Overemphasis on recovery inhibits community transformation and creates resilience traps

The paper addresses a central question in resilience research: how can community resilience be operationalized to include not only rapid recovery (“bouncing back”) but also adaptation and systemic transformation after climate disasters? The context is the rising frequency and cost of billion-dollar disasters in the U.S., alongside urbanization, aging infrastructure, climate change, and policies that prioritize recovery over mitigation. Prevailing operational models emphasize engineering resilience—minimizing deviation and returning quickly to the status quo—often measured via restoration of critical services like electricity. The authors argue this recovery-centric focus risks maladaptation by reinforcing existing vulnerabilities and inequities. The study’s purpose is to jointly quantify recovery and transformation, identify community risk factors that either catalyze or inhibit transformation, and reveal “resilience traps” (factors predictive of recovery that do not enable positive transformation). Using Hurricane Michael (2018) in Florida as a case study, the study measures recovery via power restoration and introduces a method to detect transformation by shifts in community similarity networks.

Resilience is conceptualized across disciplines (social sciences, ecology, urban planning, engineering) as including capacities to bounce back and to adapt/transform. Operational community-level models have advanced but largely emphasize engineering resilience—rapid return to pre-disruption states—guiding decision frameworks and metrics focused on service restoration. This emphasis may promote maladaptation, for instance by disaster aid and insurance policies that can exacerbate wealth and racial inequalities and sustain vulnerabilities. Prior work has used composite indices to track resilience or vulnerability temporally, but such indices can be opaque and recovery-focused. The literature also links key social factors to disaster outcomes, including income inequality, poverty, mobility, housing tenure, and access to communication and transport. The authors position their contribution as integrating recovery modeling with a transformation-oriented, network-based approach to capture systemic reorganization beyond recovery.

Study scope and data: 67 Florida counties affected by Hurricane Michael (2018). Recovery is proxied by restored access to electricity using county-level power outage/restoration data. A large set of county-level risk factors (environmental opinions, sociodemographic, economic, housing, mobility) primarily from ACS and related sources are considered. Population and hazard exposure are controlled for in recovery modeling.

Recovery modeling (engineering resilience): An ensemble-of-trees predictive framework is used to relate county risk factors to recovery outcomes (restored electricity). Variable selection follows a three-stage procedure (VSURF with random forests) to identify the smallest set of predictors most informative for recovery via out-of-bag error permutation importance. Twenty risk factors are retained as key predictors of recovery, and their relative importance and direction of effect are reported. The model is tuned and selected via cross-validation.

Transformation modeling (Contrastive Community Networks, CCN): The CCN is built on Self-Organized Maps (SOM), an unsupervised dimensionality reduction/projection method that maps counties into a low-dimensional relational network where proximity reflects similarity across the selected risk factor portfolio. Using the 20 recovery-important risk factors (excluding storm exposure variables), data are standardized (mean 0, SD 1), and a SOM is trained (hexagonal, toroidal 5×8 grid). Each county is mapped to a baseline node representing its peer similarity cluster.

Transformation detection and thresholds: For each county and each risk factor, the factor value is perturbed incrementally (0.01 SD) across plausible bounds (e.g., scaled [0,1] for proportions; for unbounded variables, 1.5× observed min/max range). At each step, a new SOM is trained on the perturbed dataset, and the county’s node assignment is tracked. Transformation is defined as the first perturbation that remaps the county to a non-baseline node, indicating a shift to a different peer set. The perturbation magnitude at tipping is the transformation threshold; the Euclidean node distance between original and new nodes is the transformation trajectory length (degree of reorganization). Risk factors with no tipping across the perturbation range are deemed non-transformational for that county. Thresholds are computed across all counties, enabling comparison of which factors are more conducive to transformation.

Interpretation: Risk factors with positive thresholds on normatively positive attributes (e.g., public transport use) indicate potential for positive transformation, while positive thresholds on normatively negative attributes (e.g., higher inequality) indicate risk of degradation. Integration with recovery modeling allows identification of resilience traps: factors that contribute to recovery but offer no possibility of positive transformation.

• Integration of recovery and transformation reveals resilience traps: 55% (11 of 20) of risk factors most predictive of recovery showed no potential to trigger positive transformation in any Florida county analyzed.
• Transformation catalysts (lowest average transformation thresholds across counties) are: 1-year housing tenure (mean ~10.3%), income inequality (mean ~15.8%), and internet access (mean ~43.9%). Despite strong transformative potential, their recovery-importance ranks are relatively low (17th, 11th, and 18th, respectively), highlighting a mismatch between recovery and transformation priorities.
• In Bay County (heavily impacted by Michael), only 8 of 20 (40%) risk factors could trigger transformation. Positive-threshold examples include public transportation commuting, within-state relocation, racial inequality, income inequality, working from home, and 1-year housing tenure; negative-threshold examples include within-county relocation and renter-occupied housing. Thresholds ranged from ~11% (1-year housing tenure) to ~260% (public transportation commuting) for factors permitting transformation in Bay County. A ~29% increase in income inequality (Gini) would cause negative transformation (degradation risk) in Bay County.
• Across all counties, income inequality had the smallest mean transformation threshold (~6.25%), while public transportation commuting had the largest mean threshold (~12,042%), indicating substantial changes would be needed in transit mode share to shift peer similarity.
• Recovery-important but non-transformational factors (potential resilience traps) include variables related to poverty and socioeconomic status (e.g., income deficit, income through interest), language/ethnicity proxies (speaking API, speaking Spanish), education (GED, bachelor’s degree), and mobility/commute patterns (commuting alone, walking commute), among others. Their recovery-importance ranks span 4th to 20th.
• Recovery model results corroborate known patterns: commuting alone associates negatively, while walking commute associates positively with engineering resilience; education-related variables show negative association with engineering resilience, likely reflecting prioritization of dense urban cores in post-disaster power restoration.
• Policy implication: Solely prioritizing recovery-linked factors risks entrenching inequities and maladaptive states; incorporating transformation-catalyst factors in funding criteria (e.g., 1-year housing tenure, income inequality) can improve long-term resilience outcomes.

The findings demonstrate that operationalizing resilience solely as rapid recovery can create resilience traps: conditions that are predictive of faster restoration yet do not enable communities to transition to more sustainable or equitable states. By combining a recovery model with CCN-based transformation detection, the study shows substantial divergence between factors that expedite recovery and those that catalyze systemic change. For instance, poverty-related metrics (income deficit) and wealth-linked measures (income through interest) aid in predicting recovery but offer no transformation potential, reflecting how aid and insurance regimes can reinforce the status quo and widen inequalities. Conversely, housing tenure, income inequality, and internet access emerge as leverage points for transformation, aligning with literature on risk perception/social capital (tenure), vulnerability and outcomes (inequality), and communication capacity (internet). The approach also distinguishes positive transformation from degradation, highlighting that increases in normatively negative attributes (e.g., inequality) can tip communities toward worse peer profiles. For policy, integrating transformation metrics into disaster aid criteria (e.g., FEMA public assistance matching) could redirect investments toward communities capable of both recovering and transforming, mitigating maladaptation and promoting equitable resilience.

This study advances the operationalization of community resilience by jointly quantifying recovery and transformation. Using Hurricane Michael in Florida, it identifies a set of recovery-critical risk factors and, via Contrastive Community Networks, measures transformation thresholds and trajectories as communities’ similarity profiles shift. Key contributions include: (i) evidence that 55% of recovery predictors are potential resilience traps, (ii) identification of transformation catalysts (1-year housing tenure, income inequality, internet access) with relatively low thresholds, and (iii) a practical framework to assess tipping points for transformation versus degradation. The results caution against recovery-only policies that risk maladaptation and advocate incorporating transformation-oriented metrics into disaster funding and planning. Future research could extend CCN analyses to multiple hazards and regions, integrate additional structural and governance variables, and evaluate longitudinal outcomes of policy changes that target transformation catalysts.