Climate change impacts in the Pacific Northwest

Climate change is often perceived as a future threat, yet substantial evidence indicates that anthropogenic forces have already influenced weather and wildfire regimes. For instance, average annual temperatures in the PNW have increased by almost 1.1°C (2°F) since 1900 [6]. Projections suggest further increases of 2.6°C (4.7°F) under a low emissions scenario and up to 5.6°C (10°F) under a high emissions scenario by the 2080s, relative to the period from 1950 to 1999. This escalating warming will exacerbate heat waves, as evidenced by the June 2021 Heat Dome during which temperature records were exceeded by over 5°C (9°F) in many PNW locations [10].

Precipitation predictions under climate change are more complex; forecasts indicate longer intervals between rainfall events, potentially exacerbating drought conditions and increasing wildfire risk across the PNW [6]. Summertime precipitation is expected to decline in the PNW; however, fall, winter, and spring forecasts are more uncertain [11].

Warmer and drier conditions in the PNW are also likely to lengthen wildfire seasons, leading to more frequent and severe wildfires compared to the previous century [7]. In fact, it is estimated that anthropogenic climate change is already responsible for approximately 50% of the burned area in the United States due to increased fuels and higher temperatures [12]. Further, wildfire smoke risk is projected to increase across coastal Oregon, eastern Washington, and southwest Idaho [13]; particulate matter 2.5 (PM2.5), a component of wildfire smoke linked to human health concerns, is also expected to increase by 55% to 190% over the next century in the contiguous United States. [14,15].

Regional impacts of climate change have and will continue to influence visitation patterns in PPAs [16–18]. We model the influence of regionally important weather- and smoke-related variables to assess how recent climate changes have already impacted visitation to PNW PPAs. These insights may help adapt visitor use management to short- and long-term regional climate change.

Evaluating the impacts of climate change on visitation

Extensive research has modeled the impacts of weather and climate on PPA visitation. Much of this work focuses on a single effect such as temperature or wildfire smoke on a specific activity, location, or setting [18–21]. More studies are beginning to assess the influence of multiple climate and weather effects on visitation; among them, temperature, precipitation, wind and extreme weather are the most common [22–24].

Temperature is often identified as a key predictor of visitation; several studies have found maximum temperature to be the most influential predictor of visitation [17,21,25]. Additionally, some research has found that visitation generally increases with temperature until a threshold is reached, typically when mean monthly temperatures exceed 25°C (77°F) [17,19] or when daily maximum temperatures exceed 33°C (91.4°F) [26]. Mean temperature has also been used to study climate effects on peak visitation season length in U.S. national parks [19,27,28].

The effect of precipitation on visitation has been mixed. Yu et al. [29] found that storms and precipitation were powerful predictors of tourism in Alaska and Florida; others found precipitation to be a poor predictor of visitation in Utah national parks [17]. Precipitation can reduce participation in water-based recreation [30] and some evidence suggests that increasing precipitation decreases visitation to U.S. national forests [31,32]. Precipitation also affects wildfire; extended dry periods can increase the severity and extent of wildfire events [33].

Wildfire and smoke also affect visitation, though these impacts are less studied compared to temperature and precipitation [34]. Recent research suggests that wildfire and associated smoke lead to very slight, if any, declines in monthly visitation at federally managed recreation sites in the United States [34–36]. The authors propose that visitors to U.S. federal PPAs exhibit a tolerance for smoke-related air pollution. To date, no work has assessed the impact of wildfire smoke on visitation to other jurisdictions of PPAs, such as state or local parks.

A majority of existing research focuses on the impact of weather and wildfire smoke on visitation to federally managed PPAs, national parks in particular [24]. Far fewer studies explore this relationship in state or local PPAs [37,38]. State and local parks play an important role in the outdoor recreation landscape; in the United States, nearly 75% of Americans live within walking distance to a local park and about one third of time spent recreating outdoors takes place in state parks [39,40]. Additionally, recent studies modeled these dynamics on monthly time scales or longer [19,34–36]. There is a need for more research that explores these relationships in state and/or local PPAs and on shorter temporal scales, as weather and smoke may affect visitors differently on smaller time scales and in other settings [24].

Our research addresses this gap by modeling the effects of weather, air quality, and wildfire smoke on visitation to federal, state, and local PNW PPAs. By understanding how weather and smoke influence visitation patterns on shorter time scales, we can help inform jurisdiction-specific management responses to changing climatic conditions. These findings may help optimize staffing during extreme weather, protect public health during wildfire and smoke, support climate-stressed ecosystems, and guide the development and maintenance of infrastructure to support evolving visitor use.

Monitoring visitor use with mobile device data

Monitoring visitor use across managerial jurisdictions of PPAs is limited by the lack of consistent visitor use data across temporal and spatial scales. While monthly visitation data for U.S. national parks are available through the National Park Service (NPS) Integrated Resource Management Application (IRMA) Visitor Use Statistics Portal, the Forest Service provides estimates only every five years [5,41]. Moreover, no federal agency collects and publicly shares weekly visitation data. This challenge is amplified in state and local PPA as many of these settings do not collect data on visitor use, or if they do, it is often not publicly available. Addressing this gap requires innovative data sources, including remotely collected data from fitness tracking applications, social media data, or mobile devices [42–44].

Mobile device data present a solution for estimating visitor use where traditional data fall short or may not exist, such as in local parks and forests, sites with porous boundaries, or on daily or weekly scales [45,46]. Mobile device data have been used to estimate visitor use in individual parks, national parks, networks of urban-proximate PPAs, and water-based recreation sites across the United States. [46–49]. Past research has documented strong correlations between mobile device data and monthly count data in national parks [50,51]. Similarly, Creany et al. [45] found that mobile device data were well-correlated with bicycle and pedestrian use (ρ values between 0.74 and 0.811, p < 0.01) in a network of locally managed urban-proximate nature reserves. Given the scarcity of weekly data among federal, state, and local PPAs, we opt to use mobile device data due to its widespread geographic availability and fine-scale temporal resolution.

Research objectives

This study aimed to model how short-term weather and air quality, including wildfire smoke, affect visitor use in the PNW. We investigate how mean weekly temperature (minimum, maximum, and maximum squared), precipitation, air quality index (AQI), and particulate matter 2.5 concentration (PM2.5) affected visitation to federal, state, and local PPAs in the PNW during an extended peak visitation season from 2017 to 2021. To date, very few studies examining the impact of climate change or weather on visitation have occurred in Washington, Oregon, or Idaho [24]. We sought to construct a more complete picture of how weather and air quality (including wildfire smoke) influence visitation to different managerial jurisdictions of PPAs; we believe this is necessary to sustain ecological integrity and high-quality recreation opportunities in PPAs within and beyond the PNW in an era of climatic change.

Study area

The PNW states of Idaho, Oregon, and Washington contain diverse ecosystems representing boreal forests, oak woodlands, dry conifer forests, alpine tundra, and high deserts; each of which is host to PPAs that provide outdoor recreation opportunities. Our study focuses on 91 PNW PPAs that provide outdoor recreation opportunities to PNW residents and tourists alike.

We selected study sites from the Protected Areas Database of the United States (PAD-US), the official national dataset for U.S. terrestrial and marine protected areas managed by the U.S. Geological Survey [52]. The PAD-US dataset encompasses many public land settings including wilderness areas, national parks, recreation centers, and cemeteries. We focused on settings that offer outdoor recreation opportunities; we excluded public land units not primarily focused on outdoor recreation (e.g., community recreation centers, sports fields, cemeteries, community centers, pools, and sports complexes), as well as wildlife areas due to seasonal closures. Finally, we limited our selection to sites larger than 10 acres to obtain reliable visitation estimates from the mobile device data, resulting in 2,643 potential study sites.

We used stratified random sampling to select 91 PPAs ensuring equal representation across managerial jurisdictions, states, and ecosystems (Table 1, Fig 1) (also see S1 Table and S1-S3 Figs) [53]. We included three managerial jurisdictions of PPAs: federal (including national parks, national monuments, U.S. Forest Service units, Bureau of Land Management units, national scenic areas), state (state parks, state recreation areas), and local (city parks, county parks, regional parks). Geographic context is also thought to moderate the influence of weather and smoke on visitation; we considered it through ecosystem type, elevation, and total PPA area [23,35]. We evaluated ecosystem type through ecoregions, or geographic areas containing distinct assemblages of natural communities, which are often used to model ecological conditions at the landscape scale [54,55]. Past work has demonstrated that the influence of weather on visitor behavior is ecoregion-dependent [56]. We included elevation as it moderates the influence of climate change on visitation by providing multiple conditions for visitor experiences in parks [17]. We thought the same may be true for PPA area and included this as a covariate.

Download:

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

Fig 1. Map of 91 federal, state, and local parks and protected area (PPA) study sites overlaid on Level 2 ecoregion.

Point locations are used in lieu of polygons for visual effect. Base layers source: Esri. “Light Gray Canvas” [vector]. Scale Not Given. “Light Gray Canvas Basemap.” Feb 19, 2025. https://www.arcgis.com/home/item.html?id=979c6cc89af9449cbeb5342a439c6a76. (February 24, 2025).

https://doi.org/10.1371/journal.pclm.0000537.g001

Download:

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

Table 1. Counts of parks and protected areas included in this study by state and managerial jurisdiction.

https://doi.org/10.1371/journal.pclm.0000537.t001

Study period

Our study focused on the extended peak visitation season from May 15 to October 15 across five consecutive years, from 2017 to 2021 (S4 Fig). This period captures the traditional summertime peak visitation season, defined as June through September for many northern latitude PPAs, as well as the shoulder season that includes part of spring and fall [5]. We used mobile device data to estimate visitor use in our study areas. These data first became available in 2016; however, our study began in 2017 to mitigate potentially confounding effects from two events in 2016: the National Park Service Centennial and the Find Your Park campaign, both of which significantly increased visitation [57]. We used weekly data as this is the smallest temporal scale for estimating vehicle volume (our metric of visitation) without compromising data quality [58].

Using mobile device data to estimate visitor use

We modeled the relationship between weather, air quality, wildfire smoke and PPA visitation using mobile device data. We used data from StreetLight Data Inc., a transportation and urban-planning service that provides aggregated, preprocessed, and de-identified human mobility data derived from mobile devices equipped with Global Positioning Systems (GPS) Location Based Services (LBS) [58]. We used a data product that aggregates mobile device signals in vehicles within PPA polygons and represents the average daily vehicle volume over a given time period; the raw data represent the average daily vehicular visitation over a given week. We multiplied the average daily vehicle volume by seven to estimate the total weekly vehicle volume. The vehicle volume metric represents visitors who travel to or from a destination using a vehicle and does not include pedestrians. As a result, visitation to local PPAs may be systematically underestimated due to the higher rates of non-vehicular travel to local PPAs.

We used weekly vehicle volume to estimate total weekly visitation in each PPA; this measurement is reported as a “count” of visitors. Counts are often used to model visitation in recreation management research when comparing visitation among similar sites [17,21]. However, when modeling visitation in sites with different visitation levels, it is common to convert counts to proportions to ensure that differences in visitation levels do not impact the results [19,37]. This conversion is particularly important for our data given that many of our federal sites exhibit substantially higher visitation than many of the state and local sites. Thus, the response variable used here was calculated as the proportion of weekly visitation to the five-year average (mean) annual visitation for each site (S5 Fig).

Validation must be performed when using alternative sources of visitor use data, including those from mobile devices [42]. We validated the mobile device data, aggregated by month, with monthly visitation data from 8 NPS sites, because these are the only sites in our sample with publicly available visitation data [5]. Spearman correlations for each of the 8 parks indicated that the mobile device data are well-correlated with official NPS monthly visitation counts (n = 160, ρ = 0.927) (S6, S7 and S8 Figs). Some discrepancies exist between the official NPS counts and the mobile device data, such as in Olympic National Park during 2020, when official park closures due to COVID19 may have displaced visitors to unofficial park entrances. We were unable to validate state and local visitation estimates due to a lack of publicly available data in these settings. At least one study has found that mobile device are an appropriate measure of visitor use compared to trail counter data in four southern California urban-proximate local parks [45].

Predictor variables

We obtained data on weather, air quality, wildfire smoke, and other covariates from open-source data repositories and software packages (Table 2). We downloaded temperature and precipitation data from Daymetr through the Daymetr R package, which provides daily weather and climate data on 1-km grids [59,60]. We extracted daily temperature and precipitation data from Daymetr using the UTM coordinates for each study site; these coordinates were identified based on the primary parking lot associated with the site on Google maps. We calculated the weekly means for each parameter using the daily data within each week of our study period (Tables 3-5). Finally, we computed the quadratic form of maximum temperature in R to model the temperature threshold effect identified in previous work [17,19,26].

Download:

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

Table 2. Descriptions of variables used in the generalized estimation equations models for federal, state, and local park and protected area (PPA) settings. Sample sizes relate to the weekly measurement for each study site over the time period and are as follows: federal (n = 3,258), state (n = 3,331) and local (n = 3,169). U.S. EPA = U.S. Environmental Protection Agency, U.S.G.S. = U.S. Geological Survey; QGIS = Q Geographic Information System.

https://doi.org/10.1371/journal.pclm.0000537.t002

Download:

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

Table 3. Statistical summary for weekly weather and air quality variables in federal parks and protected areas (PPA) over the study period (2017 - 2021) in the Pacific Northwest, United States; (n = 3,258 weekly measurements). AQI = air quality index; PM 2.5 = particulate matter with diameter <2.5 μm.

https://doi.org/10.1371/journal.pclm.0000537.t003

We obtained air quality and smoke data from the U.S. Environmental Protection Agency (EPA) AirData service [61]. This data portal provides public data for various air quality parameters; we chose Air Quality Index (AQI) as a general measure of air quality, and particulate matter 2.5 (PM2.5) due to its ability to proxy wildfire smoke events [35,36]. AQI is a general measure of air pollution that often ranges from 0 to 500 and represents ground-level ozone, particulate matter, carbon monoxide, sulfur dioxide, and nitrogen dioxide [62]. In practical terms, air quality worsens as AQI increases. We also consider PM2.5 (particulate matter with diameter <2.5 μm), which is another measure of air quality and specifically a component of wildfire smoke that has been linked to human health concerns [14]. It is estimated that wildfires in the American West accounted for 50% of the PM2.5 pollution in 2020 compared to less than 20% in 2010 [63].

Air quality data from the AirData service featured inconsistent measurements every two, three, or four days. We calculated the average (mean) air quality parameter for each week using these measurements. Further, air quality data are only measured at EPA sensor locations, requiring interpolation to estimate the parameters at our study sites. We used Kriging to interpolate the air quality data, a common geostatistical method used to estimate continuous data, like air pollution, over broad geographic areas [36,64]. Finally, we extracted the interpolated air quality values to the UTM coordinates (mentioned above) for each week in the study period in QGIS version 3.16 [65].

We included several additional variables to control for confounding effects. We used a binary COVID-19 variable to distinguish the COVID-19 season (2020) from non-COVID-19 seasons (2017–2019, 2021). We considered 2020 as the only COVID-19 season given the widespread PPA closures that occurred during this time [49]. We also included a binary variable to distinguish weeks containing U.S. federal holidays as they often have higher visitation [31]. Finally, we captured the geographic setting through elevation (mean and range within the PPA polygon), level 2 ecoregion, and PPA area.

Modeling approach

We used generalized estimating equations (GEEs) to model the impact of the climate-affected variables on the proportion of weekly visitation at federal, state, and local PPAs. GEEs are a regression method for longitudinal or clustered data where observations within clusters (such as data collected from the same PPA) are not independent [66]. GEEs take this correlation structure into account and provide unbiased estimates of population-averaged coefficients [67]. Further, GEEs can provide less biased results compared to mixed effects models, where incorrect assumptions about the data distribution can lead to misleading estimates [68]. However, GEEs should only be used when the number of clusters exceeds 20; ours are n = 22 and n = 30 for temporal and spatial clusters, respectively [69].

We constructed three separate GEE models to examine the influence of weather and smoke on the proportion of weekly visitation at federal, state, and local PPAs using the ‘geepack’ package in R, version 4.3.2 [59,70]. Sample sizes were as follows: federal (n = 3,258), state (n = 3,331) and local (n = 3,169). Given that we used proportion data as the response variable, we specified a binomial distribution and the associated logit link function [71]. We assessed model fit using the Quasi-likelihood under Independence Model Criterion (QIC) to determine the best of four correlation structures for each model: independence, exchangeable, autoregressive, and unstructured and selected the model with the lowest QIC for the final analysis [72] (Table S2).

The standard estimates produced by the GEE model represent the log-transformed population-averaged coefficients for each managerial jurisdiction. We exponentiated the standard estimates to obtain the odds ratio, which indicates the magnitude and direction of the predictors’ effect on the response variable (proportional weekly visitation) [73]. To improve interpretability of the odds ratio, we computed an additional result called visitation effect, which represents the percent change in response variable when the predictor increases by one unit (for continuous variables) or when the condition is true (for categorical variables); it was calculated as (-1*(1- Odds Ratio)) *100.

Weather and air quality results

We found that a one-unit increase in mean weekly AQI, a general measurement of air quality that includes many air pollutants, corresponded with increased weekly visitation at federal PPAs by 0.4% (p < 0.01) but was not a significant predictor of weekly visitation at state or local PPAs (Fig 2, Tables 6-8). A one-unit increase (µg/m³) in mean weekly PM2.5, a component of wildfire smoke, corresponded with decreased weekly visitation at federal and local PPAs by 0.9% and 0.6%, respectively (p < 0.001). A one-unit increase (°C) in mean weekly minimum temperature corresponded with increased weekly visitation in all three managerial jurisdictions (federal: 1.8%, p < 0.01; state: 2.6%, p < 0.001; local: 3.2%, p < 0.001). A one-unit increase (°C) in mean weekly maximum temperature corresponded with increased weekly visitation at federal PPAs by 2.7% (p < 0.05) and at local PPAs by 6.3% (p < 0.001). A one-unit increase in mean weekly maximum temperature squared (to indicate a maximum temperature threshold effect) corresponded with decreased weekly visitation at local PPAs by 0.1% (p < 0.001). Finally, a one-unit increase (mm) in mean weekly precipitation corresponded with decreased weekly visitation in all three managerial jurisdictions (p < 0.001); at federal PPAs by 5.4%, at state PPAs by 3.7%, and at local PPAs by 2.3%.

Download:

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

Table 7. Generalized estimating equation (GEE) model results on the influence of weekly weather conditions and air quality on weekly visitation to state parks and protected areas (PPA), including state parks and state recreation areas included in this study. The odds ratio, calculated as exp(Standard Estimate), indicates the magnitude and direction of the predictors’ effect on the response variable (proportional weekly visitation). The visitation effect, calculated as -1*(1-Odds Ratio)*100 indicates the percent change in the response variable when the predictor increases by one unit (for continuous variables) or when the condition is true (for categorical variables). Significant results (p<0.05) in bold. AQI = air quality index; PM 2.5 = particulate matter with diameter <2.5 μm.

https://doi.org/10.1371/journal.pclm.0000537.t008

Download:

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

Table 8. Generalized estimating equation (GEE) model results on the influence of weekly weather conditions and air quality on weekly visitation to local parks and protected areas (PPA), including city parks, county parks, and regional parks included in this study. The odds ratio, calculated as exp(Standard Estimate), indicates the magnitude and direction of the predictors’ effect on the response variable (proportional weekly visitation). The visitation effect, calculated as -1*(1-Odds Ratio)*100 indicates the percent change in the response variable when the predictor increases by one unit (for continuous variables) or when the condition is true (for categorical variables). Significant results (p<0.05) in bold. AQI = air quality index; PM 2.5 = particulate matter with diameter <2.5 μm.

https://doi.org/10.1371/journal.pclm.0000537.t007

Covariate results

A one-unit (m) increase in elevation range corresponded with increased weekly visitation at federal PPAs by 0.001% (p < 0.05) (Table 6). The Cold Deserts ecoregion exhibited 11.3% lower weekly visitation compared to the Western Cordillera at federal PPAs (p <0.01) (Table 6); the Marine West Coast forest exhibited 9.9% lower weekly visitation compared to the Western Cordillera at state PPAs (p < 0.01) (Table 7). A one-unit (sq. m) increase in PPA area corresponded with increased visitation at local settings, but the effect was smaller than 0.0001% and was rounded to zero in our results (p < 0.05) (Table 8). During weeks that contained a federal holiday, visitation increased at state PPAs by 9.3% (p < 0.001) and local PPAs by 10.8% (p < 0.001) (Tables 7 and 8). COVID-19 (marked as the year 2020) corresponded with increased visitation at state PPAs by 6.9% (p < 0.001); it had no effect in federal or local PPAs (Tables 6-8).

Effect of air quality and wildfire smoke on visitation

The effect of air quality and wildfire smoke on visitation varied based on managerial jurisdiction. We observed a 0.4% increase in weekly visitation to federal PPAs with a one-unit increase in AQI; this was not evident in state or local PPAs. Likely, visitors do not prefer higher AQI in federal PPAs. Instead, AQI tolerance may increase in settings that require more planning and travel, such as national parks [74]. This finding corroborates past research suggesting that visitation to federal PPAs is not substantially affected by smoke [34–36].

The AQI result is different from our PM2.5 findings, which suggest that wildfire smoke may lead to visitation declines. Our results indicate that a one-unit increase in PM2.5 concentrations corresponded with reduced visitation at federal and local PPAs by 0.9% and 0.6%, respectively. This may indicate that general air quality does not affect weekly visitation, but degraded air quality from wildfire smoke in particular may decrease visitation at federal and local PPAs. It is imperative to consider the units of measurement when interpreting these results, given that PM2.5 concentrations can increase 100-fold during wildfire events, potentially instigating substantial visitation declines.

This finding contradicts work by Clark et al., [34] who did not observe significant changes in visitation due to wildfire smoke across 32 national parks between 1980 and 2019. This discrepancy may be caused by the differing spatial and temporal scales in our respective studies; Clark et al. [34] examined monthly visitation and smoke across the American West; we focused on weekly scales in the PNW. Additionally, our study included data from 2020 and 2021, two historic wildfire years in the PNW, which they did not consider [33,75]. Another difference lies in the pollutants analyzed; we measured PM2.5, which includes both wildfire smoke and anthropogenic emissions, whereas Clark et al. [34] examined black carbon.

More research on the impact of air quality and wildfire smoke on visitation would be useful; future studies could compare different smoke-related metrics to determine which are most indicative of visitation trends. PM2.5 is likely a more accurate predictor of smoke compared to AQI; this is evidenced by the growing body of research using PM2.5 to model smoke-related air pollution [35,36] and the fact that 50% of the PM2.5 pollution in the American West was attributable to wildfires in 2020 [63]. Future research could clarify if and how smoke tolerance shifts as wildfires intensify with climate change.

Additionally, future research could explore thresholds of tolerance for AQI and PM2.5; for example, people may not alter their visitation behavior until these are above certain thresholds. More research on how people interpret these data would also be useful. AQI is presented to people on websites and applications as both a number and category (e.g., 0–50 is “good,” 51–100 is “moderate,” 101–150 is “unhealthy for sensitive groups,” etc.). It is unknown if people focus more on the number or the category. If people focus more on the category, then people may perceive AQI changes differently depending on whether or not they cross categories (for example, people could perceive a change from 90 to 110 to be much more impactful than a change from 110 to 130, even though they both represent a 20 unit change, because one would alter the category of the AQI and the other would not). Future research may consider binning numeric AQI and PM2.5 data into categories given by the EPA to see if there are thresholds that affect visitation.

Effect of temperature and precipitation on visitation

We found that visitation increased by 1.8% to 3.2% when mean weekly minimum temperature increased by 1°C in all three managerial jurisdictions. However, the effect of minimum temperature was smaller than that of maximum temperature in federal and local PPAs; where a 1°C increase in maximum temperature increased visitation by 2.7% and 6.7%, respectively. Maximum temperature was not significant in state PPAs, possibly due to the prevalence of overnight camping in these settings, which is more sensitive to minimum temperature [76]. Our findings align with previous research identifying maximum temperature as a key factor influencing PPA visitation [17,21,25].

We did not find a maximum temperature threshold effect in federal or state PPAs despite being observed in other settings, including Utah national parks and Canadian provincial parks [17,26]. This may be due to the relatively mild climate of the PNW which attracts visitors from warmer locations; these visitors may have a greater tolerance for hot weather. Visitors to federal and state PPAs may exhibit a higher tolerance for heat or a lower ability to quickly change leisure plans compared to visitors to local PPAs. Local PPAs exhibited a slight maximum temperature threshold effect, indicating a hot weather sensitivity among these visitors. This may be due to the type of visitors that frequent local PPAs; these individuals may have more flexibility to change leisure plans during hot weather compared to state or federal PPA visitors, who often travel farther to reach their destination and plan trips farther in advance [74,77]. Likewise, advance reservations are required in many state and federal PPAs and may discourage cancellations despite unfavorable conditions. Future research could clarify whether a maximum temperature threshold emerges in the wake of increasing temperatures in PNW federal and state PPAs.

Given these results, we anticipate that overall visitation may increase at all three PNW PPA settings as average temperatures rise with projected climate change. This trend may be especially pronounced in local and federal PPAs, where both increasing minimum and maximum temperatures resulted in heightened visitation; however, our results also indicate that visitation may decline in local PPAs during periods of extreme heat. Importantly, increased minimum temperatures in the shoulder season months (May, September, and October) may increase visitation during important phenological events like animal breeding, rearing, plant germination, and seed dispersal, potentially exacerbating risks to species already struggling to adapt to climate change [9].

Past work shows inconclusive results on the impact of precipitation on PPA visitation [17,29]. However, we found it to be a strong predictor of decreased visitation across all three types of PPAs when assessed at a weekly scale. This aligns with past findings suggesting that studies at smaller temporal scales may detect variables aside from temperature as important predictors of visitation [17]. Further, our results suggest that precipitation may lead to substantial visitation declines, given that a 1-millimeter increase in precipitation reduced weekly visitation by 2.3% to 5.4%, and in the PNW, 10 millimeters of rain in a single event is rather common. These findings indicate that peak-season visitation to all types of PPAs may increase as summers become drier [6,11]. It remains unclear how changing precipitation patterns will impact shoulder season visitation in the PNW due to variability in fall and spring precipitation predictions [11].

Limitations

These results illustrate how mobile device data can provide valuable estimates of visitor use in situations where traditional visitation data do not otherwise exist or are not publicly available [45,78]. However, some limitations of mobile device data require careful consideration. For instance, the data product we used aggregates mobile device signals within vehicles. It likely underestimates visitation in areas where walking or cycling are common modes of transport, as in local PPAs. However, visitor use would likely be systematically underestimated across all local PPAs in our study, making comparisons among them viable. Representation is another common limitation of mobile device data as not everyone owns a GPS-enabled mobile device, however, the Pew Research Center reports that nearly 85% of Americans own and use a GPS enabled smartphone [79].

A strength of mobile device data is the ability to examine visitor use estimates across large spatial extents and different PPA managerial jurisdictions. However, such conditions make validation of mobile device data more challenging In our study, due to data availability, we were only able to use NPS visitor use estimation data for validation. When possible, mobile device data should be calibrated or validated with on-site counts to determine site-specific accuracy [48]. Per the perspectives of Monz et al. [48], despite these limitations in mobile device data, we view these data as representing an estimate of visitor use not an exact count - and that the limitations of mobile device data discussed are systematic across all sites included in our study. Future studies could examine alternative approaches for calibrating mobile device data across managerial jurisdictions when on-site visitor use estimations are not available for all locations.

Finally, our findings are specific to shoulder- and peak-season visitation in PNW PPAs. Future research could examine these relationships during the off-peak season to understand how temperature and precipitation impact visitation year-round.

Management implications

Our results provide further evidence that PNW PPA visitation will likely shift under future climate scenarios; local PPAs may experience the greatest increases in overall visitation as temperatures warm. Unlike in local PPAs, we observed no maximum temperature threshold effect in federal and state PPAs, suggesting that visitors to these settings may exhibit a higher tolerance for heat. Managers of federal and state PPAs could consider improving infrastructure, services, and resources to accommodate visitors during extreme heat.

All three types of PPAs exhibit substantial reductions in visitation with increasing precipitation, implying that overall visitation could increase as summers become drier [6,11]. However, heavy rainfall events during the shoulder season may become more severe; potentially threatening PPA infrastructure including roads, trails, and shelters [8,11]. Infrastructure damage associated with heavy rainfall has already prompted visitation declines in PPAs of the American West [80]; managers could consider enhancing the resilience of recreation settings to handle extreme weather and maintain long-term accessibility and safety [81,82].

The anticipated increase in visitation may also exacerbate ecological disturbance including vegetation damage, soil compaction, and wildlife disruption [9,83]. PPA managers may need to take additional measures to protect ecological integrity by enhancing the durability of infrastructure in vulnerable areas, reducing visitor impacts to sensitive areas and species, and using appropriate vegetation to increase the resilience of recreation settings to climate-related stressors [81]. The operating budgets of PPAs may need to expand to accommodate the increased visitation and associated impacts from extreme weather and wildfire events [37].

Finally, our results contribute to a growing body of knowledge indicating that PPA visitation does not respond linearly or predictably to wildfire smoke and poor air quality. Contrary to recent research, our findings indicate that significant reductions in visitation to federal and local PPAs may occur during smoky periods [34–36]. PPA managers may consider strategies for enhancing visitor wellbeing during periods of smoke, including site-wide closures or encouraging the use of protective facial coverings.

As shifting weather and wildfire regimes continue to impact visitor use in PPAs, both visitor expectations and management practices must adapt. Visitors may need to alter the timing or location of visits due to extreme weather or suboptimal conditions [16]. They may also need to alter their expectations of PPA settings, the availability of activities, and recalibrate expectations of crowding and congestion in highly visited PPAs [81]. PPA managers may consider preparing for shifts in overall visitation numbers, especially significant increases during shoulder seasons, which may be most pronounced in local PPAs. To sustainably manage these changes, it may be necessary to expand operating budgets to support increased demands on PPA resources and infrastructure. While difficult, maintaining the dual mandate of conservation in a climate-changed world is a worthy challenge of PPA managers in the PNW and elsewhere.