Beyond warnings and shelters: local institutions and trust build cyclone resilience in Bangladesh

Abstract

Tropical cyclones (TCs) severely threaten coastal Bangladesh, making it crucial to understand how communities respond to disaster preparedness and warning systems, as well as the effectiveness of local institutions in mitigating risks and aiding recovery. This study assessed community response and institutional effectiveness through 279 household surveys, 28 focus group discussions, and 30 key informant interviews across seven TC-affected districts in Bangladesh. Results show traditional warning systems remain most effective, but digital and mass media channels show significant spatial gaps. Equity analysis revealed substantial disparities, with male-headed and farming households having 11–13 percentage points higher warning reach than female-headed and fishing households, respectively. Evacuation rates correlated with shelter quality, while non-evacuation was driven by safety concerns, economic constraints, and false alarm fatigue. Multivariable regression confirmed shelter quality as a strong predictor of evacuation. Spatial analysis indicated significant clustering of exposure and sectoral damages. Damage was severe and uneven: crop losses reached 91% in Barguna, fishing asset losses 94% in Patuakhali, and housing damage 80% in Bhola. Unmet needs for water and livelihoods were widespread, and institutional trust varied. The findings underscore the need for integrated, context-sensitive strategies combining better infrastructure, inclusive communication, and accountable governance to strengthen resilience.

Introduction

Tropical cyclones (TCs) are among the deadliest and costliest natural hazards in the world, causing destructive impacts on ecosystems¹, lives², and livelihoods³. Each year, TCs expose 150 million people to their destructive forces⁴, and the increasing intensity of TCs⁵ is expected to further exacerbate global destruction⁶. Under global warming, annual global TC exposure is projected to increase by 33 million people for every 1 °C rise in global mean surface temperature⁷. Given the substantial impacts of TCs, including their resultant storm surges and extreme rainfall on society^2,8,9, identifying strategies for risk reduction, such as preparedness, emergency response mechanisms and post-TC recovery, is of crucial importance for minimizing loss of life, reducing economic damage, and enhancing the resilience of vulnerable communities.

Raising awareness and educating communities about natural disasters are crucial for plummeting vulnerability and strengthening preparedness¹⁰. While the use of technology helps lessen disaster impacts, it is inadequate without active community participation. To discourse this, national and international disaster management agencies are distributing varying levels of resources to guide communities in preparing for different hazards and are increasingly emphasizing community engagement as a key strategy. By strengthening disaster resilience through widespread education, skill-building, and effective risk communication, these efforts empower communities and encourage their active participation in disaster management processes¹¹. The Sendai Framework for Disaster Risk Reduction 2015–2030¹², building on the foundation of the Hyogo Framework for Action 2005–2015¹³, identified five priority areas (e.g., preparedness) and highlighted the importance of taking coordinated actions across all levels and sectors to strengthen preparedness and ensure effective response efforts.

Bangladesh, highly prone to TCs, storm surges, and floods, has developed a multi-tiered disaster management framework involving Union, Upazila, District, and National Disaster Management Committees (DMCs)¹⁴. Union DMCs (UDMCs) focus on community-level awareness, disaster education, and resource mobilization, while Upazila and District DMCs coordinate regional strategies and support. At the national level, the Ministry of Disaster Management and Relief oversees policy implementation and fosters cross-sector collaboration, aligning efforts with global frameworks like the Sendai Framework. To institutionalize disaster risk reduction, the government enacted the Disaster Management Act (2012) and developed the Standing Orders on Disasters (SOD) to guide coordination and response¹⁵. A critical component of this system is the Cyclone Preparedness Program (CPP), jointly managed by the government and the Bangladesh Red Crescent Society, which ensures early warning dissemination and timely evacuation during emergencies¹⁶. Complementing these efforts, NGOs and international agencies have played an instrumental role in community-based disaster preparedness, constructing shelters, and supporting livelihood recovery programs¹⁷.

The Great Bhola Cyclone in 1970, and Cyclone Gorky in 1991 are among the top three deadliest TCs in earth’s history¹⁸, resulting in approximately 0.30–0.50 million and 0.147 million deaths, respectively¹⁶. In recent decades, despite a decreasing trend in death tolls¹⁶, cyclone SIDR in 2007, cyclone Aila in 2009 and cyclone Amphan in 2020 caused approximately $10 billion in combined losses, equivalent to 3.5% of the national GDP. The profound historical toll of TCs in Bangladesh-from immense fatalities of Bhola Cyclone (1970) to the massive economic devastation caused by more recent TCs (e.g., Amphan 2020)- underscores the critical importance of understanding the efficacy of DMCs. As the country faces projections of increased cyclone frequency and intensity, with 22% of its coastal area deemed high-risk¹⁹, the role of DMCs becomes ever more vital. These local institutions are the crucial link between early warning systems and community action, translating alerts into life-saving evacuations and ensuring that preparedness, response, and recovery efforts are both efficient and inclusive. By examining their effectiveness, we can identify gaps in communication, resource allocation, and support for vulnerable populations, thereby strengthening the entire disaster management framework. Ultimately, the ability of DMCs to perform effectively not only determines the short-term reduction in casualties but also underpins long-term resilience, safeguarding both lives and livelihoods against future climate-induced disasters.

While previous studies have extensively examined disaster risks²⁰, vulnerability²¹, impacts²², preparedness²³, and adaptation strategies¹⁷ in Bangladesh, a significant knowledge gap remains regarding the operational efficiency and decision-making processes of DMCs in disaster risk reduction. Empirical evidence is notably scarce regarding how DMCs coordinate early warnings, mobilize at-risk communities, manage shelters, allocate resources, and navigate socio-political dynamics during crises. Closing this knowledge gap is essential to strengthening resilience, as the effectiveness of DMCs directly influences both immediate disaster outcomes and long-term adaptive capacity—especially in a rapidly changing climate where efficient, equitable, and locally grounded risk governance is increasingly vital.

In this study, we conducted 279 household surveys, 28 focus group discussions (FGDs) and 30 key informant interviews (KIIs) across 14 unions from the 7 southernmost sub-districts of coastal Bangladesh. The study area has endured numerous catastrophic TCs and storm surges in recent decades, each leaving a trail of destruction and compounding vulnerabilities²⁴. Located in the lower estuary of the world’s largest delta, the Ganges-Brahmaputra-Meghna delta, this study region was severely impacted by Bhola Cyclone in 1970 and Cyclone Gorky in 1991, the two top-deadliest TCs in the world. In recent two decades, several frequent TCs, including Cyclone Sidr in 2007, Aila in 2009, Mahasen in 2013, Komen in 2015, Mora in 2017, Bulbul in 2019, Amphan in 2020, Yaas in 2021, and Remal in 2024 brought widespread flooding, damaged infrastructure, and exacerbated salinity intrusion, while triggered destructive tidal surges that submerged villages and agricultural lands^{16,21,22,25,26}. These recurrent TCs have not only caused immediate loss of life and property but have also eroded long-term resilience, pushing already vulnerable communities deeper into poverty. Despite the presence of nearly 6000 cyclone shelters in the country, strong early warning systems, and local to national level DMCs, significant gaps persist in reaching early warning to communities, ensuring timely evacuations, effective post-disaster recovery, and sustainable resilience-building measures. The following two objectives have been established to improve our understanding on the local dynamics of disaster governance.

(i) to assess the spatial patterns and socio-ecological determinants of TC impact, community vulnerability, and recovery needs across the coastal region of Bangladesh.

(ii) to evaluate the effectiveness and equity of the current early warning and institutional response systems in mitigating TC risks and supporting post-TC recovery.

Results

The results section first details TC exposure levels across coastal unions, showing high perceived impact with spatial variability. It then evaluates warning dissemination efficacy, identifying the CPP as most effective while revealing significant gaps in digital and mass media channels. Subsequent analyses examine sector-specific damage patterns, evacuation-shelter quality relationships, reasons for non-evacuation, post-TC unmet needs, and institutional performance disparities across districts—providing a comprehensive assessment of disaster impacts and response system effectiveness throughout coastal Bangladesh.

TC exposure levels

Based on the household survey data, we observed notable variation in self-reported TC exposure levels across the study area, with union-level averages ranging from 4.33 to 4.79 on a 5-point scale (Fig. 1). Unions including Sonakata (4.79), Nishanbaria (4.75), and Char Kukri Mukri (4.75) reported the highest perceived exposure, indicating these communities reported the most intense TC impacts. In contrast, Gulishakhali (4.33) and Khontakata (4.35) reported the lowest ratings, though still reflecting high overall exposure. The mean exposure rating across all unions was 4.61, demonstrating that households consistently perceived the TC event as severe, while still revealing meaningful spatial gradients in perceived intensity (Fig. 1). Spatial clustering of exposure was statistically significant (Global Moran’s I = 0.38, p < 0.01), indicating that unions with similar exposure levels tend to be geographically clustered, particularly in offshore islands.

Fig. 1: Spatial distribution of self-reported cyclone exposure levels.

Map shows the average exposure rating (on a scale of 1-5) reported by households in each union, following the recent TCs. The low exposure to high exposure is shown by orange (low exposure) and red (high exposure).

Efficacy of TC warning dissemination channels

Analysis of warning channel efficacy across seven coastal districts reveals the CPP as the singularly most effective dissemination channel, maintaining >72% reach universally and peaking at 90% in Bhola (Fig. 2a, b). Mass media (TV/radio) demonstrates significant geographic disparity, ranging from high effectiveness in Satkhira (67%) and Bagerhat (65%) to critically low penetration in Patuakhali (30%) and Bhola (35%). Digital channels (SMS/social media) show consistently poor performance overall (17–40%), indicating lower reach as primary warning systems (Fig. 2a, b). Religious institutions emerge as unexpectedly important communication hubs in southern coastal regions (Patuakhali: 40%, Bhola: 45%), while Union Parishad systems show moderate complementary reach (27.5–55%). Informal word-of-mouth channels remain consistently low (10–25%), confirming their inadequacy as systematic warning solutions (Fig. 2a, b). These findings identify concerning coverage gaps in mass media-reliant districts and affirm that human-centric, ground-level dissemination is strongly associated with inclusive early warning systems despite technological advancements.

Fig. 2: District-wise comparison of warning reach by source and channel.

Comparison of average reach (%) by warning sources (a, b) and channel types (c) across districts. Bars are colored by warning sources (a) and channel types (c), with values rounded to the nearest whole number displayed on top of each bar. Heatmap (b) showing the distribution of average reach (%) across districts and warning sources. The intensity of the color represents the percentage reach, with darker shades indicating higher values. The legend at the side provides a color gradient scale for reference.

The results indicate that traditional channels, including CPP, Union Parishad, Religious Institutions, and Informal sources, consistently achieve higher average reach percentages compared to modern channels, such as Mass media and SMS/social media, across all districts (Fig. 2c). Bhola exhibits the highest reach for traditional channels at 52%, followed by Patuakhali at 47% and Barguna at 39%, while Bagerhat has the highest reach for modern channels at 52%. In contrast, districts like Patuakhali and Bhola show significantly lower reach for modern channels, both at 26% (Fig. 2c). These findings highlight the continued reliance on traditional communication methods in disseminating warnings.

Damage of TCs

The post-cyclone damage revealed substantial losses across multiple sectors, with varying impacts across the seven coastal districts (Fig. 3). Crop losses were most severe in Barguna (91%) and Satkhira (88%) (Fig. 3a), while fishing asset losses peaked in Patuakhali (94%) and Bhola (91%) (Fig. 3b). Livestock losses showed a different pattern, being highest in Bhola (46%) and Barguna (40%) but considerably lower in other districts, particularly Pirojpur (13%) (Fig. 3c). Structural damage to houses was most extensive in Bhola (79.5%) and Barguna (72.5%), with relatively lower impacts in Bagerhat and Pirojpur (both 47.5%) (Fig. 3d). Overall, Barguna and Bhola consistently emerged as the most severely affected districts across multiple damage categories, while Pirojpur generally experienced the lowest impact levels, particularly for livestock and housing damage.

Fig. 3: Multi-sector post-cyclone damage assessment.

Spatial distribution of post-cyclone damage across multiple livelihood sectors in seven coastal districts, showing percentage losses in a crops, b fishing assets, c livestock, and d housing infrastructure. The intensity of the color represents the percentage losses, with darker shades indicating higher values. The legend at the right side provides a color gradient scale for reference.

The distribution of TC-induced damage across sectors exhibited significant but varying degrees of spatial autocorrelation, as measured by Global Moran’s I (Fig. 3). Fishing asset loss showed the strongest spatial clustering (I = 0.45, p < 0.01), with severe losses concentrated in the central coastal districts of Patuakhali and Bhola, reflecting the geographical concentration of marine fishing livelihoods (Fig. 3b). Crop loss (I = 0.42, p < 0.01) and house damage (I = 0.39, p < 0.05) also demonstrated significant positive clustering, suggesting that unions with high agricultural or structural impacts were often geographically adjacent, which may relate to shared exposure to saline intrusion and high wind zones (Fig. 3a, c). In contrast, livestock loss showed a weaker and non-significant spatial pattern (I = 0.18, p = 0.08), suggesting its distribution was more dispersed and influenced by household-specific protection strategies rather than broad geographic factors (Fig. 3d). These results confirm that the spatial heterogeneity of TC impacts is not random but follows distinct sector-specific geographic patterns.

Evacuation and shelter quality

Post-cyclone evacuation rates exhibited considerable variation across the surveyed unions, ranging from 33.0% to 68.0% (Fig. 4a). Spatial clustering was moderate but significant (Global Moran’s I = 0.25, p < 0.05). The highest evacuation rates were observed in the unions of Gabura and Sonakata (both 68.0%), followed by Nishanbaria (62.0%) and Char Kukri Mukri (61.0%). Conversely, the lowest propensity to evacuate was recorded in Lata Chapli (33.0%). The mean evacuation rate across all unions was 52.1%, indicating that just over half of all households complied with evacuation advisories.

Fig. 4: Spatial patterns of evacuation and shelter ratings.

Spatial distribution of household evacuation rates (a), shelter ratings (b) and their association (c) in seven districts. Evacuation rate in respective unions is shown as percentage and shelter rating is shown on a scale of 1-5.

Average shelter ratings, as reported by evacuees, varied significantly across unions (Fig. S1a) and districts (Fig. S1b), from a low of 1.8 to a high of 3.4 on a 5-point scale (Fig. 4b). The shelters in Char Kukri Mukri received the highest average rating (3.4), followed by Sonakata (3.3) and Munshiganj (3.2). In contrast, shelters in Char Manika (1.8) and Lata Chapli (1.9) were perceived as inadequate, scoring well below the midpoint of the scale. The overall mean shelter rating was 2.7, indicating a generally moderate perception of shelter quality across the region.

The scatter plot analysis reveals a clear positive correlation between shelter quality and evacuation rates across the seven coastal districts (Fig. 4c). Districts with higher shelter ratings were associated with higher evacuation compliance among residents. Barguna exhibited the highest evacuation rate (65%), and Satkhira showed second highest evacuation rate (58%), corresponding with their top-tier shelter rating. Conversely, Patuakhali showed the lowest evacuation rate (40%) alongside the poorest shelter rating (2.25/5). Notably, Pirojpur and Bhola showed moderate to good performance in both dimensions, while Khulna and Bagerhat clustered in the lower-left quadrant.

Reasons for non-evacuation

Based on the analysis of evacuation behavior across coastal districts, notable spatial variations were observed in the reasons for not evacuating during disasters (Fig. 5a). For example, safety and security along with false alarm fatigue emerged as key barriers in Patuakhali, while economic concerns and safety and security were the primary challenges in Khulna. In Bhola, underestimation of risk and shelter distance were the most common reasons for non-evacuation. Similarly, economic concerns and distrust in aid dominated in Bagerhat, while livestock protection and privacy concerns for women were the prominent obstacles in Barguna. In Pirojpur, political biasness and distrust in aid were significant factors, whereas false alarm fatigue and shelter distance stood out in Satkhira. In total, 13 reasons for non-evacuation of households during disasters were recorded from the household surveys (Fig. 5b).

Fig. 5: Mapping household-reported barriers to evacuation.

Spatial distribution of barriers to evacuation across the unions (a), the heatmap (b) showing the percentage of households reporting reasons for non-evacuation across district level.

Post-TC unmet needs

The distribution of the most frequently reported unmet need varied widely across unions, revealing distinct post-disaster recovery challenges (Fig. 6a). Drinking water emerged as the most critical unmet need, identified as the primary issue in five unions (Munshiganj, Gabura, Dakshin Bedkashi, Koyra, and Sapleza). This highlights a widespread failure in the restoration of essential water infrastructure following the cyclone. The need for housing repair was the most common in four unions (Royenda, Gulishakhali, Sonakata, and Char Kukri Mukri), indicating severe damage to the housing stock in these areas. Agricultural inputs were the primary concern in Nishanbaria and Khontakata, underscoring the cyclone’s devastating impact on livelihoods dependent on farming. Similarly, the demand for fishing nets in Char Manika points to the specific livelihood needs of fishing communities. Other critical needs included food aid in Lata Chapli, reflecting issues with short-term sustenance, and veterinary support in Dhulasar, indicating a focus on livestock-based recovery.

Fig. 6: Mapping post-cyclone recovery gaps.

Spatial distribution (a) of primary unmet needs reported by households across unions, and heatmap (b) showing the percentage of households reporting each unmet need, by district. The distribution map illustrates the most frequently cited post-cyclone recovery need in each union, revealing distinct geographical patterns.

The heatmap visualizes the distribution of unmet needs across districts (Fig. 6b), highlighting areas where disaster response might have been insufficient or specific needs were not adequately addressed. Global Moran’s I = 0.32, p < 0.05, indicating significant clustering of similar unmet needs in neighboring unions. Districts with darker red tiles, such as Satkhira, Barguna, and Khulna, show a higher concentration of unmet needs across multiple categories. For instance, in Khulna, Barguna, and Satkhira, “drinking water” was an unmet need for over 40%, 35%, and 28% households, respectively. “Agricultural input” was the prominent unmet need in Bagerhat (30%) and Pirojpur (33%), and “fishing net” was the dominant unment need in Bhola (30%) districts. These findings indicate that “drinking water”, and economic needs (agricultural inputs, fishing nets, food aid) are widespread issues, affecting multiple districts, while other categories, such as “housing repair”, “health care”, and “livestock,” are more localized. Districts with lighter red tiles, such as Patuakhali and Barguna, report fewer unmet needs overall, suggesting a relatively more effective disaster response in these regions (Fig. 6b).

Institutional performance

The performance of DMC reveals notable district-specific disparities across three key metrics (warning, aid and trust) (Fig. 7). Bhola emerges as the highest performer, demonstrating notably effective early warning dissemination reaching 90% of households and efficient aid distribution at 64% (Fig. 7). Barguna and Satkhira report the highest levels of institutional trust at 3.5 and 3.2 respectively on a 5-point scale, indicating stronger community confidence in local authorities. However, concerning gaps remain particularly evident in Pirojpur, which shows a critical trust deficit at 2.0, and Patuakhali, where aid distribution fails to reach most households with only 37% coverage. Khulna shows a concerning weakness in its early warning system, with only 72% of households receiving alerts—one of the lowest rates among the districts—which suggests a critical gap in this fundamental layer of protection. Conversely, Bagerhat demonstrates a notable strength in aid distribution, reaching 66% of households. However, this operational effectiveness in aid delivery does not fully translate into public confidence, as both Khulna and Bagerhat report moderate-to-low institutional trust levels (2.8 for Khulna and 2.5 for Bagerhat) (Fig. 7). The analysis reveals that effectiveness in one domain does not necessarily translate to others, highlighting the need for targeted interventions that address specific institutional weaknesses in each district rather than applying uniform solutions across the coastal region.

Fig. 7: Distinct vulnerability patterns across coastal Bangladesh.

Radar chart illustrating the comparative vulnerability profiles across seven coastal districts of Bangladesh. Each axis represents a key vulnerability indicator, with values normalized for comparison. The plot reveals distinct patterns of relative strengths and weaknesses in disaster preparedness, impact, and recovery capacity among the districts.

Determinants of evacuation, damage, and aid receipt

Multivariable regression models identified key factors associated with disaster outcomes (Table 1). Each one-point increase in shelter quality rating was associated with 41% higher odds of evacuating (Odds Ratio [OR] = 1.41, p = 0.005). A sensitivity analysis excluding the highest-exposure union (Sonakata) yielded substantively similar results, with the shelter rating coefficient for evacuation changing by only -4%, confirming the robustness of these associations. Island location was associated with 6.6% higher crop losses compared to mainland areas (β = 6.64, p = 0.056). Male-headed households had 61% higher odds of receiving aid compared to female-headed households (OR = 1.61, p = 0.069), while each percentage point increase in crop loss was associated with 1% lower odds of aid receipt (OR = 0.99, p = 0.088), suggesting potential targeting inefficiencies (Table 1).

Table 1 Determinants of key disaster outcomes from multivariable regression models

Equity in warning and aid distribution

Significant disparities were found in early warning reach and post-disaster aid receipt across population subgroups (Table 2). Male-headed households received warnings at rates 11.8 percentage points higher than female-headed households (76.6% vs. 64.8%), representing a substantial gender gap in information access. Similarly, farming communities received warnings at rates 12.5 percentage points higher than fishing communities (82.5% vs. 70.0%). Despite their higher vulnerability, island communities showed 3.3 percentage points lower warning reach compared to mainland areas, though they received 7.1 percentage points more aid (Table 2).

Table 2 Equity gaps in warning reach and aid receipt across population subgroups

Discussion

The study findings illustrate a comprehensive profile of TC impacts across coastal Bangladesh, capturing perceived exposure and warning dissemination efficacy through triangulated quantitative (household surveys) and qualitative (FGDs and KIIs) data sources.

The reported high mean exposure rating (4.61/5) is indicative of the widespread severity of the previous TC events, yet the spatial gradient—from Sonakata (4.79) to Gulishakhali (4.33)—suggests localized variability in TC intensity or community preparedness and their responsiveness. These spatial patterns align with historical storm trajectories and physical vulnerability (e.g., proximity to the coast, elevation)¹⁶ but also underscore the role of subjective risk perception in shaping exposure assessments²⁷. Communities in low-lying offshore islands in our study (e.g., Char Kukri Mukri) consistently reported higher exposure (Fig. 1), reflecting their heightened geophysical susceptibility²⁸. FGDs strongly corroborated the quantitative exposure findings. In the Sonakata union of Barguna district, participants vividly described the severity of TCs: “The wind was so strong that our tin roofs were flying like papers. We had never seen such destruction before” (FGD-17). Similarly, residents of Char Kukri Mukri island in Bhola district emphasized their heightened vulnerability: “Being on this island, we always face the worst. The water comes from all sides, and we have nowhere to run” (FGD-25). These narratives align with the higher exposure ratings recorded in offshore and coastal communities in our study (Fig. 1). Our study region lies in the Ganges-Brahmaputra-Meghna delta, the world’s largest delta and a global hotspot for TC activity²⁴. Bangladesh’s coastline is struck by at least one TC annually and an extreme TC every three years²⁹. Due to this high exposure, the country accounts for over half of all global TC-induced deaths³⁰.

Our results demonstrate notable success in early warning system foundations, with high warning reception rates (Fig. 2) aligning with Bangladesh’s integrated dissemination network combining formal mechanisms with community volunteers³¹. This consistency with global literature supports the view that robust early warning systems are cornerstone elements of successful disaster risk reduction^32,33. The traditional warning source ‘CPP’ emerged as the most reliable channel, exceeding 72% reach across all districts and peaking at 90% in Bhola (Fig. 2). Key informants strongly supported these findings, with a CPP volunteer coordinator explaining: “Our network reaches every household through trained volunteers who know each family personally. This human connection makes the difference” (KII-17). A local official corroborated the exceptional Bhola performance: “CPP volunteers here are highly trained and trusted. They use megaphones, door-to-door visits, and even mosque announcements during prayer times” (KII-21). FGDs revealed the underlying trust mechanism driving CPP effectiveness. Participants in Satkhira noted: “When the CPP volunteer comes to our house and tells us about the danger, we believe immediately. They have saved us many times before” (FGD-03). This personal trust factor explains the consistently high CPP performance across all study areas, reaffirming the well-documented effectiveness of community-based, human-centric warning systems in Bangladesh^30,31.

Significant geographic disparities in mass media effectiveness—from 67% in Satkhira to 30% in Patuakhali—highlight infrastructural inequalities. Contrasting FGD responses illuminated these variations: Satkhira participants reported: “We always listen to radio/television for weather updates. Most tea stall has a radio or television, and people gather to listen or watch the news” (FGD-01), while Patuakhali FGD revealed constraints: “TV doesn’t work properly here. Power goes out frequently, and the signals are weak during storms” (FGD-21). This variability aligns with studies emphasizing digital divides and electricity dependency in coastal regions^23,34, contrasting with other research emphasizing radio and television importance in recent TC (Amphan in 2020) warning dissemination³⁰. The persistently low performance of digital channels (SMS/social media; 17–40%) was contextualized by a DMC member: “Mobile networks become unreliable during severe weather. Even when towers function, many elderly residents don’t trust SMS warnings or lack smartphones for social media” (KII-03). Community voices corroborated these technical limitations: “My son reads the SMS for me, but during the storm, the phone had no network” (FGD-26). These findings corroborate concerns about technological barriers in rural and resource-limited settings³⁴ and may stem from limited internet access, digital literacy, and trust in online sources among vulnerable populations^23,35.

Religious institutions unexpectedly emerged as key communication nodes in Patuakhali and Bhola, inconsistent with earlier studies^23,34 that overlooked their disaster communication role. A religious leader (mosque Imam) explained: “During disasters, people come to the mosque not just for shelter but for information. We coordinate with local authorities (union chairman, CPP, UDMC) to announce warnings after prayers” (KII-18). This constitutes a novel and highly replicable pathway for last-mile warning dissemination, particularly in contexts with strong community trust in religious leadership. The use of mosque loudspeakers (miking) during prayers or community gatherings presents a low-cost, high-reach, and culturally resonant strategy that could be formally integrated into early warning systems. FGD participants supported this finding: “The mosque microphone reaches everyone in the village. When the Imam announces danger, we know it’s serious” (FGD-28). This reflects the cultural and social centrality of religious spaces in coastal communities during crises, highlighting the importance of leveraging trusted community networks for risk communication³⁰. For programmatic implementation, disaster management authorities could provide basic training and equipment (e.g., megaphones, standardized message templates) to religious leaders. However, leveraging these institutions requires careful consideration of inclusion risks. Reliance on a single religious institution could unintentionally exclude religious minorities or non-practicing community members. Therefore, such an approach must be part of a multi-channel strategy and include specific safeguards to ensure warnings reach all demographic groups through complementary channels.

The moderate reach of Union Parishad systems (27.5–55%) revealed institutional-community disconnects. While officials emphasized efforts—“We have established DMC and conduct regular awareness meetings” (KII-01)—FGD participants expressed skepticism: “The Union Parishad office is far from our village. They make announcements, but we don’t always hear or trust them” (FGD-22). This disconnection highlights last-mile communication challenges despite established local governance roles³¹. Informal word-of-mouth channels remained consistently low (10–25%), confirming their inadequacy as systematic warning solutions.

These findings underscore the continued reliance on traditional, human-centric dissemination methods despite technological advancements^30,31. The triangulation of quantitative and qualitative data reveals that warning system effectiveness is associated not only with technological reach but also with trust, accessibility, and cultural appropriateness. The success of CPP volunteers and emergence of religious institutions as communication nodes emphasize the critical role of locally embedded, trusted intermediaries in disaster risk communication, while infrastructure-dependent channels remain constrained by systemic inequalities in coastal Bangladesh.

Our findings reveal critical insights into evacuation behavior and the association between shelter quality and disaster response actions across coastal Bangladesh. The substantial variation in evacuation rates (33% to 68%) and shelter quality ratings (1.8–3.4/5) (Fig. 4) underscores the spatial heterogeneity in both protective infrastructure and community preparedness (Fig. 5).

The positive correlation between shelter quality and evacuation rates (Fig. 4c) demonstrates that infrastructure quality is strongly associated with evacuation compliance. Districts with higher-rated shelters (e.g., Barguna, Satkhira) achieved significantly higher evacuation rates (65% and 58%, respectively), while those with poor shelter conditions (e.g., Patuakhali: 2.25/5 rating) experienced markedly lower evacuation participation (40%). This relationship was strongly validated through qualitative evidence. In Barguna, where shelter quality ratings were highest, FGD participants expressed confidence in evacuation decisions: “The cyclone shelter here has proper toilets, clean water, and enough space. When the warning comes, we don’t hesitate to go there with our families” (FGD-18). Similarly, participants in Satkhira noted: “Our shelter has been renovated recently. It has separate rooms for women, good ventilation, and the roof doesn’t leak anymore. That’s why people trust it” (FGD-02). Household surveys and qualitative findings are consistent with previous research emphasizing that shelter accessibility, safety, and basic amenities are key factors in evacuation decision-making^16,23. The consistently low ratings of shelters in unions like Char Manika (1.8) and Lata Chapli (1.9) reflect inadequate WASH facilities, overcrowding, and poor structural conditions—factors previously identified as barriers to shelter utilization during TCs^34,35,36. Conversely, in Patuakhali, where shelter ratings were lowest, FGD participants explained their reluctance: “The shelter building is very old and overcrowded. The toilets don’t work properly, and there is no separate space for elderly and women” (FGD-23).

The spatial heterogeneity in reasons for non-evacuation (Fig. 5) further illustrates how localized socio-economic and cultural factors shape risk response. The prominence of ‘safety and security concerns’ in Patuakhali and Khulna resonates with studies documenting fears of theft, property damage, and gender-based violence in evacuation scenarios^16,34. FGDs revealed the multifaceted nature of security fears that drive evacuation avoidance. In Patuakhali, participants articulated specific concerns: “When we leave our house empty and go to shelter, thieves know no one is home. Last time, my neighbor lost his fishing nets and tools” (FGD-24). Women’s groups expressed additional vulnerabilities: “In the crowded shelter, we worry about our daughters’ safety. There are many unknown men, and at night it’s difficult to keep watch” (FGD-05). These narratives demonstrate how security concerns encompass both property protection and personal safety dimensions. KIIs corroborated these community concerns. A local police officer acknowledged: “During evacuation periods, property crimes do increase in empty villages. We try to patrol, but our resources are limited during storms” (KII-19). A shelter management committee member noted: “Families, especially those with teenage girls, prefer to stay home rather than risk problems in overcrowded shelters where we can’t guarantee everyone’s safety” (KII-04).

Economic concerns, particularly in Khulna and Bagerhat, reflect households’ reluctance to abandon livelihoods and assets—a well-documented barrier in agrarian and fishing-dependent communities^31,37. FGDs revealed the economic calculations underlying evacuation decisions. In Khulna, fishermen explained: “Our boat is tied at the ghat. If we go to shelter and the storm damages it, we have no way to earn money for months” (FGD-06). FGD (farming community) in Bagerhat expressed similar concerns: “During last two cyclones, I did not go to shelters. Because I had 40 ducks, 3 cows and 5 goats. The shelter is far from my house. Taking all my livestock to shelter would not be feasible. If I leave them to go to shelter, they might die or be stolen. Then how will I feed my family after the cyclone?” (FGD-09). Local leaders contextualized these livelihood protection strategies. A cooperative society leader explained: “For poor families, losing their few assets means long-term disaster worse than the cyclone itself. They calculate that staying and protecting their property is worth the risk” (KII-08). An agricultural extension officer added: “Farmers know that if they lose their livestock or stored grain, no relief program will fully compensate them. So, they choose to stay and protect what they have” (KII-05).

The persistence of ‘false alarm fatigue’ in Patuakhali and Satkhira underscores the diminishing credibility of repeated warnings without subsequent severe impacts—a phenomenon observed in other cyclone-prone regions³⁸. Community voices illustrated the erosion of warning credibility over time. In Satkhira, elderly participants reflected: “In the last five years, they announced ‘great danger’ four times, but only one storm was really bad. Now when they say evacuate, we think maybe it’s another false alarm” (FGD-04). Younger residents in Patuakhali echoed similar sentiments: “The weather forecast always says ‘severe cyclone’ but then the storm goes to India, Myanmar or weakens. We waste time and money going to shelter for nothing” (FGD-22). Officials acknowledged the credibility deficit in warning systems. A meteorological officer admitted: “Weather prediction has uncertainty, but people expect 100% accuracy. When our forecasts are cautious and the impact is less severe, communities lose trust in future warnings” (KII-25). A CPP coordinator noted: “Volunteers face difficult situations when communities say ‘you cried wolf too many times.’ We have to rebuild trust after every false alarm” (KII-02).

Likewise, ‘distrust in aid’ (particularly in Bagerhat and Pirojpur) highlights institutional legitimacy deficits, which have been shown to reduce compliance with official directives³⁹. FGD revealed deep skepticism about post-disaster assistance. In Bagerhat, participants shared experiences: “After the last cyclone, relief was given only to people who knew the UP chairman personally. We evacuated but got nothing, while some non-evacuees received tin sheets” (FGD-10). Pirojpur residents expressed political concerns: “Some people think evacuation orders are political games. They say the government wants to show they care, but real help never comes” (FGD-13). Institutional representatives recognized these legitimacy challenges. A DMC member reflected: “Past irregularities in relief distribution created lasting distrust. Now people think if they evacuate and lose property, they won’t get fair compensation” (KII-09). An NGO coordinator added: “Communities remember broken promises from previous disasters. This memory affects their willingness to follow official evacuation orders” (KII-14).

Notably, ‘gender-specific barriers’ such as privacy concerns for women (prominent in Barguna) and ‘culturally embedded factors’ like livestock protection reflect socio-cultural complexities often overlooked in top-down evacuation campaigns. Like previous studies in coastal region of the country^16,34, FGDs highlighted unique challenges in evacuation settings. In Barguna, despite relatively higher evacuation rates, women articulated concerns: “Even in better shelters, there’s no proper place for women during menstruation or for breastfeeding mothers. We feel embarrassed and uncomfortable” (FGD-19). Cultural concerns emerged in discussions: “Our elderly mother-in-law refuses to go to shelter because she says it’s not proper for women to sleep in public places with strangers” (FGD-20). Religious and community leaders provided cultural context for evacuation resistance. An imam explained: “Some families worry about maintaining purdah (modesty) in crowded shelters. For conservative families, this is a serious religious and cultural concern” (KII-26). A women’s group leader noted: “Livestock protection is often women’s responsibility, especially poultry and small animals. Women feel guilty leaving these animals that provide family income” (KII-27). These findings align with feminist political ecology perspectives emphasizing the gendered dimensions of disaster vulnerability^4,40,41.

The regional variation in driving factors—‘underestimation of risk’ in Bhola versus ‘political bias’ in Pirojpur—suggests that hyper-local socio-political contexts significantly influence evacuation decisions. Community discussions revealed diverse risk assessment approaches. In Bhola, participants expressed confidence in local knowledge: “Our grandfathers could predict cyclones by looking at clouds, bird behavior, and wind patterns. Sometimes our traditional knowledge is better than the weather office” (FGD-27). This contrasted sharply with political concerns in Pirojpur: “Some people don’t evacuate because they think it’s a government show. If you support the opposition, you don’t want to make the ruling party look good” (FGD-14). Key informants contextualized these political dimensions. A local journalist observed: “In some areas, disaster response becomes politicized. People’s evacuation decisions are influenced by their political allegiance rather than just safety concerns” (KII-15). Conversely, an elderly leader in Bhola explained: “Older community members have survived many cyclones using traditional knowledge. They trust their experience more than modern forecasting systems” (KII-22).

Inconsistencies of our findings with previous studies—such as the elevated importance of religious institutions in shaping risk perception in some unions—may stem from differences in survey timing, cyclone severity, or local leadership dynamics^16,31,36. Research timing emerged as a critical factor in key informant discussions. A university researcher noted: “Risk perceptions change rapidly after disasters. Surveys conducted immediately after impact versus months later can yield different results about evacuation barriers” (KII-28). A long-term development worker added: “Leadership changes in religious and community institutions affect their influence on disaster response. What we observe today might be different next year” (KII-29). Furthermore, the observed variations in shelter distance as a barrier (e.g., in Bhola and Satkhira) may reflect inconsistencies in shelter placement planning relative to population density. In Bhola, participants explained: “The nearest shelter is 4 km away, and we have no transport during storms. With elderly parents and small children, it’s impossible to walk that far in strong wind” (FGD-28). Satkhira residents faced different challenges: “The shelter is close, but the road floods quickly. By the time we decide to evacuate, the path is already underwater” (FGD-01).

These results collectively support the view that evacuation behavior is not merely a function of hazard intensity but is linked to the interplay of physical infrastructure, socio-economic constraints, cultural norms, and institutional trust. The qualitative evidence demonstrates that quantitative patterns of non-evacuation reflect complex decision-making processes where households weigh multiple, often competing priorities. A mixed-gender focus group captured this complexity: “Evacuation is not a simple yes or no decision. We think about safety, but also about our animals, our house, our dignity, whether help will really come, and whether the warning is true. All these things matter” (FGD-25). The need for community-tailored evacuation strategies rather than one-size-fits-all approaches was emphasized by practitioners. A disaster management specialist concluded: “Our research shows that effective evacuation strategies must integrate shelter improvement programs with community-led communication campaigns that address localized barriers and leverage trusted local institutions. Generic approaches fail because they don’t account for the specific social, economic, and cultural contexts that drive people’s decisions” (KII-30). This reinforces recent calls for locally led adaptation and micro-level planning that account for community-specific perceptions and barriers, challenging the homogenous implementation of disaster risk reduction policies across diverse coastal communities⁴².

The observed damage patterns across sectors and districts highlight notable spatial heterogeneity (Fig. 3). Crop losses were most severe in Barguna (91%) and Satkhira (88%), consistent with their agrarian economies and saline intrusion after cyclones^43,44. Fishing asset losses peaked in Patuakhali (94%) and Bhola (91%), reflecting the concentration of marine fishing activities in these regions³⁶. The high structural damage in Bhola (79.5%) and Barguna (72.5%) aligns with their high physical exposure and housing vulnerability³¹. However, the relatively lower livestock losses in Pirojpur (13%) contrast with earlier studies that reported higher livestock mortality in similar events³⁴. This discrepancy may be attributed to pre-TC livestock relocation efforts or community-based animal sheltering programs in certain districts. Geographical features may have also partially shielded Pirojpur from the TC’s most intense impacts. FGDs with farmers and fishers reinforced these observations. Farmers in Barguna and Satkhira reported, “The saltwater ruined our paddy fields. Even years after the cyclone, we could not grow anything.” (FGD-02). Fishers in Patuakhali and Bhola echoed these sentiments, stating, “We lost not only our boats but also fishing nets, which are expensive to replace.” KIIs with local government officials corroborated the findings, with an official from Bhola adding, “Housing damage was extensive because many homes here are built with lightweight materials that cannot withstand strong winds” (KII-23). These findings are indicative of the systemic vulnerability of Barguna and Bhola, which may relate to their geographic location, economic dependency on climate-sensitive sectors, and inadequate adaptive infrastructure. They also highlight the importance of pre-cyclone measures, such as livestock relocation, in mitigating losses, as evidenced in Pirojpur.

Post-TC unmet needs reveal a significant gap in the disaster management cycle (Fig. 6). While DMCs excel in early warning dissemination, their recovery and response mechanisms appear less effective. The widespread need for drinking water, housing repair, and agricultural input highlights systemic weaknesses. In particular, the prominence of drinking water as a critical unmet need across regions like Khulna, Barguna, and Satkhira aligns with studies documenting the vulnerability of water systems to saline intrusion and infrastructure damage^17,43,44. FGDs in these regions emphasized the impact of this issue, with participants reporting, “We had to walk miles to find drinkable water because all the ponds and wells were contaminated after the cyclone” (FGD-03). KIIs with NGO workers in Khulna further confirmed these challenges, with one respondent stating, “Restoring water infrastructure takes time, and often, it is not prioritized during the initial recovery phase” (KII-06).

The spatial variation in unmet needs also reflects differential institutional capacity. Housing repair needs were concentrated in unions like Royenda, Gulishakhali, and Sonakata, where structural damage was most severe. This consistency between damage patterns and unmet needs was highlighted in FGDs, where participants stated, “Even months after the cyclone, we are still living under makeshift roofs” (FGD-11). A KII with a district-level official in Barguna attributed these delays to funding shortages, noting, “We prioritize immediate relief, but housing reconstruction requires sustained resources that we often lack” (KII-12).

Livelihood-related needs, such as agricultural inputs in Nishanbaria and Khontakata and fishing nets in Char Manika, underscore the economic vulnerability of coastal households. This finding aligns with studies emphasizing that post-cyclone support often neglects livelihood restoration³⁸. FGDs with farmers in Bagerhat and Pirojpur revealed frustrations, with one participant stating, “We received some food aid, but no one helped us restart farming. Without seeds and fertilizers, how can we recover?” (FGD-12). Similarly, fishers in Bhola reported, “We need nets and boats to earn a living, but aid rarely includes these” (FGD-26). These accounts highlight the inadequacy of standardized relief packages and the need for livelihood-specific recovery support.

The persistence of food aid as a critical need in Lata Chapli and veterinary support in Dhulasar further illustrates how unmet needs often reflect local economic dependencies—fishing communities need nets, agricultural communities need seeds, and livestock-rearing communities need veterinary care. This specificity underscores the importance of contextually appropriate response strategies rather than standardized relief packages³⁹. These findings collectively suggest that current institutional response mechanisms may be insufficiently adaptive to local contexts and needs. The consistent reporting of basic needs like water and food indicates failures in the immediate relief phase, while the prominence of livelihood-related needs points to longer-term recovery gaps. Future response efforts should prioritize: (1) rapid restoration of water infrastructure; (2) targeted support for housing reconstruction in high-damage areas; (3) livelihood-specific recovery packages; and (4) strengthened local institutional capacity to ensure equitable and efficient response across regions.

The regional patterns of unmet needs reveal inconsistencies with previous research. While some studies highlight effective government-NGO collaboration in post-cyclone responses^38,45,46, our findings indicate significant gaps, particularly in remote or heavily impacted areas. KIIs with NGO representatives in Khulna suggested that these gaps are often due to coordination challenges and logistical constraints, with one respondent noting, “Reaching the most affected areas is difficult, especially when resources are limited” (KII-06). Conversely, the relatively fewer unmet needs in parts of Patuakhali and Barguna may reflect stronger local governance or NGO presence, as noted by a local Union Parishad member in Barguna, “We worked with NGOs to distribute aid quickly, which helped reduce unmet needs” (KII-13).

Among the seven districts, Bhola stands out as the highest-performing district, reaching a notable 90% reach in early warning dissemination (Fig. 7), consistent with its strong CPP presence and operational efficiency. This finding aligns with previous studies^28,47 that underscore the effectiveness of Bhola’s community-based early warning systems in mitigating disaster risks. However, Khulna’s relatively low cautionary reach (72%) underscores a critical gap in this vital defense layer. This weakness may be ascribed to logistical encounters, limited coverage of CPP volunteer, or broken communication infrastructure (e.g., limited road networks) in this district²³. FGDs in Bhola stressed the efficiency of early warnings: “We were alerted well in advance, and volunteers ensured everyone in the community received the warning,” noted one respondent (FGD-27). Conversely, FGDs in Khulna revealed significant shortcomings: “The warnings arrived late, and many households did not receive them at all,” reported a community leader (FGD-07). KIIs also emphasized these discrepancies, with one CPP volunteer in Bhola stating, “Our team is well-trained and actively engaged with the community before and during cyclones” (KII-24). While a government official in Khulna acknowledged, “The lack of volunteers and communication tools limits our ability to disseminate warnings effectively” (KII-07).

Aid distribution further highlights the inconsistency in institutional performance. Bagerhat reveals noteworthy strength, reaching 66% of households, followed by Bhola (64%). Conversely, low aid coverage in Patuakhali (37%) indicates substantial operational and logistical challenges in reaching the affected communities. These observations align with earlier studies in the coastal region of the country^38,46, recognizing resource constraints (e.g., limited aid) and harmonization inefficiencies between UDMCs and NGOs as crucial barriers to effective aid distribution in affected regions. FGDs in Bagerhat emphasized the district’s functioning strengths: “Aid reached us quickly after the cyclone, and the distribution process was well-organized,” said one participant (FGD-12). In contrast, FGDs in Patuakhali reported frustration among affected families: “We waited for months, but most of the aid never arrived,” lamented a farmer (FGD-24). KIIs with local officials provided insight into these discrepancies. A government officer in Bagerhat ascribed their success to tight management with NGOs and community leaders: “We worked closely with local organizations to ensure timely and efficient aid distribution” (KII-10). While in Patuakhali, a UDMC member acknowledged, “Our limited resources and poor logistics made it difficult to reach remote areas” (KII-20).

Community trust in UDMCs varied widely across districts, with Barguna and Satkhira reporting the highest levels (3.5 and 3.2, respectively). This higher score in trust underscores strong community confidence in UMDC members, likely due to transparent decision-making, shelter management, and timely assistance in evacuation, and primary input support (e.g., seed and food aid). However, Pirojpur displayed a critical trust deficit (2.0), which may stem from perceived disorganizations in aid distribution and inadequate community engagement. Likewise, Khulna and Bagerhat reported moderate-to-low trust levels (2.8 and 2.5, respectively), despite operational strengths in areas like aid distribution. This suggests that operational performance alone is inadequate to build community trust. FGDs in Barguna and Satkhira revealed high community trust in local institutions: “The authorities were transparent and kept us informed throughout the disaster,” said one participant in Barguna (FGD-20). Conversely, FGDs in Pirojpur reflected a lack of trust: “We don’t see the authorities until after the cyclone, and even then, the help is minimal,” noted a respondent (FGD-15). KIIs offered supplementary context, with a UDMC member in Barguna emphasizing the importance of community engagement: “We involve the community in decision-making, which builds trust and ensures better outcomes” (KII-11). In contrast, an elected leader in Pirojpur admitted, “We struggle to maintain community trust due to limited resources and delayed responses” (KII-16).

Our equity analysis reveals concerning disparities in disaster response systems. The 11.8 percentage point gender gap in warning reception suggests that early warning systems may be less effective at reaching female-headed households, potentially due to gender-based differences in mobility, social networks, or access to communication devices. This finding aligns with feminist political ecology perspectives emphasizing gendered dimensions of disaster vulnerability. The higher aid receipt by male-headed households (61% higher odds) despite similar evacuation rates raises questions about equity in post-disaster resource distribution.

The regression models provide statistical evidence supporting several observed patterns. The strong association between shelter quality and evacuation (OR = 1.41, p = 0.005) suggests that infrastructure investment may be an important factor for protective behavior, independent of gender or location. The higher crop losses on islands (β = +6.64%, p = 0.056) validate their heightened biophysical vulnerability. Surprisingly, higher crop damage was associated with lower aid receipt (OR = 0.99, p = 0.088), suggesting potential inefficiencies in needs-based targeting or that the most damaged households face barriers in accessing aid systems.

Our quantitative and qualitative findings disclose that steady performance in one domain (e.g., warning dissemination or aid distribution) does not essentially translate into elevated institutional trust. This underscores the need for a holistic approach to disaster management that discourse district-specific limitations. For instance, while Bhola surpasses early warning dissemination, its relatively moderate trust levels (3.0) suggest the need for enhanced community engagement and transparent communication. Similarly, underperforming districts like Patuakhali and Pirojpur require targeted interventions to advance both operational effectiveness and community confidence. FGDs and KIIs steadily emphasize the importance of modifying disaster management approaches to local contexts. For example, FGDs in Bhola emphasized the role of well-trained volunteers, while KIIs in Patuakhali pointed to the need for better logistical planning. These insights reinforce the importance of context-specific interventions, such as expanding CPP coverage in Khulna, improving aid coordination in Patuakhali, and fostering community engagement in Pirojpur.

Establishing better coordination among local authorities (UDMCs and UzDMCs), NGOs, and community stakeholders, alongside investments in infrastructure (e.g., renovating vulnerable cyclone shelters) and capacity-building of the officials of the responsible departments, can help address these gaps and safeguard more equitable and effective disaster management across vulnerable areas of the country. By integrating FGD and KII findings into disaster management planning, policymakers can better align institutional responses with community needs, eventually improving resilience in susceptible districts.

The effectiveness of DMCs in Bangladesh is not a simple binary but a spectrum. They have built a remarkably efficient system for warning dissemination, a monumental achievement. Yet, their effectiveness is curtailed by deeper, structural issues: trust deficits, economic vulnerabilities, and uneven recovery support. Future policies must therefore evolve beyond improving hardware and infrastructure. They must develop trust through community engagement, address economic barriers through social protection schemes linked to evacuation, and empower UDMCs to develop context-specific solutions. Our research affirms that the next frontier in disaster risk reduction lies not just in predicting the hazard, but in thorough understanding of the human landscape it impacts.

A deeper analysis of the DMCs’ operational constraints reveals why their on-the-ground impact is often less than their formal mandate would suggest. The primary challenges are structural and resource-based. First, chronic underfunding and logistical shortfalls critically hamper response and recovery phases. As evidenced by the low aid coverage in Patuakhali (37%), UDMCs frequently lack the budgetary autonomy, storage facilities, and transportation to distribute relief equitably and swiftly after a disaster. This creates a gap between warning dissemination and tangible support, eroding community trust. Second, there is often a significant capacity and coordination gap. Many UDMC members are part-time volunteers or local officials with multiple responsibilities, lacking dedicated training in needs assessment, supply chain management, or community engagement. Furthermore, coordination between Union and Upazila-level DMCs and with NGOs can be ad-hoc, leading to duplicated efforts in some areas and complete neglect in others. Finally, a persistent top-down approach can limit effectiveness. While DMCs are designed for local action, their plans and resources are often dictated by higher administrative tiers, leaving little flexibility to address hyper-local unmet needs—such as the specific demand for fishing nets in Char Manika or veterinary support in Dhulasar. This mismatch between standardized procedures and contextual realities explains the discrepancy between the DMCs’ success in the uniform task of warning dissemination and their variable performance in the complex, adaptive tasks of response and recovery. Addressing these limitations requires moving beyond structural existence to ensuring functional resourcing, capacity building, and genuine community integration in the disaster management cycle.

Policy must formally recognize and resource the most effective communication channels identified—the CPP and religious institutions. This involves: increasing the number of trained CPP volunteers in low-coverage areas (e.g., Khulna); providing them with advanced communication tools; and officially integrating mosque and temple committees into the early warning dissemination chain, complete with training and simple equipment like megaphones or battery-powered radios. Shelter improvement programs must be prioritized and tailored. Investment should focus not just on constructing new shelters, but on upgrading existing ones in low-rated areas (e.g., Patuakhali) with gender-segregated spaces, reliable WASH facilities, and secure storage for belongings and small livestock. This directly addresses the security and cultural barriers to evacuation. Post-cyclone aid must evolve beyond generic food aid. Our data calls for pre-positioned, sector-specific recovery kits: seeds and salt-tolerant saplings for farmers in Barguna and Satkhira; fishing net and boat repair materials for communities in Patuakhali and Bhola. This approach would directly reduce the economic fear that deters evacuation. Given the spatial variation in evacuation barriers (e.g., false alarm fatigue in Satkhira, political distrust in Pirojpur), UDMCs should be empowered and funded to develop union-specific communication campaigns. These would directly address local misconceptions and leverage trusted local voices to deliver credible warnings and evacuation orders.

Future research should track how specific policy interventions—like shelter upgrades or the integration of religious leaders into warning systems—impact community trust and evacuation compliance over multiple cyclone seasons. This would move beyond cross-sectional correlation to establish causality. A rigorous economic analysis comparing the cost-effectiveness of high-tech warning dissemination (e.g., satellite SMS) versus strengthening human-centric networks (CPP, religious institutions) is needed to guide optimal resource allocation. Qualitative and survey research should delve deeper into the negotiation of risk within households, particularly along gender and generational lines, to design more nuanced communication and shelter management strategies that address the concerns of all members. Research should explore the potential of hybrid systems that combine the reach of digital alerts (SMS, social media) with the verification and motivation provided by trusted human intermediaries, testing models to overcome the digital literacy and infrastructure barriers we identified.

This study has several limitations that should be considered when interpreting the findings. First, the cross-sectional design establishes associations but cannot definitively establish causal relationships between predictors and outcomes. Second, data based on self-report are subject to recall bias (especially for details of past cyclone events) and potential social-desirability bias (e.g., in reporting evacuation behavior or institutional trust). We mitigated these risks through methodological triangulation—corroborating survey responses with qualitative data from FGDs and KIIs—which provided consistency checks and contextual depth. However, triangulation reduces but does not eliminate these inherent biases. Third, while we controlled for key covariates, unmeasured confounding from factors like detailed social capital, exact distance to roads or shelters, or specific historical surge exposure may persist. Finally, the purposive sampling of highly exposed unions ensures relevance for vulnerable areas but limits generalizability to less exposed coastal regions. Future longitudinal studies incorporating objective measures and environmental data would help address these limitations.

In conclusion, this study demonstrates that effective disaster risk management in coastal Bangladesh hinges on integrating institutional systems with local social contexts. The CPP achieved over 72% warning reach, peaking at 90% in Bhola, while religious institutions emerged as a key node, reaching 45% of households in some districts. However, substantial equity gaps—including an 11.8 percentage point deficit in warning reach for female-headed households—undermine inclusivity. Shelter quality is a strong predictor of evacuation, and damages show significant spatial clustering. Building resilience requires a paradigm shift toward context-sensitive, community-tailored strategies. This includes: 1) investing in shelter quality and location based on spatial clustering patterns; 2) formally integrating trusted networks like religious institutions into warning systems with inclusion safeguards; 3) designing equity-focused aid programs that actively reach female-headed and fishing households; and 4) developing livelihood-specific recovery packages. Future research should prioritize longitudinal studies to measure intervention impacts, intra-household decision-making analysis to address gendered barriers, and cost-benefit analyses comparing communication channels.

Methods

Study area description

This study focuses on seven highly disaster-prone districts in southwestern and southcentral (Satkhira, Khulna, Bagerhat, Pirojpur, Barguna, Patuakhali, and Bhola) Bangladesh (Fig. 8). These districts are positioned in the lower estuary of the Granges-Brahmaputra-Megna delta, the largest delta system in the world²⁴. From each district, the southernmost sub-district (upazila) was selected purposively, and from each of these, the two most exposed unions were chosen for the study. These fourteen selected unions are among the regions most affected by disasters (TCs and storm surges) in recent decades, having borne the impact of TCs including ‘Sidr’ (2007), ‘Aila’ (2009), ‘Mora’ (2017), ‘Bulbul’ (2019), and ‘Remal’ (2024)^16,21,22,26. Furthermore, these regions are severely impacted by TC- and surge-induced environmental disturbances. Unions in the southwestern districts (e.g., Gabura, Munshiganj, Koyra, and Dakshin Bedkashi) face severe soil and water salinization^43,44. Conversely, unions in the south-central districts (e.g., Char Kukri Mukri, and Char Manika) are highly degraded by erosion and inundation^48,49.

Fig. 8: Location of the study area in coastal Bangladesh.

Study area map showing the location of Bangladesh (a), seven exposed coastal districts (b), and 14 selected unions (c) from these districts (2 unions from each district).

Our selected regions experience a tropical monsoon climate, characterized by high temperatures and heavy rainfall. The average annual temperature ranges between 25 °C and 35 °C, with peak heat exceeding 40 °C during the pre-monsoon months (April–May). Annual precipitation varies from 1800 mm to 3000 mm, with most rainfall occurring during the monsoon season (June–September). TCs typically strike between April–May and October–November and are often accompanied by storm surges that inundate vast coastal areas. Due to their high exposure to the Bay of Bengal, the selected regions are among the most susceptible to the adverse impacts of sea level rise. Several studies have reported sea level rise at rates ranging from 4.0 mm/year in the south-central region (e.g., islands near the mouth of the Meghna River) to 7.8 mm/year in the southwestern region (e.g., the Sundarbans mangrove forests in Khulna)⁴⁹. Sea level rise along the country’s coast is projected to reach 32 cm by 2050 and 88 cm by 2100⁵⁰.

The population in these regions is mostly rural and reliant on climate-sensitive livelihoods, including agriculture, fishing, shrimp farming, and daily wage labor⁵¹. Poverty levels are high, with many households living below the national poverty line. Landlessness and inadequate access to freshwater, medical service, and education worsen susceptibility⁴³. The population density is moderate to high, with Muslim and Hindu communities inhabiting in mixed settlements. Relocation to cities is common due to recurring disasters and livelihood uncertainty⁵². Due to frequent catastrophic events, soil and water salinity have amplified significantly⁵³, reducing farming productivity. Rice cultivation has declined⁴⁴, while saline-water shrimp farming has expanded⁵⁴, altering traditional farming systems. Freshwater scarcity is a major concern, as saline intrusion contaminates ponds and groundwater, forcing communities to rely on rainwater harvesting or distant freshwater sources⁴³.

Data collection

We employed a mixed-methods approach, including household surveys, FGDs and KIIs for collecting the data from the households, communities and stakeholders in the selected. These data were collected during March–July 2023 using three semi-structured questionnaires. We employed a multistage purposive sampling in selecting the districts (highly disaster prone), sub-districts (southernmost sub-district), unions (most exposed to cyclones and heavily affected by disasters in recent decade). For selecting the unions, we consulted with the Upazila DMC members in the respective sub-districts, and the highly rated unions were considered as our study unions.

Based on the information from the UDMC members, we identified households for surveying using a systematic transect walk. Transect lines were established along the riverbank, traversing the areas deemed most exposed to cyclones, riverine flooding and storm surges based on recent Union Parishad’s Risk Reduction Action Plan (RRAP prepared by UDMC), and prior historical impact data. Along each transect, households were selected at systematic intervals (e.g., every 10^th household) to ensure spatial coverage and reduce selection bias. From each district, we collected information from at least 40 households using a semi-structured questionnaire. A total of 287 household surveys were conducted, of which 8 surveys were excluded because of incomplete information. Thus, 279 household surveys were considered for analysis. Using this questionnaire, information on community awareness of DMC roles, experiences during past disasters, evacuation efficiency, shelter accessibility, and post-disaster support received were collected. The surveys also documented persistent challenges in disaster response and recovery, providing valuable insights into the ground-level effectiveness of existing disaster management systems.

A semi-structured questionnaire was employed to collect information on community-level perceptions regarding the role of DMCs in disaster preparedness, early warning, emergency response, evacuation, shelter management, livelihood disruptions, post-disaster recovery process. We collected this information from 28 FGDs, 2 from each union. Each FGD consists of 6–8 participants, including farmers, fishermen, elected representatives, NGO representatives, religious leaders, local elites, business communities, and CPP representatives. We tried to ensure the participation of youth, elderly, and women in FGDs. Each FGD was 40-60 min long and data was collected using local language.

We collected valuable information from the key informants located in respective unions and sub-districts. A semi-structured questionnaire was used to collect the required information related to disaster risk reduction interventions. We collected information from 30 KIIs from the union, sub-district, districts and national levels. A wide range of participants were selected for data collection, including DMC members at union and sub-district level, NGO staff in DRR sector, community leaders, religious leaders (mosque Imam), schoolteachers, university teachers, local elites, journalists, meteorologists, and CPP representatives. These interviews offered institutional perspectives on coordination mechanisms between different stakeholders, resource limitations affecting preparedness efforts, and local adaptations to improve early warning systems. The qualitative data from these interviews helped identify systemic gaps in disaster governance that may not be apparent through quantitative surveys alone.

Informed consent and ethical approval

Ethical considerations were strictly followed, with informed consent obtained from all participants and data anonymization procedures implemented to protect respondent identities.

Ethical consent was obtained from the Research Review Board of German University Bangladesh. Approval number is GUB/ETP/res/2017-10/54(6). This research adhered strictly to the ethical guidelines of the World Medical Association Declaration of Helsinki regarding human subjects.

Measurement of key constructs

Key outcome and explanatory variables were constructed as follows:

Institutional trust: A composite index was created to measure community trust in local disaster management institutions. The index was calculated as the mean score of responses to three questions assessed on a 5-point Likert scale (1 = Strongly Distrust to 5 = Strongly Trust): (1) How much do you trust the UDMC to act in your community’s best interest during a cyclone?; (2) How much do you trust the UDMC to distribute relief aid fairly?; and (3) How much do you trust the information provided by the UDMC about cyclone warnings and evacuation? The scale demonstrated acceptable internal consistency (Cronbach’s α = 0.78).

Aid coverage: This was defined as the percentage of households in a given union that reported receiving any form of post-cyclone assistance (including food, cash, housing materials, or agricultural inputs) in response to the question: “Did your household receive any assistance (food, cash, materials) from any source (government, NGO, community) after the last major cyclone?”.

Shelter rating: The perceived quality of cyclone shelters was measured as the mean score (1–5, where 1 = very poor and 5 = very good) from evacuees’ ratings of the nearest shelter’s condition, space, cleanliness, and facilities.

Non-evacuation reasons and unmet needs: These variables were recorded as categorical responses from pre-defined lists developed through preliminary fieldwork and literature review. The exact question wording and full list of response categories are provided in Supplementary Information 1.

Data processing and analysis

Raw survey data were cleaned to ensure consistency and accuracy. This process involved handling missing values through descriptive diagnostics, employing mode imputation for minor, random missingness in categorical variables, and excluding entries with critical data gaps. Data recoding was performed to convert categorical responses into factors and to validate numerical ranges. Furthermore, composite indices were constructed from related questions to measure latent constructs such as institutional trust and perceived warning efficacy. Summary statistics (frequencies, proportions, means, standard deviations, medians, and interquartile ranges) were calculated for all key variables to describe the central tendencies and distributions of the sample population. This provided an overview of household demographics, cyclone exposure levels, evacuation behaviour, and unmet needs.

Spatial data processing was conducted to integrate and analyze the geographical dimensions of the survey data. This involved joining the household survey data to union-level administrative shapefiles using common identifiers, such as union names, within R using the sf package⁵⁵, thereby creating enriched spatial layers for analysis. Subsequently, choropleth maps were generated to visualize the spatial distribution of key variables, including average cyclone exposure, evacuation rates, shelter ratings, and the prevalence of unmet needs across the study unions. Finally, these maps were analyzed to identify spatial clusters—hotspots and coldspots—of high disaster impact and low response, which revealed critical geographic patterns in vulnerability and post-disaster recovery. Spearman’s correlation⁵⁶ coefficients were calculated to quantify the strength and direction of linear relationships between the correlation between shelter rating and evacuation rate at the union level.

To assess the geographic patterning of key variables, we performed spatial autocorrelation analysis using the Global Moran’s I statistic. This measure quantifies whether features with similar values are clustered (I > 0), dispersed (I < 0), or randomly distributed (I ≈ 0) across the study area. The analysis was conducted at the union level for variables including cyclone exposure, sectoral damage percentages, evacuation rates, and primary unmet needs. We used the centroids of the union boundaries from our study area shapefile to create a spatial weights matrix based on queen’s contiguity (where unions sharing a boundary or a corner point are considered neighbors). The moran.test function from the spdep package in R was applied to calculate the Moran’s I statistic and its associated p-value, testing the null hypothesis of spatial randomness. Significance was assessed at p < 0.05.

We identified the significant differences in shelter quality perceptions across unions and districts. Prior to selecting a statistical test, the normality of residuals was formally assessed using a Shapiro-Wilk test⁵⁷ (Table 3) and visual inspection of a Q-Q plot⁵⁸ (Fig. 9), which confirmed a significant deviation from normality, justifying the use of non-parametric methods. A non-parametric Kruskal-Wallis test⁵⁹ was first employed to assess the overall significance of differences in median shelter ratings across unions and, separately, across districts. Upon finding a statistically significant result from the Kruskal-Wallis test (p < 0.05), a post-hoc Dunn’s test⁶⁰ was applied for pairwise comparisons to identify which specific unions or districts differed from each other. The Bonferroni method⁶¹ was used to adjust the p-values in the Dunn’s test. The following packages were applied to the analysis: the stats package for the Kruskal-Wallis and Shapiro-Wilk tests; the dunn.test package for post-hoc pairwise comparisons with Bonferroni adjustment; and the car package for checking model assumptions. Data visualization was conducted using ggplot2 within the statistical software R⁵⁵.

Fig. 9: Assessment of residual normality for shelter rating ANOVA models.

Normality assessment using quantile-quantile (Q-Q) plots of the residuals from the one-way ANOVA models for shelter rating by the households across the (a) Unions and (b) Districts. The solid red line indicates the expected normal distribution; points closely following the line suggest normality of residuals.

Table 3 Results of Shapiro-Wilk normality tests on the residuals of the one-way ANOVA models for shelter rating

To assess distributional equity, we calculated absolute gaps in warning reach and aid receipt across three dimensions: gender (male- vs. female-headed households), simplified livelihood (farmer vs. fisher based on district-level predominance), and location (island vs. mainland unions). To identify determinants of key outcomes, we fitted multivariable regression models: 1) a logistic regression predicting evacuation, 2) a linear regression predicting crop loss percentage, and 3) a logistic regression predicting aid receipt. Predictors included shelter rating, gender, island location, and crop loss levels. A sensitivity analysis excluded the highest-exposure union to test robustness.

Qualitative data from FGDs and KIIs notes were analysed using thematic analysis⁶². This process involved systematically coding the transcripts to identify recurring themes, concepts, and patterns related to key issues such as challenges in warning reception, evacuation decisions, and unmet needs. These codes were then grouped into broader thematic categories—such as “barriers to evacuation” and “reasons for distrust in institutions”—to provide rich, contextual explanations that complemented and elucidated the quantitative findings.

Demographic and socio-economic characteristics

Household size

The average household size among survey respondents varies across the seven coastal districts, with Bhola reporting the largest households at 6.5 persons, followed by Barguna at 6.3 persons. Moderately sized households were observed in Khulna and Patuakhali (both 5.8 persons), Bagerhat (5.6 persons), and Satkhira (5.4 persons), while Pirojpur had the smallest average household size at 5.2 persons (Fig. 10a).

Fig. 10: Socio-demographic profile of surveyed households across seven coastal districts.

Distribution of household size (a), the percentage of gender of the respondents (b), family type (c), respondents’ age (d), their education level (e), and main occupation (f) across the seven study districts.

Gender distribution

The gender distribution of our household surveys across the seven coastal districts reveals a consistent pattern of male dominance (Fig. 10b). Male-headed households constitute the majority in all districts, ranging from 58% in Satkhira to 78% in Barguna. Conversely, female respondents are most prevalent in Satkhira (43%), followed by Bhola (33%), while the lowest proportions appear in Barguna (23%) and Bagerhat and Pirojpur (both 30%) (Fig. 10b).

Family type

Extended families were the predominant household structure in most districts, particularly in Patuakhali (64%), Khulna (61%), Barguna (59%), Satkhira (58%) and Bagerhat (53%). In contrast, nuclear families constituted the majority in Bhola (62%), followed by Pirojpur (54%) (Fig. 10c).

Age group

The distribution of respondents age indicates that middle-aged and older adults comprise the majority of our respondents, with the 50–60 age group representing the largest cohort in Khulna (32.5%), Patuakhali (32.5%), and notably in Bhola (35.9%) (Fig. 10d). Similarly, the 40–50 age group was dominant among respondents in Bagerhat (40%) and Barguna (35%). Younger respondents aged 18–30 were consistently underrepresented across all districts, with the lowest proportion in Pirojpur (2.5%) and the highest in Satkhira (10%). The elderly population (aged over 60) was most prominent among respondents in Bagerhat (25%) and Satkhira (22.5%) (Fig. 10d).

Education

The educational attainment of household survey respondents reveals substantial disparities across the seven coastal districts, with illiteracy rates notably high in Bhola (46.15%), Patuakhali (42.5%), and Barguna (40%), while primary education represents the highest level achieved for 37.5% of respondents in Pirojpur and 35% in Bagerhat (Fig. 10e). Secondary education attainment is relatively moderate across most districts, peaking in Pirojpur (30%), though progression to higher secondary education remains limited, with the highest rates in Patuakhali (20%) and Satkhira (15%). Respondents holding a degree or above are nearly absent in Barguna and Bhola (0%) and remain below 3% in Satkhira, Pirojpur, and Patuakhali, with only Khulna (7.5%) and Bagerhat (5%) showing slightly higher proportions (Fig. 10e).

Occupation

The occupational distribution of household survey respondents reveals distinct livelihood specializations across the coastal districts, with farming representing the primary occupation in Pirojpur (52.5%), Bagerhat (45%), and Satkhira (40%), while fishing dominates in Bhola (56.4%), Barguna (47.5%), and Patuakhali (42.5%) (Fig. 10f). Day labor constitutes a supplementary income source across all regions, peaking in Patuakhali (15%) and Khulna (12.5%), whereas business activities remain limited, not exceeding 10% in any district. Notably, wood collection is only reported in Bagerhat (5%), Satkhira (7.5%) and Khulna (12.5%), reflecting localized mangrove forests dependencies, while expatriate work is most significant in Pirojpur (10%), indicating higher international migration from this district (Fig. 10f).