A computational approach to analyzing climate strategies of cities pledging net zero

The increasing number of cities setting net-zero emissions goals necessitates a deeper understanding of their strategies. Existing analysis is hampered by the incomplete and heterogeneous nature of city climate policy documents. While numerous reports document the landscape of net-zero target setting, a systematic global analysis is lacking. Previous studies often focus on specific regions (e.g., Europe) or individual cities, hindering broader comparisons. This paper addresses this gap by analyzing a large corpus of climate action plans from cities with net-zero targets. The research questions explore how cities articulate their plans to achieve these targets and whether certain textual patterns are predictive of more ambitious goals. The paper also examines the sectoral distribution of climate actions and potential trade-offs between different approaches. The study acknowledges the lack of a universally accepted definition for net-zero emissions at the city level, leading to varied interpretations and incomparable targets across different jurisdictions. The diversity in how cities define and prioritize climate action, often prioritizing local adaptation or mitigation approaches, further complicates systematic analysis. The paper utilizes machine learning and natural language processing to overcome these challenges and analyze a large dataset of climate strategy documents to gain global insights into cities' climate strategies.

The literature review highlights the growing number of cities committing to net-zero emissions, spurred by the IPCC's 1.5°C report and global climate action initiatives. Several reports have documented the landscape of net-zero target setting, but these analyses often lack the global scale and systematic approach needed for in-depth comparative analysis. Prior studies on climate action plans are frequently limited in scope, focusing on specific regions (like Europe) or single cities. While some research examines the content of climate action plans, it is often constrained by data availability and inconsistent documentation. Existing work emphasizes the heterogeneity and comparability challenges of net-zero targets due to varied definitions, scopes, and emissions coverage. Studies on mitigation and adaptation strategies reveal prioritization differences between Global North and South cities. Some studies have examined the links between mitigation and adaptation policies, demonstrating both distinct approaches and potential synergies between them.

The study collected climate policy strategy documents from 318 cities that had declared net-zero emissions reduction targets or committed to equivalent targets. The data was compiled from publicly available documents, obtained through internet searches and incorporating data from previous studies. The documents vary widely in format and length. A machine learning-based NLP approach was used to analyze the text data. The preprocessing steps involved optical character recognition (OCR) for image-based documents, translation of non-English documents, and removal of stop words. The text data was then converted into a numerical representation using tf-idf (term frequency-inverse document frequency) feature extraction of 1-grams and 2-grams. The analysis focused on two main objectives: (1) predicting economy-wide net-zero targets and (2) identifying key themes in cities’ climate action plans. For the first objective, a logistic regression model was used to predict whether a city had set an economy-wide net-zero target based on the text features. L1 regularization was applied to prevent overfitting and enhance model interpretability. The model's performance was evaluated using out-of-sample accuracy metrics. For the second objective, a key term-based topic analysis was performed. Keyword lexicons were created for nine key sectors (land use, industry, buildings, transportation, electricity, heating, waste/pollution, climate impacts, offsets). Word2vec similarity was used to expand these seed terms and a topic vector for each city was generated. Factor analysis was then used to identify underlying themes explaining how cities address different climate issues. Statistical tests were used to explore relationships between city characteristics (area, population, climate zone, region) and their climate strategy focus. The methodology employed several open-source tools like pytesseract, Google Translate API, scikit-learn, and spacy.

The logistic regression model identified text patterns predictive of ambitious, economy-wide net-zero targets. These patterns were grouped into four themes: (1) specific, quantitative metrics (percentage reduction, target years, etc.); (2) identification of emission reduction sources (specific technologies, materials, etc.); (3) governance mechanisms (mayoral involvement, fees, energy plans); and (4) human-centered approaches (inclusive strategies, community engagement). The key term-based topic analysis revealed that cities most frequently emphasized energy-related actions in their climate action plans, reflecting the largest emission sources. However, this focus often came at the expense of addressing other sectors, such as land-use and climate impacts. Correlation analysis showed trade-offs between certain topics. For instance, there was a negative correlation between emphasis on energy and discussions of land-use and climate impacts. Factor analysis reduced the nine identified topics into two dominant factors: "Ecology" (pollution/waste, land-use, impacts) and "Infrastructure" (heating, buildings, energy, transportation). A quadrant plot based on these factors revealed different city approaches. Quadrant II cities showed a balanced focus on ecology and infrastructure, while Quadrant I cities emphasized infrastructure at the expense of ecology. Quadrant III cities showed minimal focus on both, possibly due to less detailed planning documents. Quadrant IV cities prioritized ecology over infrastructure. Some cities demonstrated a more balanced approach. Further analysis suggested that climate zones and geographic regions might influence the emphasis on ecological versus infrastructure approaches. Tropical zones showed a stronger focus on ecology, while arid climates tended towards infrastructure. European cities showed a relatively lower emphasis on ecological considerations compared to other regions.

The findings indicate that successful climate action plans require specificity, quantification, and strong governance structures. While focusing on energy-related emissions is understandable given their significance, the observed trade-offs suggest opportunities for improved integration and cross-sectoral synergies. The dominance of energy-related actions might distract from other important mitigation areas, particularly consumption-based emissions. The negative correlation between energy focus and attention to land-use and climate impacts highlights a need for more integrated approaches. This integration could lead to more efficient resource use and capitalize on potential synergies between mitigation and adaptation efforts. The lack of significant relationships between city characteristics (population, area) and strategy focus underscores the need for further investigation into the underlying drivers of climate action planning choices. However, differences based on climate zone and geographic region offer intriguing avenues for future research. The study's findings contribute to a growing body of literature highlighting the need for more comprehensive and integrated climate action planning, including the integration of spatial planning and a broader scope of emissions consideration.

This study presents a novel, scalable approach using NLP to analyze the climate strategies of cities with net-zero targets. The findings highlight the importance of specific, quantitative metrics and strong governance for ambitious climate action. However, the overemphasis on energy-related actions at the expense of other crucial sectors underscores the need for more integrated and holistic approaches to climate planning. Future research should explore the causal relationships between plan characteristics, implementation actions, and actual emission reductions. Further research is needed to understand the factors driving the observed regional variations in climate action priorities.

The study is limited by its sample, which is predominantly composed of cities in the Global North, potentially limiting the generalizability of the findings. The analysis is also restricted to cities with publicly available climate action plans, excluding those without such documents. Moreover, the study focuses on the content of plans rather than their implementation, limiting its ability to assess the effectiveness of these strategies. The reliance on publicly available documents may also introduce bias, as the completeness and quality of documentation may vary across cities.