2.1. Study Area and Governance Context

Surabaya is a coastal metropolis with a complex drainage hierarchy that blends legacy kampung layouts, newly paved developments, and dense road networks connected to secondary and primary channels. Low relief and episodic high-intensity rainfall create frequent conditions for rapid surface runoff and nuisance ponding, particularly where tertiary inlets are clogged or where microtopography funnels water toward bottlenecked segments. Municipal responsibilities are distributed across public work units for drainage and roads, environmental agencies for blue–green spaces, and subdistrict offices for mobilizing neighborhood preparedness and cleanup days. This distributed governance architecture means that actionable analysis must speak to different users: engineers who schedule inlet cleaning and gully-pot vacuuming, planners who site pocket green infrastructure, and community leaders who convene preparedness meetings. By working at the Kelurahan scale and reporting patterns as coherent clusters rather than as isolated cells, our framework aims to bridge these constituencies and encourage synchronized actions ¹³.

2.2. Data Sources and Preprocessing

We assembled two families of indicators—spatial exposure and social capacity—along with administrative boundaries and ancillary layers that helped interpret the patterns. The exposure family comprises observed or reported inundation points and extents compiled from municipal situation reports, field visits, and curated community submissions during the 2019–2025 seasons; proximity to the drainage network measured along the road graph to capture how rapidly water can access channels; land cover with a focus on imperviousness proxies, such as built-up fractions and road density; local slope and micro-relief derived from a hydrologically corrected digital elevation model to identify flow accumulation tendencies; and proximity to sensitive and critical facilities, including schools, clinics, and markets, where shallow ponding triggers outsized social costs. Each layer was projected onto a common coordinate system, clipped to the city boundary, and harmonized with neighborhood polygons using area-weighted aggregation or network-based distance summaries. We reduced duplication where two indicators encoded similar mechanisms, for example, using either the built-up ratio or road imperviousness, rather than both, when collinearity was evident ¹⁴.

The social capacity family was developed using a rapid assessment protocol designed for low respondent burden and high comparability across neighborhoods. Modules included household preparedness behaviors, such as the presence of basic protective equipment, awareness of early-warning cues, and knowledge of safe routes; adaptive capacity proxies, such as access to savings, micro-insurance, or mutual-aid mechanisms; and recovery markers, including the typical time to restore mobility or reopen small businesses after heavy rain. Enumerators worked with neighborhood leaders to ensure inclusive coverage and avoid repeatedly sampling the most vocal households. To protect privacy and reduce the risk of re-identification, all records were de-identified at the source, no exact addresses were retained, and only aggregated summaries at the Kelurahan level were used in the analysis. Qualitative notes captured issues such as vendor encroachment over inlets or recurring debris sources, and served as contextual annotations during interpretation ¹⁵.

Administrative boundaries for Kelurahan and subdistricts were validated against recent municipal shape files, and road graphs were simplified to remove dangling segments unrelated to runoff conveyance. All spatial joins and aggregations were scripted so that the pipeline could be rerun when new data arrived. Before indicator construction, we performed basic quality checks, including range screening, visual inspection of outliers, and cross-layer plausibility (for example, confirming that the reported ponding points do not systematically fall in parks or open water unless they correspond to known depressions or blocked culverts).

2.3. Indicator Definitions, Directionality, and Normalization

This study was designed to support municipal planning and neighborhood preparedness and employs a multi-stage analytical framework. The methodology integrates spatial exposure and community capacity indicators through composite index construction, followed by hotspot detection and prioritization logic to form a cohesive decision-making structure. For exposure layers representing adverse conditions, such as inundation frequency, flow accumulation propensity, or distance to functioning drains, the values were inverted or rescaled such that higher standardized values consistently represented better resilience conditions. Capacity indicators were coded in their natural directions, with larger values indicating stronger preparedness, greater adaptive resources, or shorter recovery. All indicators were normalized to the unit interval using min-max scaling within a citywide distribution ¹⁶. Although z-score normalization is widely used in resilience studies, bounded min-max normalization was selected to improve interpretability and communicate results to municipal stakeholders and decision-makers. To guard against the undue influence of extreme observations, we capped distributions at the 2nd and 98th percentiles prior to the min-max transformation, noting any neighborhoods with truncated values so that analysts could revisit data provenance rather than allowing tails to dominate ¹⁷.

Conceptually, we grouped capacity into three dimensions corresponding to concrete program levers. Preparedness captures anticipatory actions such as understanding risk maps, knowing whom to contact for assistance, or keeping gutters clear before storms. Adaptive capacity aggregates resources that allow households to cope with disruptions, including financial buffers and strong social ties that mobilize labor and information. Recovery reflects the speed and reliability with which normal activity resumes after ponding and is measured through self-reported times to reopen stalls or reestablish safe mobility for children and the elderly. By distinguishing these dimensions, we enabled departments to act on the specific levers they control while observing an integrated composite at the neighborhood scale.

2.4. Composite Resilience Index and Weighting

Neighborhood flood resilience was defined as a composite function of reduced spatial exposure and increased social and adaptive capacity. The composite index \(R\) was constructed as the average of the four dimension scores of one exposure component and three capacity components—after bringing each to a \([0,1]\) scale. The composite flood resilience index was calculated using the following equation:

2.5. Hotspot Detection and Spatial Coherence

Although composite indices are useful for summarization, decisions in cities are often triggered by clusters of concern rather than by isolated outliers. For coherent surface geographies where exposure remains high and capacity is weak, we computed local indicators of spatial association on both the inverted exposure dimension and the composite resilience index. Neighborhoods that appear low on RRR and are surrounded by similarly low neighbors form “cold spots” of resilience that merit concerted attention because interventions in one area may reinforce benefits nearby. In parallel, we examined the proximity of known bottleneck inlets, road geometry that channels sheet flow into depressions, and micro-relief features revealed by the digital elevation model (DEM) to understand why clusters form where they do. This interpretation step supports measures that address mechanisms rather than simply responding to symptoms.

2.6. Translating Maps to Measures Through Suitability and Feasibility

To move from diagnosis to action, we overlay two pragmatic lenses: suitability of blue–green interventions and feasibility under municipal routines. Suitability is derived from a land-surface screen that favors cells near bottlenecked inlets or along corridors with repeated reports of shallow ponding, on slopes sufficiently gentle for bioretention to function, and on or adjacent to public right-of-way, where implementation hurdles are lowest. We did not attempt a full hydraulic sizing exercise; instead, we identified candidate cells where small rain gardens, curb extensions, or porous patches could plausibly intercept runoff before it accumulates in depressions. Feasibility considers whether the proposed measures align with scheduled maintenance, whether traffic and vendor patterns allow temporary work, and whether Kelurahan offices have active neighborhood groups that can support risk communication. The overlay yields a short list of measures that can be initiated within a planning cycle, recognizing that larger capital projects may proceed over longer horizons ³.

2.7. Prioritization Logic at the Neighborhood Scale

As resources are limited, we allocated attention using a simple but communicable matrix that crosses exposure and capacity. Neighborhoods that sit in the high-exposure, low-capacity quadrant are placed in Priority A; those with high exposure but medium capacity are placed in Priority B; and those with medium exposure and low capacity are placed in Priority C. This scheme avoids perverse incentives, where areas with entrenched exposure but strong capacity monopolize investments, and it highlights locations where a small improvement in preparedness or a targeted micro-retrofit could unlock disproportionate benefits. This classification scheme helps prevent disproportionate allocation of resources to areas with high exposure but already strong capacity, thereby ensuring that limited interventions are directed toward locations with the greatest combined vulnerability. Moreover, the matrix provides community leaders with a language to advocate for specific support, for example, arguing for an early-warning refresh and gully-pot cleaning ahead of the monsoon, when preparedness is flagging and inlets are known to clog.

2.8. Sensitivity Analysis and Robustness Checks

We report three forms of robustness evidence. First, we examined how neighborhood ranks changed under the \(\pm 25\)% weight perturbations described earlier; high Spearman rank correlations with the baseline imply that prioritization is insensitive to reasonable weight choices. Second, we implemented leave-one-indicator-out tests for each dimension to identify where a particular proxy may drive the classification; neighborhoods that flip categories under such tests are flagged for closer scrutiny during stakeholder review. Third, we conducted a sanity check against independent observations by comparing low-resilience neighborhoods with independent notes on inundation persistence and service disruptions, not to claim causal validation but to ensure that the composite signal is not at odds with lived experience captured in field notes.

2.9. Ethics, Data Protection, and Reproducibility

All human-facing instruments and procedures followed institutional ethical guidelines. Respondents provided informed consent, their participation was voluntary, and no personal identifying information was retained in the analytical files. We stored de-identified data in a restricted workspace and released only neighborhood aggregates in maps and tables. Spatial scripts and indicator construction notebooks are organized such that the entire pipeline can be rerun when new seasons of data become available, which is central to the “city-ready” ethos of the framework. We included a reproducibility note with software versions and a brief changelog documenting updates to layers or indicator definitions so that municipal users could track differences across annual cycles.

2.10. GenAI Use Disclosure

An AI assistant was used to help structure the manuscript in English and ensure clarity and coherence across sections. The study design, data curation, spatial analysis, indicator construction, and interpretation of the findings remain the sole responsibility of the authors. All sources were cited and no proprietary or confidential information was provided to the assistant.

3.1. Respondent Profile and Spatial Coverage of Evidence

The survey achieved citywide coverage with responses drawn from 29 districts across Surabaya and a total sample of approximately 800 participants, which provided both breadth and sufficient density for neighborhood-scale interpretation. Three districts, Krembangan, Mulyorejo, and Rungkut, contributed a particularly large share of observations and thus featured prominently in the spatial signal reported below. The gender composition skews toward female respondents at around 62% of respondents, which is consistent with community participation patterns in neighborhood meetings and preparedness activities, while the age structure is anchored by the 17–25 age group at approximately 38%–39%, followed by a balanced representation of adults in their late twenties to forties and a smaller share of older residents. Education is tilted toward senior secondary and tertiary qualifications, which tends to raise baseline awareness of flood issues but does not guarantee consistent protective behaviors on its own, a point we revisit when linking attitudes to practices. The religious and occupational mix reflects the city’s diversity, with a notable number of small business owners and service workers whose livelihoods are sensitive to shallow, short-duration ponding. Together, these features suggest that the dataset is well placed to capture how everyday disruptions intersect with social capacity and that the results are not dominated by a single enclave or demographic.

3.2. Exposure Patterns and Neighborhood-Scale Clustering

The mapped exposure surface demonstrates a familiar geography in which recurrent shallow ponding aligns with dense built-up corridors, tertiary inlet bottlenecks, and micro-depressions visible in digital elevation data, particularly along sections of the Krembangan waterfront approaches and pockets of Mulyorejo and Rungkut, where rapid development has outpaced inlet maintenance. When we invert the exposure and overlay the composite resilience index at the Kelurahan level, statistically coherent clusters emerge rather than isolated outliers. Neighborhoods with repeated reports of water standing beyond one hour after peak rainfall form low-resilience cold spots when their capacity scores are also weak, whereas areas adjacent to well-maintained secondary drains can present medium exposure while maintaining higher composite scores because preparedness behaviors and local information flows allow more rapid recovery. Importantly, the cluster structure helps move the discussion away from single intersections or anecdotal hotspots and toward contiguous neighborhoods, where coordinated maintenance, micro-retrofits, and communication can reinforce one another ¹⁹.

The normalized exposure map (Fig. 1) visually quantifies the physical burden of flooding across the city’s 31 districts. Higher exposure values (indicated in red) are concentrated in the northern and eastern sectors, specifically within the industrial and dense residential corridors of Krembangan, Mulyorejo, and Rungkut. These hotspots correlate with areas of low elevation and high surface impermeability, where the drainage network frequently reaches its capacity during peak rainfall events.

Equation \(\eqref{eq:1}\) summarizes the flood resilience profile across the 29 districts of Surabaya, combining the exposure scores with three dimensions of community capacity: preparedness, adaptive capacity, and recovery into a composite index. This classification highlights distinct geographies of concern. Districts such as Krembangan, Mulyorejo, and Rungkut consistently emerged with high exposure values coupled with weak capacity indicators, yielding composite resilience scores below 0.50 and placing them in Priority Class A. These areas reported recurrent ponding, longer recovery times, and limited preparedness routines, underscoring the need for immediate low-regret interventions ²⁰.

Other districts, including Wonokromo and Tegalsari, exhibited high exposure but moderate preparedness and adaptive capacity, producing composite scores ranging from 0.53 to 0.55. These are categorized as Priority Class B, where modest investments in preparedness, communication, and targeted inlet maintenance can unlock faster recovery before capital-intensive work is scheduled.

Several peripheral districts, such as Sukolilo, Genteng, and Tandes, demonstrate medium-to-high exposure but stronger social and adaptive indicators, with composite scores around 0.57–0.59. These areas are therefore placed in Priority Class B, reflecting a relatively better balance between hazard and capacity, although preparedness must be continuously reinforced to sustain resilience.

Districts such as Kenjeran, Tambaksari, Asemrowo, and Bulak demonstrate combinations of medium exposure with weaker preparedness and adaptive resources, resulting in Priority Class C. Although not the most exposed, these areas remain vulnerable because limited community resources and slower recovery trajectories mean that even moderate events can lead to prolonged disruptions.

Therefore, the full citywide map does not present a uniform pattern but a patchwork of priorities. High-density waterfronts and industrial corridors in the north and east concentrate Priority A clusters, while more central mixed-use districts fall into Priority B, and a group of peripheral neighborhoods displays Priority C patterns. This distribution suggests that different policy levers must be sequenced geographically: intensive maintenance and micro-retrofits in Priority A, preparedness reinforcement in Priority B, and social capacity-building in Priority C. The classification offers municipal planners a transparent and reproducible basis for targeting low-regret measures where they are most needed and for aligning neighborhood-level preparedness with citywide drainage planning.

3.3. Capacity Dimensions, Knowledge, Attitudes, and Behaviors

Across the city, the knowledge base about flood causes and local risk is relatively strong, which is an unsurprising result considering the frequency of seasonal disruptions and the presence of tertiary education among the respondents. Attitudes were broadly pro-prevention, with most residents endorsing inlet cleanliness and routine pre-storm checks as community responsibilities; however, the translation of attitudes into routine behaviors was uneven. In several Kelurahans inside the Krembangan-Rungkut arc, households report understanding early-warning cues and knowing whom to contact; however, the self-reported frequency of gutter clearing, waste separation to reduce inlet debris, and pre-event checks remain inconsistent. This divergence between knowing and doing matters because neighborhoods with similar exposure but more regular maintenance report noticeably shorter recovery times. The pattern highlights the role of micro-practices that are easy to overlook in planning documents but decisive in the hours after intense rainfall ²¹.

The adaptive capacity map (Fig. 2) illustrates a community’s ability to adjust to and moderate the impact of flooding. Although central districts, such as Genteng and Tandes, exhibit higher adaptive scores (darker orange to red) owing to stronger social capital and resources, the map identifies a critical “capacity gap” in several northern districts. Despite high awareness, these areas demonstrate lower scores (blue to green), reflecting a lag in transforming knowledge into consistent protective physical adaptations at the household level.

3.4. Psychosocial Resilience and Recovery Experiences

Psychosocial indicators provide a nuanced picture. Social ties, mutual aid, and informal information channels are widely present and are often mobilized during disruptions, contributing positively to adaptive capacity and helping smooth access to temporary shelters or shared equipment ⁴. Simultaneously, respondents pointed to stressors that accumulate over repeated events, particularly among small traders and caregiving households who juggle income loss with additional safety duties for children and elders. Several neighborhoods would benefit from simple but targeted counseling and risk communication refreshers that normalize protective routines and reduce anxiety during heavy rain episodes. Measured through self-reported time to resume mobility and reopen small businesses, the recovery dimension varies across clusters in ways that align with preparedness practices and inlet maintenance rather than headline exposure alone, reinforcing the composite view of resilience rather than a hazard-only lens.

Spatial variations in the duration of post-event disruptions are captured in the recovery capacity map (Fig. 3). The map highlights that recovery is not solely a function of water depth. Districts with well-established mutual aid networks and informal information channels, such as those in the southern and western peripheries, demonstrate higher recovery scores. Conversely, lower scores in the northeast reflect areas where livelihoods are more sensitive to prolonged ponding, leading to slower returns to normalcy for small businesses and residents.

3.5. Infrastructure and Service Disruption Impacts

The infrastructure story is less about catastrophic failures and more about the cumulative burden of nuisance floods that interfere with transportation, schooling, and small-scale commerce. The reported impacts concentrate along transport routes where shallow but persistent ponding reduces throughput, causes detours, and spreads knock-on delays into adjacent neighborhoods. Sensitive facilities such as schools and clinics experience intermittent access issues that amplify social costs beyond what water depth alone would suggest. Where tertiary inlets are obstructed by debris or vendor encroachment, even well-sized secondary channels cannot drain surface water rapidly, leading to a perception that flooding is “everywhere” when the operative mechanism is a short chain of small bottlenecks. These observations support the intervention logic focused on inlet-level maintenance, micro-regrading at driveway mouths that block gutter flow, and pocket blue–green installations positioned to intercept runoff upstream of known depressions ⁹.

The preparedness capacity map (Fig. 4) benchmarks the proactive measures adopted by residents, including routine gutter cleaning and waste management. A significant cluster of low preparedness (blue shades) is visible in the eastern waterfront, where the survey data suggest that despite high flood frequency, the translation of risk awareness into routine “mundane maintenance” remains inconsistent, creating a bottleneck for overall resilience.

3.6. Composite Index Results and Priority Classes

When the capacity dimensions are combined with the inverted exposure score, the composite resilience index ranks neighborhoods in a way that is both intuitive and operational ²². Priority A neighborhoods occupy quadrants of high exposure and low capacity, often where preparedness behaviors are inconsistent and informal waste practices interfere with inlets. These neighborhoods are found in parts of Krembangan near tidal influences and in select Rungkut corridors where rapid paving has created hard-edged flow paths. Priority B covers high-exposure areas with moderate capacity, where a modest investment in preparedness and communication can unlock faster recovery, even before physical work is deployed. Priority C identifies places with medium exposure but low capacity, often dormitory neighborhoods with limited volunteer bandwidth, where risk communication, community drills, and simple household-level measures can yield immediate dividends while the city schedules inlet work. The citywide map does not present a single sweeping front but a patchwork of clusters where targeted and synchronized measures are likely to compound ²³.

The composite flood resilience index map (Fig. 5) integrates the previous four dimensions into a single metric using an equal-weighting formula. This map reveals the “true” resilience landscape of Surabaya, where low-resilience zones (red) emerge not only where the water is deepest but also where high exposure meets low social capacity. This integrated view allows planners to see beyond the hazard and identify neighborhoods where technical drainage solutions must be paired with social intervention.

3.7. Sensitivity and Robustness of Rankings

The equal-weighted baseline is intentionally transparent; however, the rankings prove resilient to plausible changes. When we perturb the weights by plus or minus a quarter for each dimension while preserving a unit sum, the Spearman correlation between the perturbed and baseline ranks remains high, indicating that the ordering of neighborhoods is not an artifact of a particular weighting choice. A leave-one-indicator-out analysis reveals a handful of Kelurahans whose classification is sensitive to the presence or absence of a single proxy, typically in cases in which two dimensions sit near a class boundary. These neighborhoods are flagged for discussion during stakeholder sessions so that local knowledge can arbitrate whether classification should be nudged by observed maintenance patterns or recent works not yet reflected in the data. Finally, a sanity check against independent field notes on standing water duration and access disruptions reveals broad agreement with the low-resilience clusters, lending further confidence that the composite signal captures lived experiences rather than merely reproducing statistical artifacts ⁶.

3.8. From Maps to Measures Within Municipal Routines

Translating a spatial signal into action leads to a short list of feasible measures within routine budgets and operating practices. In Priority A clusters, the first order of business is to restore conveyance through coordinated inlet cleaning, debris management near vendor stands, and rapid micro-drain maintenance timed ahead of the monsoon. In parallel, pocket blue–green installations, such as curb-extension bioretention and small rain gardens, are located along corridors that feed the depressed cells identified on the exposure surface so that runoff is intercepted before it accumulates ²⁴. In Priority B areas, a paired strategy of risk-communication refresh and targeted porous patches near schools and clinics helps secure near-term benefits, while the city sequences larger works. Priority C focuses on rebuilding preparedness muscle memory through neighborhood drills, early-warning refreshers, and simple household measures, such as portable barriers and gutter hygiene, ensuring that medium exposure does not cascade into prolonged recovery times. This coordinated approach is not novel. Instead, it reflects a deliberate focus on reliable actions that city crews and neighborhood groups can repeat and scale without specialized equipment.

Finally, the priority class map (Fig. 6) translates the composite index into an actionable decision-support tool. By classifying districts into Priority A (red), Priority B (yellow), and Priority C (green), the map provides a clear hierarchy for municipal resource allocation. Priority A districts, concentrated in the north and east, are flagged for immediate structural and non-structural interventions, whereas Priority B and C areas are categorized for preparedness reinforcement and long-term capacity building, respectively.