Large-scale text analysis of strategy documents finds that climate change actions vary among actors at different scales – and that there may be room for better coordination.

This content is cross-posted from the SpringerNature Sustainability Community blog – read the original version here

The 2015 Paris Agreement formally recognized the contributions of non-state actors (“NSAs” – cities, business, civil society, etc.) towards global climate change governance. Since this time, diverse NSA networks have continued to spring up all over the world to try and fill in the gaps in climate change mitigation left open by country-level mitigation commitments. These networks have served critical functions in global climate change governance – such as pooling and distributing resources – and have also contributed their own emissions reductions pledges and mitigation data.

These NSA efforts have made the global climate governance picture incredibly complex, and to get a better sense of how these actors fit into the big picture we wanted to understand two main things. First, whether cities, regional governments, businesses and countries are taking different types of actions; and second, whether actions are connected or disconnected. In other words, do different types of actors appear to be coordinating? These questions are important to understand from a governance standpoint, as they can potentially indicate whether complex and polycentric governance systems are operating in mutually reinforcing and beneficial ways (Ostrom, 2010), or whether there is governance fragmentation (Biermann, 2009) that might result in conflict and lost momentum down the road.

To try and make sense of it all, we collected data on over 9,000 actors and their climate strategy documents. We then leveraged Natural Language Processing (NLP) and topic modeling methods to run through these thousands of climate strategy documents from cities, regions, countries, and companies. Topic modeling is an “unsupervised technique,”  which means that the researcher doesn’t predetermine categories or themes and then search and code documents for them. This approach allowed us to “discover” a coherent topic framework that emerged from the collection of strategy documents where overlap between topics was minimized and our own implicit biases and logical priors didn’t affect the analysis.

Figure 1 from paper - per document per topic probabilities show estimated proportion of words from actors' documents generated from one of 30 topics

Figure 1. Per-document-per-topic probabilities for four groups of climate actors show the estimated proportion of words from actors’ documents that are generated from one of 30 topics in the topic model.

Themes and patterns in climate action

We discovered that strategies related to climate change adaptation, as opposed to mitigation, are overwhelmingly likely to be a part of country-level strategies (and mostly “developing” countries rather than “developed” countries), rather than cities, regions, or companies. Among NSA groups, climate adaptation and vulnerability topics were relatively unlikely to be covered in climate strategy documents. Additionally, out of 30 strategies identified by our topic modeling approach, just one was related to climate change adaptation, where all others were primarily related to climate change mitigation. This result indicates that there is a potential gap in the climate governance picture related to NSA-level contributions to climate change adaptation measures.

Compared to the strategies covered in country- and region-level climate strategy documents, city- and company-level planning documents showed a much greater diversity of approaches. For companies, most of the topics covered in strategy documents relate to technological approaches towards climate change mitigation, such as green building standards, and construction efficiency standards for heating, ventilation, and cooling (HVAC) systems. Additionally, company-level strategy documents were much more likely to formally incorporate carbon offsetting into their mitigation strategy than any other type of actor. City-level planning documents also cover technological and efficiency solutions, but are more likely to also contain topics related to ‘soft policy’ approaches like citizen awareness-building and public awareness campaigns. Both city and company-level strategies tend to overlook mitigating consumption-based or supply chain (“scope 3”) emissions – and accelerating scope 3 emissions commitments will be critical to limiting the carbon footprint of every actor both within and beyond their immediate physical boundaries.

Our analysis indicates that different types of actors are indeed (mostly) taking fundamentally different approaches to address climate change concerts, but does this mean that climate governance is becoming increasingly fragmented?

Not necessarily – but further study will be needed to determine more conclusively whether the global climate governance picture is increasingly characterized by fragmentation and strategy gaps or rather a kind of coherent hierarchy or division of labor among types of actors. We find some evidence to support either case. Because we find a high degree of strategy similarity within actor groups, this may mean that different types of actors are effectively coordinating with their peers. However, because we find relatively few strategic links across actor levels, as well as between actors located in different regions, this may also be pointing to missed opportunities for coordination.

In looking at the geography of commitments, we find that many city-level strategy commitments are located in European cities, and we know that at least some of these are the result of actor coordination across scales (e.g., EUCoM, 2016), evidencing perhaps both strong coordination within the EU and large gaps or opportunities outside of the EU to replicate or learn from such governance strategies. We also find that companies tend to have the highest proportion of strategic links with actors outside their own geography, pointing to another area where lessons could be learned by other actors looking to more effectively coordinate policy inter-regionally.

Looking Ahead

Mitigating the worst effects of climate change will take a high level of coordination within and across all types of actors and geographies. We have demonstrated here how using text-based data processing strategies such as NLP and topic modeling to point to both gaps and coordination in the existing actor network, and hope that these strategies can be continually leveraged and improved upon to provide a more coherent picture of global climate governance. Whether global climate governance is more appropriately characterized as cooperative or fragmented is still an open question, although this research provides a step towards shedding light on how diverse actors are taking climate action and whether similarities exist.

Read the full paper here.


Biermann, F., Pattberg, P., van Asselt, H., & Zelli, F. (2009). The Fragmentation of Global Governance Architectures: A Framework for Analysis. Global Environmental Politics9(4), 14–40.

EUCoM. (2016). The Covenant of Mayors for Climate and Energy Reporting Guidelines (p. 72). EU Covenant of Mayors & Mayors Adapt; Joint Research Centre of the European Commission.

Ostrom, E. (2010). Polycentric systems for coping with collective action and global environmental change. Global Environmental Change20(4), 550–557.