2.1 Cross-mapping of the 17 SDGs to the European Green Deal Policies
The mapping of the 17 SDGs to the EGD was done by reading all the policy documents selected (Table 1) and highlighting phrases that were semantically linked to the objectives of one or more SDGs. The sample of policies selected was based on expert judgement aiming to cover all relevant thematic policy areas: policy and strategy documents published after the EGD launch in December 2019 were retrieved, from the official website of European Union law (EURlex). For the search we using keywords covering the thematic areas of the EGD, namely Biodiversity, Building and renovating, Clean Energy, etc., as summarized in Table 1.
To decide whether there is a meaningful connection or not, we referred to the description of the 169 individual objectives defining the 17 SDGs (United Nations, 2017), and looked for keywords or synonyms. A similar matching approach was followed by Sachs et al. 2021, at a higher level though, as the 17 SDGs were mapped to the EDG's nine broad thematic policy areas, rather than specific policy documents, as herein. However, apart from the overall characterization of the degree of connection, we also assign a score, in a proportional way, that refers to the level of connection of each Policy to the SDGs, using a 4-point scale:
-
0: The Policy document does not interact with the specific SDG,
-
1: The Policy document enables the SDG outcomes,
-
2: The Policy document reinforces the SDG outcomes,
-
3: The Policy document directly affects the SDG outcomes.
The criterion for score assignment was the number of excerpts of a policy document that were linked to each SDG. Specifically, we assign a score of 3 to the SDGs with the maximum number of relevant phrases from the policy document, a score of 1 to the SDG with the least number of such phrases, and a score of 2 to those in between. Score 0 was received by the SDGs without connection to Policies.
Table 1
Mapping of Policies/Strategies to The European Green Deal Policy Areas. These 22 significant policy and strategy documents were published in 2020-21 in support of the implementation of the EGD.
EGD Policy Area
|
Name of Policy/Strategy
|
Biodiversity
|
· Biodiversity Strategy for 2030
· Circular economy action plan
· Blue economy strategy
|
Building and renovating
|
· A Renovation Wave for Europe – Greening our buildings, creating jobs, improving lives
|
Clean energy
|
· Hydrogen Strategy
· Offshore Renewable Energy Strategy
· Methane Strategy
· Energy poverty recommendation
|
Climate action
|
· European Climate Law
· European Climate Pact
· Adaptation Strategy
· Stepping up Europe’s 2030 climate Ambition
|
Eliminating pollution
|
· Chemicals strategy for Sustainability
|
From Farm to Fork
|
· Farm to Fork' strategy
|
Sustainable industry
|
· Industrial strategy
· Updating the 2020 New Industrial Strategy: Building a stronger Single Market for Europe’s recovery
|
Sustainable mobility
|
· Smart Mobility Strategy
|
Overarching
|
· Fit-for-55
· Strategy for Financing the Transition to a Sustainable Economy
· Annual Sustainable Growth Strategy (ASGS) 2021–7 flagship areas
· The European economic and financial system: fostering openness, strength, and resilience
· Directing finance towards the European Green Deal
|
2.2 Cross-mapping through a Machine Learning (ML) Method
In parallel with the above method, a Deep Learning Model was developed. Its added value is the capability of capturing the semantic similarity between policies and SDGs. The main advantage of using ML, is the speed and preciseness of the process. It is a smart and accurate tool that could reveal hidden connections between texts, that are not easily noticed by the human eye. Moreover, the findings of the "human" approach could be validated. Such ML algorithms have been extensively used in the literature for accurate sentiment analysis (Maulud et al., 2021; Trappey et al., 2020).
Deep learning refers to extensive neural networks with many layers (deep) that “allow computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction” (LeCun et al., 2015). In 2017, Google Research introduced The Transformer, a deep learning model based on attention mechanisms, dispensing with recurrence and convolutions entirely (Vaswani et al., 2017). This innovation led to the development of a wide range of models based on transformers, allowing the processing of entire sequences without the need for labelled data in pre-training.
In this work, we fine-tuned a pre-trained transformer-based model to find the similarity score of each policy document with each SDG. This is called BERT (standing for “Bidirectional Encoder Representations from Transformers”) and was introduced by Google Research in 2018 (Devlin et al., 2018). BERT is a bidirectional transformer pre-trained by using masked language modelling objective and next sentence prediction. Therefore, they are more advanced than Standard Language Models which are unidirectional, thus limiting the architectures that can be used for pre-training.
A pre-trained model supports the effort of defining the similarity score between each one policy document and SDG’s – Targets - Indicators definitions, trying to compensate for the ambiguity of Natural Language. Data used for BERT pretraining come from the Toronto Book Corpus (c.800 million words) and Wikipedia (c.2,500 million words). For the purposes of the study, the “bert-base-uncased” (12-layer, 768-hidden, 12-heads, 110M parameters) model was used.
To fine-tune the pretrained BERT model, the latest available OSDG Community Dataset was used, along with the description of the targets and indicators of each Goal[1]. The OSDG Community Dataset is the result of the work of more than 1,000 volunteers from all over the world using the OSDG Community Platform, who label sentences and paragraphs according to their relation with each SDG. Each labelling exercise is a binary decision problem, in which volunteers decide whether specific text excerpts relate to a proposed SDG or not. A ratio is then calculated as follows:
$$agreement=\frac{\left|{Labels}_{positive}- {Labels}_{negative}\right|}{{Labels}_{positive}+ {Labels}_{negative}}$$
The dataset is formed after text excerpts of paragraphs deriving from public documents, such as reports, policies, and publication abstracts. Furthermore, some documents originate from UN-related sources (e.g. SDG-Pathfinder and SDG Library). The released dataset (OSDG, 2021) constitutes of 32,115 labelled document excerpts and it contains the referred SDG, the number of volunteers that classified the connection to the SDG as negative, the number of volunteers that classified the connection to the SDG as positive and the agreement score based on the formula:
For the purposes of the presented research, data used were pre-selected using the following criteria:
• \({Labels}_{positive}\) > \({Labels}_{negative}\), as we needed only to use data related to an SDG.
• \(agreement\) > 0.6, as we needed to be sure that the volunteers agreed to the labelling.
This pre-selection process produced 14,280 excerpts, in which minor corrections were made, such as separation of combined words and replacements of wrong letter. In the final set of excerpts, we added the indicators and the descriptions for each of the 169 targets of the SDGs, retrieved from Ritchie et al. (2018). This process led to a total of 15,083 text excerpts for model fine-tuning.
The number of text excerpts used for each SDG are shown in Table 2:
Table 2
Text excerpts used for each SDG
SDG
|
1
|
2
|
3
|
4
|
5
|
6
|
7
|
8
|
9
|
10
|
11
|
12
|
13
|
14
|
15
|
16
|
17
|
Number of extracts used for fine-tuning
|
970
|
726
|
1639
|
1993
|
1948
|
1132
|
1391
|
759
|
605
|
385
|
1081
|
232
|
991
|
613
|
480
|
66
|
72
|
The model was developed in Python, using PyTorch and Scikit-Learn. 80% of the text excerpts were used as training data and 20% as testing data. Adam Optimizer was used as an optimizer, while the Cross-Entropy Loss was chosen as the optimization criterion. Adam (short for Adaptive Moment Estimation) is a popular optimization algorithm used in machine learning for optimizing the parameters of a model during training. Entropy Loss is a term used in machine learning and specifically in the context of training neural networks for classification tasks. Next, the model was trained for 10 Epochs with a learning rate of 10− 5, resulting in an accuracy score of 0.889. Considering that the training data used for each SDG are not of the same size and that a text excerpt is most probably linked to more than one SDG, the accuracy score is acceptable.
Our objective is to get an answer to this question: “What is the probability that X policy document is linked to the Y SDG?”. Scores are calculated by the model described above. The higher the score the bigger the probability for a scanned policy to be linked to a given SDG.
After getting our first results, we noticed that some scores are extremely high due to the high relevance of the policy to an SDG. Of course, this is not wrong, but the dominance of some SDGs over the others makes the interpretation of the results more difficult. So, we took another step and for each policy, we excluded the highest value. Then, we re-ran the process and obtained the scores distributed in a more uniform manner.
2.3 Mapping EGD Policies to the 6 SDGs Transformations
The SDGs and the Paris Agreement on Climate Change (2015) require governments to implement major transformations with the input of civil society, the scientific community, and business. Governments need to engage in robust multi-sector collaboration to meet their commitments and adequately prioritize the 17 SDGs in their national policies.
Sachs et al. (2019) proposed Six Transformations necessary to achieve the 17 SDGs, and calculated each one’s relationship with the SDGs using scores for 0 to 3. These transformations refer to: 1. Education, Gender, and Inequality; 2. Health, Wellbeing, and Demography; 3. Energy Decarbonization and Sustainable Industry; 4. Sustainable Food, Land, Water, and Oceans; 5. Sustainable Cities and Communities; and 6. Digital Revolution for Sustainable Development. They are all interconnected and mutually reinforcing and achieving them will require collaboration and action across sectors and at all levels of society.
Based on the mapping of policy documents to the SDGs, and the mapping of the SDGs to the six transformations, it is possible to estimate the linkage of Policies directly to the Transformations, by following three steps:
-
Step 1: For each Transformation, calculate the simple average of the contribution of SDG in each transformative category (Table 3), as given by Sachs et al. (2019).
-
Step 2: Multiplying the Table derived from step 1 with the scores of mapping policies to the SDGs with the human approach. A 22x6 matrix is obtained (Table 4), showing the extent that each policy contributes to each Transformation.
-
Step 3: Using the data from Table 4, a Sankey diagram is produced, using SankeyMATIC, to visualize the influence of EGD Policies on the six Transformations (Fig. 1).
Table 3
Average SDG contribution to the 6 Transformations.
|
1. Education, Gender, and Inequality
|
2. Health, Wellbeing and Demography
|
3. Energy Decarbonisation and Sustainable Industry
|
4. Sustainable Food, Land, Water, and Oceans
|
5. Sustainable Cities and Communities
|
6. Digital Revolution for Sustainable Development
|
SDG 1-No poverty
|
2,00
|
2,00
|
1,33
|
2,00
|
1,50
|
2,00
|
SDG 2-Zero hunger
|
2,00
|
3,00
|
1,33
|
3,00
|
1,50
|
2,00
|
SDG 3-Good health and well-being
|
1,67
|
3,00
|
2,33
|
3,00
|
1,50
|
2,00
|
SDG 4-Quality education
|
1,67
|
2,00
|
1,00
|
1,00
|
1,50
|
2,00
|
SDG 5-Gender equality
|
2,00
|
3,00
|
1,33
|
2,00
|
1,50
|
1,00
|
SDG 6-Clean water and sanitation
|
1,00
|
0,00
|
2,00
|
3,00
|
2,50
|
1,00
|
SDG 7-Affordable and clean energy
|
1,67
|
0,00
|
2,33
|
1,00
|
1,00
|
2,00
|
SDG 8-Decent work and economic growth
|
2,33
|
2,00
|
2,00
|
2,00
|
1,50
|
2,00
|
SDG 9-Industry, innovation, and infrastructure
|
2,00
|
1,00
|
2,00
|
1,00
|
2,50
|
3,00
|
SDG 10-Reduced inequalities
|
2,00
|
2,00
|
1,67
|
2,00
|
2,00
|
2,00
|
SDG 11-Sustainable cities and communities
|
1,00
|
1,00
|
2,67
|
2,00
|
3,00
|
2,00
|
SDG 12-Responsible consumption and production
|
1,33
|
1,00
|
2,67
|
3,00
|
2,00
|
2,00
|
SDG 13-Climate action
|
1,67
|
0,00
|
2,67
|
3,00
|
2,50
|
2,00
|
SDG 14-Life below water
|
1,33
|
0,00
|
2,00
|
3,00
|
1,00
|
1,00
|
SDG 15-Life on land
|
1,33
|
0,00
|
2,33
|
3,00
|
1,00
|
1,00
|
SDG 16-Peace, justice, and strong institutions
|
1,00
|
1,00
|
1,33
|
1,00
|
0,50
|
1,00
|
SDG 17-Partnerships for the goals
|
1,00
|
0,00
|
0,67
|
1,00
|
0,00
|
2,00
|
Table 4
Link between Policies and the Six Transformations.
|
1. Education, Gender, and Inequality
|
2. Health, Wellbeing and Demography
|
3. Energy Decarbonisation and Sustainable Industry
|
4. Sustainable Food, Land, Water, and Oceans
|
5. Sustainable Cities and Communities
|
6. Digital Revolution for Sustainable Development
|
European Climate Pact
|
34
|
26
|
43
|
50
|
39
|
39
|
Directing finance towards the European Green Deal
|
19
|
10
|
25
|
29
|
21
|
20
|
European Climate Law
|
40
|
24
|
51
|
59
|
41
|
46
|
Fit for 55
|
43
|
24
|
57
|
55
|
50
|
54
|
A New Industrial Strategy for Europe
|
42
|
27
|
49
|
50
|
39
|
50
|
EU Hydrogen Strategy
|
36
|
21
|
44
|
38
|
39
|
45
|
7 technology flagship Areas, ASGS for 2021
|
43
|
33
|
53
|
50
|
46
|
51
|
Chemicals strategy for Sustainability
|
32
|
19
|
47
|
52
|
36
|
38
|
EU Strategy to reduce methane emissions
|
31
|
24
|
36
|
40
|
30
|
37
|
A Renovation Wave for Europe
|
30
|
14
|
43
|
39
|
36
|
39
|
EU Commission Recommendation on Energy Poverty
|
26
|
18
|
28
|
29
|
26
|
28
|
EU Strategy to harness the potential of offshore renewable energy for a climate neutral future
|
32
|
14
|
42
|
37
|
34
|
41
|
Smart Mobility Strategy
|
34
|
20
|
46
|
45
|
39
|
43
|
Updating the 2020 New Industrial Strategy: Building a stronger Single Market for Europe’s recovery
|
42
|
27
|
49
|
50
|
39
|
50
|
EU Biodiversity Strategy for 2030
|
39
|
26
|
47
|
56
|
34
|
42
|
Stepping up Europe’s 2030 climate Ambition
|
42
|
26
|
56
|
54
|
47
|
50
|
EU Strategy on Adaptation to Climate Change
|
57
|
40
|
70
|
77
|
61
|
65
|
Circular Economy Action Plan
|
38
|
23
|
49
|
55
|
41
|
43
|
Farm to Fork Strategy
|
41
|
31
|
48
|
58
|
37
|
45
|
The European economic and financial system: fostering openness, strength, and resilience
|
26
|
15
|
30
|
28
|
21
|
32
|
The EU's Blue Economy for a Sustainable Future
|
30
|
18
|
40
|
46
|
35
|
35
|
Strategy for Financing the Transition to a Sustainable Economy
|
36
|
20
|
44
|
44
|
36
|
42
|
2.4 The value of Natural Capital in Europe and its link with the SDGs achievement
The 17 SDGs represent a baseline framework of sustainable development for future generations. The goals are deeply interconnected, which means that failure to any one of them hinders progress on others. According to the OECD, there is a huge funding gap in the implementation of the SDGs, estimated at USD 4.2 trillion per year, which was further amplified by the COVID-19 recession (OECD, 2020).
Another estimation, that included the cost of meeting growing commitments under the Paris Agreement and the cost of creating financial inclusion and prosperity for large parts of the world, found that the actual financing gap is likely to double or more, estimating it to be between USD 8.4 trillion and USD 10.1 trillion, which equates to almost 9–11% of global GDP in 2021 (Patel et al., 2020). This means that it is increasingly urgent for governments to develop comprehensive, sustainable, and inclusive approaches to finance the SDGs.
Money should not come solely from public finances, but from private sector, too. The motivation of private companies to adopt a holistic environmental strategy and invest seriously in SDGs, will come when they realize that a sound sustainability performance, generally implies a good financial performance as well (Koundouri et al., 2022).
Natural Capital provides a wide range of services called ecosystem services, which make the economy functional, therefore companies must realize the interaction among all types of Capital (Natural, Human, Produced), and how much dependent they are on each of them. This involves quantifying their impact on Natural Capital, Human Capital and Produced Capital to help them develop an appropriate strategy to address their business risks and opportunities.
Recognizing the importance of natural capital in the transition to sustainability and the need to help all stakeholders understand the value of nature and its contribution to society, we provide a valuation of the European Ecosystem Services in order to shed light on the full cost associated with the transition from the status quo to the complete achievement of the 17 SDGs, focusing on three main types of ecosystems: terrestrial, marine, and freshwater. The empirical analysis is aimed at first deriving the economic value of EU ecosystems, and then, building on the results, the study integrated the unit value of ecosystems with the SDG index. This enables the study's second goal, which is to quantify the social-economic value derived from shifting from the status quo of ecosystems to full SDG achievement.
The valuation of Ecosystem Services was done with the method of Benefit Transfer, using meta-regression analysis. Benefit transfer is a widely used approach for the estimation of economic values for ecosystem services by transferring available information from studies already completed in another location and/or context (Johnston, 2015).
Empirical valuation studies were used to obtain measures such as public’s willingness to pay (WtP) to enjoy ecosystemic services in specific biogeographical regions. The primary literature was retrieved from the publicly accessible database EVRI (Environmental Valuation Reference Inventory), using Europe, and publication dates between 2012–2022, as selection criteria. The search initially returned 212 studies, which after screening resulted in 165 studies to be used for data extraction (more details in the Supplementary Material).
The studies were separated by ecosystem typology: Terrestrial, Marine, and Freshwater and each one was used to extract specific parts information. An excel file filled-in with the following data:
-
Study details: Title, authors year of publication etc.
-
Willingness to Pay (WtP) or Willingness to Accept (WtA): Continuous variable expresses the average WtP in EUR in an annual basis.
-
Ecosystem: Categorical variable for the typology of ecosystem considered in the study. The categorization of the Mapping and Assessment of Ecosystems and their Services (MAES) Typology for ecosystem classification (Zhongming, 2015) was followed: Terrestrial [Forest (42 studies), Cropland (18), Heathland and Shrub (1), Sparsed vegetated land (1), Urban (15), Grassland (6), and Inland and Wetlands (3)], Freshwater [Rivers and Lakes (14)], and Marine [Marine and Coastal (65)].
-
Type of Ecosystem Service: Dummy variables referring to Cultural, Provisioning, Regulating, Supporting services.
-
Survey design: Categorical variable describing the different methods for data collection, e.g., Computer-aided individual interviews, focus groups, in-person interview, etc.
-
Data year: Year of data collection
-
Valuation method: Categorical variable indicating the method used to develop the analysis, i.e., Contingent valuation, Choice experiment, Actual Expenditure/Market price, Count data model, Hedonic Price Method, Hedonic Property, Meta-analysis, Replacement costs, Travel cost method. In our final dataset, we have 76 Choice Experiment (CE) studies and 67 CVM studies and 22 studies from studies using revealed preferences.
-
Location: Categorical variable for the geographical area in which the analysis has been developed
-
Country: Categorical variable for the European country in which the analysis has been developed
-
Biogeographical and marine regions: Dummy variables indicating the specific biogeographical and marine regions of European Union in which the study has been developed, namely Alpine, Atlantic, Black Sea, Boreal, Continental, Macaronesian, Mediterranean, Pannonian, Steppic. The categories are those used for reporting under Article 17 of the Habitats Directive (92/43/EEC).
-
Value elicitation method: Categorical variable indicating the typology of elicitation used in the study
-
Age: Continuous variable indicating the average value of age of the sample population (in years). In case of missing values, we used the mean age per country in the reference year was used, as derived from EUROSTAT database
-
Income: Continuous variable indicating the average annual income of the sample population (in EUR). In case of missing values, the average income per country in the reference year was used, as derive from EUROSTAT database, or from EU-SILC and ECHP surveys.
-
Gender: Percentage of males and females in the sample population.
-
Education: Percentage of people in the sample, with high education level. In case of missing values, we used the percentage of population attended tertiary education in the country and the year of reference, as provided by EUROSTAT.
After gathering and cleaning up the dataset, we estimated the WtP, using the meta-regression model given by the formula:
𝑌𝑖 = 𝛾 + 𝛽′𝑋𝑖 + 𝜀𝑖 (1)
Where \(Y\), is dependent variable (in our case, WtP), \(i\) refers to observations gathered from the studies, \(\gamma\) is the intercept of regression, \(\beta {\prime }\) is the vector of parameters to be estimated as slopes of the matrix of the explanatory variables \({X}_{i}\); \(\epsilon\) is the error term. For further details on the results of the regression, please refer to the supplementary material.
Finding a balance between socioeconomic development and ecosystem services is a crucial challenge for sustainable development (McCartney, 2014). To gain a high-level understanding of how WtP for ecosystem services relates to the achievement of 17 SDGs, for the 27 countries of the European Union, we calculated the correlation of the SDG scores per country, as provided by the UNSDSN Sustainable Development Report Europe 2021, and the Marginal WtP per country, calculated previously.
[1] For SDG16 and SDG17 we used expressions linked to these goals from the human approach, as the OSDG Community Dataset does not include texts for these goals, yet.