This study examined the association between weather events and service utilization among unhoused populations using 2-1-1 call data in southern Louisiana from 2014 to 2023. By analyzing over 20,000 calls, we identified notable patterns in help-seeking behavior among 2-1-1 users and different weather events. While temperatures were significantly associated with changes in service utilization, consistent declines, with reductions during active cyclone events and recovery periods during tropical cyclones, possibly due to power outages, disrupted networks, or increased utilization of other resources (e.g., shelters, emergency services), highlighting a complex response pattern. This finding challenges conventional assumptions about disaster response. It suggests the presence of multiple barriers, including disrupted communication infrastructure, evacuation dynamics, or strategic reliance on alternative support networks, which may influence the utilization of 2-1-1 among the unhoused. In contrast, temperatures exhibited consistent associations, particularly at the highest and lowest temperatures, with more 2-1-1 calls related to housing and shelter concerns. These results have implications for disaster preparedness and response strategies targeting service-utilizing unhoused populations, who face disproportionate exposure to climatic events [8, 9] and simultaneously encounter structural barriers to accessing traditional support systems [10, 11].

We observed reductions in 2-1-1 call volume during active cyclone events and recovery periods, with non-significant trends during pre-cyclone and immediate-aftermath phases. Beyond individual decision-making, systemic factors may create additional barriers to 2-1-1 access during cyclones, including power outages, telecommunications disruptions, and cell tower damage that physically prevent calling. During declared emergencies, alternative service pathways typically become available that are not accessible during non-disaster periods. Emergency shelters open with fewer barriers to entry, and emergency services are more actively deployed in affected areas (HUD [30]). Additionally, disasters often activate informal community support networks, including family, friends, faith communities, and neighborhood connections, which provide temporary assistance without requiring formal service systems [31]. Temporary expansion of the safety net may also reduce reliance on 2-1-1 as the primary point of access to services.

Additional explanations may account for this phenomenon. Doran et al. [10] found that emergency department visits for newly unhoused individuals increased following Hurricane Sandy. This suggests that individuals with recently disrupted housing may turn to medical facilities rather than social services during disasters, possibly due to limited awareness of alternative services. Complementing this perspective, Settembrino and Allen [32] documented how chronically homeless individuals often develop sophisticated self-reliance strategies through years of negotiating life on the economic margins. Their qualitative research revealed that many unhoused individuals deliberately choose to “ride out” hurricanes in improvised shelters (i.e., tents) rather than engage with formal emergency services, reflecting lived experience and a preference for autonomy over perceived institutional restrictions. This finding suggests that unhoused populations often assess and respond to risk in dynamic ways [33, 51], which may help explain reduced 2-1-1 call volume during cyclone periods in this marginalized population.

Our temperature analysis revealed a pronounced U-shaped association between weather extremes and service utilization, with both the coldest and warmest temperature deciles showing significantly increased 2-1-1 calls compared to moderate conditions. The distributed lag non-linear modeling corroborated this U-shaped temperature-response pattern. The association was more pronounced during extreme cold, with significantly elevated housing and shelter requests. This pattern is consistent with documented temperature-related health vulnerabilities among unhoused populations [9, 34] and suggests that temperature variations may influence service-seeking behavior among those connected to formal support systems and the need for shelter during these temperatures.

Our computational text analysis of call narratives revealed associations between weather conditions and specific service needs among unhoused populations. Mental health mentions increased as temperatures rose from the coldest to the warmest deciles, while showing significant decreases during cyclone periods compared to non-cyclone conditions. Similarly, rising temperatures were generally associated with increased mentions of housing/shelter, as well as healthcare concerns. These temperature-dependent patterns suggest shifts in seasonal vulnerability [34], while the cyclone-related decrease in mental health mentions, despite consistent proportions of other service needs, indicates a potential reprioritization of immediate survival needs during disaster phases. Our findings highlight how climate conditions shape the volume and content of help-seeking behavior, with implications for seasonally tailored intervention strategies that address these shifting priority landscapes.

The demographic profile of unhoused 2-1-1 callers, predominantly women (65.3%) and working-age adults (55.0%), reveals complex social dynamics underlying service utilization patterns beyond simple exposure risks, and aligns with general 2-1-1 utilization patterns, whereby adult women account for the majority of 2-1-1 clients [4, 5, 35,36,37]. These patterns also align with [48] observation that homelessness represents a social position where vulnerability factors intersect, with ethnicity, gender, health, and sexuality creating nuanced risk profiles. The higher proportion of female callers reflects documented gendered differences in help-seeking behavior, potentially driven by caregiving responsibilities, heightened exposure to gender-based violence, or more extensive social networks [38]. Future work incorporating intersectional analysis may demonstrate how help-seeking patterns are shaped by complex survival strategies that challenge simplistic narratives of vulnerability during disaster contexts [39].

Traditional disaster response frameworks often overlook the complex realities faced by marginalized populations, particularly those experiencing homelessness, whose risk is shaped more by social, economic, and political conditions than by natural hazards alone [39]. While models such as the Protective Action Decision-Making framework [32, 49, 50] provide valuable insights into threat response, they may not fully capture the decision-making processes of individuals living in chronic precarity [50]. Rather than reflecting passive vulnerability, reduced service utilization during extreme weather events illustrates the agency of unhoused individuals navigating structural barriers to safety [40, 41]. Our findings reinforce Settembrino and Allen’s [32] call to move beyond top-down notions of vulnerability by recognizing localized, experience-driven interpretations of risk [51]. For individuals experiencing homelessness, immediate survival needs often outweigh concerns about weather events, highlighting the need for risk frameworks that center on structural and situational constraints. Policy implications include developing seasonally tailored interventions for temperature-dependent mental health and housing needs, reducing access barriers through non-traditional outreach, investing in year-round resilience efforts, and coordinating community-informed services [32, 51].

Our findings have direct implications for enhancing 2-1-1 services as one critical access point within the broader emergency response infrastructure. The temperature-dependent patterns we observed, particularly the U-shaped association with increased calls during both extreme cold and heat, suggest opportunities for proactive resource allocation and targeted outreach by 2-1-1 providers during forecasted temperature extremes. The distinct patterns observed during tropical cyclones, showing decreased call volume despite presumably elevated need, highlight the importance of multimodal service delivery that does not rely solely on phone-based systems during disasters, when infrastructure disruptions may limit 2-1-1 accessibility. For 2-1-1 programs specifically, these findings suggest several operational improvements: (1) implementing weather-triggered surge staffing 2-1-1 protocols during temperature extremes; (2) developing partnerships with emergency shelters and emergency services to create seamless referral pathways during cyclone events when 2-1-1 access may be limited; (3) tailoring resource databases to reflect seasonal shifts in service needs, particularly the increased mental health mentions during warmer temperatures; and (4) recognizing that 2-1-1 serves as one node in a larger network that includes emergency departments, homeless shelters, faith-based organizations, and informal community support systems. Understanding utilization patterns among service-connected individuals enables 2-1-1 programs to better serve this population while acknowledging that parallel efforts are needed to reach those not currently connected to formal services.

Our study offers several methodological contributions to understanding the relationship between extreme weather events and service utilization among unhoused populations. First, utilizing 2-1-1 call data from a decade (2014–2023) provides a longitudinal perspective on help-seeking behavior during weather events. This approach addresses traditional limitations in homelessness research, which often rely on cross-sectional data collection that fails to capture temporal dynamics [8]. Additionally, real-time 2-1-1 call data to reflect service needs represents an innovative approach to monitoring vulnerable populations during environmental conditions. Unlike traditional surveillance systems, which often experience significant reporting delays, 2-1-1 data captures immediate help-seeking behavior, enabling more responsive intervention strategies during emerging crises [3]. Second, our mixed-methods approach, combining negative binomial regression, distributed lag non-linear modeling, and natural language processing, provides a more comprehensive understanding of service utilization than previous studies. The integration of statistical modeling with computational text analysis of narrative call records offers both quantitative trends and qualitative insights into the lived experiences of unhoused individuals during extreme weather events, addressing the shortcomings of single-method approaches that may overlook critical contextual factors [42].

Despite these strengths, some limitations should be considered. First, 2-1-1 data only captures those aware of and able to access telephone services, potentially underrepresenting the most vulnerable populations, such as individuals without phones, those unaware of 2-1-1 services, or those with severe mental illnesses [3, 43]. Our administrative data also limits differentiation between sub-populations of homelessness, which may experience weather differently. Secondly, temperature exposure assignment may introduce misclassification, as unhoused individuals may be more mobile than housed populations, and their reported ZIP code location may not reflect where they experienced the day’s weather conditions. While this method provides the best available approximation given the constraints of 2-1-1 specialist-collected data, the potential for exposure misclassification should be considered when interpreting the results. Exposure misclassification may also occur from cyclone exposure which was assigned uniformly across the entire study area rather than ZCTA-specifically, which may not capture spatial variations in storm impacts across our study region Thirdly, the geographic scope of our study was limited to southern Louisiana and a limited number of tropical cyclone events (Table 1), which may limit the generalizability of our findings to other regions. Fourth, societal changes over the study period, such as policy shifts and evolving awareness of 2-1-1 services, may confound observed trends. We also cannot determine what proportion of the unhoused population is familiar with 2-1-1 services or likely to use them, which affects the interpretation and generalizability of our findings. Fifth, NLP narrative analysis and counselor-coded categories capture distinct dimensions of service needs, and unadjusted versus adjusted estimates may differ due to confounding potential explaining differences in our results. Finally, while we identify associations between weather and service utilization, causal inferences remain limited due to the potential presence of unobserved variables that influence help-seeking behavior. Future research should integrate primary data collection and comparative geographic analyses to enhance generalizability and address these limitations, and should include other data sources, such as the Department of Housing and Urban Development’s Continuum of Care Program shelter reports, to validate 2-1-1 as a proxy for vulnerable populations, such as the unhoused.