Assessing Seasonal Climate Forecasts to PredictTemperature-Related Mortality in Cyprus

Seasonal climate predictions offer an important opportunity to anticipate weather patterns several months ahead, enabling proactive planning and risk management across sectors that are sensitive to climate variability, including agriculture, energy production, and disaster response. However, their use in public health planning remains limited, especially in regions vulnerable to extreme weather events, such as the Eastern Mediterranean. This study investigates the potential application of seasonal forecasts in predicting temperature-related mortality risks in Cyprus—a Mediterranean island increasingly affected by summer heat extremes. Although extreme cold temperatures are less common and often understudied in the region, they still pose significant health threats, particularly among vulnerable populations.

We analyze daily temperature data from ERA5-Land reanalysis and national mortality records for the period 2004–2016 to quantify the health impacts of both heat and cold extremes. Statistical and machine learning approaches, including Distributed Lag Non-Linear Models (DLNMs) and Generalized Additive Models (GAMs), are applied to estimate relative risk profiles and attributable fractions of mortality. To evaluate the potential applicability of seasonal forecasts in impact modeling — including health risks — we use temperature data derived from simulations produced with the Advanced Research WRF (WRF-ARW) model. These runs are ERA5-driven and are used as reference hindcast simulations ("perfect model") developed under the PREVENT Horizon project. PREVENT aims to establish a framework for improving the quality and usability of seasonal climate information across the Mediterranean region, building on recent efforts to enhance climate predictions for decision-making. Hindcasts at three- and six-month lead times for the 2004–2016 period are compared with ERA5-Land results to evaluate their ability to reproduce observed temperature–mortality associations. Preliminary results highlight both the potential and limitations of this approach. While some consistency in risk estimation was observed—particularly at shorter lead times—discrepancies remain, especially for forecasts extending further into the future emphasizing the need for further refinement of seasonal models. These results highlight the potential of integrating seasonal climate information into public health early warning systems and underscore its value in enhancing climate resilience across the Eastern Mediterranean and other climate-vulnerable regions.

Acknowledgments: The work was supported by PREVENT project. This project has received funding from Horizon Europe programme under Grant Agreement No: 101081276.