Early engagement and co-benefits strengthen cities’ climate commitments

Cities account for the majority of energy-related CO2 emissions and are pivotal for achieving global climate targets. Since the 1990s, city networks such as ICLEI and C40 have advanced ambitious urban climate policies. Recent compound crises—COVID-19, geopolitical conflict, and intensifying climate disasters—raise concerns about whether cities can sustain and advance climate action under pressure. Prior work suggests crises, especially economic ones, often erode environmental commitments, but most evidence is national-level; city-level analyses are sparse and often limited to specific actions or geographies. The study asks how COVID-19 affected cities’ climate actions and finance globally, and which factors are associated with sustaining or enhancing commitments, including engagement in longer-term green recovery. The purpose is to generate global, generalizable evidence to inform how to bolster the durability of urban climate policy through crises.

Existing literature generally finds negative crisis effects on environmental policy, with some evidence of reconfiguration of economy–environment relationships. City-level research has focused largely on initial adoption of climate targets and policies rather than responses during crises. Influences on local climate policy adoption identified in prior studies include exposure to environmental stress (e.g., hazards, air pollution, climate-related health impacts) and early engagement with climate/sustainability (e.g., membership in climate networks, business collaboration, alignment of economic development with sustainability). Studies also highlight roles of fiscal capacity, governance type, government response capacity, and public/business support. There is limited evidence on disaster impacts on urban climate actions and on policy termination, with some cases from Japan and the U.S. on solar adoption. This study builds on these strands to examine how such factors moderate cities’ responses to COVID-19 vis-à-vis climate action, finance, and green recovery.

Data: The study uses self-reported 2021 CDP-ICLEI Unified Reporting System data from 793 cities worldwide. Key dependent variables are nominal categories capturing reported COVID-19 impacts on (1) overall emphasis on climate action and (2) climate finance available for climate action (increased, decreased/reduced, no change, other). A third dependent variable is a count of green recovery interventions (number of selected synergies between COVID-19 recovery measures and climate action across 11 listed options). Responses in 'other' were reviewed and reclassified where appropriate.

Explanatory variables: Guided by the literature, two main categories were operationalized. Environmental stress: (a) number of climate-related health issues reported; (b) number of medium-high/high-probability climate hazards; (c) pre-COVID air pollution (tropospheric NO2). Early engagement: (a) membership in at least one international climate network (from ClimActor); (b) presence of pre-2021 short-term GHG/climate targets; (c) collaboration with business on sustainability projects; (d) number of sustainability opportunities identified. COVID-19 impact controls: (a) change in economic activity proxied by VIIRS nighttime lights (difference vs counterfactual 2020); (b) national COVID-19 mortality per population; (c) national Government Response Index (OxCGRT). Additional controls: population density; pre-COVID city economic activity (log VIIRS NTL, 2019); national GDP (log); national liberal democracy index (V-Dem).

Data sources and processing: City coordinates and network participation were obtained via the ClimActor dataset; geospatial rasters included VIIRS nighttime lights (2019–2021), SEDAC/CIESIN population density, and NASA/OMI tropospheric NO2. Due to incomplete administrative boundary data, circular buffers (0.25° and 0.50° radii) around city centroids were used to compute zonal statistics, selecting measures correlating best with boundary-based statistics where available. National-level mortality (Our World in Data), OxCGRT government response (averaged display index), World Bank/OECD GDP, and V-Dem liberal democracy were integrated. CDP long-format responses were reshaped to city-level for analysis; variables were standardized or recoded as needed.

Statistical analysis: For climate action and climate finance outcomes (nominal), multilevel multinomial logistic regressions were estimated using Stata GSEM with random intercepts at the country level to account for clustering (likelihood-ratio tests supported multilevel specification). Results are reported as marginal effects with confidence intervals. For the count of green recovery interventions, multilevel negative binomial models were estimated. Robustness checks included alternative model specifications (clustered/unclustered), alternative operationalizations (e.g., categorical NO2; network membership by specific networks; alternative thresholds for 'short-term' targets; categorical NTL change) and sensitivity analyses; results were broadly consistent. A post hoc ANOVA tested associations between numbers of green recovery interventions and reported changes in climate action/finance.

• Overall patterns (n=793): More cities reported increased climate action (39.2%) than decreased (14.4%) due to COVID-19; for finance, more reported reduced finance (30.5%) than increased (22.4%). About 40.4% (action) and 41.5% (finance) reported no change.
• Geography: East Asia had the highest share reporting increased action (63.8%) and increased finance (48.9%). Decreased finance was most common in Africa (63.3%) and the Middle East (57.1%) among those regions' samples. Europe and North America most often reported no change.
• Action–finance linkage: Action and finance responses were associated (χ²=301.68, p<0.001), yet 11.3% of cities reported increased action alongside reduced finance (most prevalent in Africa and the Middle East, then Latin America). These cities tended to have lower economic productivity (NTL), lower national GDP, weaker liberal democracy, and higher likelihood of climate network membership.
• Environmental stressors: Each additional climate-related health issue was associated with higher probability of increased action (marginal effect 0.025; p=0.003) and increased finance (0.023; p=0.001), and lower probability of 'no change' (action −0.032, p<0.001; finance −0.025, p=0.006). Higher pre-COVID NO2 associated with increased finance (0.022; p=0.010). Number of climate hazards was not significantly related to action or finance changes.
• Early engagement: Climate network membership reduced the probability of decreased action (−0.087; p=0.017) and increased 'no change' in action (0.138; p=0.002), with no significant effect on finance. Collaboration with business increased probability of increased action (0.101; p=0.021) and increased finance (0.090; p=0.024). Identifying more sustainability opportunities reduced likelihood of decreased action (−0.012; p=0.038) and increased finance (0.011; p=0.026). Pre-2021 short-term GHG targets were not significant.
• COVID-19 impacts and governance: Economic impact (NTL change) and national mortality rates were not significantly related to action or finance. Stronger national Government Response Index scores were associated with increased action (0.008; p=0.001) and weakly with increased finance (0.004; p=0.062).
• Macroeconomic and political context: Higher national GDP was negatively associated with increased action (−0.027; p=0.040) and increased finance (−0.027; p=0.012). Higher liberal democracy reduced probabilities of decreased action (−0.002; p=0.029) and reduced finance (−0.005; p<0.001), while increasing 'no change' in both (action 0.004, p=0.001; finance 0.004, p=0.003), suggesting more stable (unchanged) commitments.
• Green recovery: 42.6% of cities reported no green recovery interventions; regions with the highest shares implementing at least one were Africa (73.3%), Middle East (71.4%), and Southeast Asia & Oceania (64.6%). Most implementing cities reported one intervention (24.8%); 7.94% reported more than five. More climate-related health issues (0.166; p<0.001), climate network membership (0.413; p=0.044), business collaboration (0.586; p=0.007), and more sustainability opportunities (0.111; p=0.001) were positively associated with the number of green recovery interventions; climate hazards showed weak positive association (0.072; p=0.082). Government response was weakly positive (0.024; p=0.094). Liberal democracy was negatively associated (−0.012; p=0.044). ANOVA showed numbers of green recovery interventions were significantly associated with reported changes in action (F=24.55, p<0.001) and finance (F=35.38, p<0.001).

The findings indicate that, despite pandemic-related pressures, many cities sustained or increased climate actions in the short term, even amid widespread reductions in climate finance. In parts of the Global South, increased action without parallel finance cuts was relatively common in the short term, yet higher engagement in green recovery suggests potential for longer-term investment realignment. In contrast, cities in higher-income, liberal democracies tended to maintain stable (unchanged) action and finance, and were comparatively less likely to implement green recovery measures, signaling lower short-term ambition.

Two reinforcing levers appear central for resilient climate commitments during crises: (1) co-benefits, especially links between climate and health, which are associated with increased action and finance and more green recovery interventions; and (2) early engagement, including climate network participation, business collaboration, and identification of sustainability opportunities, which correlate with sustained or increased action and green recovery. National governance capacity, as proxied by COVID-19 response stringency/effectiveness, also aligns with stronger city-level climate responses.

These dynamics suggest that embedding climate within broader urban agendas—health, economic development, mobility, ecosystem restoration—can stabilize and advance climate action during shocks. Coalitions that align municipal governments, businesses, and transnational networks appear to help maintain commitments. The global scope of the dataset extends previous largely regional or sector-specific studies, offering generalizable insights into how urban climate action weathers compound crises.

This study provides global empirical evidence that urban climate action often persists—and can even intensify—during crises, though finance constraints are common. Early engagement in climate and sustainability and attention to co-benefits, particularly health, are associated with more durable action, better-preserved finance, and greater uptake of green recovery measures. National governance quality during the pandemic further supported stronger city responses. Policy and practice should therefore: strengthen ties to climate networks; cultivate business collaboration; systematically identify and leverage sustainability co-benefits; and integrate green recovery into broader development strategies. Future research should examine causal pathways, assess how to align fiscal and regulatory incentives to embed sustainability in economic priorities (e.g., tax credits, subsidies), and investigate how to raise ambition in high-income, liberal democracies that currently tend toward stability rather than enhancement. Building richer, longitudinal, and globally comparable city-level datasets will be essential to track and sustain climate action through future crises.

Key limitations include reliance on self-reported city data, which reflect officials’ perceptions and may introduce recall or interpretation bias, and uneven data availability—especially in developing countries—limiting completeness and longitudinal comparability. Geospatial proxies (e.g., nighttime lights for economic activity) and centroid buffers were used due to incomplete boundary data, which may introduce measurement error. Some variables (e.g., city capacity, public support) lack direct, globally consistent indicators and are only partially captured by proxies. Causality cannot be inferred from the observational design despite robustness checks and multilevel modeling.