key: cord-0280144-j31rtvin authors: Clark, Matt; Killion, Alexander; Williamson, Matthew A.; Hillis, Vicken title: Breakpoint models shown no evidence of thresholds in recreational response to increasing wildfire smoke in the American West date: 2022-04-21 journal: bioRxiv DOI: 10.1101/2022.04.21.489032 sha: f112ebb6f76a86f0ae6a61e9a003b4bff90bb294 doc_id: 280144 cord_uid: j31rtvin Ambient wildfire smoke in the American West has worsened considerably in recent decades, while the number of individuals recreating outdoors has simultaneously surged. Wildfire smoke poses a serious risk to human health, especially during long periods of exposure and during exercise. Here we aggregate data on black carbon, a major component of wildfire smoke, and recreational visitation in 32 U.S. national parks from 1980 - 2019 to examine how visitors respond to wildfire smoke. We hypothesize that visitor response may exhibit a threshold effect where ambient smoke reduces visitation after a critical level, but not before. We develop a series of breakpoint models to test this hypothesis. Overall, these models show little to no effect of ambient smoke on visitation to the 32 parks tested, even when allowing for critical thresholds at the extreme upper ranges of the smoke data. This suggests that wildfire smoke does not significantly alter behavior of park attendance. This finding has implications for the management of recreation areas, public health, and climate change adaptation broadly. In some regions of the American West, wildfires now account for 50% of all 20 atmospheric PM 2.5 , compared to less than 20% in 2010 [19] . As any level of ambient 21 PM 2.5 increases community morbidity, public health agencies increasingly call for 22 individuals to limit outdoor recreational behavior when wildfire smoke is present 23 [20] [21] [22] . 24 While calls to limit outdoor recreation during wildfire smoke events have become 25 ubiquitous in the summertime American West, a survey of federal land managers 26 indicated that they feel they have a shortage of information describing how air quality 27 actually affects the recreational behavior of their visitors [23] . Reviews of scientific 28 literature come to a similar conclusion, pointing to a dearth of research describing how 29 individuals respond to low air quality (although see [24] ) and how climate change is 30 affecting the recreation landscape [25, 26] . 31 This article uses estimates of black carbon and visitation data collected from 32 national parks in the American West from 1980-2019 to answer whether individuals 33 alter their recreation behavior in response to ambient wildfire smoke. We hypothesize 34 that the effect of ambient smoke on national park visitation may be nonlinear, only 35 showing a measurable impact in the upper ranges of the observed smoke values. 36 Threshold effects such as this are common in ecology and human behavioral sciences 37 [27] [28] [29] . Toms and Lesperance (2003) show that breakpoint models are an effective tool 38 to account for these thresholds and make inferences about the effect of a predictor 39 before and after a particular threshold in natural systems [30] . Here we develop a series 40 of three hierarchical breakpoint models to determine if a threshold effect is present in 41 national park visitation response to ambient smoke and if so, what the effect of smoke is 42 on visitation post-threshold. 43 Materials and methods 44 Data collection 45 Visitation data 46 We obtained monthly visitation data through the National Park Service (NPS) Visitor 47 Use Statistics Portal. These data are generally collected via car counters, permit 48 information, and concessionaire reporting. Methods of collection vary from park to park. 49 Unit specific information can be found on the National Park Service Visitor Use 50 Statistics Portal. Although collection methods are inconsistent across the sample, all 51 data was taken as reported. We retrieved monthly data for all national park units in the 52 Intermountain and Pacific West regions for years 1980 through 2019. 53 While wildfires have seen a dramatic increase across the United States over the last 54 four decades, changes in wildfire smoke have been most concentrated in the westernmost 55 regions of the country [19] . In the contiguous United States, the western regions are 56 the only areas with a significant smoke burden where the majority of the smoke they see 57 originates from wildfires within their own borders [31] . Therefore, as to not confound 58 our analysis with unpredictable smoke events which originate from distant wildfires, we 59 limit our sample to the national parks in the American West (i.e. Intermountain and 60 Pacific West NPS regions). The 32 national parks included in the two regions in our sample have different high 62 seasons for visitation. Wildfire smoke occurs year-round, but primarily in the Summer 63 months. In some seasons, for some parks, there is a natural visitation drop driven by 64 temperature alone that coincides with highest levels of wildfire smoke. To ensure that 65 we estimate visitation changes driven by smoke rather than seasonality, we subset our 66 data to only the three month high season associated with each park. For example, only 67 Summer (June, July, Aug.) data were used for Yellowstone National Park and only 68 Winter data (Dec., Jan., Feb.) were used for Death Valley National Park. Smoke data 70 To detect times of high concentrations of black carbon associated with smoke, we used 71 data from the second Modern-Era Retrospective analysis for Research and Applications 72 [32] . MERRA-2 is a NASA atmospheric reanalysis that begins in 1980 and replaces the 73 original MERRA reanalysis using an upgraded version of the Goddard Earth Observing 74 System Model, Version 5 data assimilation system [33] . MERRA-2 provides mean 75 monthly measurements of black carbon starting in 1980 at a spatial resolution of 0.625°76 x 0.5°. We used the monthly black carbon column mass density measurements (kg/m −2 ) 77 and calculated the mean of those pixels that intersected national park boundaries to 78 represent wildfire smoke at the park level; Fig 1 [ 34] . We also correlated the column 79 black carbon values to the amount of black carbon only at the surface layer (i.e., the 80 measurements closest to the ground; S1). Analyses 82 Visitation to US national parks has increased sharply in recent years [35] . We first [36] . In this particular instance we 90 chose k = 1 to reduce issues of collinearity and identifiability among autoregressive 91 predictors [37] . 92 We formulated this baseline autoregressive model using a Bayesian hierarchical 93 framework. This approach accounts for between-park variation in visitation trends while 94 acknowledging the interconnectedness of these trends through partial pooling [38] . For 95 each park (j) we estimate a parameter value (ν) representing the trend in visitation 96 from one year to the next, in order to predict the visitation for each month in our data 97 set (i). Finally, we estimate both global (α0) and park specific (β0 j ) intercepts to yield 98 equation 1 below. We modeled these data using a negative binomial distribution to allow 99 us to estimate an overdispersion parameter (ϕ) rather than assuming the dispersion is 100 equal to the mean. We fit this model using 7 Markov chains run for 5,000 iterations. This baseline autoregressive model showed adequate diagnostic statistics, exhibiting 102 ample mixing of Markov chains,R values equal to 1 for all parameter estimates, and the 103 absence of divergent transitions after warm up. We examined the within-sample 104 predictive capacity of this autoregressive-only model to be sure we sufficiently 105 accounted for baseline changes in visitation before building on this model to test for the 106 effects of smoke on visitation. This baseline model allowed us to account for 107 approximately 37% of all variation in the data, with the bulk of the error occurring 108 when actual visitation was very high or very low (Fig 2) . We estimated the effect of smoke on visitation before (β1) and after (β2) each 123 breakpoint (BP). Just as with the autoregressive term (ν), we estimated unique 124 parameter values for each park (j) in a hierarchical framework. The complete model 125 used for each of the three breakpoint values is then as seen in Eq (2). Each model was 126 run using 7 Markov chains for 5,000 iterations. Following the recommendation given by Gelman et al. (2008) for producing stable, 128 conservative estimates, we specified long-tailed regularizing priors for the autoregressive 129 term, the global intercept value, and the overdispersion parameter [40] . We used 130 normally distributed regularizing priors for estimating the standard deviation parameter 131 (σ) and the pre and post-breakpoint smoke effects to allow for more efficient sampling 132 while generating within-sample predictions at high smoke values. As the goal of this paper is inference, we test just the suite of breakpoint models 134 described above, which were developed a priori to evaluate our hypothesis that smoke 135 would exhibit a non-linear threshold effect on park visitation [41] . We do not test our 136 models against other candidate models using information criterion metrics or cross 137 validation, as the objective of such tests is prediction rather than inference and we 138 would not expect our models to be predictively valid strictly speaking [41, 42] . All three breakpoint models showed adequate model diagnostics, including well-mixed 141 Markov chains,R values of 1 for all estimated parameters, and the absence of divergent 142 transitions after warm up [43] . We then take the coefficient estimates produced from 143 our models as reliable for inference regarding our a priori hypothesis. While some between-park variation exists, our overall study shows no evidence for have opposite signs (Fig 3) . The post-breakpoint, slope 2, parameter estimates 150 exclusively overlap zero at a 90% credibility interval (Fig 4) . The uncertainty in these 151 estimates increases dramatically as we use more extreme breakpoint values and the 152 amount of data post-breakpoint decreases. We note however, that even parks with Visually examining the marginal effect of ambient smoke on park recreational visits 157 provides a similar intuition as above (Fig 5) . We see little difference in the effect of 158 smoke before and after the hypothesized thresholds. In addition, we descriptively show 159 high levels of visitation even under very high levels of ambient smoke. In Redwood response in all 32 parks in our sample are considerably more narrow than the estimates 171 produced via the breakpoint models (Fig. 6) . Still, the 90% credibility intervals national parks in the American West is not driving any measurable change in visitation 177 even at sustained and dangerous levels. We thus conclude that visitors are not altering 178 recreation behavior in response to smoke and therefore no overall effect or critical 179 threshold for adaptation is detectable in these data. We did not detect visitor adaptation to increasing wildfire smoke in national parks in 182 the American West. This result is troubling both specifically for visitor health in U.S. 183 national parks and for climate change adaptation broadly. As discussed above, wildfire 184 smoke significantly increases community morbidity even under acute exposure. In 185 concert with showing no overall trends in behavioral adaptation, these data provide 186 specific instances of historic smoke events where visitation did not deviate from normal 187 (Fig 5) . The highly variable nature of wildfires, their relationship to climate change, and the 189 great distances that smoke can travel during and after wildfire events make it difficult 190 for individuals to plan around smoke events [31, 44, 45] . In conjunction, U.S. national 191 parks draw a great number of non-local visitors, many of whom are coming from other 192 states or countries and are likely visiting for the first time [46] . We speculate that these 193 visitors are less likely to change their plans due to wildfire smoke than individuals 194 recreating locally or repeatedly in one location. We propose that these unique features 195 of national parks and wildfire smoke make visitors particularly unable to adapt to 196 changing climatic conditions. Based on our findings here, we suggest that a regional or national level policy 198 limiting visitation during dangerous smoke events may be necessary to protect would-be 199 visitors to U.S. national parks [47] . Presently, there is considerable variation in the way 200 states react and plan for climate change and associated hazards [48] . Considering our findings more generally, humans must adapt quickly to the new 202 realities of our increasingly variable climate if we are to continue to thrive over the 203 coming decades [49] . Climate change is already dramatically altering the 204 social-ecological landscapes in which humans have learned to operate [50] . This 205 research contributes to a broader body of work showing that despite increasing 206 awareness, as a society we still fail to respond adequately [51] . Even when adaptive 207 strategies and their benefits are known, we do not translate strategies into action [52] . 208 While research has shown that extreme events tend to spur climate adaptation on the 209 part of governments, individual response is much less consistent [53] [54] [55] . Therefore, 210 while we speculate that the system of recreational visitation to U.S. national parks may 211 be particularly problematic, we suppose that the trends observed in this study may be 212 characteristic of individual response to many changing environmental conditions. Limitations and future research 214 A key limitation of this study is the lack of fine scale visitation data to US national 215 parks. Daily visitation data would allow us to identify a more complete picture of if and 216 how visitors respond to smoke events day to day. Current data at the monthly scale 217 allows us to investigate the big picture trends in visitor behavior, but future research 218 would benefit greatly from daily count data. It is possible that individuals postpone 219 visits to national parks by days or weeks when smoke is present, behavior which our 220 current study would be unable to capture. It is also possible that this trend would not be observed for recreation on all public 222 lands or for all climate change-induced hazards. Other paths for future research may The results presented here indicate that individuals are not modifying their behavior to 227 adapt to worsening wildfire smoke events. It is unclear if this finding is representative of 228 individual climate adaptation generally or is unique to this system. Regardless of the 229 generality of our findings, this study has specific implications for management of U.S. (kg/m −2 ). We use just surface level density measurements in our analyses. The surface 238 and column-wide densities are correlated at a value of 0.85. Contingent Pacific-Atlantic Ocean influence on multicentury wildfire synchrony over western North America Increasing western US forest wildfire activity: sensitivity to changes in the timing of spring Impact of anthropogenic climate change on wildfire across western US forests Ensemble projections of wildfire activity and carbonaceous aerosol concentrations over the western United States in the mid-21st century Human-caused climate change is now a key driver of forest fire activity in the western United States A new climate regime in northeast pacific ecosystems Climate change presents increased potential for very large fires in the contiguous United States Warming weakens the night-time barrier to global fire Projected Effects of Climate and Development on California Wildfire Emissions through 2100 Validation analyses of an operational fire monitoring product: The Hazard Mapping System The Mortality and Medical Costs of Air Pollution: Evidence from Changes in Wind Direction Estimated Global Mortality Attributable to Smoke from Landscape Fires Long-and Short-Term Exposure to PM2.5 and Mortality Excess of COVID-19 cases and deaths due to fine particulate matter exposure during the 2020 wildfires in the United States Black carbon lofts wildfire smoke high into the stratosphere to form a persistent plume Recommendations for reporting "black carbon" measurements The changing risk and burden of wildfire in the United States Wildland fire smoke and human health Critical Review of Health Impacts of Wildfire Smoke Exposure Summary and analysis of approaches linking visual range, PM2.5 concentrations, and air quality health impact indices for wildfires Air quality and the visitor experience in parks and protected areas Air pollution and visitation at US national parks Climate Change Adaptation Strategies and Approaches for Outdoor Recreation Air Quality and Visitor Behavior Ecological thresholds: a survey The concept and utility of 'ecological thresholds' in biodiversity conservation Chains of cooperation: Threshold effects in collective action Piecewise regression: a tool for identifying ecological thresholds Connecting smoke plumes to sources using Hazard Mapping System (HMS) smoke and fire location data over The Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2) NASA's Modern-Era Retrospective Analysis for Research and Applications tavgU 2d lnd Nx: 2d, diurnal, time-averaged, single-level, assimilation What does the future hold for US National Park visitation? estimation and assessment of demand determinants and new projections The Stan Core Library A Bayesian analysis of linear regression models with highly collinear regressors Statistical Rethinking: A Bayesian Course with Examples in R and Stan Scaling regression inputs by dividing by two standard deviations A weakly informative default prior distribution for logistic and other regression models A practical guide to selecting models for exploration, inference, and prediction in ecology Information theory and an extension of the maximum likelihood principle Wildfire Policy and Public Lands: Integrating Scientific Understanding with Social Concerns across Landscapes Regional relationships between climate and wildfire-burned area in the Interior West, USA Who visits the US National Parks? An analysis of park visitors and visitation Policy Adoption and Policy Intensity: Emergence of Climate Adaptation Planning Evaluating climate change adaptation efforts on the US 50 states' hazard mitigation plans Climate Change 2021: The Physical Science Basis. Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change Our future in the Anthropocene biosphere Limiting climate change requires research on climate action Meta-analyses of factors motivating climate change adaptation behaviour A comprehensive review of climate adaptation in the United States: more than before, but less than needed Tracking global climate change adaptation among governments Are we adapting to climate change? Acknowledgments 240 We would like to thank Dr. Mojtaba (Moji) Sadegh for initial thoughts on the selecting 241 appropriate smoke data for this analysis. Additionally, we would like to thank the entire 242 "EcoStats" group at Boise State University for ongoing support on all things related to 243 quantitative ecology.