This repository contains data to reconstruct the exposure-response functions (ERF) of temperature-related mortality by five 5 age groups in 854 cities in Europe. These ERFs have been derived in the study by Masselot et al. 2023, Excess mortality attributed to heat and cold: a health impact assessment study in 854 cities in Europe, The Lancet Planetary Health (https://doi.org/10.1016/S2542-5196(23)00023-2). An associated semi-replicable GitHub repository is available at https://github.com/PierreMasselot/Paper--2023--LancetPH--EUcityTRM to reproduce part of the analysis and the full results, as well as to provide technical details on the derivation of these ERFs. Note: This updated version contains revised data after the correction of an error in the code related to the computation of the age-specific baseline mortality rates. Details about the error can be found in the GitHub repository linked above. This correction only affects the figures of excess mortality (found in the results.zip archive) while the ERFs are negligibly affected. The originally published results can be found in V1.0.0 of this repository. Extraction of the ERFs The ERFs are provided as coefficients of B-spline functions that can be used to reconstruct the ERFs, along with variance-covariance matrices and quantiles from location-specific temperature distributions. The parametrisation associated with these coefficients is a quadratic B-spline (degree 2), with knots located at the 10th, 75th and 90th percentiles of the temperature distribution. In R, the associated basis can be constructed using the dlnm package, with a temperature series x, as follows:

library(dlnm) basis <- onebasis(x, fun = "bs", degree = 2, knots = quantile(x, c(.1, .75, .9))) The main files associated with ERFs are the following: coefs.csv: The B-spline coefficients for each age group and city. vcov.csv: The variance-covariance matrix of the coefficients in each city and age group. It is provided here as the lower triangular part of the matrix with names indicating the position of each value (v[row][column]). In R, assuming x is a row of this file, the matrix can be reconstructed using xpndMat(x) after loading the mixmeta package. coef_simu.csv: 1000 simulations from the distribution of each city and age-specific coefficients. Useful to derive empirical confidence intervals for derived measures such as excess deaths or attributable fractions. tmean_distribution.csv: The city-specific temperature percentiles representing the distribution of the data derived from the ERA5-Land dataset. Health impact assessment results results.zip: A summary of the results from the health impact assessment reported in the analysis. The dataset includes several impact measures provided in files representing different geographical levels, including city, country and regional level. Different files are also provided for age-group specific or all age results. Additional data We provide additional data that are useful to reproduce or extend the analysis. Please note that due to restrictive data-sharing agreements for the mortality series, only a part of the code is reproducible. See the associated GitHub repository for more details. metadata.csv: City-specific metadata used to create the ERFs and perform the health impact assessment. additional_data.zip: contains further data used to replicate the second stage of the analysis and the final health impact assessment. It includes the full city-level daily temperature series (era5series.csv), the detail of extracted metadata for available years (metacityyear.csv), a description of the city-level characteristics (metadesc.csv), and the first-stage ERF coefficients for all available city and age-groups (stage1res.csv). Additionally, the file meta-model.RData contains R object defining the second-stage model that can be used to predict new ERFs.

This repository contains the data and results from the paper Estimating future heat-related and cold-related mortality under climate change, demographic and adaptation scenarios in 854 European cities published in Nature Medicine (https://doi.org/10.1038/s41591-024-03452-2).

It provides projections of excess death rates and burden for the period 2015-2099 for five age groups in 854 cities across 30 countries, under three Shared Socioeconomic Pathway (SSP) scenarios, and four adaptation scenarios. The results include point estimates for five-year periods and four global warming levels, along with 95% empirical confidence intervals.

The fully reproducible analysis code using the data and producing the results included in this repository is provided in GitHub. The results can be visualised and explored in a dedicated Shiny app.

Content

This repository contains three zip files, each with an internal codebook:

• data.zip: contains the input data necessary to run the analysis. It includes historical and projected daily temperature at the city level, age-group specific projections of population and survival rates at the country level, and exposure-response functions extracted from another Zenodo repository (https://doi.org/10.5281/zenodo.10288665). This file also include a script showing how each dataset was extracted for the purpose of this projection study.
• results_csv.zip: contains the full results from the health impact projections. It includes one file for each combination of geographical level (city, country, region or European wide) and scale of reporting (five year periods or global warming levels).
• results_parquet.zip: contains the same information as the results_csv.zip but in a parquet format. This allows for more efficient storage and data reading.

It is recommended to only download results_csv.zip for a quick exploration of the results, or only results_parquet.zip when the results are to be loaded into a software for deeper analysis.

These data contain estimates of temperature-related human mortality, as well as the associated economic assessments, related to urban heat islands for 85 European cities over the years 2015-2017. They are based on temperature-mortality relationships from Masselot et al. 2023 and 100m resolution UrbClim urban climate model simulations of near-surface air temperature (De Ridder et al. 2015, Hooyberghs et al. 2019), re-gridded to 500m resolution.

Details of the methodology are provided in the associated paper:

Huang, W.T.K. et al. Economic valuation of temperature-related mortality attributed to urban heat islands in European cities. Nat Commun 14, 7438 (2023). https://doi.org/10.1038/s41467-023-43135-z

And associated core analysis code is available on GitHub at https://github.com/hkatty/Paper_UHI_mortality_Europe (doi:10.5281/zenodo.8429209).

The content of the files are as follows:

spatial_timeseries zip files: These contain the most unprocessed attributable fraction estimates, with the exposure-response relationships applied to the modelled temperature, prior to any further processing.

uhi csv files: These are tables of the average mortality and years of life lost, as well as associated economic assessment, related to urban heat islands for each city. They are identical to Tables S4-S11 in the supplementary materials of the above paper.

spatial_maps_time_averaged_diff_from_rural.zip: Spatial maps showing the difference from the rural average for each day and grid box, then averaged over time.

data_urbanruralavg_timeseries.nc: Time series of urban and rural averages, as well as the difference between the two (i.e. the urban heat island effect).

avg_diff_from_rural_urbanrural.nc: The above timeseries file temporally aggregated.

simulated_urbanruraldiff_timeseries.zip: Time series of urban-rural difference in attributable fraction for 1000-member ensembles representing uncertainties in the exposure-response relationships as captured by Monte Carlo simulations.

simulated_urbanruraldiff_averaged.zip: The above simulated timeseries temporally aggregated.

Some variables explained:

fAF = forward attributable fraction (i.e. fraction of total mortality associated with a single day's temperature, cumulative over lag time)

fAD = forward attributable deaths (i.e. equivalent to fAF but for number of deaths)

tas = temperature

heat_ex = average over heat extreme days (i.e. the warmest 2% days in 2015-2017 for the city)

cold_ex = average over cold extreme days (i.e. as heat_ex but for the coldest 2% days)

heat = average over days warmer than the age-dependent optimal temperature

cold = average over days colder than the age-dependent optimal temperature

heat_count = number of days warmer than the optimal for the age group, note that for combined 2085.1 and 2085.5 age groups, days are counted if it is considered warm for at least one age group (therefore heat_count + cold_count ≠ total days over period)

cold_count = number of days colder than the optimal for the age group

rural = rural average

imd = land imperviousness

popden = population density

age groups:

20 = 20 to 44
45 = 45 to 64
65 = 65 to 74
75 = 75 to 84
85 = 85 and over
2085.1 = all above age groups combined, weighted by the local population age structure
2085.5 = all above age groups combined, weighted by the 2013 European standard population age structure

References:

De Ridder, K., Lauwaet, D., and Maiheu, B., (2015): UrbClim – A fast urban boundary layer climate model. Urban Climate, 12, 21–48. https://doi.org/10.1016/J.UCLIM.2015.01.001.

Hooyberghs, H., Berckmans, J., Lauwaet, D., Lefebre, F., and De Ridder, K., (2019): Climate variables for cities in Europe from 2008 to 2017. Copernicus Climate Change Service (C3S) Climate Data Store (CDS). https://doi.org/10.24381/cds.c6459d3a.

Masselot et al. (2023): Excess mortality attributed to heat and cold: a health impact assessment study in 854 cities in Europe, The Lancet Planetary Health, https://doi.org/10.1016/S2542-5196(23)00023-2.

The death cases and age-standardized mortality rates from pancreatic cancer, as well as their temporal trends by sex and super-region from 1990 to 2019.