Improving adaptation to wildfire smoke and extreme heat in frontline communities: evidence from a community-engaged pilot study in the San Francisco Bay Area

Exposure to climate hazards is increasing, and the experiences of frontline communities warrant meaningful and urgent attention towards how to mitigate, manage, and adapt to hazards. We report results from a community-engaged pilot (November 2021–June 2022) of N = 30 participants in four frontline communities of the San Francisco Bay Area, California, USA. The study region is an area where low-income, non-English-speaking residents are inequitably exposed and vulnerable to wildfire smoke, extreme heat, and other climate hazards. Building from a yearslong partnership of researchers, community organizations, and community members, we report the feasibility of a project piloting (1) instruments to monitor indoor air quality, temperature, and participant sleep health, and (2) interventions to improve indoor air quality and support protective behaviors. Data collection included experience-based survey data (via in-person administered surveys and a smartphone application) and interviews about heat and air quality, as well as data from an air monitoring protocol. Results cover the prevalence of hazard exposure and protective actions among participants. We discuss throughout methods for conducting and evaluating a community-engaged pilot, particularly by using a community ambassador program. Implications include the feasibility of community-engaged research projects, including discussion of resources required to accomplish this work.


Introduction
Environmental extremes like extreme heat (Tuholske et al 2021) and wildfire smoke (Burke et al 2021) are escalating in places like the American West due to climate change and other factors 11 . These exposures can result in acute health impacts such as cardiorespiratory events, hospitalization, and premature death (Stowell et al 2019, Leibel et al 2020, Aguilera et al 2021, Liu et al 2021. Frontline communities are more exposed (Cardona et al 2012 IPCC) due to a legacy of racism (Tessum et al 2021), lower income (Wing et al 2022), less access to high-quality healthcare (Lang et al 2016), location (Rappold et al 2017), and language barriers (Méndez et al 2020). Reducing illness, suffering, and early mortality, especially in frontline communities (Trujillo-Falcón et al 2021), will require deliberate intervention to make protective actions more accessible and widespread, advancing climate equity and resilience (Fitzgerald 2022).
Our team of researchers from Stanford University and community organization partners from Climate Resilient Communities and El Concilio of San Mateo County developed a community-engaged research program 12 'Our Communities, Our Bay' in the San Francisco Bay Area, USA. The program aims to understand and support frontline communities in managing, mitigating, and adapting to climate change-related hazards ('climate hazards'). To inform the design of the program's multi-year longitudinal field study, a pilot study with 30 participants was conducted November 2021 through June 2022 focused on extreme heat and wildfire smoke. Participants were recruited by the program's community ambassadors ('ambassadors') who tested (1) low-cost monitoring sensors (air quality and sleep quality), (2) a smartphone application ('app') called 'Our Communities' , and (3) survey instruments administered in-person and via the app. Evaluation of measures and procedures were assessed through interviews and an ambassador listening session.
Increasingly, research has focused on engaging frontline communities in environmental research (Fernandez-Bou et al 2021). Community-engaged research is a process by which partnerships can identify meaningful and pressing questions and codevelop study design (Key et al 2019, Wong-Parodi 2022. Collaborating can generate new views of 11 Wildland fires are increasing due to the combination of climate change and societal decisions about management (Pechony and Shindell 2010), like the development of the wildland-urban interface. 12 Principles of community-engaged research include the structure of the partnership, relationships between actors, and a focus on delivering intermediate and long-term outcomes to improve equity and justice, although each of these attributes can vary on a continuum. See (Key et  problems and solutions, yielding improved outcomes and better science (Medin et al 2017) by lending voice in the process to those excluded in the past (Ortiz et al 2020). While there is increased interest in community-engaged research (DePrince et al 2022), it is not always clear what the process entails or how challenges are incorporated back into study design. In this pilot study, we evaluated how technological and behavioral interventions for improving individual adaptation to wildfire smoke and extreme heat may be used in frontline communities. We report on operating procedures, successes, and challenges from a community-engaged research method, and we detail how the project was accomplishable by collaboration with community leaders.

Prior work
The pilot design is rooted in several years of exploratory work in California conducted in San Mateo County, Santa Clara County, and Northern California by our team. Project selection was guided by local needs and priorities as determined by community organizations in the region who became key partners. Exploratory work began in 2019 with a low-cost wearable air quality monitoring sensor and smartphone app with 18 residents from East Palo Alto and an online survey of 265 California residents impacted by the 2018 wildfire season. Participants for the monitoring and app study were recruited with community partner assistance and focused on learning more about residents' experience with climate hazards, risk perceptions, protective behaviors, health impacts, and usability of wearable air quality sensors and a smartphone app as a decision support and data collection tool. Participants for the online survey were recruited through Amazon's MTurk platform and provided information regarding how vulnerable populations (e.g. households with low-income, children under the age of five, and/or residents suffering from respiratory illnesses) viewed and managed wildfire smoke risks (Wong-Parodi 2020). From 2019 through 2020, an interview study of 45 Northern California residents was conducted to better understand the role of experience, social influences, and protective behaviors during the devastating 2018 and 2020 wildfire seasons (Santana et al 2021). We conducted a second interview study of 40 low-income, hard-to-reach Bay Area residents recruited with community partner assistance to understand how they perceived and coped with the compounding risk of climate hazards and COVID-19 (Fischer et al in preparation). Finally, we conducted a third interview study of 15 practitioners in Northern California and 15 community members of North Fair Oaks in 2021 to better understand opportunities and barriers for a community-level intervention, clean air centers, to protect against wildfire smoke exposure (Treves et al 2022).

Broad aims
As discussed previously, project selection was shaped by collaboration between researchers and community organizations focused on serving frontline communities. Similar partnerships between academic institutions, independent research institutes, and community advocacy organizations have been deployed for other community engaged research studies (Balazs and Morello-Frosch 2013). Within our partnership, a key priority for ongoing work was whether interventions can support household adaptation to wildfire smoke, extreme heat, and other climate hazards in the region, including flooding. Some interventions are more engineered or technological, including weatherization retrofits to add insulation and improve building envelope and air filtration systems to decrease infiltration from outdoor pollution and heat. Other interventions are primarily behavioral, including educational interventions to increase the adoption of protective action and reduce exposure to climate hazards. Amid this suite of possible actions in which communities and policymakers may consider investing, two key pieces of knowledge are missing to demonstrate realized outcomes.
The first piece of knowledge is how each proposed intervention works in realistic settings, including demonstrations of how communities of interest respond to and use proposed interventions in their daily lives as opposed to models or controlled lab settings, and if they are implementable without causing harm. This test of application is crucial for each of the proposed interventions, but the second piece of knowledge is equally important: how do interventions work together? Whether interventions work in tandem for the same communities in which they are tested is an important needed step in demonstrating feasibility. This work is a further step towards unifying technological and behavioral interventions into holistic approaches designed to work in frontline communities.
The present pilot study focuses on this first piece of knowledge by testing several interventions in the same population. The pilot was primarily focused on improving measurement quality and operating procedures that could scale to a study without compromising positive experiences for participants. With the results of the present pilot, we intend to further explore the second question in a longitudinal field study. The longitudinal study aims to characterize how residents in San Francisco Bay Area frontline communities adapt to climate hazards, and to experimentally evaluate the effectiveness of interventions in increasing the adoption of adaptation behaviors over two to three years and across several hazards. To meet these goals, the study pilot was intended to develop, test, and improve protocols (recruitment, enrollment, and intervention deployment) and materials (intervention materials, and study questionnaires).

Study area and population
In the densely populated region of the San Francisco Bay Area, residents vary widely on metrics of vulnerability, including socioeconomic status, race and ethnicity, immigration status, languages spoken at home, and access to services. Large numbers of residents who have one or more characteristics of social vulnerability reside in San Mateo County (Community Vulnerability Index 2022). For example, in 2021, 86.8% of the city of East Palo Alto in San Mateo County were non-white, non-Hispanic or Latino, 68.4% spoke a language other than English at home, and 11.4% lived in poverty. Frontline communities such as these already suffer disproportionately from exposure to environmental toxins (Colborn et al 2014, Davies et al 2018, disease burden (e.g. asthma, cancer, respiratory ailments) (Beck et al 2014, Benmarhnia et al 2017, and other communitylevel stressors (e.g. lack of access to healthcare, unemployment, etc).
Community partners directed the research team to four frontline communities within San Mateo County where our partners had strong on-theground relationships and recognition as respected leaders. The sample area was defined as eligible residents of the municipalities of Redwood City and East Palo Alto, census-designated area North Fair Oaks, and neighborhood Belle Haven of municipality Menlo Park (figure 1). These communities were chosen for their exposure to climate hazards, including their high risk of flooding (Bick et al 2021). The region's shortage of housing supply (Elmendorf et al 2020) and older housing stock-in some cases, a majority of which was built before 1960 (Chapple et al 2020)-contributes to residents' increased vulnerability. Households may lack sufficient cooling during extreme heat (Samuelson et al 2020) and have high infiltration to wildfire smoke (Liang et al 2021) and top-down rainfall (Velterop et al 2022). Discriminatory housing practices in the area made these communities home to many working-class, non-White families, among whom are communities of immigrants from Mexico, Central America, Tonga, and Samoa.

Recruitment
Our team developed an ambassador program to compensate community members for recruiting and enrolling their neighbors in the study. Residents engaging fellow residents to advance research and public health, also known as promotoras de salud, is a longstanding model in the region with demonstrated effectiveness at reaching vulnerable groups (Chugg et al 2021). Past community engaged research has demonstrated the benefits of community ambassadors in California as research collaborators beyond study recruitment, including in policy impact stages (Minkler et al 2010). Ambassadors were recruited through the community-based organizations. Project recruitment began in November 2021, and after a delay during the COVID-19 Omicron surge, project enrollment began in March 2022. For more details see the supplementary data.

Interventions
The research team designed the pilot to trial offthe-shelf interventions in frontline communities. A smartphone app was adapted from the U.S. Environmental Protection Agency's Smoke Sense app (Rappold et al 2019) as an educational and behavioral intervention for this study. Smoke Sense uses a citizen science model of collecting self-reported data about wildfire smoke experiences along with air quality maps, aligning with calls for citizen science to integrate with community-engaged, participatory research methods to improve environmental and population health (Katapally 2019). In this study, the updated app was used for survey data collection and realtime information delivery (air quality, smoke, and fire information), with two educational modules for air quality awareness and smoke exposure mitigation. Educational modules included a home screen with current and projected Air Quality Index (AQI) based on user ZIP code, trivia games testing knowledge of air quality facts, and an air quality map; Rappold et al (2019) report on these features in more detail. Behavioral components were added to the existing Smoke Sense app to include messages designed with frontline communities in mind. Messages were informed by research in behavioral science and tested before being added to the app. App features are intended to increase the use of air quality indices and behaviors that can reduce exposure, but these materials are often developed without the concerns of frontline communities in mind; for example, recommendations about not spending time outside cannot be reasonably adhered to for those who work outside, and suggestions that indoor air quality is better than outdoor air quality may not be true with high levels of infiltration in low-quality housing. The app was available in English and Spanish for both Android and iOS platforms, distributed through the public app marketplaces. All participants were to pilot the smartphone app, and each was invited with a unique access code to use the app. No personally identifiable information was stored in the app. To pilot an intervention that could improve indoor air quality, 15 participants used Medify Air portable air cleaners (MA-25) with high efficiency particulate air (HEPA) filters. Evidence suggests that portable HEPA filters, when used with closed doors and windows, can reduce the concentration of indoor air particulates caused by wildfire smoke (Fisk and Chan 2017). However, less is understood about the effectiveness of air cleaners in behaviorally realistic contexts like homes, and more evidence is needed to inform behavioral interventions to maintain their efficacy. For example, recommended operational parameters, like keeping windows closed, may not be realistic to individual behaviors, or lowquality housing may have air infiltration from outside that nullifies the improvements from an air cleaner. Participants were asked to place the air cleaner in the room in which they spend the most time.

Instruments
This study piloted two sensors for measuring air quality: one for measuring PM 2.5 concentrations and temperature (indoor PurpleAir sensors) inside participants' homes, and another for measuring indoor and outdoor PM exposure (RTI's MicroPEM) (figure 2). Indoor PurpleAir sensors were chosen because they are low-cost compared to other models, contribute to crowdsourced air quality data, and measure temperature. All 30 participants piloted PurpleAir sensors and were informed to keep sensors in central locations away from air vents, humidifiers, and cooking areas. MicroPEMs were chosen as reliable wearable instruments for measuring personal exposure to particulate matter (PM 2.5 and PM 10 ) indoors and outdoors (Cho et al 2016) and were piloted by six of the 30 participants. This instrument can identify microenvironments where exposure may be worse, including when working outdoors or when at other indoor locations such as work or while in transit.
Exposure to environmental hazards like wildfire smoke can negatively impact sleep quality (Integrated Science Assessment for Particulate Matter 2009). To measure sleep quality and other indicators of health (heart rate and breathing disturbance), Withings Sleep Tracking Mats ('sleep mats') were chosen as low-cost and nonintrusive instruments demonstrated in prior research to provide reliable measurement (Edouard et al 2021), and piloted by nine of the 30 participants.

Surveys
Survey development began in Spring 2021. Through meetings and document revision, the researchers and community partners collaborated on the concepts that were most important to ask, as well as how to best ask questions in respectful and culturally appropriate language. Surveys were written instruments designed to be self-administered by participants, or administered with the assistance of a study ambassador or personnel who could read questions and collect responses from a set of mostly closeended options. Three surveys were piloted: one survey administered at the time of enrollment, and two surveys administered through the app. The surveys asked about housing conditions, past experiences with climate hazards, and health. All surveys were written and administered in English and Spanish, with the intention of all 30 pilot participants completing each survey. We report survey results from this pilot to demonstrate the (1) prevalence and need for ongoing work focused on adaptive actions for heat and wildfire smoke in frontline communities and (2) feasibility of survey methods for our longitudinal approach. More information about survey items is provided in the supplementary data.

Evaluation plan
Evaluation occurred during and after data collection and was intended to update procedures for the longitudinal study. Primarily, evaluation was focused on improving measurement quality, participant experiences, and ambassador experiences. The research team led evaluation of measurement quality from indoor air quality monitors, personal exposure monitors, sleep mats, and survey instruments. The research team worked closely with project partners for evaluation of participant and ambassador experiences.
Five participants were recruited for in-depth interviews in March 2022 (mid-study), and 27 participants completed exit interviews from May through June 2022. In both interviews, participants offered reflections and areas for improvement on study enrollment, study equipment, and the smartphone app. For more details see the supplementary data.
The project team hosted a listening session with ambassadors in April 2022 with two main goals: first, to improve the installation and troubleshooting procedures for indoor air sensors and sleep mats; second, to cover intended changes for the full study, including a plan to implement a physical intervention of building weatherization (e.g. improving insulation, adding weatherstripping to windows and doors) in participating households (see supplementary data). Community partners and ambassadors explained to the research team how a weatherization intervention could not be ethically implemented as it could produce unintended harms. They detailed how participants who rent their homes would need to contact their landlords for permission, and that this could expose participants to extra surveillance by their landlords. Additionally, these weatherization improvements could be a subtext for landlords to increase rents, which could contribute to the displacement of participants and gentrification of these neighborhoods. The project team subsequently re-designed the study to not include a weatherization intervention.

Participant characteristics
The following results draw from surveys collected from 30 pilot participants. Most participants (25, 83.3%) spoke Spanish as their primary language at home. Three participants (10%) lived in households with children under five years of age, and four participants (13.3%) lived in households with adults over 65 years of age; both are markers of vulnerability. Additional description of participants regarding household characteristics, housing, access to resources, health, and demographics are included in the supplementary data.

Understanding and managing exposures
Wildfire smoke. When asked how participants knew whether air quality was bad, fewer reported having looked at the AQI than relying on environmental or social cues (figure 3). Participants also reported several negative effects of wildfire smoke events on their health and well-being based on past experiences, including the 2020 wildfire season when the air was smoky for many days. To protect themselves from smoky air, 24 participants (80%) reported wearing some type of mask. However, only 11 participants (36%) used a N95 or KN95 mask, which is effective protection against smoke. Six (20%) participants reported not wearing a mask. Two participants (6.6%) used a HEPA room purifier. All participants reported adopting a combination of several out of eight protective actions during wildfire smoke events (figure 4, panel (A)), including staying inside and wearing masks. While most participants reported taking adaptive action, some action provided no protection and could be maladaptive. For instance, three participants (10.0%) reported wearing a damp cloth around their mouth to breathe through, which provides little protection from wildfire smoke.
Heat. As with wildfire smoke, based on past experiences including heatwaves in the summer of 2020, most participants reported negative health effects (figure 3). To protect themselves from extreme heat, all participants reported taking at least one protective action ( figure 4, panel (B)), and most participants (n = 28, 93.3%) reported taking more than one action out of thirteen actions. The most common actions reported were those to make the individual more adaptive to the heat: spending less time outside (n = 20), wearing light layers (n = 19), and drinking 'lots of ' water (n = 16). Participants also reported actions that made their homes more comfortable, including keeping windows covered with drapes or shades (n = 19), avoiding cooking with their stove or oven (n = 16), and using a fan in their home (n = 14).  Few participants reported using air conditioning in their homes (n = 8) 13 , corresponding with the small number who reported having air conditioning.

Smartphone app
During the study period, the research team deployed two app-based surveys paired with a short message service (SMS) notification. Real-time data monitoring revealed slow survey completion rates, which prompted the study team to extend data collection for each survey over multiple weeks. In-depth interviews revealed key problems with the study app deployment, including participants not having the app downloaded at enrollment (interviewer notes, 3 March 2022; interviewer notes, 11 March 2022) and not knowing how to navigate the app to take surveys: 'No, I do not see [the surveys]' (verbatim, 7 March 2022). With this information, the project team developed a plan to collect any outstanding pilot survey data by phone or in-person.
Ultimately, n = 27 (90%) of participants completed app surveys, either in real-time on their smartphones or during exit interviews. Conversations during exit interviews illuminated new challenges for deploying the survey app among this sample, including participants reporting that it felt important to have face-to-face interaction with the researchers to answer questions (interviewer notes 9 June 2022, 14:00) and others who reported having younger family members teach them how to navigate the app (interviewer notes, 17 May 2022). These challenges for implementation prompted an overhaul of the app implementation for the longitudinal study: increased ambassador training for app installation and walk-throughs, increased research team support for troubleshooting before survey deployment, twoweek survey data collection windows, and more frequent notifications in-app and via SMS when surveys are deployed.

Indoor air cleaner
Ambassadors deployed air cleaners in participant homes with instructions on best use. Upon collecting study equipment, study personnel learned some air cleaners went unused during the pilot, as collected materials included three cleaners with plastic film still on filters. Although these interventions were unused or could not be used correctly, several participants reported in exit interviews being glad to try the air cleaners. The interest these participants had to try something that could improve their indoor air quality and health bolstered the team's interest in this intervention.
From this pilot, the study team revised plans for air cleaner deployment by instead delivering air cleaners directly from study personnel. Additionally, this challenge identified the opportunity for behavioral support to make intended interventions meaningfully protective.

Indoor ambient air monitoring
Community ambassadors installed indoor PurpleAir sensors in the 30 participant homes. Of the 30 deployed sensors, 15 (50.0%) sent partial or complete data (figure 5). Eleven did not transmit data during the pilot period. In analysis, the team identified Wi-Fi connectivity issues as a major driver of failed connectivity during participant enrollment.
At the conclusion of the pilot, participants were given summary information about their indoor air quality. Participants with complete data were interested and provoked by what they saw, including reflecting on why their air quality was worse than other households in the study (interviewer notes, 9 June 2022, 12:00) and how their behavior might help with air quality: 'made her feel good about her behavior of opening windows on clear days' (interviewer notes, 17 May, 2022). In homes where PurpleAir sensors did not transmit data, participants were 'surprised' (verbatim, 9 June 2022, 13:50) and 'disappointed' (verbatim, 1 June 2022, 19:00;verbatim, 9 June 2022, 14:00) to not receive summary information regarding their indoor air quality. Challenges with incomplete data led to increased ambassador training and the development of a realtime connectivity protocol for air quality monitors. For the updated protocol, trained members visit enrolled households where data transmission is either spotty or incomplete and troubleshoot disconnected sensors. Lessons learned from the pilot led the research team to increase budget and team hours to ensuring sensor connectivity immediately following participant enrollment and continue our monitoring and troubleshooting capacity throughout the longitudinal study.
PurpleAir sensors contribute to crowdsourced knowledge of local air quality. Our pilot communities had fewer publicly available sensors than neighboring communities within the same county. There were a total of 409 indoor PurpleAir sensors and 657 outdoor PurpleAir sensors within San Mateo County during the pilot; however, few were within the pilot communities, as sensors are inequitably distributed in frontline communities in northern California (Kelp et al 2022, Kramer et al 2023, perhaps because of sensor costs.

Personal exposure monitoring
Six participants who received the MicroPEM were given the device turned on with a battery life that would last over the following 48 h. After deployment, initial evaluation of collected accelerometer data in the MicroPEM files suggested that all participants carried the monitors, and four out of six (66.7%) devices were likely carried with participants during much of their day.
The MicroPEM real-time measurements showed significant within-and between-person variabilities for PM exposures. Pilot results suggest that participants need additional clear directives to make best use of this monitoring equipment for measuring differences between exposures in participants' homes, places of work, and other routine locations. Results informed the research team's decision to table the deployment of MicroPEMs for the longitudinal study, with the team considering an additional pilot to test a mail-out protocol (Ha et al 2020).

Sleep health monitoring
Ambassadors installed sensors ('sleep mats') where participants slept. For this sensor, participants were particularly curious about its functioning, the data collected and its meaning in connection to their quality of sleep; one participant asked specifically in an exit interview why a study about the environment would be measuring sleep, or why these topics were related (interviewer notes, 4 May 2022). In our listening session, ambassadors reported several participants raised concerns about the monitors' intrusiveness and the protections for their privacy; participants needed a high level of trust in the ambassador to consent to this data collection.
Takeaways from this implementation were focused on both technical challenges and information gaps between teams to be resolved in the full study design. As noted above, sensors did not send reliable data for all participants. As both sleep mats and PurpleAir sensors transmit data wirelessly, concerns about internet connectivity were priorities for the longitudinal study. The research team also developed a monitoring system to understand closer to real-time which sensors were not transmitting data, improving our troubleshooting. The community partner team re-trained ambassadors on installation protocol to reduce both technical challenges and information gaps that may interfere with data transmission (e.g. unplugged sensors). For instance, after learning that several participants reported not understanding what each intervention or instrument did in their homes, subsequent training focused on providing ambassadors and participants more complete information.

Recruitment and enrollment
During interviews, participants reported positive experiences with the ambassadors, who they described as kind, well-prepared, and who made them feel confident about participating. The positive experience of participants was a key strength of using an ambassador program for the study and strengthened the project team's commitment to this model going Note: Connectivity of deployed indoor air quality monitoring (PurpleAir) sensors. Grey squares are days with incomplete data transmission, and days with complete data transmission are colored according to AQI category (legend).
forward. Subsequent changes to the enrollment procedure focused on what participants had identified as ambassadors' strengths. Ambassadors were essential for establishing credibility and working with participants, and members of the research team could not have performed these roles to the same effect. For the longitudinal study, the research team sought increased funding to better compensate ambassadors for their work recruiting and enrolling participants. Additionally, this new funding and changing compensation structure was designed to increase the follow-up time ambassadors spent as the point of contact for participants.

Conclusion
Building on years of partnership between researchers, community organizations, and community members, this work aims to understand how communityengaged research can be used to investigate how frontline communities are adapting to climate hazards.

Pilot successes and challenges
Community engagement drove the success of our collaboration, both in determining how to proceed with interventions and how to engage with the community with whom we want to conduct this work. Several project challenges led the research team to rescope our focus: fears of contributing to community displacement encouraged us to drop a weatherization protocol, difficulties deploying personal exposure monitors led us to pause this measurement, connectivity problems for air sensors led us to invest more in realtime troubleshooting, and concerns from participants regarding surveillance and discomfort with sleep mats required further investment in improving communication. Through these challenges, a pivotal success was the investment in an ambassador program.
The pilot illuminated what resources we required to accomplish community-engaged research. The team of ambassadors were instrumental to pilot success, and the research team sought additional funding to continue working with them for the longitudinal study. Day-to-day troubleshooting and tasks were managed by a leadership team that included university faculty and research staff with a team of community organization leadership. This project also benefited from the involvement of students over two academic years. Defining clear operational tasks and communicating updates among the team was a challenge, and weekly project meetings, shared documentation in a cloud-based folder, and a Slack channel for ad hoc communication were all crucial to our success. Just as crucial was an investment in working in Spanish, with additional consideration paid to Tongan and Samoan following the leadership of project partners, to best reach frontline communities in our study area. Investing in both written materials and in-person live translation during meetings improved our ability to share ideas and understand each other to improve our work.

Ongoing work and future directions
Interventions piloted were focused on individual-and household-level behaviors that could reduce shortterm exposure to wildfire smoke and extreme heat and improve health and well-being. In our ongoing work, we will explore a range of climate hazards in the region, including flooding, and further consider how interventions can support individuals managing, mitigating, and adapting to co-occurring hazards (e.g. wildfire smoke during an extreme heat event). Ongoing work is underway to better understand how behavioral and technological interventions can together deliver benefits in frontline communities reducing exposure to climate hazards. The research team is using the developed pilot materials, operating procedures, and trained personnel for a randomized controlled trial in the same communities. The randomized controlled trial uses piloted materials and processes, including the community ambassador model, smartphone app, and air cleaners, to measure outcomes using the same piloted instruments (PurpleAir monitor, sleep mat, and surveys). Behavioral interventions are message-based using the smartphone app, which could promise many avenues for finding those avenues most likely to support protective action (e.g. positive reinforcement, social norms). Ongoing work will evaluate how behavioral interventions can improve how technological interventions are used, thereby contributing to reduced exposure. Future work could also evaluate additional physical interventions, including crafting ethical ways to implement building improvements in frontline communities (e.g. weatherization); pursuing this work ethically means partnering with and centering community throughout.
Connecting the process of designing, deploying, and evaluating this pilot is a yearslong commitment to partnership and community engagement, enduring well beyond funding cycles. Committing to community-engaged research has been especially important when protocols fail (e.g. smartphone app surveys, sensor data transmission) or could cause harm (e.g. pausing work during the Omicron outbreak, dropping a study arm for weatherization). Working closely with community partners facilitated access for the researchers who would otherwise be unable to reach frontline community members while also ensuring the research aligned with community desires. Engagement improved the research methods deployed, including improved questionnaire wording. Work in the broader sciences supports the lessons we have learned: that partnership with a diverse set of research team members deploying mixed methods can learn more with communities otherwise not included in research, thereby generating new and useful knowledge (Medin et al 2017).
Conducting community-engaged research means prioritizing relationships because the process and findings are enriched by collaboration; this pilot generated unexpected insight, encouraged rethinking and reframing, and improved our ability to reach shared goals because it is community-engaged.
Results of our community-engaged pilot are providing additional evidence for ongoing policy concerns led by the projects' community teams. Efforts include investing in community climate teams or block action teams, which can mobilize neighbors responding to climate hazards. One additional goal of community partners in this project is keeping lowincome, non-White residents in their homes, resilient to environmental hazards. Steps towards this goal may include increased renter protection, paired with increased eligibility for renters to use low-income home energy assistance programs to pay for energy costs associated with clean indoor air.

Data availability statement
The data cannot be made publicly available upon publication because they contain sensitive personal information. The data that support the findings of this study are available upon reasonable request from the authors.