Exploring disaster impacts on adaptation actions in 549 cities worldwide

Cities are projected to face increasing losses from extreme hazard-related disasters (floods, storms, wildfires). Learning from past disasters is crucial to enhance preparedness and reduce risk, making cities key arenas for adaptation decisions. Prior research shows mixed evidence on whether disasters drive local adaptation: some find strong effects, others suggest interactions with additional factors or no relationship, and most evidence is based on single/small-N cases in Europe/North America. The study articulates three perspectives: (1) disaster frequency and severity (fatalities, affected population, economic loss) may heighten attention and pressure for action but may also trigger blame avoidance, reduce urgency after perceived response success, or overwhelm communities; (2) disasters can shape different adaptation action types—specific (targeting hazards recently experienced), expansive (targeting hazards not recently experienced), generic preparedness (hazard-agnostic readiness measures), and other actions (climate- or environment-related but not disaster-specific); and (3) national adaptive capacity attributes (political stability, local government power, stakeholder diversity, meritocracy) can enable or constrain translation of disaster experience into actions. The research assesses how disaster frequency, severity, and adaptive capacity relate to these action types across 549 cities worldwide.

The literature offers competing views on the disaster–adaptation link: some studies report disasters catalyze adaptation, others find effects conditional on context or no effect. Evidence has been largely limited geographically and methodologically. Theoretical expectations include agenda-setting and public pressure after high-impact events, but also blame avoidance and capacity overload. Frameworks of adaptive capacity emphasize political stability, stakeholder inclusion, meritocracy, and local government power as mediators. Prior work suggests recent events and city size can influence climate action. The study builds on this by disaggregating adaptation actions, distinguishing multiple severity metrics, and testing conditioning effects of adaptive capacity.

Data: The study integrates EM-DAT (International Disaster Database) records of 673 natural hazard events (2013–2018) with CDP (Carbon Disclosure Project) data on 3,604 city adaptation actions (2018–2019) spanning 243 action categories in 549 cities across 69 countries. Hazard types include drought, earthquake, extreme temperature, flood, landslide, mass movement, storm, volcanic activity, and wildfire. Linkage and geocoding: EM-DAT is country-level; the authors manually geocoded subnational affected administrative units (first-order) to link hazards to cities’ regions using Natural Earth and ArcGIS administrative boundary datasets. CDP provides city and country names enabling regional assignment. Actions were classified by whether they address specific EM-DAT hazard types via direct effects. Action typology: Actions were coded as (i) Specific (target hazards that occurred in the city’s region), (ii) Expansive (target hazards not occurring in the region), (iii) Generic preparedness (hazard-agnostic disaster readiness), and (iv) Other (climate/social/environmental actions not directly disaster-related). An inter-coder reliability test on 20 actions showed >70% agreement among three coders. Measures: Disaster frequency is the number of events per region (2013–2018). Severity measures: average economic damages (inflation-adjusted to 2021 USD), affected population, and fatalities. Authors also constructed normalized measures: frequency as a ratio to country total events; damages normalized by 2017 GDP per capita; fatalities/affected normalized by city population. Baseline measures use 15-year country running averages (1998–2012) to capture deviations from historical norms. Time lag is the mean years between disasters and reported actions. Analysis: Bivariate OLS regressions tested associations between single disaster metrics and counts of actions by type. Multiple OLS models combined frequency and severity and included controls: time lag, city population, and national adaptive capacity (political stability, meritocracy, stakeholder diversity, local government power from QoG and V-Dem). Moderation analyses tested interactions of frequency or damages with adaptive capacity attributes and city population for specific, expansive, and general actions. Variance inflation factors were <5, suggesting multicollinearity was acceptable.

Descriptive: Of 3,604 actions (2018–2019), 47% specific (n=1,691), 17% expansive (n=624), 16% generic preparedness (n=575), 20% other (n=714); actions span all world regions, over half in high-income countries. Bivariate associations (absolute measures; Table 1):

• Frequency: positive for specific (b=0.615, p<0.01); negative for expansive (b=-0.231, p<0.01); not significant for general or other; positive for all actions combined (b=0.557, p<0.05).
• Economic losses: positive for specific (b=0.322, p<0.05), general (b=0.103, p<0.1), and all actions (b=0.588, p<0.05); not significant for expansive/other.
• Affected population: weak negative for other (b=-0.235, p<0.1); otherwise not significant.
• Fatalities: not significant for any type in absolute models; in normalized models some weak negative effects appeared. Baseline (15-year) measures yielded few significant effects, except frequency predicting specific actions (b=0.845, p<0.05). Combined models: Adding multiple disaster metrics marginally improved fit; frequency’s positive specific effect weakened when controlling for damages; frequency’s negative effect on expansive actions strengthened (b≈-0.307, p<0.01). The interaction frequency × damages was not significant. Controls and mediators: City population and political stability positively correlated with most action types (except ‘other’). Time lag was negative for specific and expansive actions, indicating actions follow more recent events. Stakeholder diversity showed significant negative associations mainly with specific actions. Moderation (Table 2; absolute damages):
• Frequency × adaptive capacity: largely insignificant across action types (some weak effects for expansive).
• Damages × adaptive capacity: Significant positive interactions for specific actions with political stability (b=0.126, p<0.05) and meritocracy (b=0.412, p<0.01); negative with stakeholder diversity (b=-0.472, p<0.05). For general actions, positive interactions with political stability (b=0.068, p<0.01), meritocracy (b=0.125, p<0.05), and stakeholder diversity (b=0.299, p<0.01). For expansive actions, damages × stakeholder diversity was positive (b=0.215, p<0.05). Simple slopes: Economic damages predict more specific actions in high political stability (b≈0.48) and high meritocracy (b≈0.63) contexts; effects vanish at low levels. Damages predict more general preparedness in high stability, meritocracy, and stakeholder diversity contexts; effects are absent at low levels. At low stakeholder diversity, damages associate with more specific actions; at high diversity, this specific effect disappears. Overall: Disaster frequency/severity has limited influence on city adaptation actions once controls are included. Economic damages matter more than human losses and primarily under high adaptive capacity. Frequency tends to reduce expansive actions. A wealth effect is evident: cities in affluent, high-capacity countries experience larger absolute economic losses and have greater governance capacity, creating both incentives and opportunities for adaptation.

The study addresses whether disasters spur urban adaptation and finds overall modest and context-dependent effects. Disaggregating adaptation into specific, expansive, generic preparedness, and other actions reveals distinct patterns: higher disaster frequency tends to crowd out expansive actions, plausibly because attention and resources focus on known hazards rather than unexperienced threats. Economic damages, more than human toll, are associated with increases in specific and generic preparedness actions, but primarily where political stability and meritocracy are high, suggesting institutional capacity conditions how loss experiences translate into action. City size and recency of events further shape action, highlighting temporal windows of opportunity. Stakeholder diversity shows mixed effects—potentially enhancing generic preparedness while complicating specific actions—consistent with inclusion versus veto dynamics. Collectively, the findings nuance the notion that disasters routinely catalyze adaptation, showing that institutional context and action type are critical.

The paper provides a global, city-level assessment linking disaster frequency and severity to different adaptation action types. It shows that: (i) disasters’ frequency and human impacts generally do not robustly predict adaptation; (ii) economic damages can spur specific and generic preparedness actions, contingent on adaptive capacity (political stability, meritocracy); (iii) increased frequency can reduce expansive adaptation; and (iv) larger cities and shorter time since events are associated with more action. A wealth effect likely underlies observed patterns, as high-income, high-capacity contexts both incur larger recorded damages and possess resources to act. The study underscores the need to disaggregate adaptation actions and incorporate institutional conditioning. Future research should employ finer-grained, sub-national indicators of adaptive capacity and local impacts, expand coverage in the Global South, and unpack causal mechanisms such as interest mobilization, learning, and governance processes that translate disaster experience into different adaptation pathways.

EM-DAT includes only larger events (threshold-based), omitting smaller hazards that may trigger adaptation. The design assumes disasters in 2013–2017/2014–2018 influence actions in 2018/2019, excluding same-year reactions and potentially introducing spurious timing if long gaps exist; CDP lacks adoption dates for actions. Geocoding relies on manual interpretation of EM-DAT’s subnational descriptors, introducing potential location error. Adaptive capacity and other controls are measured at the national level; sub-national variation could not be captured, risking measurement error. The CDP sample is skewed toward high- and upper-middle-income countries, raising equity and generalizability concerns. Model fits are modest and many effects are sensitive to measurement choices. Human loss measures showed limited effects across specifications.