Main

Climate change is significantly increasing the frequency and intensity of extreme weather events, posing severe threats to both the environment and human health. As the planet continues to warm, we are witnessing an increase in the occurrence and severity of phenomena such as heatwaves, storms, floods, and wildfires. These events not only devastate ecosystems and economies but also pose significant risks to human health [1]. The health consequences are multifaceted, encompassing immediate and long-term effects. The repercussions include substantial loss of life, disability, and diminished well-being, driven by a complex interplay of physical injuries, the spread of infectious diseases, and the worsening of non-communicable diseases such as respiratory and cardiovascular conditions [2].

Extreme weather events also influence human behavior [3–4], which in turn might affect human health indirectly [5]. These behavioral shifts, often driven by the need to seek safety or comfort, has the potential to create compound risks for human health during ongoing public health emergencies [1]: for example, during the first three years of the Covid-19 pandemic, unprecedented heatwaves drove many people indoors, leading to a substantial number of excess Covid-19 cases and deaths [6–7]. This complex interplay between environmental stressors and behavioral responses highlights the potential for extreme weather events to amplify public health crises by creating conditions that facilitate the spread of infectious diseases. Despite these critical observations, the cascade effects of localized or permanent behavioral changes due to extreme weather events on the dynamics of infectious diseases remain largely underexplored. Addressing this gap is essential for enhancing our ability to predict and effectively respond to future epidemics in a world that has warmed approximately 1.3–1.5°C compared to the 1850–1900 average [8].

Wildfires are among the most disruptive events for human behavior, severely impacting air quality across vast regions [9] and posing significant concerns for public health. While much of the public health focus has been on the physiological effects of wildfires [10]—such as direct mortality due to short-term exposure to fine particulate matter (PM_2.5) air pollution [11], and synergistic effects of PM_2.5 exposure on respiratory infectious disease susceptibility and severity [12]—there has been less attention to how wildfire-induced changes in behavior might influence the spread of airborne diseases. However, perceptions of deteriorating air quality and government advisories [13] often lead individuals to reduce outdoor activities. This shift results in increased indoor activity, where conditions such as close social proximity, limited ventilation, and the accumulation of respiratory droplets can facilitate the transmission of airborne diseases [14,15]. From a public health perspective, with the increasing frequency and severity of wildfires, it has become crucial to understand how disruptions in outdoor activities during wildfires might indirectly impact disease dynamics in indoor settings.

To answer this question, we concentrated on the 2020 wildfire season, one of the most severe in recent history. In August 2020, thunderstorms sparked multiple wildfires across California, Oregon, and Washington [16]. By early September, a record-breaking heatwave and intense katabatic winds accelerated fire growth and spread smoke across Oregon and Washington state. To assess the fires’ impact on human behavior, we analyzed a weekly time series of the relative tendency for people to stay indoors versus outdoors across U.S. counties. This mobility metric, called the indoor activity seasonality index, was derived from mobile phone data [15]. While September is typically a period of low viral circulation for most airborne diseases, in Seattle (King County)—the only county with a substantial number of COVID-19 cases at the time—the effective reproduction number (R(t)) increased significantly approximately 10 days after peak wildfire-induced indoor activity (Fig A in S1 Text). This finding reinforces the plausibility of a link between increased indoor crowding and heightened transmission risk. Although multiple factors could have contributed to this trend, it underscores the need for a deeper understanding of the indirect epidemiological effects of extreme weather events.

By integrating mobility data with air quality indicators and a mechanistic epidemic model, we aimed to quantify the behavioral changes induced by wildfires and their implications for viral transmission, thereby informing public health strategies in response to wildfire events.

Results

Wildfires induce behavioral changes

During the 2020 wildfire season, numerous counties along the West Coast of the United States experienced significant deteriorations in air quality, as reflected by sharp increases in the Air Quality Index (AQI). These spikes were primarily due to smoke from extensive wildfires in California, Oregon, and Washington. The wildfire smoke was carried by atmospheric circulation, spreading poor air quality over vast regions, including areas far from the actual fires (Fig 1A).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Air Quality level during the West Coast wildfire season in 2020.

(a) Air Quality Index extracted for PM2.5 particle pollution for all counties in Oregon (OR), Washington (WA), and California (CA). Horizontal line at 150 indicates the Air Quality Index (AQI) threshold for health concerns; (b) The map highlights in dark red the most affected counties in Oregon (OR) and Washington (WA) that were included in our study. The orange counties represent unaffected areas used as a baseline. Both affected and unaffected counties are part of the Northern Indoor Seasonality Cluster (in dark grey). Base layer: https://www2.census.gov/geo/tiger/GENZ2020/shp/cb_2020_us_county_20m.zip.

https://doi.org/10.1371/journal.pclm.0000542.g001

Our analysis of AQI data revealed that counties in Oregon (OR) and Washington (WA) were particularly hard-hit, with AQI levels exceeding 150 during the 2020 wildfire season, as depicted in Fig 1A. Through much of September, indeed at least 8 large wildfires, each of 100,000 acres (400 km2) or more, were burning in Washington and Oregon, with 3 in Washington and 5 in Oregon. Although California also experienced elevated AQI levels over a more extended period, the specific air quality impact on September 10th was less severe compared to Oregon and Washington. We therefore focus our analysis on the most affected counties in OR and WA state (Fig 1B). In contrast, counties not affected by wildfire smoke generally maintained AQI levels below 50, indicative of normal air quality conditions.

Focusing on the counties with AQI levels exceeding 150 during the 2020 wildfire season, we examined the disruption in human behavior in such locations during the wildfires. As shown in Fig 2, our analysis revealed a sustained increase in indoor activity in the most affected counties for approximately four weeks during the wildfire events. In contrast, unaffected counties with healthy AQI levels exhibited no such increase during the same period (Fig B in S2 Text). The counties compared belong to the same Northern Indoor Seasonality Cluster (Fig 1B), meaning they share similar seasonal patterns of indoor and outdoor activity due to environmental aspects. Comparing counties within the same cluster is essential for isolating behavioral changes specifically related to the wildfire events, rather than broader seasonal trends based on exogenous factors. Refer to the Methods section for more details on county selection. Using a regression discontinuity approach (see Methods) [17], we found that Washington and Oregon states experienced an average increase in indoor activity of 10.8% and 14.3%, respectively. Notably, we observed a significant rise in indoor activity in all affected counties except for Marion County, Oregon (γ5% < 0), as shown in Table 1. Major cities were the most impacted, with indoor activities increasing by 11% in Portland (Multnomah County) and 16% in Seattle (King County).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Regression Discontinuity Coefficients by Affected U.S. County with 90% Confidence Intervals.

https://doi.org/10.1371/journal.pclm.0000542.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Shift in human behaviors due to wildfire smoke.

The figure shows the indoor activity seasonality index between July 1, 2020, and November, 01, 2020 in the 10 selected affected counties: Multnomah County (Portland City), Washington County, Clackamas County, Lane County, Marion County in Oregon state, and King County (Seattle city), Spokane County, Yakima County, Clark County, Thurston County in Washington state. In each subplot, we also show in yellow the median and 95% CI of the indoor activity seasonality index for the unaffected counties that we used as a baseline. The violet curves represent the AQI of the affected counties during the studied period.

https://doi.org/10.1371/journal.pclm.0000542.g002

Wildfire-response behavior shapes respiratory pathogen circulation

To assess how wildfire-induced changes in behavior impact the circulation of respiratory pathogens, we incorporated the indoor activity seasonality index—reflecting disruptions during wildfire events—into an infectious disease transmission model (described in the Methods section). This model enabled us to evaluate the effects of wildfires on various airborne respiratory diseases by examining a range of generation times, from 2 days (for flu-like illnesses) to 25 days (for pertussis-like diseases). We focused on Washington County, one of the areas most impacted by wildfire smoke, and compared the relative peak incidence of diseases in this area to that in unaffected counties. Our findings show that increased indoor activity due to wildfires significantly impacts disease spread, with this effect diminishing as the generation time lengthens (Fig 3A). For diseases with shorter generation times (less than one week), such as COVID-19 and influenza, we observed a notable increase in relative peak incidence (+2) and greater variability in disease spread between affected and unaffected counties. This suggests that the risk of increased epidemic activity is more pronounced when the disease dynamics has a temporal scale that is comparable with the duration of the human mobility disruptions.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. The cascade impact of the shift in human behavior on airborne diseases.

A) Relative variation in peak incidence between Washington County, OR, and unaffected counties for different generation times [20] with R₀= 1.3. The boxplot median represents the median relative peak incidence across unaffected counties, and the blue vertical lines denote the average generation times for various airborne respiratory diseases. B) Relative variation in peak incidence for the selected 10 affected counties compared to the unaffected ones for three different reproduction ratios R0 = 1.3 (flu-like w/ interventions), R0 = 1.5 (flu-like). and, R0 = 3.0 (COVID-19-like). C) Relative variation in peak incidence when masking interventions are put in place, with R₀= 1.3.

https://doi.org/10.1371/journal.pclm.0000542.g003

By simulating the disease dynamics in any affected county, we also tested different reproduction ratios (R₀) consistent with seasonal respiratory viruses and currently circulating strains of SARS-CoV-2 (Fig 3B). Results indicated that peak incidence in wildfire-affected counties surpassed that of unaffected counties, with the highest relative peak incidence in Washington County, OR (+1.98). This elevated peak incidence is attributed to the increased indoor activity and, in some counties, generally higher seasonal indoor behavior compared to baseline levels. The impact of increased indoor activities on disease spread diminished with higher R₀ values. Similar results were observed for the relative attack rate (Fig C in S3 Text). We also examined a scenario incorporating 50% population immunity, representing an upper bound for seasonal influenza. Estimating population-level immunity is inherently challenging, as it depends on both vaccination coverage and prior infection—factors that fluctuate across regions and seasons. While adult influenza vaccination coverage in the U.S. typically falls just below 50%, actual immunity is further constrained by modest vaccine effectiveness (ranging from 20% to 60% over the past two decades) [18]. Even under this assumption of 50% immunity, wildfire-affected counties still experienced a relative peak incidence above 1, underscoring the persistent and measurable effect of wildfire-driven behavioral disruptions on respiratory disease transmission—even in populations with substantial pre-existing immunity (Fig D in S3 Text).

Discussion

Our study provides novel insights into the indirect effects of climate change on public health by illustrating how wildfire-induced behavioral changes can exacerbate the spread of airborne diseases. The findings highlight the interconnectedness of environmental, behavioral, and epidemiological factors, emphasizing the need to consider the climate-driven behavioral shifts to characterize disease dynamics and design policies for disease prevention and control. Wildfires stand out among climate events for their profound disruption of human behavior, primarily due to the extensive reach of wildfire smoke. Carried over long distances by atmospheric currents, this smoke affects not only nearby areas but also distant communities, as was evident during the 2020 wildfire season [9,16]. While the direct health impacts of wildfire-related air pollution have been well-documented [11], there has been less attention given to the indirect effects—specifically, the increased time spent indoors during periods of smoke-induced poor air quality, which may enhance the transmission of airborne diseases in indoor settings. As global climate change is expected to intensify wildfire activity, leading to longer fire seasons and larger areas burned [19], it is crucial to understand these behaviorally driven risks. This knowledge is vital in shaping public health strategies and preparing for potential epidemic surges induced by extreme events.

To address this gap, we provided quantitative evidence on how the 2020 wildfire season altered human behavior, increasing the risk of pathogen transmission in indoor environments. Leveraging behavioral and air quality data, we first systematically assessed the disruption of seasonal outdoor activities across U.S. counties that experienced significant spikes in unhealthy Air Quality Index (AQI) levels. Our analysis reveals that declining air quality led to a reduction in outdoor activities, disrupting established behavioral patterns. Following the most intense wildfire events on September 10, indoor activity surged by 10.8% in Washington and up to 14.3% in Oregon. Even urban areas farther from the burn sites, such as Portland and Seattle, saw significant increases in indoor activity, with rises of 30% and 16%, respectively. These shifts in behavior were likely driven by concerns over unhealthy air quality, prompting self-isolation to avoid air pollution and adhere to public health guidelines.

By integrating the indoor seasonality index into an epidemic model, we show how behavioral shifts foster environments that facilitate the spread of airborne pathogens. By capturing the increased tendency of people to remain indoors versus outdoors during wildfires in the indoor activity index, we characterize broader variations in exposure risk by combining multiple transmission-related factors. These include elements that influence host contact, susceptibility, and transmissibility [15]. For instance, heightened indoor activity can lead to prolonged airborne exposure (e.g., spending extended time in confined spaces), increasing the risk of respiratory transmission. Higher indoor density can further enhance droplet transmission due to close-range interactions. Additionally, indoor environments with poor ventilation, elevated pollutant levels, limited sunlight, and low humidity can weaken immune responses, potentially increasing susceptibility. Climate-controlled spaces may also promote viral survival by maintaining low humidity and supporting viral spread. While the indoor activity index does not isolate these factors, it provides an aggregated measure of heightened exposure risk due to behavioral responses, based on indoor versus outdoor activities. This framework enables us to assess the impact of seasonal disruptions, such as those caused by wildfires, in comparison to typical seasonal exposure risks. By incorporating the seasonal index as a linear modulator of the force of infection in an SIR model, we can analyze the effects of these seasonal disruptions on disease dynamics. Our simulations indicate that the surge in indoor activity during wildfires significantly alters the transmission dynamics of airborne diseases, with varying effects depending on the pathogen’s intrinsic transmissibility and generation time. Diseases with shorter generation times, such as COVID-19 and influenza, exhibited more pronounced increases in peak incidence, as their rapid transmission aligns with the short-term mobility disruptions. In contrast, diseases like mumps and pertussis, which have longer infectious periods, were less impacted by these localized behavioral changes. Furthermore, variations in the pathogen’s basic reproduction number (R₀) reveal that the degree to which increased indoor activity affects disease spread is closely tied to the pathogen’s transmissibility. It is important to note that our model assumes a fully susceptible population, which reflects a particular epidemiological scenario. This assumption is most applicable to the early stages of novel pathogen emergence, as seen with SARS-CoV-2, or to highly mutable viruses such as Influenza A virus subtype H3N2, for which antigenic drift can erode existing immunity, even among previously exposed or vaccinated individuals. To test the robustness of our findings, we also explored an extreme scenario with 50% population immunity and found that the increase in peak incidence during wildfires remained substantial, reinforcing the significance of behavior-driven risk even in partially immune populations. These findings underscore the importance of considering disease-specific characteristics when developing public health strategies for extreme climate events.

To mitigate the compound effects of wildfire on infectious disease dynamics for flu-like diseases, our study highlights the importance of indoor masking as a preventive measure. We find that even a modest increase in mask use (10%) can reduce the rise in peak incidence during wildfires. In heavily impacted areas like Washington County, OR, and Yakima County, WA, masking rates above 50% may be required to effectively mitigate disease transmission. These findings have immediate policy implications, suggesting that public health guidelines should integrate recommendations for mask use during wildfire events to reduce the increase of flu-like disease circulation. Future guidelines should address the dual challenges of poor air quality and increased risk of disease transmission, promoting strategies that balance the need for respiratory protection with the need to prevent airborne pathogen spread. As wildfires and other extreme weather events become more frequent with climate change, it is essential to develop comprehensive public health strategies that account for these interconnected risks.

While our study provides valuable insights, it is important to acknowledge several limitations. Our analysis is based on a specific wildfire season in the United States, which may limit the generalizability of the findings. As more behavioral data on indoor activities become available, future research should seek to validate these results across diverse geographic contexts and different wildfire seasons. Additionally, this work relies solely on the ratio of indoor-to-outdoor activities, neglecting indoor activities at home, which could also impact epidemic circulation. Nevertheless, our primary objective is to elucidate the relationship between shifts in indoor behavior and respiratory disease spread in crowded public settings, to inform control policies. Future research should explore the role of various indoor environments in disease transmission during wildfires, including the impact of residential settings on epidemic dynamics. Lastly, our model does not incorporate public health data on infection rates, which limits the ability to directly correlate behavioral changes with disease outcomes. Further studies incorporating such data are urgently needed to explore the broader implications of climate change on specific diseases more comprehensively.

Our research provides a foundation for future investigations into the indirect public health impacts of environmental disruptions on human behavior. It underscores the urgent need to address the secondary effects of climate change on public health, with a particular focus on infectious disease transmission. By revealing how behavioral shifts induced by wildfires can elevate disease spread, our study advocates for the inclusion of behavioral responses in public health strategies. As climate change continues to alter our environment, it is crucial for public health interventions to evolve accordingly, equipping us to tackle the multifaceted and interconnected risks of a warming world.

Methods

Our analysis proceeded in two steps. First, we characterize weekly human behavior patterns during the 2020 wildfire season in the Western United States in both affected and unaffected areas to identify significant behavioral shifts. Second, we integrated these behavioral changes into a mechanistic model to assess their impact on many airborne diseases.

Detecting disruptions of indoor activity seasonality

Human behavior was quantified using the indoor seasonality index by county, as introduced in [15], which captures the tendency of individuals to engage in indoor versus outdoor activities. This index, derived from SafeGraph (now called Advan Patterns) Weekly Patterns mobility data spanning 2018–2021, tracks activity based on mobile phone users’ visits to over 4.6 million points of interest (POIs) in the United States, excluding home locations. POIs were categorized according to their primary indoor or outdoor nature using the six-digit North American Industry Classification System (NAICS) codes. For example, schools, hospitals, and grocery stores were classified as primarily indoor, while parks and cemeteries were classified as outdoor. Approximately 90% of the POIs were classified as indoor, 6.5% as outdoor, and 3.5% as unclear, as in the case of gas stations, which could have both indoor and outdoor features. To quantify the indoor seasonality index, a specific metric was defined as follows:

where and represents the number of visits to indoor and outdoor points of interest, respectively, in week t in the US county i.

This metric is then mean-centered to provide a relative measure of indoor activity seasonality that is comparable across all counties.

To detect shifts in indoor activity patterns, we applied a regression discontinuity (RD) approach [17]. RD is a statistical method that estimates the causal effect of an intervention by exploiting a sharp change in the relationship between a continuous assignment variable and an outcome variable. In our study, RD is based on the premise that indoor activity seasonality in each county remains consistent throughout the 8-week study period, except for the impact of the wildfire event. We employed a local linear regression model for the RD analysis, estimating the wildfire event’s effect, denoted as γ, which captures the difference in outcomes before and after the event. Additional details on the linear regression model are provided in S4 Text.

Infectious disease model and scenario analysis

We assessed the impact of behavioral shifts on epidemic circulation using a deterministic Susceptible-Infected-Recovered (SIR) compartmental model for each county. The model does not depend on county population. The compartmental model simulated the daily spread of respiratory pathogens from seven days before the AQI alert to 14 days after. We defined the force of infection for a county as [15]. β₀ is a constant parameter that takes into account the overall transmissibility, while is indoor seasonality index that accounts for season behavioral patterns in the county and potential disruption of human mobility caused by wildfires. We calculated the relative variation in disease incidence by comparing the modeled peak incidence and attack rate in affected counties with that in unaffected ones.

The model explores various reproduction numbers (R₀) equal to 1.3, 1.5, and 3 These values were selected because 1.3 and 1.5 are typical for seasonal respiratory viruses like influenza and SARS-CoV-2, while 3.0 approximates the basic reproduction number of the wild-type SARS-CoV-2. Additionally, we explored generation times ranging from 2 to 12 days to evaluate their impact on airborne diseases ranging from flu-like illnesses to pertussis-like illnesses [21].

To model the effect of mask-wearing, we incorporated a reduction factor m into the force of infection to account for masking. The reduced force of infected is defined as follow:

Where m = σρ+(1 − ρ), with σ set to 0.6, representing the estimated reduction in respiratory infection due to mask-wearing [22], and ρ representing the fraction of the population wearing masks during the wildfire event. We explored various values for ρ ranging from 0.1% to 50%. Mask-wearing was considered only on days with unhealthy are quality (AQI > 150). Additional details on the model’s mathematical framework and simulation algorithm are provided in S6 Text.

Supporting information

Acknowledgments

We gratefully acknowledge the Complexity72h workshop, held at IFISC in Palma, Spain, from 26 to 30 June 2023, where the initial design and a preliminary report of this study were developed.