The Paris Agreement is a crucial framework for global climate action, with countries outlining their climate goals and strategies through Nationally Determined Contributions (NDCs). While existing research has primarily focused on assessing the mitigation targets within NDCs, the broader textual content of these documents has received little systematic analysis. This content includes not only mitigation targets but also national contexts, implementation plans, fairness, and transparency. However, the lack of transparency and comparability in NDCs, particularly regarding specific policies, financing, and adaptation measures, poses challenges to achieving global climate goals. To address this, Ivan Savin, Lewis C. King, and Jeroen van den Bergh employed Natural Language Processing (NLP) to systematically analyze the full textual content of NDCs, aiming to uncover deeper discourse and explore the focus and evolution of climate actions across different countries.

Source of the Paper

The paper was co-authored by Ivan Savin (ESCP Business School, Institute of Environmental Science and Technology at Universitat Autònoma de Barcelona), Lewis C. King (Institute of Environmental Science and Technology at Universitat Autònoma de Barcelona), and Jeroen van den Bergh (Institute of Environmental Science and Technology at Universitat Autònoma de Barcelona, Vrije Universiteit Amsterdam). It was published in March 2025 in the journal Nature Sustainability under the title “Analyzing Content of Paris Climate Pledges with Computational Linguistics.” The research was funded by the European Research Council (ERC) and is part of the “María de Maeztu” Programme for Units of Excellence by the Spanish Ministry of Science and Innovation.

Research Process

1. Data Collection and Preprocessing

The research team collected all NDC documents available on the Climate Watch platform and the United Nations Framework Convention on Climate Change (UNFCCC) NDC registry as of May 31, 2023. A total of 309 NDCs were gathered, including 167 initial submissions and 142 updated versions. For non-English documents, the team used the DeepL translation service to convert them into English. Data preprocessing included tokenization, removal of punctuation and stop words, lemmatization, and the generation of bi-grams. The final dataset consisted of 7,599 unique words and 539,902 word occurrences.

2. Topic Modelling

The study employed Structural Topic Modelling (STM) to analyze the NDC texts. STM is a topic modelling method that combines natural language processing and machine learning, clustering texts into distinct themes and analyzing topic distributions based on document features such as GDP and emission intensity. To address the limited number of texts, the team segmented each NDC into blocks of approximately 1,000 words, resulting in 1,280 text blocks. Using the STM model, the team identified 21 topics and grouped them into 7 thematic clusters.

3. Topic Analysis and Clustering

The research team calculated the distribution of the 21 topics across each NDC and conducted clustering analysis of 167 countries based on topic similarity. Using Euclidean distance and hierarchical clustering, the countries were grouped into 9 clusters. Additionally, regression analysis was used to explore the relationship between topic distribution and country characteristics such as GDP, emission intensity, and vulnerability.

Key Findings

1. Topic Identification and Distribution

The 7 thematic groups identified and their respective proportions were: Development (25%), Implementation and Planning (21.5%), Mitigation Targets (20.3%), Policies and Technologies (11.3%), Impacts of Climate Change (10.7%), Agriculture and Ecosystems (7.4%), and Stakeholders (3.8%). The Development group, covering sustainable, economic, and rural development, was the most prevalent in NDCs. The Mitigation Targets group focused on greenhouse gas accounting, emission scenarios, and target reporting.

2. Country Clustering Analysis

Based on topic distribution, the research team grouped countries into 9 clusters. For example, a cluster (C1) comprising Australia, Canada, and the UK focused on Policies and Technologies, while another cluster (C2) including the EU and the US emphasized Mitigation Targets. Developing countries such as Brazil, India, and Small Island Developing States (SIDS) prioritized sustainable development and climate change impacts.

3. Relationship Between Topics and Country Characteristics

Regression analysis revealed that high-GDP countries focused more on mitigation targets, while developing countries emphasized development and adaptation. For instance, OPEC countries highlighted economic development, while SIDS focused on climate vulnerability.

4. Evolution in the NDC Update Process

The study found that updated NDCs showed improvements in transparency and specificity, particularly in detailing mitigation targets and policies. However, attention to topics such as international support and green energy technologies declined, possibly due to the expanded focus on other areas in updated NDCs.

Research Conclusions

The study systematically analyzed the full textual content of NDCs using natural language processing, uncovering the focus and evolution of climate actions across countries. It revealed that NDCs encompass more than just mitigation targets, including development, adaptation, and implementation. High-income countries prioritized mitigation targets, while developing countries framed NDCs as part of sustainable development. The study also highlighted the lack of transparency and comparability in NDCs, recommending standardized formats to enhance credibility and transparency.

Research Highlights

1. Systematic Analysis: The first study to systematically analyze the full textual content of NDCs using natural language processing, filling a gap in existing research.
2. Topic Identification and Clustering: Identified 21 topics and grouped them into 7 thematic clusters, providing new perspectives on NDC content.
3. Relationship Between Topics and Country Characteristics: Revealed the relationship between topic distribution and country characteristics through regression analysis, offering insights for differentiated climate policies.
4. Evolution in the NDC Update Process: Found that updated NDCs improved in transparency and specificity, providing a valuable reference for assessing climate action progress.

Research Significance

The study not only provides new tools and methods for understanding NDC content but also offers critical insights for global climate policy formulation and evaluation. By enhancing the transparency and comparability of NDCs, the research helps bridge the gap between the goals of the Paris Agreement and implementation progress, advancing global climate action.