Climate Vulnerability and Economic Development in Europe (1995-2024)

The project aims to analyze if richer European countries are less vulnerable to climate extremes than poorer ones. The considered period ranges from 1995 to 2024 and every country of the EU26 was considered except Malta.

Research Question

Are richer European countries less vulnerable to climate extremes than poorer ones?

The underlying hypothesis is that richer economies can better adapt to climate extremes, and therefore experience lower damages relative to their population or economic size.

Data Source

Yearly datasets concerning Gross Domestic Product (GDP), Household Income, Government Expenditure, Population, Climate Disasters and Climate Economic Losses were applied and are available inside the Data folder.

• The annual GDP dataset is available at Gross domestic product (GDP) and main components (output, expenditure and income), provided by Eurostat. The unit of measurement is Chain linked volumes (2010), million euro over the period 1995-2024;

• The annual Household (HH) Income dataset is available at Mean and median income by household type, provided by Eurostat. The Mean equivalized net income expressed in Euro for the total of the HH types was considered over the period 1995-2024. Computing of missing values was necessary due to the lack of several values. The applied computing process is explained in the Methodology section;

• The annual Government Expenditure dataset is available at Gross domestic product (GDP) and main components (output, expenditure and income), provided by Eurostat. The Final Consumption Expenditure Of General Government with unit Chain linked volumes (2010), million euro was considered over the period 1995-2025;

• The annual total Population dataset is available at Population on 1 January by age and sex, provided by Eurostat. No distinction of age and sex was applied, over the period 1991-2025;

• The annual Climate Related Economic Losses dataset is available at Climate related economic losses by type of event, provided by Eurostat over the period 1995-2024. Values are related to the sum of:

  • meteorological (storms, avalanches);
  • hydrological (floods);
  • climatological events (heatwaves, cold waves, droughts, forest fires).

Unit of measurements are million Euros and Euros per inhabitant, both at 2022 constant prices. The events are based on the classification by International Council for Science (ICSU). Originally, Eurostat receives the data from the CATDAT of RiskLayer through the European Environmental Agency (EEA) under institutional agreement, with numbers adjusted to account for inflation. Although the base years differ (2010 for GDP and 2022 for climate losses), both variables are expressed in real terms. Therefore, no additional deflation was applied, as the analysis focuses on relative comparisons rather than absolute levels;

• The Climate Disasters dataset is provided by EM-DAT. The dataset contains the climate disaster and its impacts. It was filtered for the each of the EU27 countries over the period 1995-2024. According to the Climate Related Economic Losses dataset, the considered climate disasters are:
  • Drought;
  • Wildfire (Forest Fire, Land Fire (Brush, Bush, Pasture), Wildfire (General));
  • Mass Movement (dry)(Avalanche (dry), Landslide (dry), Rockfall (dry), Sudden Subsidence (dry));
  • Flood (Coastal Flood, Flash Flood, Flood (General), Ice Jam Flood, Riverine Flood);
  • Extreme Temperature (Cold wave, Heat Wave, Severe Winter Conditions);
  • Storm (All Classifications).

The datasets contains values also that explain the Human Impact of each climate disaster as follow:

• Total Deaths: includes confirmed fatalities directly imputed to the disaster plus missing people whose whereabouts since the disaster are unknown and so they are presumed dead based on official figures;
• No. Injured: when the term "injured" is written in the source. Any related word like "hospitalized" is considered as injured. If no precise number is given, such as "hundreds of injured," 200 injured will be entered (although this figure is probably an underestimate);
• No. Affected: If only the number of families affected or houses damaged are reported, the figure is multiplied by the average family size for the affected area (×5 for developing countries, ×3 for industrialized countries, according to the UNDP country classification);
• No. Homeless: mentioned whenever it is found in reports. If only the number of families that are homeless or houses that are destroyed are reported, the figure is multiplied by the average family size for the affected area (x5 for developing countries, x3 for industrialized countries, according to the UNDP country list);
• Total Affected: is the total of injured, affected, and homeless people. This project assumes Human Impact = Total Deaths + Total Affected as a human-related proxy of the climate disasters. The missing values in the variables "Total Deaths" and "Total Affected" were replaced with zero. This assumes that missing entries correspond to events with no reported human impacts. This approach avoids discarding disasters with incomplete reporting but may lead to a slight underestimation of total impacts. This lead to around 14% of the events (138/985) with assumed no reported human impacts. The Year associated to each climate disaster is the column Start Year in the dataset.

Methodology

The study applied the following procedure:

(a) Manipulation of each dataset through Python;

(b) Analysis of the manipulated Climate Disasters and HH Income datasets;

(c) Merge of all the singular datasets in a final one;

(d) Computing of the visualizations and following analysis of the results.

(a) All the scripts where datasets are manipulated are in the Scripts folder. To manipulate the data, Python packages Pandas and NumPy were applied. The manipulation consisted in the following steps:

• Dataset import;
• Missing values check;
• Data filter of the data to have year range [1995-2024], limits included, and countries of the EU26 (No Malta);
• For specific datasets, additional variables were calculated such as the GDP per capita (€/person), Human Impact (Total Deaths + Total Affected), Fatality Rate (Total Deaths/Human Impact);
• Export of the elaborated dataset in .csv format in the Code_Output folder;

(b) The analysis of the manipulated Climate Disasters and HH Income datasets is performed using RStudio and several libraries. Here it is explained what the analysis consisted of:

• HH Income dataset: first, the needed packages are imported. Then it proceeded to deal with the HH Income gaps present in the dataset. In the Notes folder there is a .txt file called "HH_Income_Countries_Gaps" that lists the missing year for each country over the considered period. For countries that have small inside gaps(< 4 years) a linear interpolation based on the available previous and following years is conducted, whereas for countries that misses the income for previous years, this is calculated by adjusting the next year's income (available) down by GDP growth, which ensures that the trend of early years is consistent with macroeconomic growth, not just constant values. Then, for both the procedures a visual check through plots is conducted to confirm the results and the final dataset is exported in the Code_Output folder;

• Climate Disasters dataset: first, the needed packages are imported. Then every climate disaster is analyzed in terms of Total Deaths, Total Affected, Human Impact and Fatality Rate, with a chart as well. After this, annual EU26 aggregates are computed to calculate normalized indexes for Total Deaths, Total Affected and Total Number of Climate Disasters based on year 1995 for further analysis. The indexes are then plotted. The final step is the export of each visualization;

(c) The merge allows to have a final dataset with all the needed variables for each country over the considered period. Additionally, the Climate Damages per Capita [€/person] and the Climate Vulnerability Index (Climate Economic Losses/GDP) are computed;

(d) The final charts are available in the folder Visualizations folder inside the Output folder. The plots are performed with RStudio. The next step is the results' analysis.

NOTE: Due to the backward extrapolation, it is not possible to calculate values of the year 1995 for countries that did not reported it or for which it is not available.

Key Findings

The analysis of the Climate Disaster dataset for the EU26 countries, in terms of EU aggregates, over the observed period shows the following:

• The climate extremes have different impacts in terms of total casualties and total affected people. While extreme temperatures showed high death numbers but less affected, floods and storms showed the opposite. Instead, wildfires did not show any predominance;
• The average deaths count per disaster trend shows a slight increase from 2019. This might be due to the addition of the COVID-19 pandemic effect, even though it was not analyzed in this project. The average deaths per disaster peak happened in 2003, where a big portion of Europe was affected by very high extreme temperatures. On the other hand, the average affected number per disaster trend shows a slight decrease over the observed period. Still, it is possible to notice a small increase after 2020, which might be related to the COVID-19 pandemic. The peak for this measure happened in 1999, when France was heavily hit by the storms "Lothar" and "Martin" that had consequences on other European countries. In general, it seems that the average deaths number increased whereas the average affected number decreased over the observed period;
• Referring to the singular extreme events, the file Extremes_Highest_Country_Year in the Notes folder shows in which country and year the extreme event with the highest number of Total Deaths, Total Affected and Human Impact happened for each considered type of event.

The analysis of the relationship between averaged Climate Damages per capita and averaged GDP per capita for the selected countries over the observed period pointed the following:

• The countries with the highest averaged GDP per capita over the period of time are Luxembourg, Denmark, Ireland, and the Netherlands. On the other side, Bulgaria, Romania, Latvia and Poland are the countries with the lowest averaged GDP per capita;
• The trend is flat, even though it is possible to notice a very slight positive slop suggesting a weak positive relationship between GDP per capita and climate damages per capita. Slovenia showed the highest Damage per capita due to the effects of the 2023 flood and 2014 extreme temperatures. On the other hand, Finland showed the lowest Damage per capita while recording a high GDP per capita. In general, data did not show a clear distinction between higher/lower GDP per capita and Climate Damages per capita. Therefore, a higher GDP per capita does not necessarily turn into lower Climate Damages per capita. While there are countries with high GDP per capita and low Climate Damages per capita such as Finland, data shows that there are countries with both high GDP per capita and Climate Damages such as Germany, Italy and Austria. Therefore, the initial hypothesis of the project is not supported by the data;
• The result is related to many factors: country extension, territory's characteristics, population, types of extreme events that happened. Nonetheless, in absolute terms it does not seem that a higher GDP per capita means lower Climate Damages per capita.

The analysis of the Climate Damages per capita for the selected countries over the observed period shows the following:

• a small increase for the last years. This increase might be influenced also by the COVID-19 pandemic effect, where countries probably had to focus more on the epidemic rather than on climate extremes event. Generally, the reported annual total number of climate disasters seems to slightly increase over the observed period. The reasons might be many, from a better accuracy, higher, a broader monitoring or other factors such as the climate change feedbacks or impacts. Anyway, the project did not analyzed deeper the reasons;
• 2023 was the year with the highest number of reported climate extreme events for the selected European countries (65). 2022 was the third highest year in terms of the same value over the observed period (52) and the average of the latest years (2018-2024) is almost 45 extreme evens per year, which appear to be in an increasing trend. This might result in a heavier pressure and slowly weakening the resilience to climate extremes.

The analysis of the Climate Vulnerability Index for the selected countries over the observed period shows the following:

• the index is measured dividing the averaged Climate Economic Losses by the averaged GDP;
• results show that European countries with lower GDP have a higher Climate Vulnerability. Slovenia has the highest Climate Vulnerability index value (climate disaster in 2014 and 2023). In general, Slovenia, Romania, Czechia, and Croatia are classified as "High Vulnerability" countries to climate extremes;
• Results might be influenced by the reported GDP value of the country. Anyway, results show that countries with low GDP or similar characteristics that reported a low Climate Vulnerability Index.

Therefore, results showed a complex picture. While a clear relationship showing that countries with higher GDP per capita did not experience lower Climate Damages per capita was not reported, some countries with low GDP recorded a higher Climate Vulnerability Index.

Overall, the results suggest that economic development alone is not sufficient to explain climate vulnerability, which appears to depend on a combination of geographic, structural, and policy-related factors.

Further analysis with the additional variables available in this project can lead to more detailed and insightful results.

Limitations

• This project only on specific climate extremes (storms, avalanches, floods, heatwaves, cold waves, droughts, forest fires), ignoring other major ones. The EM-DAT public dataset concerns several others climate extremes, therefore data for those events are available in that dataset;
• Reporting data regarding the damages, people affected or casualties due to climate extremes is extremely complicated and hard. Therefore, the reported data might underestimate the real impacts and limit the precision of this analysis;
• The estimation process applied in for the gaps in the Household Income dataset does not bring the exactly accurate data. While the calculated results with both the linear interpolation and the backward interpolation were visually checked, the true values might have been over- or underestimated;
• Filling the missing values for the reported climate extremes which reported no Total Deaths and/or Total Affected with 0 assumed that the corresponding events did not reported any human impact. This might underestimate it actually. As mentioned above, reporting damages and effect of climate extremes is complicated and might lead to missing data. However, filling these gaps with 0 avoided to discard several events (around 14%) with no reported human impacts for the selected climate extremes, countries and period;
• Applying GDP per capita as a measure of social welfare and economic growth has several limitations. As van den Bergh, 2008 noted, GDP per capita is very often appointed as measure of the "standard of living". The author contested, though, its role and efficiency in providing a real measure of the economic reality. Particularly, GDP per capita does not capture the majority of social costs and omits external costs, states that individual welfare is affected by various income-dependent factors which means that aggregating individual incomes to obtain GDP is not always a robust indicator of social welfare and several other limitations. On the other hand, the author points out that GDP is an efficient indicator to express different economic variables with only one indicator and allows to quickly compare countries. This study applied GDP per capita, but other economic indicators are available in the Data folder which can be used to perform analysis from other perspective.

Repository Contents

Data folder containing:

• a folder for each dataset with the data in .csv format. The folders are:
  • Climate Disasters' records for each EU country (Malta included, Liechtenstein as well) for the period 1995-2024;
  • annual Climate Economic Losses for each EU country (Malta included) for the period 1995-2024;
  • annual GDP for each EU country (Malta included) for the period 1995-2024;
  • annual Government Expenditure for each EU country (Malta included) for the period 1995-2025;
  • annual Household Income for each EU country (Malta included) for the period 1995-2024;
  • annual Population for each EU country (Malta included) for the period 1991-2025

Notes folder containing:

• the .txt file that lists and explain the procedure to fill the missing value for the HH Income dataset

Output folder containing two folders:

• Code_Output folder, which contains all the exported datasets in format .csv from the scripts after their manipulation and analysis;
• Visualizations folder, which contains:
  • Climate Disasters Impacts in the EU (Index 1995 = 100);
  • Climate Disasters' Total Deaths and Total Affected for the EU plot;
  • Average Deaths per Climate Disaster in the EU;
  • Avg. GDP per capita vs Avg. Climate Damage per capita (EU countries);
  • Damages_Capita folder with two visualizations:
    • Climate Damages per Capita in EU Countries(log scale);
    • Climate Damages per Capita in EU Countries;
  • Vulnerability_Index folder with two visualizations:
  • Climate Vulnerability Index by Country;
  • Climate Vulnerability Index by Country(High Vulnerability/Other Countries)

Scripts folder which contains all the scripts

Bibliography

• EM-DAT Climate Disasters (https://public.emdat.be/)
• European Environmental Agency (EEA) (https://www.eea.europa.eu/en)
• Eurostat (https://ec.europa.eu/eurostat/databrowser/explore/all/all_themes?lang=en&display=list&sort=category)
• Jeroen C.J.M. van den Bergh, The GDP paradox, Journal of Economic Psychology, Volume 30, Issue 2, 2009, Pages 117-135, ISSN 0167-4870, https://doi.org/10.1016/j.joep.2008.12.001