Wildfires

In the Text4Hope program, individuals self-subscribe to receive daily supportive text messages for 3 months by texting “COVID19HOPE” to 393939. The messages are aligned with a cognitive behavioral framework, with content written by mental health therapists as well as our research team members (authors MH and VIOA). The following is an example of the messages sent: “When bad things happen that we can’t control, we often focus on the things we can’t change. Focus on what you can control; what you can do to help yourself (or someone else) today” [29 (link)]. The messages are preprogrammed into an online software that delivers messages at 9 AM each morning. At the onset of the first message, respondents are welcomed to the service and are invited to complete an online baseline survey capturing demographic information; COVID-19–related self-isolation/quarantine information; and responses on the Generalized Anxiety Disorder-7 (GAD-7) scale [30 (link)], Perceived Stress Scale [31 (link)], and the Patient Health Questionnaire-9 (PHQ-9) [32 (link)]. Survey questions were programmed into SelectSurvey.net, an online survey tool operated by the Alberta Health Services Evaluation Services Team. No incentives are offered to respondents. Participation in the program is entirely voluntary, and completion of the survey was not a prerequisite requirement to receive supportive text messages. Subscribers may opt out at any time by texting “STOP” to 393939. Survey responses will be stored within our regional health system (Alberta Health Services) Select Survey account, and data will be exported, stored, and maintained by the Research and Evaluation team within our health region. The supportive SMS text messaging project subscriber recruitment plan was based on the success of a Text4Mood program in Alberta that was launched in response to the Fort McMurray wildfire disaster in 2016. Text4Hope has been the subject of a wide-exposure communications campaign (TV, radio, internet, and print media), including the local provincial mental health foundation, the single provincial government health care provider Alberta Health Services (AHS). Additionally, Text4Hope was the subject of a specific COVID-19 mental health support media release by the Provincial Chief Medical Officer [33 ]. Ethics approval has been granted by the University of Alberta Health Research Ethics Board (Pro00086163).

A pooled analysis or meta-analysis was not performed because of the heterogeneity of the study populations across geographic locations and the heterogeneity of exposures (ie, PM_2.5 from vehicle traffic and wildfires). The review presented the primary findings in summary of evidence tables for each key question and tabulated the preponderance of evidence that found significant associations (ie, 19 of 24 studies on preterm birth and air pollution found a significant association). The overall number of births were included in the review, and the Table lists the number of births reviewed for each key question, and the mean (SD) births among these studies. This was presented to highlight the large populations being studied that provided credence to the significant findings. The degree of risk was identified for significant associations as a range with median. These data were tabulated and calculated with an Excel spreadsheet (Microsoft).

iLand (the individual-based forest landscape and disturbance model) was developed to simulate the dynamic interactions between climate change, vegetation dynamics, and disturbances (Seidl et al. 2012 (link)). It operates at the grain of individual trees, for which it simulates competition for resources spatially explicit in space and time. Landscape-scale processes such as the dispersal of seeds or the spread of disturbances are simulated explicitly over extents of several tens of thousands of hectares. iLand is a process-based model, in which stand-level gross primary production is simulated based on a light use efficiency approach, and combined with ecological field theory for determining the resource availability for every tree (Seidl et al. 2012 (link)). The effects of environmental constraints on vegetation development are accounted for at daily time steps. Individual tree mortality is calculated based on carbon starvation, and regeneration of new seedlings depends on the local presence of seeds, light, and a favorable abiotic environment. iLand tracks ecosystem carbon stocks and fluxes, and is able to simulate detailed forest management interventions via an agent-based management model (Rammer and Seidl 2015 (link)). The model has previously been parameterized and tested for ecosystems in Central and Northern Europe as well as Western North America, and was successfully applied to simulate decadal to millennial scale forest dynamics for landscapes between 2500 and 25,000 hectares.
iLand is particularly suited to study disturbance interactions as it operates at a fine spatial and temporal grain while being able to simulate disturbance processes spatially explicit at the landscape scale. Wildfire and wind disturbances have been included in the model in previous efforts. Wind damage is modeled at the level of individual trees with wind disturbance events being simulated iteratively, dynamically accounting for changes in forest structure during the course of a storm (Seidl et al. 2014a (link)). Both upwind gap size and local shelter from neighboring trees are considered explicitly, and critical wind speeds for uprooting and stem breakage are distinguished in the model. Tree response to wind is derived from empirically parameterized turning moment coefficients (Hale et al. 2010 ). Besides the dynamically simulated forest structure and composition, major drivers of wind disturbance are wind speed and direction, storm duration, and soil frost (the latter influencing the anchorage of a tree). For details on simulating wind disturbance in iLand as well as a sensitivity analysis and thorough test against independent data we refer to Seidl et al. (2014a (link)).

We developed an ML-based air quality forecast modeling framework that consists of two independent ML models, in order to predict O₃ at Kennewick, WA (Fan et al., 2022 (link)). The first ML model (ML1; Supplementary Figure 1a) consists of a random forest classifier and a multiple linear regression model: the RandomForestClassifier and RFE functions in the Python library scikit-learn are used (Pedregosa et al., 2011 ). The second ML model (ML2; Supplementary Figure 1b) is based on a two-phase random forest regression model: the RandomForestRegressor function in the Python library scikit-learn is used (Pedregosa et al., 2011 ). More details of our ML modeling framework are available in Dataset and Modeling Framework section in Fan et al. (2022 (link)).
In this study, we use the same ML models to predict the O₃ and PM2.5 at various AQS sites in the PNW. To better fit the local conditions, the model is trained at each individual site. Hourly O₃ and PM2.5 predictions are used to compute maximum daily 8-h running average (MDA8) O₃ mixing ratio and 24-h averaged PM2.5 concentrations, as these are the requirements of the National Ambient Air Quality Standards (NAAQS). Due to the different sources of PM2.5 during wildfire and cold seasons in the PNW, the model is trained separately for two seasons at each site. The feature-selection module from the functions listed above are used to select the features at each site to train the models. For ML2, the weighting factors vary at each site, which are computed based on the local input data.
Given ML models can be subject to overfitting and can be sensitive to issues in the training dataset, we account for these issues in our modeling setup. To avoid overfitting, we limit five features in the model training, and use 10-time 10-fold cross-validation to evaluate our model. Our training datasets are air quality observation, which are generally imbalanced: a highly polluted event or an extremely clean event is a rare event. Haixiang et al. (2017 (link)) shows that imbalanced training data may lead a bias toward commonly observed events. To alleviate the imbalance problem, we apply several methods such as turning on the balanced_subsample option in the function of the random forest model and using multiple linear regression and second phase random forest regression in the modeling system.

In the PNW, currently there are 47 AQS sites with O₃ observations, 138 sites with PM2.5 observations. Similar to the ML modeling framework for Kennewick, the training dataset for this multi-site ML models included the previous day's observed O₃ or PM2.5 concentrations, time information (hour, weekday, month represented as factors), and hourly meteorological forecast data from twice-daily ensemble WRF forecasts extracted at each AQS site. The WRF meteorology data was provided by the twice-daily ensemble forecasts with 4 km horizontal resolution, produced by the University of Washington (UW, https://a.atmos.washington.edu/mm5rt/ensembles/).
The UW ensemble system applies multiple physical parameterizations and surface properties to the WRF model simulations, and the ensemble forecasts could improve the forecast skill for some cases (Grimit and Mass, 2002 (link); Mass et al., 2003 (link); Eckel and Mass, 2005 (link)). To utilize the varying settings for the meteorology simulations, we input the multi-member WRF ensemble forecasts for the air quality forecasts in the PNW.
The evaluation of O₃ predictions in this paper covers May to September from 2017 to 2020 and PM2.5 predictions cover two seasons, wildfire season (May to September) and cold season (November to February) from 2017 to 2020. While wildfires can affect both O₃ and PM2.5 concentrations significantly, wood burning from stoves during cold season is a significant source of PM2.5 in populated areas, so we look at only PM2.5 for cold season. To identify the characteristics of each individual site, the models are trained for each monitoring site with archived 4 km WRF forecasts and observations. For the model evaluation, we used the archived operational WRF data, which is a single ensemble WRF member from UW forecasts. The observations and archived WRF data are available at 30 sites for O₃ and more than 100 sites for PM2.5, and there are 12 sites where both O₃ and PM2.5 are measured.

Institutional Review Board approval was obtained at the University of California, Santa Barbara. Recruitment began in November 2020 and continued through April 2021. The initial plan was to recruit participants who originally participated in a longitudinal, multi-site survey study regarding the psychosocial adjustment of young adults’ following devastating hurricanes and wildfires that occurred in 2017–2018. They were recent graduates and had agreed to be contacted again for future related studies. Those residing in California and Florida who self-identified as Latinx or Hispanic were invited to participate in the study via email and phone. Only eight participants were recruited from the longitudinal, multi-site survey study; therefore, the Institutional Review Board (IRB) protocol was modified to recruit other Latinx emerging adults in California and Florida. We continued with data collection from these two regions to be consistent with the existing sample and it provided the opportunity to obtain diversity in participants’ experiences during the pandemic given different state regulations in the pandemic. Although our aim was not to compare experiences between regions, we considered how varied settings, COVID-19 protocols, and context influenced Latinx emerging adults in California and Florida. The study was advertised on social media (Facebook and Instagram) and through word of mouth. Those interested in participating were screened and completed a brief phone briefing prior to being enrolled in the study. In order to meet criteria for the study participants had to identify as Latinx, reside in California and Florida, and be willing to partake in a focus group. In the consent form sent to each participant through a Qualtrics survey, the participant was informed about the study procedures. All participants were compensated for their participation with a $15 Amazon gift card.
A total of 35 Latinx individuals were recruited for this study but four participants who initially expressed interest were unable to be reached. Thus, the final sample consisted of 31 participants. By study design, participants were between the ages of 18–29 years (M = 23.32, SD = 2.73). In addition, 66.7% of participants were bilingual in English and Spanish. Over half of the participants (56.7%) reported that they had a family member that was considered an essential worker and a large percentage (83.3%) reported having family members that they considered to be at greater risk of contracting COVID-19. Over half (63.3%) were living in multigenerational homes at the time of the study. Additional information about the participant characteristics is displayed in Table 1.