A Comparative Analysis of the Temperature‐Mortality Risks Using Different Weather Datasets Across Heterogeneous Regions

Abstract New gridded climate datasets (GCDs) on spatially resolved modeled weather data have recently been released to explore the impacts of climate change. GCDs have been suggested as potential alternatives to weather station data in epidemiological assessments on health impacts of temperature and climate change. These can be particularly useful for assessment in regions that have remained understudied due to limited or low quality weather station data. However to date, no study has critically evaluated the application of GCDs of variable spatial resolution in temperature‐mortality assessments across regions of different orography, climate, and size. Here we explored the performance of population‐weighted daily mean temperature data from the global ERA5 reanalysis dataset in the 10 regions in the United Kingdom and the 26 cantons in Switzerland, combined with two local high‐resolution GCDs (HadUK‐grid UKPOC‐9 and MeteoSwiss‐grid‐product, respectively) and compared these to weather station data and unweighted homologous series. We applied quasi‐Poisson time series regression with distributed lag nonlinear models to obtain the GCD‐ and region‐specific temperature‐mortality associations and calculated the corresponding cold‐ and heat‐related excess mortality. Although the five exposure datasets yielded different average area‐level temperature estimates, these deviations did not result in substantial variations in the temperature‐mortality association or impacts. Moreover, local population‐weighted GCDs showed better overall performance, suggesting that they could be excellent alternatives to help advance knowledge on climate change impacts in remote regions with large climate and population distribution variability, which has remained largely unexplored in present literature due to the lack of reliable exposure data.

S12. Absolute differences in mortality fractions for heat and cold estimated by the GCD exposure datasets and the weather station, plotted against regional characteristics for Switzerland Figure S13. Absolute relative risk (RR) for heat and cold estimated by the GCD exposure datasets and the weather station, plotted against regional characteristics for Switzerland Figure S14. Absolute differences in mortality fractions for heat and cold estimated by the GCD exposure datasets and the temperature monitor, plotted against regional characteristics for England and Wales Figure S15. Absolute differences in relative risk (RR) for heat and cold estimated by the GCD exposure datasets and the weather station, plotted against regional characteristics for England and Wales Tables   Table S1. Descriptive statistics of the mortality data and regional-level indicators (area, population) and characteristics of the temperature datasets (weather station, local and global gridded climate datasets (GCD)) in England andWales (1993-2006) and Switzerland (1989Switzerland ( -2017 Table S2. Descriptive statistics of the observed temperature by exposure dataset for the 26 regions in Switzerland, 1989 -2017 Table S3. Descriptive statistics of the observed temperature by exposure dataset for the 10 regions in England andWales, 1993 -2006 Table S4. Relative risk (RR) for all-cause mortality (95%CI) for heat (99 th percentile) by exposure dataset for Switzerland Table S5. Relative risk (RR) for all-cause mortality (95%CI) for cold (1 st percentile) by exposure dataset for Switzerland Table S6. Relative risk (RR) for all-cause mortality (95%CI) for heat (99 th percentile) by exposure dataset for England and Wales Table S7. Relative risk (RR) for all-cause mortality (95%CI) for cold (1 st percentile) by exposure dataset for England & Wales Table S8. Goodness of fit for each exposure dataset measured by the qAIC (quasi-Akaike's information criterion) averaged over the 10 regions in England and Wales Table S9. Goodness of fit for each exposure dataset measured by the qAIC (quasi-Akaike's information criterion) averaged over the 26 regions in Switzerland Table S10. Annual excess number of deaths and mortality fractions (%) for cold ( ≤10 th , ≤25 th percentile) and heat (≥75 th and ≥90 th percentile) in Switzerland between 1989 and 2017 for each exposure dataset Table S11. Annual excess number of deaths and mortality fractions (%) for cold ( ≤10 th , ≤25 th percentile) and heat ( ≥75 th and ≥90 th percentile) in England & Wales between 1993 and 2006 for each exposure dataset Table S12. Annual number of excess deaths and mortality fractions (%) for cold, (≤10 th , ≤25 th percentile) and heat ( ≥75 th and ≥90 th percentile) for the four selected regions for each exposure dataset Table S13. Summary statistics for the annual excess number of deaths, mortality fractions and associated 95% CI for heat (≥ 90 th percentile and ≥75 th percentile) in Switzerland   Table S14. Summary statistics for the annual excess number of deaths, mortality fractions and associated 95% CI for cold (≤10 th percentile and ≤25 th percentile) in Switzerland   Table S15. Summary statistics for the annual excess number of deaths, mortality fractions and associated 95% CI for heat (≥ 90 th percentile and ≥75 th percentile) for England and Wales Table S16. Summary statistics for the annual excess number of deaths, mortality fractions and associated 95% CI for heat (≤10 th percentile and ≤25 th percentile) for England and Wales

Missing data Switzerland
Missing data amounted to 0.09% of the days. For temperature series with one or two missing days, we used a 4-day moving average to impute the values. In the canton of Fribourg, a total of 250 days were missing for the year 2006 and we assigned the value of the corresponding TabsD-grid cell for these missing days.

Missing data England & Wales
Missing data amounted to 0.00% of the days.
GCD processing and population weighted series. We extracted hourly (global GCD) or daily (local GCDs) mean temperatures for each grid cell for the corresponding period covering a specific region/canton. For the former, we aggregated hourly temperature observations for each region by day and created daily mean temperature averages for all grid cells throughout the regions and cantons. All cells that intersect the canton or region were included for the analysis.
We created two pairs of population-weighted and unweighted temperature series for each GCD and region. For the unweighted series (i.e. without accounting for population distribution), we estimated the average values across the cell-specific daily mean temperatures of those grid cells intersecting the boundaries of the corresponding region.
Additionally, we created a single population-weighted daily mean temperature for each region and GCD using EOSDIS gridded population data in 2000 on a 1x1 km grid resolution (UN WPP-Adjusted Population Count, v4.11 -2000) (Centre for International Earth Science Information Network -CIESIN -Columbia University. 2018). Population estimates have been created using national census and population registries based on the highest national administrative boundary available (which corresponds to the municipality level in Switzerland and Lower Super Output Areas level in the England and Wales). We summed the total population living in the region and additionally we summed the total population that residing within each grid cell using Geographic Information System methods. Then we computed the weights in each GCD-specific cell using the ratio between the population residing in the corresponding grid cell and the total population within that region. Therefore, the contribution of the grid cell towards the full time series for a region is dependent on the population residing within the grid cell relative to the overall-region specific population. Finally, we computed weighted-mean daily series for each region using mean daily temperatures of all cells in that region and the derived weights. Thus,  Descriptive statistics of the mortality data and regional-level indicators (area, population) and characteristics of the temperature datasets (weather station, local and global gridded climate datasets (GCD)) in England andWales (1993-2006) and Switzerland (1989Switzerland ( -2017 Table S10. Annual excess number of deaths and mortality fractions (%) for cold ( ≤10 th , ≤25 th percentile) and heat (≥75 th and ≥90 th percentile) in Switzerland between 1989 and 2017 for each exposure dataset    Table S14. Summary statistics for the annual excess number of deaths, mortality fractions and associated 95% CI for cold (≤ 10 th percentile and ≤ 25 th percentile) in Switzerland ≤ 10 th percentile ≤ 25 th percentile Table S16. Summary statistics for the annual excess number of deaths, excess mortality fractions and associated 95% CI for cold (≤10 th percentile and ≤25 th percentile) for England and Wales