ANALYSIS OF WESTERN EUROPE’S CONTRIBUTION TO GLOBAL WARMING IN 2022
KABORE Houzeifa
OUEDRAOGO Barakissa
ZEBOU Boga Roch Fourier
09/10/2025
📦 Loading the necessary packages
# Chargement des bibliothèques utilisées dans l'analyse
library(FactoMineR)## Warning: le package 'FactoMineR' a été compilé avec la version R 4.4.3library(factoextra)## Warning: le package 'factoextra' a été compilé avec la version R 4.4.3## Le chargement a nécessité le package : ggplot2## Warning: le package 'ggplot2' a été compilé avec la version R 4.4.3## Welcome! Want to learn more? See two factoextra-related books at https://goo.gl/ve3WBalibrary(psych)## Warning: le package 'psych' a été compilé avec la version R 4.4.3##
## Attachement du package : 'psych'## Les objets suivants sont masqués depuis 'package:ggplot2':
##
## %+%, alphalibrary(ggplot2)
library(Factoshiny)## Warning: le package 'Factoshiny' a été compilé avec la version R 4.4.3## Le chargement a nécessité le package : shiny## Warning: le package 'shiny' a été compilé avec la version R 4.4.3## Le chargement a nécessité le package : FactoInvestigate## Warning: le package 'FactoInvestigate' a été compilé avec la version R 4.4.3library(shiny)
library(FactoInvestigate)
library(ggcorrplot)## Warning: le package 'ggcorrplot' a été compilé avec la version R 4.4.3library(GGally)## Warning: le package 'GGally' a été compilé avec la version R 4.4.3## Registered S3 method overwritten by 'GGally':
## method from
## +.gg ggplot2THEME : THE GLOBAL WARMING
Summary
This study aims to identify and prioritise the main causes of global warming in Western Europe in 2022, based on sectoral data on CO₂ emissions. A cross-methodological approach using principal component analysis (PCA), hierarchical ascending classification (HFA) and multiple linear regression was adopted. The ACP highlights a concentration of emissions around the transport, industry and fossil fuel sectors, forming a dominant dimension of carbon intensity. The HFC distinguishes three groups of countries according to their emissions profiles, while the regression confirms the significant effect of transportation and cement emissions on oil consumption. The study thus makes it possible to structure the causes of global warming, to compare national situations and to formulate appropriate recommendations to guide regional climate policies.
List of abbreviations
ACE : Annual CO₂ emissions
ACC: Anual CO₂ emissions from cement
ACI: Annual CO₂ emissions from other industry (émissions de l’industrie)
ACO: Annual CO₂ emissions from oil
ACCO: Annual CO₂ emissions from coal (émissions dues au charbon)
ACG: Annual CO₂ emissions from gas (émissions dues au gaz)
ACF: Annual CO₂ emissions from flaring (émissions dues au torchage)
ACT: CO₂ emissions from transport (émissions du transport)
ACGP: Annual CO₂ emissions growth (%)
ACL_L : Annual CO₂ emissions from land-use change (émissions dues au changement d’usage des terres)
PGP: Population growth (annual population growth in %)
CPC: CO₂ emissions per capita
Keywords
Global warming
Western Europe
CO₂ emissions
Principal Component Analysis (PCA)
Hierarchical ascending classification (CAH)
linear regression
Transport
Industry
Fossil fuels
Environmental Forecasting
Differentiated climate policies
INTRODUCTION
For several decades, climate change has been one of the major challenges facing humanity. It is scientifically recognized that this global phenomenon is largely induced by human activities, particularly since the Industrial Revolution, with a marked increase in greenhouse gas (GHG) emissions such as carbon dioxide (CO₂), methane (CH₄) and nitrous oxide (N₂O). The latest report from the Intergovernmental Panel on Climate Change confirms that concentrations of these gases in the atmosphere have reached levels not seen in the last 800,000 years, leading to a continuous rise in global temperatures. In this global context, Western Europe, due to its high level of industrialisation and urbanisation, is a particularly exposed region. Indeed, 2022 was particularly revealing of regional climate vulnerabilities, with extreme weather events, such as prolonged heat waves, intense droughts and severe rainfall events. According to the annual report of the Copernicus Climate Change Service programme , Western Europe experienced its hottest summer on record, with temperatures on average 1.4°C higher than in the pre-industrial period. These events have not only amplified the region’s environmental vulnerabilities, but also led to important socio-economic consequences, such as a significant decline in agricultural productivity, an increase in heat-related health risks, as well as increased constraints on water resources and energy infrastructure . Faced with these growing challenges, it is becoming imperative to better understand the specific causes of global warming in Western Europe for the year 2022. This study is precisely in line with this perspective, seeking to answer the following question: WHAT ARE THE WESTERN EUROPE’S CONTRIBUTION TO GLOBAL WARMING IN 2022 ? Thus, the general objective of this article is to identify and analyze the key factors responsible for the global warming observed in this specific region. This objective is broken down into three specific areas:
To identify the economic sectors and human activities that contribute most to greenhouse gas emissions in Western Europe .
Analyze regional energy dynamics in detail, in particular the continued dependence on fossil fuels and their direct impact on increasing regional emissions .
Examine the complex interactions between natural and anthropogenic climate factors to accurately assess their combined contribution to observed warming .
We formulate the main hypothesis that:
Global warming in Western Europe in 2022 is mainly due to the persistence of an energy model that is highly dependent on fossil fuels, reinforced by traffic intensity and specific sectoral structural choices, despite the efforts made towards the low-carbon transition. This hypothesis is based on several scientific studies. For example, temperature increases are strongly associated with increased burning of fossil fuels in industrial economies . Similarly, the transport and energy production sectors are responsible for more than 70% of European emissions. Finally, the International Energy Agency confirms that the slow energy transition in European countries is a key factor in the persistence of high emissions. The aim of this study is to provide a detailed and up-to-date understanding of the mechanisms responsible for global warming at the regional scale, based on solid empirical data and validated theories. The aim is to contribute to the identification of priority levers for action for an effective reduction of emissions in Western Europe.
LITERATURE REVIEW
Greenhouse gases (GHGs), mainly carbon dioxide (CO₂), methane (CH₄) and nitrous oxide (N₂O), are the main contributors to global warming. In 2020, Western Europe accounted for nearly 8% of global CO₂ emissions, despite energy transition efforts. This proportion remains significant, in particular because of the industrial intensity and the density of road traffic in this region.
In addition, data compiled in the Global Carbon Budget confirms that CO₂ emissions in Europe, although relatively stable, are struggling to decline at the pace required by the goals of the Paris Agreement . In 2022, some countries such as Germany and France even saw a slight increase in their emissions, partly due to an increased use of coal to compensate for the disruptions in natural gas supplies caused by the war in Ukraine . Energy production and consumption are the largest source of GHG emissions in the region. The share of coal in electricity generation increased in Europe in 2022 for the first time in a decade, with notable increases in Germany, the Netherlands and Spain . These countries have temporarily reactivated some thermal power plants to deal with the energy crisis.
The transition to renewable energy remains hampered by slow grid integration and the still high costs of some technologies in some Member States . Even though wind and solar power represent a growing share of the European energy mix, their intermittency poses major challenges in terms of stability of supply . The transport sector accounts for around 25% of total GHG emissions in the European Union, with a major share coming from road transport. In 2022, the post-pandemic recovery in car and air traffic contributed to an increase in emissions, despite the rise of electric mobility .
The penetration of electric vehicles remains uneven between Western European countries, with a strong contrast between Norway and countries such as Italy or Spain . In addition, air travel resumed at a rapid pace in 2022, which generated further pressure on overall emissions . The agricultural sector is a major source of GHGs, particularly through methane emissions from livestock farming and nitrous oxide from nitrogen fertilizers.
Agriculture accounted for around 12% of total EU emissions, a proportion that remains stable despite mitigation policies . In Western Europe, cattle farming is particularly emitting, particularly in France, Ireland and Germany. The study on the emission balances of the European agricultural sector shows that the intensive use of mineral fertilisers increases N₂O emissions, while contributing to water and soil pollution . In 2022, drought conditions worsened the effect of these practices, reducing the natural absorption of carbon in agricultural soils.
In addition to anthropogenic causes, certain natural dynamics aggravate global warming. For example, prolonged droughts reduce the ability of forests and soils to store carbon. The 2022 heatwaves contributed to record forest biomass losses in regions such as the Iberian Peninsula[22]. Similarly, wildfires – aided by extreme temperatures – have released significant amounts of carbon. The study published in Environmental Research Letters shows that forest fires in Western Europe doubled their CO₂ emissions in 2022 compared to the ten-year average[23].
Despite the stated ambition of the European Green Deal, current climate policies do not yet fully meet the objectives of carbon neutrality. carbon pricing mechanisms and EU emission allowances have led to a slight reduction in emissions, but national disparities undermine their overall effectiveness[24]. In 2022, the energy crisis linked to the war in Ukraine led some countries to relax their environmental constraints, extending the life of coal-fired power plants or reducing subsidies for renewable energy[25]. This shows that climate policies remain vulnerable to geopolitical shocks.
To sustainably mitigate global warming in Western Europe, several levers can be mobilized. The massive development of renewable energies, the electrification of transport, agroecology and the restoration of ecosystems are identified as priorities[26]. The introduction of carbon adjustment mechanisms at the EU’s borders, as well as inter-state cooperation for the sharing of low-carbon technologies, are also among the recommendations issued by the EEA and the JRC[27].
However, these strategies require substantial funding, strengthened political will and the active involvement of citizens to produce systemic change that is commensurate with climate challenges. The intersection of disciplines makes it possible to better understand the complexity of the causes of global warming. For example, the work of Steffen et al. (2018) highlighted the concept of “socio-environmental trajectories” in which feedbacks between society, economy and climate can lead to tipping points. This idea highlights that several critical systems (ice caps, boreal forests, ocean currents) could be close to irreversible thresholds, accentuated by European emissions[28].
Moreover, the economic dimension cannot be ignored. A study by Burke
et al. (2015) shows that economic growth could be severely penalized in
temperate regions such as Europe if the average temperature rises by
more than 2°C, leading to lower productivity, higher energy costs, and
greater social instability. Thus, the climate challenges in Western
Europe are not only environmental, but also geopolitical, economic and
health.
Finally, high-spatial resolution analyses are being developed to better
understand regional disparities. For example, van der Schrier’s work on
local climate data shows that some areas of France, Belgium and Spain
show an acceleration of warming higher than the European average[29].
This justifies a differentiated approach in policy responses and
adaptation strategies.
METHODOLOGY
Studying the causes of global warming in Western Europe requires a
rigorous scientific approach because of the multidimensional complexity
of the problem.
Indeed, several interrelated anthropogenic factors contribute to the
increase in temperatures. The massive use of fossil fuels and
deforestation emit considerable amounts of greenhouse gases (GHGs) into
the atmosphere, reinforcing the natural greenhouse effect and leading to
global warming.
In the face of this reality, it is essential to adopt advanced quantitative analysis in order to unravel the respective influence of each factor and provide objective conclusions based on the data. Recent work also highlights that multiple dimensions – demographic, economic, environmental – are involved in GHG emissions. This justifies the use of robust statistical tools to simultaneously analyze these variables and avoid any interpretation bias. Thus, the methodology implemented stems from an in-depth review of the scientific literature and aims to reliably quantify the impact of each explanatory variable on the phenomenon studied, rather than relying on qualitative intuitions
Materials used in the project
• The Excel and Word software of the Office suite were essential for the capture, organization and graphical representation of our research, observations, analyses and conclusions. Their versatility has made it easier for us to interpret and understand the results obtained.
• KoboCollect has been a valuable tool for collecting the data needed for our analysis. This user-friendly and flexible application made it easy to create and administer questionnaires, providing an efficient method for collecting data in the field.
• R Studio, as a data analysis software, was essential in carrying out our statistical analyses. Thanks to its advanced features, we have been able to perform factor analysis, clustering methods, variable cross-referencing, statistical testing and much more. Its accessibility and versatility make it an essential tool for researchers working with complex data.
• Finally, QGIS has proven to be a powerful tool for the visualization and spatial analysis of our data. As GIS software, it allowed us to create custom maps and explore the geographic relationships between the different elements of our study. Its advanced features helped us to better understand the themes addressed in our project, highlighting the geographical and spatial aspects of the data.
📁 Data import
# Vérifie le répertoire de travail
getwd()## [1] "C:/Users/Lenovo/Desktop/PROJET RTI/analyse"# Lecture des données depuis un fichier CSV. S'assurer que le fichier est dans le bon répertoire.
data <- read.csv(file = "classeur1.csv", header = TRUE, sep = ";", quote = "\"", dec = ".", row.names = 1)
data## ACC ACE ACF ACI ACG ACO ACCO
## Austria 1831572 61489110 80240 1210193 15967127 30733780 11666200
## Belgium 2456034 89002056 111066 1592600 31086884 41814188 11941285
## France 6396732 293501630 1591822 3725112 76353304 179440910 25993764
## Germany 12537957 671471550 1835065 8242323 160775700 247440220 170640270
## Ireland 1956535 36711388 200 735816 10745410 19237046 4036380
## Italy 7918827 340672300 2549430 3277906 137601070 157948580 31376494
## Luxembourg 368935 7276724 31 12425 1249039 5409675 236619
## Portugal 2228547 40687052 1174634 803425 11367924 24867992 244531
## Spain 7714822 234657020 3824197 2643633 67501400 135100960 17872014
## Switzerland 1667094 32818890 21669 168148 6015103 24515458 431417
## United Kingdom 11726342 413834780 2191720 4812964 174871650 29520260 99665548
## ACT PGP ACGP ACL_C CPC
## Austria 22330000 0.9562878 -6.491059 -1891466.60 6.783374
## Belgium 22920000 0.8081654 -6.201172 2654677.80 7.645034
## France 118910000 0.3269925 -3.961164 -24758526.00 4.428381
## Germany 147240000 0.7208754 -5.076305 -8638173.00 7.985512
## Ireland 10620000 2.5991810 -2.218407 79985.12 7.184207
## Italy 98240000 -0.2023009 1.397252 -30496900.00 5.714146
## Luxembourg 4990000 2.0166678 -13.584030 136154.25 11.138139
## Portugal 15450000 0.6972565 1.000249 1113929.20 3.905804
## Spain 86540000 0.7214810 2.439284 -10521726.00 4.906229
## Switzerland 14620000 0.8299272 -8.275461 -366839.70 3.732736
## United Kingdom 154320000 1.1344483 -8.904034 9763057.00 4.603079# Aperçu rapide des données
head(data)## ACC ACE ACF ACI ACG ACO ACCO
## Austria 1831572 61489110 80240 1210193 15967127 30733780 11666200
## Belgium 2456034 89002056 111066 1592600 31086884 41814188 11941285
## France 6396732 293501630 1591822 3725112 76353304 179440910 25993764
## Germany 12537957 671471550 1835065 8242323 160775700 247440220 170640270
## Ireland 1956535 36711388 200 735816 10745410 19237046 4036380
## Italy 7918827 340672300 2549430 3277906 137601070 157948580 31376494
## ACT PGP ACGP ACL_C CPC
## Austria 22330000 0.9562878 -6.491059 -1891466.60 6.783374
## Belgium 22920000 0.8081654 -6.201172 2654677.80 7.645034
## France 118910000 0.3269925 -3.961164 -24758526.00 4.428381
## Germany 147240000 0.7208754 -5.076305 -8638173.00 7.985512
## Ireland 10620000 2.5991810 -2.218407 79985.12 7.184207
## Italy 98240000 -0.2023009 1.397252 -30496900.00 5.714146Analysis methodology
After having clearly defined the problem and the working hypotheses,
we detailed a quantitative analysis protocol in several steps in order
to rigorously answer the question of the causes of global warming. As a
first step, we built a Correlation matrix between all the quantitative
variables retained. This correlation matrix (based on the Pearson
coefficient) is used to assess the linear relationships between two
variables at once. In concrete terms, each coefficient measures the
intensity and direction of the link (positive or negative) between a
potential factor (e.g. GDP) and a climate indicator (e.g. the level of
CO₂ emissions). We were thus able to detect significant associations
indicating that one variable could influence another, or on the contrary
to identify significant associations indicating that one variable could
influence another. Independence (lack of correlation) suggesting that
some factors do not have a direct linear relationship with emissions.
For example, a positive correlation between a country’s economic wealth
and its emissions (a trend generally) is expected to be verified in
previous studies [30] while the share of renewable energy in the mix
could show a negative correlation with total emissions[30].
These preliminary bivariate analyses guided the rest of the study by
pointing out the links to be deepened.
In a second step, we performed a principal component analysis (PCA) to explore the multivariate structure of the data and reduce the dimensionality of the problem. PCA is a statistical method that synthesizes information from several correlated variables into a smaller number of components that are independent of each other. Each principal component is a linear combination of the initial variables maximizing the explained variance. In other words, the PCA extracts the major latent factors underlying the data, which makes it possible to summarize the complexity of the climate system in a few interpretable axes. In our study, the CPA was used to identify patterns or syndromes among Western European countries in 2022 – for example, routes between countries with high coal and oil consumption and countries with a greater reliance on natural gas and renewables. Technically, we retained the first principal components whose eigenvalues were greater than 1 (Kaiser’s criterion) and which, together, accumulated most of the total variance (generally more than 70%). These components were then analyzed to interpret their physical significance, examining the contribution of each initial variable on each axis. This exploratory step provides an overview of the main combinations of causes of warming in the region, while filtering out noise and redundancies between indicators.
Finally, multiple linear regression was used to explicitly model the
relationships between the explanatory factors and the target variable
reflecting warming. We have built a model where the dependent variable
represents the level of climate impact (e.g., the volume of CO₂
emissions per capita), and where the explanatory variables include the
selected socio-economic and energy indicators (population, coal use,
etc.). The objective of this regression is twofold: (1) to quantify the
influence of each factor by controlling for the others (via the
estimated regression coefficients, accompanied by tests of statistical
significance), and (2) to identify which variables contribute most
significantly to the variations in emissions between countries. The
ordinary least squares method was used to fit the model and calculate
the explained variance share (R²). We paid particular attention to the
model diagnostics (residue independence, multicollinearity between
predictors, etc.) in order to guarantee the validity of the inferences.
In short, this modelling approach makes it possible to prioritise the
causes of warming by quantifying their respective weight in the observed
emissions. For example, if the variable “coal” comes out with a strongly
positive and significant coefficient, it will indicate that at the same
level of GDP and population, a country that consumes more coal emits
significantly more CO₂ – supporting the idea that coal is a major
contributor to warming .
Conversely, a non-significant variable would suggest that its effect is
not decisive in the framework under consideration. Supported by the
findings of the previous PCA, multiple regression thus provides
statistical validation of the key drivers of global warming in Western
Europe.
Variable selection
Taking into account the above considerations, a set of relevant variables has been selected to cover the different facets of the potential causes of regional global warming. Each variable is defined below with its abbreviation, precise meaning and role in the analysis:
Annual-co2-cement (ACC)
(ACC) represents the annual CO₂ emissions related to cement production. This is usually the CO₂ released during the process of calcination of limestone (transformation of CaCO₃ into CaO) and the combustion of fossil fuels to heat the kilns. Cement production is a carbon-intensive industrial sector. Tracking CO₂ emissions related to this sector highlights the impact of the construction industry on global warming[32]. ACC is therefore essential to distinguish the share coming from heavy industry and to assess the possibilities of decarbonization (e.g. low-carbon cements, CO₂ capture).
Annual CO₂ emissions (ACE)
(ACE) refers to the total sum of each country’s annual CO₂ emissions from all sources (industry, transport, energy production, residential, etc.). It is the main indicator for assessing a country’s overall impact on global warming. ACE makes it easy to compare the scale of emissions between different states and is often used as a benchmark to measure a country’s trajectory towards its climate commitments (e.g. Paris Agreement).
Annual-co2-flaring (ACF)
(ACF) refers to CO₂ emissions from flaring, mainly in the oil and gas sector. Flaring involves burning residual gases for safety reasons or lack of storage infrastructure. Flaring is a significant source of emissions in the hydrocarbon industry. Its evaluation provides information on the degree of production optimization (or inefficiency) and can guide flaring reduction policies (regulations, gas recovery infrastructure).
Annual CO₂ emissions from_other_industry (ACI)
(ACI) covers CO₂ emissions from industries other than cement and petrochemicals (e.g. steel, chemicals, paper mills). It is therefore a broad sectoral grouping. The other industries are a very diverse group, and ACI makes it possible to capture the overall industrial contribution, beyond cement or flaring. By identifying this contribution, we can identify avenues for decarbonisation in sectors that are often less publicised, but which can weigh heavily in national inventories [33].
Annual CO₂ emissions from gas (ACG)
ACG aggregates CO₂ emissions from the combustion of natural gas. This includes both electricity generation, residential heating, and industrial use. Natural gas is often considered a “transition fuel” because it emits less CO₂ than coal or oil per unit of energy. However, studies show that the high dependence on gas can delay the adoption of low-carbon solutions, especially in the event of methane leaks [34]. ACG makes it possible to precisely quantify the share of CO₂ attributable to gas in the energy mix.
Annual CO₂ emissions from oil (ACO)
ACO refers to the CO₂ emissions resulting from the combustion of oil and its derivatives (petrol, diesel, fuel oil). Petroleum products dominate the transport sector (road, sea, air) and are therefore a major source of CO₂ [35]. Follow ACO distinguishes the contribution of oil, in order to more effectively target public policies for the decarbonization of mobility (electric vehicles, biofuels, etc.).
Annual CO₂ emissions from coal (ACCO)
ACCO refers to the total CO₂ emissions from the combustion of coal (power plants, steel industry, heating). Coal remains the fossil fuel source that emits the most CO₂ per unit of energy produced[36]. ACCO is therefore crucial for determining the place of coal in the energy mix, assessing its impact on warming and measuring the effect of coal phase-out policies.
Co2-emissions-transport (ACT)
ACT stands for direct CO₂ emissions generated by the transport sector (road, rail, aviation, inland maritime). Transport is one of the key sectors contributing to the increase in greenhouse gases in Western Europe [37]. ACT reports on the evolution of mobility demand, the energy efficiency of vehicles, and the degree of electrification of the vehicle fleet. Analysing ACT is crucial for developing targeted public policies (public transport, soft mobility).
Population growth (annual %) (PGP)
PGP indicates the annual population growth rate. It measures the percentage increase (or decrease) in the population from one year to the next. Population growth mechanically increases demand for energy and manufactured goods, and can stimulate growth in emissions [38]. However, some countries are managing to contain their per capita carbon footprint thanks to more sober technologies. PGP is therefore an explanatory factor in assessing whether or not population growth goes hand in hand with a significant increase in total emissions.
.
Annual CO₂ emissions growth (%) (ACGP)
ACGP represents the annual percentage change in total CO₂ emissions (usually compared to the previous year). This indicator makes it possible to monitor the dynamic year-over-year emissions change [39]. A rapid increase in CPA may signal a carbon-intensive economic recovery or a massive return to certain fossil fuels, while a decrease indicates a possible decoupling between economic activity and emissions.
Annual CO₂ emissions from land-use change (ACL_L)
ACL_C includes CO₂ emissions associated with changes in land use (deforestation, conversion of natural areas into agricultural or urban areas). Land-use changes release significant amounts of carbon stored in biomass and soils [40]. Although often less visible than industrial emissions, these processes are crucial to understanding the global impact of human practices on the climate, especially in Western Europe where land artificialization remains high.
Co-emissions-per-capita (CPC)
CPC stands for CO₂ emissions per capita. Dividing total emissions by the number of inhabitants (ACE / population) allows us to measure the average carbon intensity of a country [41]. CPC is an essential indicator for international or regional comparisons, as it controls for the effect of population size. A high CPC generally suggests a high per capita consumption of fossil fuels.
Data collection:
For data collection, we developed questionnaires that were administered using the KoboCollect app (see Appendix 1).
Population Cible:
The target population of our study includes the following countries in Western Europe:
Historical analysis, also known as temporal analysis, involves looking at data in the order in which it was collected or recorded. It is used to identify trends, recurring patterns or significant events over time. To illustrate and quantify this relationship, we looked at relevant indicators related to CO₂ emissions in Western Europe, such as coal combustion or annual emissions growth. The data, spread from 2021 to 2022, were analyzed in one-year intervals to detect trends and interactions between these variables. A linear regression was then used to assess the impact of each sector on warming, revealing the necessary trade-offs between economic growth and environmental protection.
🔍 Initial exploration
# Affiche la structure des données (types de colonnes)
str(data)## 'data.frame': 11 obs. of 12 variables:
## $ ACC : int 1831572 2456034 6396732 12537957 1956535 7918827 368935 2228547 7714822 1667094 ...
## $ ACE : int 61489110 89002056 293501630 671471550 36711388 340672300 7276724 40687052 234657020 32818890 ...
## $ ACF : int 80240 111066 1591822 1835065 200 2549430 31 1174634 3824197 21669 ...
## $ ACI : int 1210193 1592600 3725112 8242323 735816 3277906 12425 803425 2643633 168148 ...
## $ ACG : int 15967127 31086884 76353304 160775700 10745410 137601070 1249039 11367924 67501400 6015103 ...
## $ ACO : int 30733780 41814188 179440910 247440220 19237046 157948580 5409675 24867992 135100960 24515458 ...
## $ ACCO : num 1.17e+07 1.19e+07 2.60e+07 1.71e+08 4.04e+06 ...
## $ ACT : int 22330000 22920000 118910000 147240000 10620000 98240000 4990000 15450000 86540000 14620000 ...
## $ PGP : num 0.956 0.808 0.327 0.721 2.599 ...
## $ ACGP : num -6.49 -6.2 -3.96 -5.08 -2.22 ...
## $ ACL_C: num -1891467 2654678 -24758526 -8638173 79985 ...
## $ CPC : num 6.78 7.65 4.43 7.99 7.18 ...# Statistiques descriptives
describe(data)## vars n mean sd median trimmed mad
## ACC 1 11 5163945.18 4308350.49 2456034.00 4877389.44 3094332.98
## ACE 2 11 202011136.36 211267599.30 89002056.00 171486025.11 121165977.22
## ACF 3 11 1216370.36 1302035.20 1174634.00 1061760.67 1622548.54
## ACI 4 11 2474958.64 2460631.98 1592600.00 2107755.22 2111892.54
## ACG 5 11 63048601.00 65974716.23 31086884.00 57490435.78 44237589.00
## ACO 6 11 81457188.09 82958367.93 30733780.00 71464352.67 17045057.83
## ACCO 7 11 34009501.98 53481840.41 11941285.00 22580848.09 17341607.48
## ACT 8 11 63289090.91 58589781.27 22920000.00 59652222.22 26583018.00
## PGP 9 11 0.96 0.76 0.81 0.91 0.22
## ACGP 10 11 -4.53 4.92 -5.08 -4.30 4.74
## ACL_C 11 11 -5720529.81 12148019.52 -366839.70 -4687998.33 4479701.85
## CPC 12 11 6.18 2.23 5.71 5.91 2.18
## min max range skew kurtosis se
## ACC 368935.00 12537957.00 12169022.00 0.52 -1.43 1299016.55
## ACE 7276724.00 671471550.00 664194826.00 0.88 -0.48 63699577.93
## ACF 31.00 3824197.00 3824166.00 0.54 -1.10 392578.38
## ACI 12425.00 8242323.00 8229898.00 1.02 0.04 741908.46
## ACG 1249039.00 174871650.00 173622611.00 0.60 -1.44 19892125.40
## ACO 5409675.00 247440220.00 242030545.00 0.75 -1.12 25012889.06
## ACCO 236619.00 170640270.00 170403651.00 1.58 1.14 16125381.61
## ACT 4990000.00 154320000.00 149330000.00 0.39 -1.73 17665483.73
## PGP -0.20 2.60 2.80 0.75 -0.20 0.23
## ACGP -13.58 2.44 16.02 -0.12 -1.16 1.48
## ACL_C -30496900.00 9763057.00 40259957.00 -0.86 -0.59 3662765.70
## CPC 3.73 11.14 7.41 0.76 -0.45 0.67# Dimensions du tableau (nombre de pays x variables)
dim(data)## [1] 11 12# Résumé des données (moyenne, min, max, etc.)
summary(data)## ACC ACE ACF ACI
## Min. : 368935 Min. : 7276724 Min. : 31 Min. : 12425
## 1st Qu.: 1894054 1st Qu.: 38699220 1st Qu.: 50954 1st Qu.: 769620
## Median : 2456034 Median : 89002056 Median :1174634 Median :1592600
## Mean : 5163945 Mean :202011136 Mean :1216370 Mean :2474959
## 3rd Qu.: 7816824 3rd Qu.:317086965 3rd Qu.:2013392 3rd Qu.:3501509
## Max. :12537957 Max. :671471550 Max. :3824197 Max. :8242323
## ACG ACO ACCO
## Min. : 1249039 Min. : 5409675 Min. : 236619
## 1st Qu.: 11056667 1st Qu.: 24691725 1st Qu.: 2233898
## Median : 31086884 Median : 30733780 Median : 11941285
## Mean : 63048601 Mean : 81457188 Mean : 34009502
## 3rd Qu.:106977187 3rd Qu.:146524770 3rd Qu.: 28685129
## Max. :174871650 Max. :247440220 Max. :170640270
## ACT PGP ACGP ACL_C
## Min. : 4990000 Min. :-0.2023 Min. :-13.5840 Min. :-30496900
## 1st Qu.: 15035000 1st Qu.: 0.7091 1st Qu.: -7.3833 1st Qu.: -9579950
## Median : 22920000 Median : 0.8082 Median : -5.0763 Median : -366840
## Mean : 63289091 Mean : 0.9645 Mean : -4.5341 Mean : -5720530
## 3rd Qu.:108575000 3rd Qu.: 1.0454 3rd Qu.: -0.6091 3rd Qu.: 625042
## Max. :154320000 Max. : 2.5992 Max. : 2.4393 Max. : 9763057
## CPC
## Min. : 3.733
## 1st Qu.: 4.516
## Median : 5.714
## Mean : 6.184
## 3rd Qu.: 7.415
## Max. :11.138📊Correlation matrix
Mathematical Analysis of Data
Data analysis and processing
The first step in the analysis is to examine the correlation matrix (Figure 1) to assess the linear relationships between the climate and energy variables considered. Pearson correlation measures the degree of linear association between two variables (with coefficients ranging from -1 to +1). A value close to +1 indicates a strong positive correlation (the two variables increase or decrease together), while a value close to -1 indicates a negative correlation (when one variable increases, the other decreases); A value close to 0 means that there is no significant linear relationship In order to identify redundancies and linear relationships between the explanatory variables of CO₂ emissions in Western Europe in 2022, a cross-correlation matrix was generated from the 12 selected variables. This visualization, in the form of a scatterplot matrix, allows a qualitative first reading of the bivariate correlations between sectoral sources of emissions, demographic data, and synthetic indicators (such as emissions per capita or emission growth).
# Affichage d'une matrice de nuages de points
pairs(data[, 1:12])
# Calcul de la matrice de corrélation
mat_cor <- cor(data[, 1:12])
# Visualisation classique de la matrice de corrélation
corrplot::corrplot(mat_cor, method = "color", type = "upper")
# Graphique plus avancé pour trois variables d'intérêt
ggpairs(data, columns = c("ACE", "ACC", "ACT"))
# Matrice de corrélation avec couleurs ordonnées
ggcorrplot(mat_cor, hc.order = TRUE, type = "upper",
title = "Matrice de Corrélation", legend.title = "Degré de corrélation",
lab = TRUE, lab_col = "black", lab_size = 2,
ggtheme = theme_dark(), outline.color = "yellow")
Visual results of the correlation matrix
Strong linear correlations
The observation of scatter plots reveals several significant positive linear relationships:
ACE – ACO / ACE – ACC / ACE – ACT : total CO₂ emissions are strongly correlated with emissions from oil, the cement sector, and road transport, respectively. This reflects an emissions structure dominated by fossil fuels and heavy industrial sectors.
ACO – ACT: this correlation suggests that transport is one of the major outlets for oil consumption.
ACC – ACO / ACC – ACT: the proximity of these variables in the matrix highlights a profile of energy interdependence in the building and infrastructure sectors.
CPC – ACE / CPC – ACT / CPC – ACO: The per capita emissions indicator is highly correlated with major sectoral emissions, indicating a high carbon intensity per capita in the countries considered.
Weak or no correlations
Some variables show little or no obvious linear relationship with the main sources of emissions:
PGP (population growth): the near absence of correlation with the other variables suggests that the demographic impact on total emissions is marginal in the short term.
ACL_L (emissions due to land use change): emissions linked to land use planning appear to be structurally independent of emissions from fossil fuels.
Potential inverse correlations
Although less visually marked, some negative relationships can be evoked:
ACGP appears to be partially opposed to emissions from coal (ACCO) or oil (ACO), which could indicate a relative slowdown in emissions from these sources.
Preliminary Scientific Discussion
These exploratory correlations highlight the centrality of fossil fuels and the industrial sector in CO₂ emissions in Western Europe in 2022. The strong linear relationships between global emissions (ACE) and sectoral variables (ACO, ACT, ACC) support the hypothesis of an energy system structured around hydrocarbons. Similarly, the strong association between per capita emissions (CPCs) and other variables reflects a high level of carbon dependence at the individual level.
The lack of significant correlation for the PGP and ACL_L variables justifies their separate treatment in principal component analysis (PCA): they are the result of dynamics that are independent of the dominant energy systems.
✅ PCA validity tests
# Vérification du déterminant de la matrice de corrélation (doit être < 1 pour faire une ACP)
det(mat_cor)## [1] 1.000539e-40# Test de Bartlett (p < 0.05 : ACP justifiée)
cortest.bartlett(mat_cor, n = nrow(data))## $chisq
## [1] 475.8648
##
## $p.value
## [1] 2.269966e-63
##
## $df
## [1] 66# Test KMO (doit être > 0.5 pour une ACP fiable)
KMO(mat_cor)## Error in solve.default(r) :
## le système est numériquement singulier : conditionnement de la réciproque = 3.73444e-18## matrix is not invertible, image not found## Kaiser-Meyer-Olkin factor adequacy
## Call: KMO(r = mat_cor)
## Overall MSA = 0.5
## MSA for each item =
## ACC ACE ACF ACI ACG ACO ACCO ACT PGP ACGP ACL_C CPC
## 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5These tests confirm that PCA is relevant to our dataset. A good KMO value and a significant Bartlett test validate the factor structure.
🧮Principal Component Analysis (PCA)
# Réalisation de l'ACP avec centrage et réduction
res.pca <- PCA(data, scale.unit = TRUE, ncp = 5, graph = TRUE)
# Valeurs propres
valeur.propre <- get_eigenvalue(res.pca)
valeur.propre## eigenvalue variance.percent cumulative.variance.percent
## Dim.1 7.2702131534 60.585109612 60.58511
## Dim.2 2.3182363158 19.318635965 79.90375
## Dim.3 1.0382009912 8.651674926 88.55542
## Dim.4 0.6142147971 5.118456642 93.67388
## Dim.5 0.3565264047 2.971053373 96.64493
## Dim.6 0.2328385516 1.940321264 98.58525
## Dim.7 0.1350604428 1.125503690 99.71076
## Dim.8 0.0277185445 0.230987871 99.94174
## Dim.9 0.0065075425 0.054229521 99.99597
## Dim.10 0.0004832564 0.004027137 100.00000# Graphique des valeurs propres
fviz_eig(res.pca, addlabels = TRUE, ylim = c(0, 75),
barfill = "blue", barcolor = "black", linecolor = "red")
Factor analysis: Circle of correlations and interpretation of principal dimensions.
Presentation of the factorial design
The principal component analysis (PCA) carried out on the explanatory variables of CO₂ emissions in Western Europe in 2022 results in a circle of correlations between the active variables projected on the first factorial plane (Dim 1 and Dim 2). This plane captures 79.9% of the total inertia of the variable cloud, divided between:
· Dim 1 : 60,6 %
· Dim 2 : 19,3 %
This two-dimensional representation makes it possible to identify latent data structures as well as highly correlated variables and their oppositions.
📌 Variable analysis
# Extraction des caractéristiques des variables projetées
res.var <- get_pca_var(res.pca)
res.var$coord## Dim.1 Dim.2 Dim.3 Dim.4 Dim.5
## ACC 0.9631309 0.1729594 0.17334884 0.076234660 -0.069851657
## ACE 0.9644745 0.2477376 -0.04267797 -0.005333488 0.072882971
## ACF 0.7793210 -0.3572245 0.16313781 0.305168563 -0.319299288
## ACI 0.9368934 0.2970049 -0.03809451 0.001496565 0.165879395
## ACG 0.9300684 0.2030425 0.12705511 -0.029377803 -0.161329041
## ACO 0.8574381 -0.1464306 -0.42005561 0.014789369 0.179858377
## ACCO 0.8066888 0.5386042 0.05269031 -0.004135643 0.212612117
## ACT 0.9536057 0.1365144 0.12168903 -0.054163455 -0.148195927
## PGP -0.5899227 0.5598297 0.04223217 0.495247994 0.004946408
## ACGP 0.3347884 -0.7704291 0.05281705 0.457639806 0.236629751
## ACL_C -0.4919022 0.5780668 0.59068549 0.086467347 0.117220380
## CPC -0.2547014 0.6313446 -0.64390052 0.221428548 -0.146855202res.var$cor## Dim.1 Dim.2 Dim.3 Dim.4 Dim.5
## ACC 0.9631309 0.1729594 0.17334884 0.076234660 -0.069851657
## ACE 0.9644745 0.2477376 -0.04267797 -0.005333488 0.072882971
## ACF 0.7793210 -0.3572245 0.16313781 0.305168563 -0.319299288
## ACI 0.9368934 0.2970049 -0.03809451 0.001496565 0.165879395
## ACG 0.9300684 0.2030425 0.12705511 -0.029377803 -0.161329041
## ACO 0.8574381 -0.1464306 -0.42005561 0.014789369 0.179858377
## ACCO 0.8066888 0.5386042 0.05269031 -0.004135643 0.212612117
## ACT 0.9536057 0.1365144 0.12168903 -0.054163455 -0.148195927
## PGP -0.5899227 0.5598297 0.04223217 0.495247994 0.004946408
## ACGP 0.3347884 -0.7704291 0.05281705 0.457639806 0.236629751
## ACL_C -0.4919022 0.5780668 0.59068549 0.086467347 0.117220380
## CPC -0.2547014 0.6313446 -0.64390052 0.221428548 -0.146855202res.var$cos2## Dim.1 Dim.2 Dim.3 Dim.4 Dim.5
## ACC 0.92762110 0.02991495 0.030049821 5.811723e-03 4.879254e-03
## ACE 0.93021109 0.06137392 0.001821409 2.844609e-05 5.311928e-03
## ACF 0.60734123 0.12760932 0.026613944 9.312785e-02 1.019520e-01
## ACI 0.87776922 0.08821189 0.001451192 2.239706e-06 2.751597e-02
## ACG 0.86502716 0.04122627 0.016143001 8.630553e-04 2.602706e-02
## ACO 0.73520008 0.02144192 0.176446715 2.187255e-04 3.234904e-02
## ACCO 0.65074675 0.29009446 0.002776268 1.710355e-05 4.520391e-02
## ACT 0.90936390 0.01863618 0.014808219 2.933680e-03 2.196203e-02
## PGP 0.34800876 0.31340928 0.001783556 2.452706e-01 2.446695e-05
## ACGP 0.11208325 0.59356098 0.002789641 2.094342e-01 5.599364e-02
## ACL_C 0.24196781 0.33416118 0.348909349 7.476602e-03 1.374062e-02
## CPC 0.06487281 0.39859596 0.414607876 4.903060e-02 2.156645e-02res.var$contrib## Dim.1 Dim.2 Dim.3 Dim.4 Dim.5
## ACC 12.7592008 1.2904187 2.8944127 9.462038e-01 1.368553334
## ACE 12.7948256 2.6474403 0.1754389 4.631294e-03 1.489911393
## ACF 8.3538298 5.5045865 2.5634674 1.516210e+01 28.595928387
## ACI 12.0735005 3.8051291 0.1397795 3.646454e-04 7.717794059
## ACG 11.8982365 1.7783461 1.5549014 1.405136e-01 7.300177225
## ACO 10.1124968 0.9249239 16.9954292 3.561058e-02 9.073391280
## ACCO 8.9508620 12.5135844 0.2674115 2.784620e-03 12.678980222
## ACT 12.5080776 0.8038949 1.4263345 4.776309e-01 6.160001784
## PGP 4.7867751 13.5192983 0.1717930 3.993238e+01 0.006862592
## ACGP 1.5416776 25.6039894 0.2686995 3.409787e+01 15.705327392
## ACL_C 3.3282079 14.4144570 33.6071100 1.217262e+00 3.854025200
## CPC 0.8923096 17.1939313 39.9352225 7.982647e+00 6.049047131# Cercle de corrélation : qualité de représentation
fviz_pca_var(res.pca, col.var = "cos2", gradient.cols = c("blue", "green", "red"),
repel = TRUE)
# Description des dimensions principales
res.desc <- dimdesc(res.pca, axes = c(1, 2), proba = 0.05)
res.desc$Dim.1##
## Link between the variable and the continuous variables (R-square)
## =================================================================================
## correlation p.value
## ACE 0.9644745 1.669574e-06
## ACC 0.9631309 1.969296e-06
## ACT 0.9536057 5.462626e-06
## ACI 0.9368934 2.128336e-05
## ACG 0.9300684 3.344898e-05
## ACO 0.8574381 7.401870e-04
## ACCO 0.8066888 2.696672e-03
## ACF 0.7793210 4.689923e-03res.desc$Dim.2##
## Link between the variable and the continuous variables (R-square)
## =================================================================================
## correlation p.value
## CPC 0.6313446 0.037222039
## ACGP -0.7704291 0.005525046# Contributions principales aux axes
fviz_contrib(res.pca, choice = "var", axes = 1:2, top = 12)
# Autre vue par contribution
fviz_pca_var(res.pca, col.var = "contrib", gradient.cols = c("blue", "green", "red"),
repel = TRUE)
Reading the correlation circle
Variables strongly correlated with the first dimension (Dim 1)
The first axis is characterized by a strong contribution of the following variables, whose vectors are aligned, close to the edge of the circle, and red to orange (cos² > 0.9):
· ACE : Total CO₂ emissions
· ACO : Petroleum Emissions
· ACT : Transportation Emissions
· CCA : Cement Emissions
· ACI : Industrial Emissions
· ACCO : Coal Emissions
These variables have a very strong positive correlation with each other, indicating that they co-vary strongly in the statistical space. They reflect a common latent factor that can be interpreted as a component of fossil sectoral carbon intensity, i.e. the concentration of emissions around fossil fuels and industrial activities.
Variables Correlated with the Second Dimension (Dim 2)
The second dimension is dominated by variables of different nature:
PGP : population growth
ACL_L : Change in land use
ACGP : growth in CO₂ emissions
Their vectors are oriented negatively on the vertical axis (Dim 2), indicating a structure independent of that of fossil emissions. Their contribution to the plan is moderate, but their angular separation from Dim 1 confirms that they are part of a distinct structural logic, possibly associated with exogenous factors, in particular socio-environmental dynamics.
Intermediate CPC Positioning
The variable CPC (emissions per capita) is located at the interface of the two axes, well represented on the plan (cos² > 0.7), and close to the sectoral variables. It can be seen as a synthetic indicator of carbon pressure per individual, reflecting both sectoral emission levels and their redistributed impact at the demographic level.
Interpretive synthesis of dimensions
| Dimension | Interpretive meaning | Dominant variables | ||||
| Dim 1 (60,6%) | Sectoral carbon intensity – dependence on fossil fuels and heavy industry | ACE, ACO, ACC, ACT, ACI, ACCO | ||||
| Dim 2 (19,3%) | Socio-environmental component – demography, land use, energy transition | PGP, ACL_L, ACGP | ||||
The orthogonality between the two sets of variables suggests the existence of distinct causal logics: on the one hand, well-identified sources of direct emissions; on the other, indirect or systemic factors influencing the trajectory of emissions, but in a less immediate way.
Analytical consequences for the research question
As the initial objective is to identify the main causes of global warming in Western Europe in 2022, this step shows that the major explanatory factors are:
Emissions from the energy sector, particularly oil, coal, and industrial emissions;
Transport, which is highly dependent on oil and directly correlated with total emissions;
Emissions per capita, a revealing indicator of the carbon pressure exerted by lifestyles.
On the other hand, variables related to population growth, land use change, or variation in the rate of emissions (CGA) are secondary factors in this PCA, justifying a more detailed analysis in other dimensions or with complementary methods (regression, clustering, etc.).
👤Analysis of individual contributions: differentiation of European countries in terms of CO₂ emissions
res.ind <- get_pca_ind(res.pca)
res.ind$coord## Dim.1 Dim.2 Dim.3 Dim.4 Dim.5
## Austria -1.979849 0.09413444 -0.18253170 -0.4973910 0.21468578
## Belgium -1.737616 0.39155175 -0.22407091 -0.4158144 0.22250333
## France 1.899033 -1.26645949 -0.77500289 -0.8049065 -0.05927246
## Germany 5.034445 2.22388342 -0.85793904 0.1526624 0.97831595
## Ireland -2.726005 0.49747332 -0.03748109 1.5290912 0.51866015
## Italy 2.756602 -2.02795935 -1.03605521 -0.1644259 -0.55400836
## Luxembourg -3.490785 2.03376025 -1.37721681 0.1050657 -0.97537339
## Portugal -1.590536 -1.64603507 1.05309839 0.2209714 0.63255943
## Spain 1.594555 -1.79366336 0.39710936 1.3020186 -0.53447445
## Switzerland -2.305255 -0.52827737 0.76483977 -1.2264982 0.30456095
## United Kingdom 2.545411 2.02159146 2.27525013 -0.2007732 -0.74815693res.ind$cos2## Dim.1 Dim.2 Dim.3 Dim.4 Dim.5
## Austria 0.9039172 0.002043439 0.0076831837 0.0570506348 0.0106284804
## Belgium 0.7422174 0.037687897 0.0123422554 0.0425033057 0.0121701687
## France 0.4852045 0.215795590 0.0808102780 0.0871667435 0.0004726786
## Germany 0.7894791 0.154049908 0.0229271603 0.0007259413 0.0298123137
## Ireland 0.6687143 0.022270336 0.0001264188 0.2104039916 0.0242076675
## Italy 0.5534852 0.299554888 0.0781850500 0.0019692389 0.0223558172
## Luxembourg 0.6327247 0.214767231 0.0984856909 0.0005731787 0.0493981079
## Portugal 0.3594082 0.384927696 0.1575574310 0.0069370198 0.0568464919
## Spain 0.3019094 0.382014253 0.0187248323 0.2012945734 0.0339196683
## Switzerland 0.6589688 0.034605963 0.0725383776 0.1865351670 0.0115020524
## United Kingdom 0.3942692 0.248693053 0.3150178494 0.0024529483 0.0340613828res.ind$contrib## Dim.1 Dim.2 Dim.3 Dim.4 Dim.5
## Austria 4.901446 0.03474935 0.29174437 3.6617012 1.17522812
## Belgium 3.775441 0.60121242 0.43963963 2.5590936 1.26237575
## France 4.509465 6.28973306 5.25935535 9.5891011 0.08958217
## Germany 31.693005 19.39427928 6.44523465 0.3449460 24.40472327
## Ireland 9.292094 0.97048582 0.01230128 34.6061937 6.85932501
## Italy 9.501852 16.12753907 9.39921698 0.4001543 7.82615141
## Luxembourg 15.237244 16.21993540 16.60850354 0.1633837 24.25813570
## Portugal 3.163349 10.62497156 9.71099593 0.7227015 10.20277453
## Spain 3.179356 12.61628475 1.38084878 25.0911989 7.28399296
## Switzerland 6.645048 1.09439381 5.12232106 22.2649048 2.36517913
## United Kingdom 8.101699 16.02641548 45.32983845 0.5966210 14.27253194# Visualisation des individus
fviz_pca_ind(res.pca, col.ind = "cos2", gradient.cols = c("blue", "green", "red"),
repel = TRUE)
Presentation of the Individuals’ Factorial Design
The graph shows the projection of Western European countries (statistical individuals) on the first factor plane of the ACP (Dim 1 and Dim 2), accounting for 60.6% and 19.3% of the total inertia. Each dot represents a country, its color indicating the quality of its representation in the plan (cos²), from 0.65 (blue) to > 0.90 (red).
The objective of this visualization is to understand the diversity of national CO ₂ emission profiles and to identify the leading or lagging countries in each of the previously identified dimensions.
Reading the main dimension (Dim 1 – Sectoral carbon intensity)
Countries strongly to the right of Dim 1 (high sectoral carbon intensity)
Germany (cos² > 0.90, in red): Extreme position on the right-hand axis, it is the largest contributor to the sectoral carbon intensity component. This reflects a high dependence on heavy industries, fossil fuels, and a high overall volume of emissions (ACE, ACO, ACCO, ACT).
Italy and France: Also on the right, with a moderate contribution (cos² ≈ 0.70), these countries have an emissions profile related to the energy and industrial sectors, but to a lesser extent than Germany.
Countries projected below Dim 2
Portugal, Switzerland, Austria, Belgium, Ireland: These countries have a negative projection on Dim 1, indicating a lower overall level of sectoral emissions, or a less carbon-intensive energy model. Their distance from the “fossil/industry” pole suggests a different strategy (energy efficiency, lower share of heavy industry, etc.).
Spain, Italy, France, Portugal: These countries show a negative contribution to Dim 2, which can be interpreted as a more stable or mature position from a socio-environmental point of view (contained growth in emissions, less dynamic demography, etc.).
Regional groupings and disparities
Group 1: Germany, United Kingdom, Luxembourg
Strong contribution to both dimensions: both industrial polluters and individual carbon-intensive countries.
Group 2: France, Italy, Spain
Intermediate profile: moderate sectoral contributions, low socio-demographic influence.
Group 3 : Portugal, Switzerland, Austria, Belgium
Low overall impact, less carbon-intensive model, potentially more advanced in the energy transition.
Scientific implications
This spatial distribution of countries in terms of factors confirms that global warming in Western Europe in 2022 is mainly due to a hard core of countries with high industrialization and fossil fuel consumption, in particular:
Germany, whose profile is dominated by the most polluting sectors,
Luxembourg, which has a structurally high per capita impact, and to a lesser extent the United Kingdom.
The other countries have differentiated trajectories, more or less aligned with the fossil or demographic component. This legitimizes a differentiated approach to climate policies, taking into account the national industrial and energy structure.
🧭 Biplots (variables + individuals)
fviz_pca_biplot(res.pca, col.ind = "cos2", col.var = "contrib",
gradient.cols = c("blue", "green", "red"), repel = TRUE)
Interpreting the PCA Biplot - Joint representation of individuals and variables
The biplot combines the projection of explanatory variables (colored arrows) and countries (blue dots) onto the first two principal components of the PCA, which account for almost 80% of the total variance (Dim 1: 60.6%, Dim 2: 19.3%). It offers a simultaneous reading of the relationships between variables and the distribution of countries according to their emissions profile.
Reading the variables (vectors)
The longest vectors pointing in the same direction - ACC, ACT, ACE, ACI, ACO, ACG, ACCO - reflect a strong correlation between these variables. Their grouping in the right-hand quadrant of the map suggests that they define a core of sectoral carbon intensity, strongly structuring the first dimension. These variables represent high-emission sectors (cement, oil, coal, transport, industry), confirming their decisive contribution to total emissions.
The variables PGP (population growth), CPC (per capita emissions) and ACL_C (land use change), oriented towards the left quadrant, are weakly correlated with the emissions core. They express structural, socio-demographic or territorial dimensions that have little to do with the dominant industrial logics, but are useful for explaining certain national variations. The ACGP variable (annual growth in emissions) is oriented downwards and stands out for its low correlation with the core emissions.
🌐 Interpretation of the results of the Hierarchical Ascending Classification (CAH) method.
res.cah <- HCPC(res.pca, nb.clust = -1, consol = FALSE, graph = FALSE)
plot.HCPC(res.cah, choice = 'tree', title = 'Hierarchical tree')
plot.HCPC(res.cah, choice = 'map', draw.tree = FALSE, title = 'Factor map')
plot.HCPC(res.cah, choice = '3D.map', ind.names = TRUE, centers.plot = FALSE, angle = 60)
The hierarchical ascending classification (HFC), based on the factor coordinates extracted from the principal component analysis (PCA), allows the countries studied to be grouped according to the structural similarity of their CO₂ emission profiles. It relies on Euclidean distance as a measure of dissimilarity and on Ward’s algorithm, which minimizes intra-class inertia.
The cross-section in the dendrogram results in a division into three homogeneous groups, which offer a hierarchical reading of the causes of regional global warming.
Isolated (Class 1) : Portugal, Switzerland, Austria, Belgium, Luxembourg, Ireland
This first group can be interpreted as countries with controlled or historically low carbon intensity. The key elements of this profile are as follows:
· Low dependence on coal and oil: these countries do not appear to be highly correlated with ACCO or ACO vectors in the ACP.
· Less polluting or better decarbonized industry: the contribution of the ACI and ACC variables is lower for these countries, which reflects either a past deindustrialization or a move towards less carbon-intensive economic sectors (services, finance, innovation).
· Limited weight of road transport in emissions: the scores of these countries on ACT are also reduced.
· Population growth is often moderate or non-existent, and land use is relatively stable.
It is a kind of model of structural energy sobriety, either by political choice or by geographical or historical configuration. This group represents the countries least responsible for regional global warming, in terms of gross emissions.
Note: Luxembourg is a relative exception here, with a high level of emissions per capita (CPC), but it is dragged into this class by its low gross sectoral contribution at aggregate level.
Leadership (Class 2) : Spain, France, Italy
This second group constitutes an intermediate profile, which could be described as transitional. These countries cumulate:
· A significant level of total emissions (ACE), although lower than Germany or the United Kingdom;
· A still real dependence on fossil fuels, in particular gas (ACG), oil (ACO) and to a lesser extent coal (ACCO), but with national policies of gradual reduction;
· Industrial and transport sectors are still important, but better supervised or gradually decarbonised (development of rail in France, partial electrification in Spain);
· A central position in the biplot, reflecting a tension between industrial heritage and energy transition efforts.
These countries are therefore in an intermediate dynamic: they have a history of high emittances, but their trajectory tends towards a moderation of emissions, without any marked break yet. They represent essential levers for European climate policies, because their switch to a low-carbon model would have a knock-on effect.
Dense (Class 3) : United Kingdom, Germany
The third class includes the two countries that contribute the most to CO₂ emissions according to the ACP, both in absolute and per capita terms:
· Germany: country with the most intensive profile, related to industrial emissions (ACI), coal (ACCO), transport (ACT) and total emissions (ACE). Despite its transition commitments, its residual dependence on heavy industry and coal in 2022 remains evident.
· United Kingdom: apparently more advanced in the transition, but still a strong emitter due to the high share of transport and a business model historically based on oil and gas. Its high level of emissions per capita (CPC) also contributes to its position in this group.
This group reflects a centralized, emitting-intensive energy structure and a dominant position in the statistical variance of emissions. These are the priority targets for a significant reduction in global warming in Western Europe.
Analytical contribution of the CAH
This classification makes it possible to statistically validate the geographical and energy typology revealed by the ACP. It provides:
· A consolidated vision by clusters, facilitating international comparisons;
· A rationale for the differentiation of climate policies, based on the specific emissions structure of each group of countries;
· A rigorous framework for targeted recommendations, e.g.: incentives for green investment in Class 2 countries, strengthened fossil phase-out plans for Class 3, support for sobriety and preservation in Class 1.
Limits and additional tracks
The classification here is carried out on the basis of the first two dimensions of PCA, which captures almost 80% of the variance, but does not allow for the exploration of the finer dimensions (national specificities, temporal effects, etc.). A complementary analysis by k-means or a classification on more dimensions could refine certain groupings (for example, the special case of Luxembourg or the place of Belgium).
📈 Linear regressions
regressionL <- lm(ACO ~ ACT, data = data)
summary(regressionL)##
## Call:
## lm(formula = ACO ~ ACT, data = data)
##
## Residuals:
## Min 1Q Median 3Q Max
## -145928823 -8134950 -6689567 35020100 79301430
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 1.611e+07 2.726e+07 0.591 0.5691
## ACT 1.032e+00 3.230e-01 3.197 0.0109 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 59840000 on 9 degrees of freedom
## Multiple R-squared: 0.5318, Adjusted R-squared: 0.4797
## F-statistic: 10.22 on 1 and 9 DF, p-value: 0.01088# Visualisation de la régression simple
ggplot(data, aes(x = ACT, y = ACC)) +
geom_point(color = "blue") +
geom_smooth(method = "lm", se = TRUE, color = "red") +
ggtitle("Régression linéaire : Effet de ACT sur ACC") +
xlab("ACT") + ylab("ACC") +
theme_minimal()## `geom_smooth()` using formula = 'y ~ x'
📊 Multiple regression and multicollinearity
regressionM <- lm(ACI ~ ACT + ACCO + ACG, data = data)
summary(regressionM)##
## Call:
## lm(formula = ACI ~ ACT + ACCO + ACG, data = data)
##
## Residuals:
## Min 1Q Median 3Q Max
## -1030605 -220055 118454 344063 508932
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.861e+05 2.755e+05 1.401 0.20384
## ACT 2.316e-02 9.947e-03 2.328 0.05273 .
## ACCO 2.910e-02 6.249e-03 4.656 0.00233 **
## ACG -5.815e-03 9.533e-03 -0.610 0.56116
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 583300 on 7 degrees of freedom
## Multiple R-squared: 0.9607, Adjusted R-squared: 0.9438
## F-statistic: 56.98 on 3 and 7 DF, p-value: 2.768e-05car::vif(regressionM)## ACT ACCO ACG
## 9.982610 3.282679 11.624370Multiple linear regression: a tool for causal analysis of CO₂ emissions
Multiple linear regression is a statistical method for modeling the relationship between a target variable (in this case, total CO₂ emissions) and several explanatory variables representing the potential sources of these emissions. In the context of this study, it is a central tool for understanding, quantifying and anticipating the causes of global warming in Western Europe.
A tool for predicting emissions
The regression model makes it possible to predict global CO₂ emissions (ACE) from the intensity of emissions from different sectors such as industry (ACI), transport (ACT), or fossil fuels (ACO, ACCO, ACG), but also from contextual variables such as population growth (PGP) or per capita emissions (CPC). This paves the way for forward-looking modelling: by simulating a reduction in coal-related emissions, for example, the direct impact on the country’s total emissions can be estimated.
A detailed reading of causal relationships
Regression is not limited to describing correlations: it measures the own contribution of each factor to emissions, controlling for the effect of other variables. For example, the specific impact of transport on emissions can be quantified, independent of that of industry. The sign (positive or negative) and the magnitude of the estimated coefficients thus make it possible to interpret the economic and environmental causal relationships in the model.
A method of identifying key variables.
The regression also provides an objective hierarchy of the causes of warming: the most significant variables (according to p-values) are those that should be targeted as a priority in climate policies. If, for example, the coefficients of ACO (oil) and ACT (transport) are high and statistically significant, this confirms that these sectors need to be the subject of rapid reduction or transformation actions.
A tool to support climate decision-making
Through the estimation of sectoral impacts, regression becomes a strategic instrument to guide public policies. It allows governments to know where to focus efforts, whether in the decarbonization of industry, energy substitution, or consumption control. It is therefore essential to prioritise investments and reforms within a constrained budgetary and time frame.
A basis for scenario evaluation
Finally, regression can be used to simulate different climate scenarios. For example, a projected decrease in coal emissions or a moderate demographic change can be introduced into the model to observe the expected evolution of overall emissions. This makes it possible to test undefined.
In this graphic:
· The horizontal axis represents the theoretical quantiles of a standard normal distribution.
· The vertical axis displays the observed quantiles (standardized residuals from the model).
· The dotted line corresponds to the ideal theoretical alignment if the residues follow a normal distribution.
The closer the points are to this line, the more plausible the assumption of normality.
Accurate chart reading
· The points are relatively well aligned with the normality line, suggesting that the residuals normality assumption is generally respected.
· Two slight differences are visible:
· The United Kingdom appears outside the expected alignment at the far left of the graph: it is a relatively strong negative residual, indicating that the model underestimates the value of ACO for this country.
· France and Spain are slightly above the line on the right: their ACO values are slightly overvalued compared to the model’s predictions.
These deviations are moderate, and do not indicate a serious breach of normalcy. The model remains statistically robust for the interpretation of the coefficients, but these observations may merit further analysis if they influence the fit excessively.
🔎 Model diagnostics
par(mfrow = c(2, 2))
plot(regressionL)
plot(regressionM)
Interpretation: These plots verify the assumptions of normality, homoscedasticity and linearity of residuals. A random and symmetrical distribution around zero is desired.
🔮Evaluating Prediction Quality: Predicted vs. Actual Values for ACT
predicted_vals <- predict(regressionM)
ggplot(data, aes(x = predicted_vals, y = ACT)) +
geom_point(color = "darkgreen") +
geom_abline(intercept = 0, slope = 1, color = "red", linetype = "dashed") +
ggtitle("Valeurs prédites vs Réelles pour ACT") +
xlab("Valeurs prédites") + ylab("Valeurs réelles (ACT)") +
theme_minimal()
This graph shows:
· The horizontal axis: the values predicted by the regression model for the ACT variable,
· The vertical axis: the observed (actual) values in the sample.
The dashed red line is the perfect identity line (reality = prediction). If all of the model’s predictions were correct, all points would be aligned to this diagonal.
Reading and interpreting discrepancies
Satisfactory overall alignment
It can be seen that the majority of the points are close to the diagonal, which indicates that the values predicted by the model are relatively consistent with the observed values. This shows a good predictive capacity of the model on the ACT variable.
Presence of significant discrepancies
Several points (top right or below the diagonal) have significant deviations, which means that the model underestimates or overestimates some transport values.
· Points above the line: The model underestimates actual emissions. The relevant observations show a higher level of transport emissions than expected.
· Points below the line: The model overestimates the emissions for these cases.
These differences can be explained by:
· Specific factors not taken into account in the model (local mobility policies, urban geography, etc.),
· Structural heterogeneity between countries,
· Or extreme values (outliers) in some cases (such as a country with a very high road density or a low-carbon policy).
Overall Accuracy Assessment
The graph suggests that:
· The model is generally robust, but not perfectly accurate.
· There are significant residuals for some observations, which could affect the reliability of the prediction for extreme countries.
· Further analysis of the residues (already initiated with the Q-Q plot) is recommended to identify possible disproportionate influences or systematic errors.
Scientific consequences for your study
This graph confirms that your model can be used to:
· Estimate transport emissions (ACT) from other known variables,
· Provide a reliable basis for scenario simulation (e.g. impact of a reduction in industrial emissions on transport),
· Support sectoral forecasts from an energy transition perspective.
However, it also highlights the need to:
· To complete the analysis with nonlinear or mixed models if you want to improve the accuracy (spline, random forest, etc.),
· Or to segment your model (by country class, as in the CAH) to reduce heterogeneity and better capture national specificities.
OVERALL CONCLUSION
The aim of this study was to identify, structure and quantify the main causes of global warming in Western Europe in 2022, based on a set of sectoral, demographic and economic variables related to CO₂ emissions. By successively mobilizing a principal component analysis (PCA), a hierarchical ascending classification (HFA), and then a multiple linear regression, the article made it possible to cross-reference exploratory, typological and causal approaches.
Factor analysis showed that almost all of the variance in total emissions (ACE) is explained by a core of highly correlated variables, related to fossil fuels (coal, oil, gas), heavy industry (cement, processing) and road transport. The first dimension extracted, representing more than 60% of the inertia, reflects a sectoral carbon intensity, independent of socio-environmental dynamics such as demography or land use.
The hierarchical classification made it possible to group the countries into three distinct profiles:
A carbon-intensive group, including Germany and the United Kingdom, which are responsible for a disproportionate share of emissions;
An intermediate group, comprising France, Spain and Italy, characterised by a partial dependence on fossil fuels and a transition trajectory that has begun;
And a group of countries with relative low emissions, marked by more sober economic structures or more advanced climate policy.
Finally, the multiple linear regression confirmed the decisive contribution of the industrial and energy sectors to oil-related emissions (COA). The statistically robust model yielded interpretable and meaningful coefficients, paving the way for reliable predictions in prospective simulations. The validation plots, including the Q-Q plot and the comparison of actual vs. predicted values, validated the quality of the model while revealing some margins of error related to non-modeled factors.
Thus, this cross-cutting approach makes it possible to prioritize the causes of global warming and to orient public policies according to differentiated national profiles. The following section makes concrete recommendations based on the results obtained.
RECOMMENDATIONS
Strengthen the decarbonisation of transport: massively develop clean public transport networks, encourage the electrification of private and logistics vehicles, and implement a dissuasive tax on fossil fuels.
Accelerate the phase-out of coal in countries that are still highly dependent (such as Germany), through a scheduled closure of power plants, the development of renewable energies and social support for the territories concerned.
Modernise heavy industry (in particular cement and industrial processing), through technological innovation (CO₂ capture, process electrification), and make public subsidies conditional on clear emission reduction targets.
Impose a minimum European carbon tax, more ambitious and harmonised, applied to the highest sectoral emissions (transport, industry, energy), to correct imbalances between Member States.
Adapt climate policies to the national profiles revealed by the hierarchical classification: targeted actions for the most intensive countries (class 3), technical support for countries in transition (class 2), and recognition of the efforts of sober countries (class 1).
Integrate per capita emission indicators (CPCs) into national targets, in order to adjust climate strategies not only on aggregate volumes but also on individual emissions intensity.
Support energy sobriety and demand management, through education, behavioural incentives, and the fight against forms of energy overconsumption in the residential and commercial sectors.
Strengthen prospective analysis through modelling: use regression models to simulate different medium- and long-term climate scenarios, in order to anticipate the effects of public policies and optimise investment choices.
BIBIOGRAPHICAL REFERENCE
[1] I. P. O. C. Change, « Climate change 2007 : Impacts, adaptation and vulnerability », Geneva, Switzerland, 2001.
[2] C. Good time et al., « The copernicus climate change service : climate science in action », Bulletin of the American Meteorological Society, vol. 103, no 12, p. E2669-E2687, 2022.
[3] K. R. van Daalen et al., « The 2022 Europe report of the Lancet Countdown on health and climate change : towards a climate resilient future », The Lancet Public Health, vol. 7, no 11, p. e942-e965, 2022.
[4] P. Dechamps, « The IEA World Energy Outlook 2022–a brief analysis and implications », European Energy & Climate Journal, vol. 11, no 3, p. 100-103, 2023.
[5] A. Rybak, J. Joostberens, A. Manowska, et J. Pielot, « The impact of environmental taxes on the level of greenhouse gas emissions in Poland and Sweden », Energies, vol. 15, no 12, p. 4465, 2022.
[6] A. M. Van Der Woude et al., « Temperature extremes of 2022 reduced carbon uptake by forests in Europe », nature communications, vol. 14, no 1, p. 6218, 2023.
[7] M. Burke, S. M. Hsiang, et E. Miguel, « Global non-linear effect of temperature on economic production », Nature, vol. 527, no 7577, p. 235-239, 2015.
[8] P. Dechamps, « The IEA World Energy Outlook 2022–a brief analysis and implications », European Energy & Climate Journal, vol. 11, no 3, p. 100-103, 2023.
[9] M. Klösch et T. Wittreich, « Can Reporting Drive Corporate Sustainability ? », Available at SSRN 5053274, 2022.
[10] P. Friedlingstein et al., « Global carbon budget 2022 », Earth System Science Data, vol. 14, no 11, p. 4811-4900, 2022.
[11] R. Jackson et al., « Global fossil carbon emissions rebound near pre-COVID-19 levels », Environmental Research Letters, vol. 17, no 3, p. 031001, 2022.
[12] P. IEA, « World energy outlook 2022 », Paris, France : International Energy Agency (IEA), 2022.
[13] D. García-Gusano et D. Iribarren, « Prospective energy security scenarios in Spain : The future role of renewable power generation technologies and climate change implications », Renewable energy, vol. 126, p. 202‑209, 2018.
[14] D. Gielen, « Critical minerals for the energy transition », International Renewable Energy Agency, Abu Dhabi, 2021.
[15] European Environment Agency, « Greenhouse gas emissions from transport in Europe », tech. Rep., EEA, 2019.
[16] A. A. Adesina et al., « Towards low-GHG emissions from energy use in selected sectors-CAETS Energy report 2022 », 2023.
[17] A. D’Amico, « Strategies and instruments for active mobility : comparison of international experiences », TeMA-Journal of Land Use, Mobility and Environment, vol. 17, no 1, p. 155-167, 2024.
[18] D. S. Lee et al., « The contribution of global aviation to anthropogenic climate forcing for 2000 to 2018 », Atmospheric environment, vol. 244, p. 117834, 2021.
[19] A. Raihan, « A review of recent advances, challenges, and potential future directions of climate-smart agriculture », présenté à Proceedings of the international conference on climate-smart agriculture, 2024.
[20] A. Leip et al., « Evaluation of the livestock sector’s contribution to the EU greenhouse gas emissions (GGELS) », 2010.
[21] J. Soussana et al., « Carbon cycling and sequestration opportunities in temperate grasslands », Soil use and management, vol. 20, no 2, p. 219-230, 2004.
[22] A. Bastos et al., « Direct and seasonal legacy effects of the 2018 heat wave and drought on European ecosystem productivity », Science advances, vol. 6, no 24, p. eaba2724, 2020.
[23] M. W. Jones et al., « Global rise in forest fire emissions linked to climate change in the extratropics », Science, vol. 386, no 6719, p. eadl5889, 2024.
[24] S. Oberthür et C. Dupont, « The European Union’s international climate leadership : towards a grand climate strategy ? », Journal of European Public Policy, vol. 28, no 7, p. 1095-1114, 2021.
[25] S. Teske, M. Feenstra, et S. Niklas, « Europe’s Energy Crisis Effect of non-technical measures to reduce energy demand to decrease fossil fuel imports », 2022.
[26] J. Rockström et al., « Safe and just Earth system boundaries », Nature, vol. 619, no 7968, p. 102-111, 2023.
[27] R. Muon, « European commission and the use of scientific knowledge : an empirical study on sustainable battery regulation of the EU Green Deal », 2023.
[28] T. M. Lenton et al., « Climate tipping points—too risky to bet against », Nature, vol. 575, no 7784, p. 592-595, 2019.
[29] D. Monteiro et S. Morin, « Multi-decadal analysis of past winter temperature, precipitation and snow cover data in the European Alps from reanalyses, climate models and observational datasets », The Cryosphere, vol. 17, no 8, p. 3617-3660, 2023.
[30] F. Hua, M. Alharthi, W. Yin, M. Saeed, I. Ahmad, et S. A. Ali, « Carbon emissions and socioeconomic drivers of climate change : empirical evidence from the logarithmic mean Divisia index (LMDI) base model for China », Sustainability, vol. 14, no 4, p. 2214, 2022.
[31] C. De Ketele and J.-B. Hasse, “Influence of the introduction of the European carbon market on emissions in the Benelux area”.
[32] K. Scrivener, M. B. Haha, P. Juilland, et C. Levy, « Research needs for cementitious building materials with focus on Europe », RILEM Technical Letters, vol. 7, p. 220‑252, 2022.
[33] T. G. Wyns, G. A. Khandekar, M. Axelson, O. Sartor, et K. Neuhoff, « Industrial transformation 2050-Towards an industrial strategy for a climate neutral Europe », 2019.
[34] J. Asker, A. Collard-Wexler, C. De Canniere, J. De Loecker, et C. R. Knittel, « Two Wrongs Can Sometimes Make a Right : The Environmental Benefits of Market Power in Oil », National Bureau of Economic Research, 2024.
[35] F. Creutzig et al., « Transport : A roadblock to climate change mitigation ? », Science, vol. 350, no 6263, p. 911-912, 2015.
[36] M. Burke et K. Emerick, « Adaptation to climate change : Evidence from US agriculture », American Economic Journal : Economic Policy, vol. 8, no 3, p. 106-140, 2016.
[37] P. Plötz et al., « Greenhouse gas emission budgets and policies for zero-Carbon road transport in Europe », Climate Policy, vol. 23, no 3, p. 343-354, 2023.
[38] B. C. O’neill, M. Dalton, R. Fuchs, L. Jiang, S. Pachauri, et K. Zigova, « Global demographic trends and future carbon emissions », Proceedings of the National Academy of Sciences, vol. 107, no. 41, p. 17521-17526, 2010.
[39] R. B. Jackson et al., « Global energy growth is outpacing decarbonization », Environmental Research Letters, vol. 13, no 12, p. 120401, 2018.
[40] R. Lal, C. Monger, L. Nave, et P. Smith, « The role of soil in regulation of climate », Philosophical Transactions of the Royal Society B, vol. 376, no 1834, p. 20210084, 2021.
[41] H. Ritchie, P. Rosado, et M. Roser, « CO₂ and greenhouse gas emissions », Our world in data, 2023.{r}