above-mentioned sectors in Switzerland. For the health sector, we show that the (observation-based) statistical relationship between temperature and mortality in combination with downscaled subseasonal temperature forecasts can be used to predict mortality attributable to heat. We demonstrate that for two densely populated areas of Switzerland (Cantons of Zurich and Geneva) and two past hot summers (2018 and 2022) this system is able to predict individual heat-related mortality peaks up to two weeks ahead and anticipate longer-lasting periods of heat-related excess mortality up to four weeks in advance. For the alpine sector, we show that individual summer heatwaves can play an important role in accelerating permafrost thawing, even though the process is driven by long-term climate change. We demonstrate this with idealized sensitivity experiments with the SNOWPACK model (a physical model that predicts the evolution of the snowpack and the ground temperature below). They indicate that both the duration of heatwaves as well as their timing within an individual summer are important for the intensity of the ground warming in permafrost regions. In summary, this project demonstrates a large potential for using subseasonal heatwave predictions for early warning systems for the health sector. For the alpine sector, it highlights the potential importance of individual heatwaves for permafrost thawing and raises the question if subseasonal heatwave predictions could support monitoring and early warning systems in high-elevation areas in some way.