The effects of climate change on outdoor recreation participation in the United States: Projections for the 21^st century

Abstract

Climate change is expected to impact individuals’ recreational choices, as changing temperatures and precipitation patterns influence participation in outdoor recreation and alternative activities. This paper empirically investigates the relationship between weather and outdoor recreation using nationally representative data from the contiguous United States. We find that across most outdoor recreational activities, participation is lowest on the coldest days (<35 degrees Fahrenheit) and highest at moderately high temperatures (80 to 90 degrees). Notable exceptions to this trend include water sports and snow and ice sports, for which participation peaks at the highest and lowest temperatures, respectively. If individuals continue to respond to temperature changes the same way that they have in the recent past, in a future climate that has fewer cool days and more moderate and hot days, our model anticipates net participation across all outdoor recreation activities will increase by 88 million trips annually at 1 degree Celsius of warming (CONUS) and up to 401 million trips at 6 degrees of warming, valued between $3.2 billion and $15.6 billion in consumer surplus annually (2010 population). The increase in trips is driven by participation in water sports; excluding water sports from future projections decreases the consumer surplus gains by approximately 75 percent across all modeled degrees of warming. If individuals in northern regions respond to temperature like people in southern regions currently do (a proxy for adaptation), total outdoor recreation trips will increase by an additional 17 percent compared to no adaptation at 6 degrees of warming. This benefit is generally not seen at lower degrees of warming.

1. Background

Climate change is widespread and intensifying and will have far-reaching implications for human activity. Among its many effects, climate change will directly impact individuals’ recreational choices, as changing temperatures and precipitation patterns influence the attractiveness of alternative activities, and these shifts may have downstream ramifications for health and wellbeing.

In this paper, we examine how climate change will impact outdoor recreation in the contiguous United States using data on time allocation from the American Time Use Survey (ATUS). We apply a partially non-parametric regression approach to document how weather has historically affected participation in outdoor recreation activities for a baseline including the present and recent past. We then overlay these estimates on leading climate models to characterize potential changes to outdoor recreation participation from changing climate. We further analyze how these projections for outdoor recreation might change under different scenarios for adaptation and acclimatization.

A number of studies have started to probe the potential impacts of climate change on particular recreation activities and associated industries. Prior work has found adverse effects of warming on winter recreation and tourism (Morris and Walls 2009; Damm, Greuell, Landgren, and Prettenhaler 2017; Wobus et al. 2017; Steiger, Damm, Prettenhaler, and Proebstl-Haider 2021; Parthum and Christensen 2022), losses to the value of beachgoing from rising seas (Pendleton, King, Mohn, Webster, Vaughn, and Adams 2011), mixed results for recreational fishing (Pendleton and Mendelsohn 1998; Whitehead and Willard 2016; Dundas and von Haefen 2021), and potential increases in warm-weather activities (Mendelsohn and Markowski 1999 and Chan and Wichman 2020).

In light of divergent results above, this study seeks to investigate a wide range of outdoor recreation activities, in the vein of Chan and Wichman (2022), Askew and Bowker (2018), and Morris and Walls (2009), among others. This approach allows for analysis of different activities under a common framework and provides novel insights on tradeoffs induced by climate change. We extend the extant literature in four key ways. First, we probe how impacts vary by region of the United States, revealing disparate impacts across geographies. Second, we analyze substitution patterns between alternative activity categories, which sheds light on the broader implications beyond participation in outdoor recreation. Third, we add depth to the future projection of climate impacts by considering how our projections might change under empirically informed scenarios of adaptation and acclimatization. Finally, our analysis showcases one way in which climate change results in net benefits.

This work is part of a broader, multi-sector project to estimate the economic impacts of climate change in the United States (Martinich and Crimmins 2019). The Climate Change Impact and Risk Analysis (CIRA) modeling framework used for this study provides a consistent set of climate and socio-economic variables to facilitate comparisons across space, time, and alternative greenhouse gas and adaptation scenarios. While previous CIRA studies have focused on specific recreational activities in the United States, such as freshwater fishing (Jones et al. 2012), coral reef recreation (Lane et al. 2013), skiing and snowmobiling (Wobus et al. 2017), and reservoir use (Chapra et al. 2017), a broader approach encompassing multiple activities has been needed.

2. Data

2.1. American Time Use Survey

For historical information on time use in the United States, including outdoor recreation participation and other activities, we use the ATUS data for the 17-year period between 2003 and 2019. ATUS is a nationally representative cross-sectional survey of Americans age 15 and older which asks respondents to report a detailed diary of how they allocated the preceding 24 hours by activity, location, and length of time (in minutes) in a telephone interview. ATUS interviewers then code the verbatim responses of respondents into specific activities using the ATUS coding lexicon. ATUS respondents are a random sub-sample of Current Population Survey (CPS) monthly respondents from the U.S. Census Bureau. Each respondent is only asked to report their time use diary one time.

Sample selection

Our method explores the interaction between weather and time use choices; therefore our sample is limited to ATUS respondents with reported locations, allowing mapping to observed weather data. County information is provided for respondents in counties with populations greater than 100,000 residents. For confidentiality purposes, the location of respondents in counties with populations less than 100,000 are identified only with broader statistical area information: Census Based Statistical Area (CBSA), New England City and Town Area (NECTA), or Primary Metropolitan Statistical Area (PMSA). We assign these respondents to the most populous county in their identified statistical area based on 2010 Census populations. County information is directly provided for 44 percent of ATUS respondents and statistical area information allows us to assign counties to an additional 35 percent of respondents. The resulting sample for our analysis includes 167,211 observations (79 percent of the full ATUS sample). We consider the influence of this method of location matching for our primary analysis (Analysis 1, described in Section 3.1) by keeping only observations with the county identified in the data and matching observations with only CBSA identified with the least populous county instead of the most populous county (Table S10). The main results hold for these tests.

The ATUS does not provide location information for specific activities therefore we cannot be sure all activities occur in the respondent’s county of residence; however, activities that occur in places with very different weather than the respondent’s county of residence are likely to be a small portion of the overall sample. Further, the ATUS manual states the survey undercounts trips away from home due to the method of reaching respondents for diary entries.

Time allocation

We use the activity codes and location codes to identify participation in outdoor recreation and categorize recreation activities by the categories shown in Table 1. First, we isolate a set of activity codes identifying recreation activities most likely to take place outdoors. Then, we remove any observations that identify the location of the activity as “gym or health club,” the one location code relevant for recreation activities in ATUS that is definitively indoors. Finally, following Chan and Wichman (2022), we create three aggregate outdoor recreation activities: all outdoor recreation, non-sport activities, and a limited set of activities. The “all outdoor recreation” category includes all activities under the “sport, exercise, and recreation” category likely to occur outdoors, including walking that is specifically identified as for the purpose of exercise (e.g., power walking or speed walking), as opposed to commuting or other non-recreational walking, which is considered travel in the ATUS coding lexicon. The “non-sport activities” category excludes team and group sporting activities that may have occurred during scheduled event times with less flexibility for rescheduling. The “limited set of activities” category includes major recreation activities such as participating in water sports (e.g., swimming, surfing, waterskiing, diving, river tubing), fishing, hiking, boating, and bicycling, which have direct matches in the valuation database introduced below.

Table 1:

Total number of observations with any time spent on each activity category by region of the United States

	Midwest	Northeast	Northern Great Plains	Northwest	Southeast	Southern Great Plains	Southwest	TOTAL, CONUS	% OF SAMPLE
Outdoor recreation, all	3,677	4,209	121	1,016	4,240	1,713	4,478	19,454	11.67%
Playing baseball	35	32	2	6	39	22	36	172	0.10%
Playing basketball	148	175	9	26	159	77	135	729	0.44%
Participating in equestrian sports	26	25	0	9	16	9	30	115	0.07%
Playing football	39	50	1	8	64	31	58	251	0.15%
Golfing	223	195	12	46	221	73	178	948	0.57%
Playing racquet sports	53	83	0	13	96	30	85	360	0.22%
Participating in rodeo competitions	0	0	0	1	0	0	1	2	0.00%
Playing rugby	2	5	0	0	0	1	0	8	0.00%
Playing soccer	53	54	2	18	60	44	119	350	0.21%
Softball	31	22	2	7	26	11	39	138	0.08%
Other	240	216	13	77	206	102	211	1,065	0.64%
NON-SPORT	2,988	3,492	86	849	3,488	1,390	3,731	16,024	9.62%
Walking	1,632	2,117	50	481	2,021	789	2,223	9,313	5.59%
Rollerblading	36	24	0	7	18	14	50	149	0.09%
Climbing, spelunking, caving	4	3	1	0	2	2	9	21	0.01%
LIMITED	1,401	1,446	39	386	1,533	626	1,541	6,972	4.18%
Participating in water sports	447	487	12	78	594	217	490	2,325	1.40%
Running	385	449	12	132	404	227	484	2,093	1.26%
Biking	225	223	10	71	218	83	279	1,109	0.67%
Fishing	155	102	4	37	164	63	53	578	0.35%
Hiking	42	72	2	38	50	12	184	400	0.24%
Boating	97	85	1	15	89	21	33	341	0.20%
Hunting	66	40	2	12	57	22	19	218	0.13%
Skiing, ice skating, snowboarding	45	62	1	21	14	5	52	200	0.12%
Indoor recreation	2,681	3,145	108	624	2,596	1,108	2,730	12,992	7.80%
Other home	34,335	38,134	1,155	7,062	38,113	16,224	31,621	166,644	100.00%
Other non-home	31,032	34,348	1,049	6,425	34,104	14,654	28,714	150,326	90.21%
TOTAL OBS	34,336	38,134	1,155	7,062	38,113	16,226	31,621	166,647

This table presents the number of ATUS observations that have any time spent on the activities presented by NCA region then total across CONUS. The final column also denotes the percent of all ATUS observations with any time spent in each category. “Outdoor recreation” activities are determined based on six-digit ATUS activity codes (with descriptions above) as well as location code (all activities not tagged as “gym or health club” are assumed to take place outdoors). “Indoor recreation” was also determined based on six-digit ATUS activity codes and includes: playing billiards, bowling, dancing, fencing, doing gymnastics, playing hockey, participating in martial arts, using cardiovascular equipment, vehicle touring/racing, playing volleyball, weightlifting/strength training, working out, wrestling, doing yoga, as well as any activity code in the outdoor recreation set that took place in a gym or health club. “Other home” activities are determined based on two-digit ATUS codes identifying: personal care, household activities, caring for household members, eating/drinking, socializing/leisure/relaxing, and phone calls. “Other non-home” activities are determined based on two-digit ATUS codes identifying: caring for non-household members, work, consumer purchases, professional services, household services, government services, religious, volunteer, traveling, and other data codes. An annualized version of this data table with population weights is available in SM Table S1.

We group all remaining activity observations into three groups: indoor recreation, other home, and other non-home (see Table 1 for detailed categorization rules). Indoor recreation includes the observations omitted from our outdoor recreation group given the location criteria described above as well as a range of other activities chosen based on their six-digit ATUS activity codes. Non-recreation time is divided into activities likely to take place in the home (e.g., personal care and caring for household members) versus away from the home (e.g., work and caring for non-household members), based on two-digit ATUS activity codes. These are categorized as “other home” and “other non-home” activities in Table 1 and Table 2.

Table 2:

Mean amount of time (minutes) spent per day per activity across full sample by region of the United States

	Midwest	Northeast	Northern Great Plains	Northwest	Southeast	Southern Great Plains	Southwest	TOTAL, CONUS
Outdoor recreation, all	11.37	11.34	11.10	13.47	11.14	11.10	12.86	11.66
Playing baseball	0.09	0.15	0.14	0.07	0.19	0.25	0.14	0.15
Playing basketball	0.64	0.87	0.88	0.44	0.87	0.71	0.50	0.72
Participating in equestrian sports	0.09	0.06	0.00	0.23	0.05	0.04	0.10	0.08
Playing football	0.25	0.21	0.34	0.19	0.30	0.47	0.29	0.28
Golfing	1.29	0.96	1.29	1.17	1.21	0.79	1.11	1.10
Playing racquet sports	0.21	0.29	0.00	0.31	0.27	0.23	0.26	0.26
Participating in rodeo competitions	0.00	0.00	0.00	0.02	0.00	0.00	0.00	0.00
Playing rugby	0.01	0.02	0.00	0.00	0.00	0.00	0.00	0.01
Playing soccer	0.28	0.25	0.30	0.35	0.26	0.49	0.62	0.36
Softball	0.17	0.10	0.68	0.11	0.09	0.18	0.20	0.14
Other	0.59	0.63	0.73	1.03	0.49	0.59	0.51	0.58
NON-SPORT	7.74	7.82	6.74	9.56	7.40	7.35	9.14	7.98
Walking	2.73	3.12	2.38	3.30	2.78	2.63	3.99	3.09
Rollerblading	0.11	0.08	0.00	0.08	0.07	0.25	0.20	0.12
Climbing, spelunking, caving	0.01	0.01	0.15	0.00	0.03	0.02	0.05	0.03
LIMITED	4.89	4.61	4.21	6.17	4.52	4.45	4.89	4.75
Participating in water sports	1.49	1.26	2.02	0.89	1.48	1.69	1.63	1.46
Running	0.67	0.73	0.52	0.99	0.60	0.90	0.98	0.76
Hunting	0.53	0.52	0.24	0.52	0.36	0.41	0.13	0.40
Biking	0.52	0.48	0.76	0.62	0.43	0.31	0.66	0.50
Boating	0.38	0.28	0.03	0.27	0.38	0.19	0.16	0.29
Hiking	0.16	0.26	0.08	0.62	0.16	0.05	0.66	0.29
Skiing, ice skating, snowboarding	0.10	0.21	0.05	0.70	0.09	0.02	0.34	0.19
Indoor recreation	6.64	6.83	7.08	7.41	5.97	5.62	7.27	6.59
Other, home	1052.00	1052.19	1049.92	1045.38	1055.44	1042.01	1047.34	1050.60
Other, nonhome	369.99	369.64	371.89	373.74	367.45	381.27	372.53	371.15

This table presents the average number of minutes spent on each activity by respondents in the ATUS sample used in this analysis by NCA region and across all CONUS. For reference, there are 1440 minutes per day. See Table 1 for description of how time categories in ATUS correspond with the categories used in this analysis. Survey weighted applied.

As presented in Table 1, 11.7 percent of the sample engaged in any outdoor recreation, 9.6 percent reported non-sport outdoor recreation activities specifically, and 4.2 percent noted activities in the limited set of outdoor recreation activities. At the individual activity level, less than 2 percent of the sample reported participation in any one activity except walking (5.6 percent). About 7.8 percent of respondents reported any time spent on indoor recreation, relative to nearly all respondents recording time spent on “other home” activities and 90.2 percent on “other non-home” activities. Table 2 additionally describes the amount of time spent on these activities. Respondents allocated an average of 11.7 minutes per day to all outdoor recreation activities, with the most time spent by respondents in the Northwest and Southwest regions. Table S2 in the supplemental materials compares these overall statistics for the sample we include in our analysis to the sample we omit for lack of location information.

Demographics and other control variables

The CPS and ATUS together also provide other demographic information for the respondents used as control variables in our analysis. This information includes age, gender, number of children, income, race and ethnicity, education level, employment status, and marital status. Our analysis also controls for details about the specific date the diary was recorded, including day of the week and whether the date was a holiday. This location identification process as well as demographic and date controls mirror the methods employed in Neidell et al. (2021). Descriptive statistics for each of these variables can be found in Table S3 in the supplemental materials.

2.2. Historical weather data

We derive historical weather variables—daily maximum and minimum temperature, precipitation, and snowfall—for the study period from the Global Historical Climatology Network (GHCN)-Daily (Menne et al. 2012). Of the more than 100,000 stations available in the GHCN-Daily dataset, 50,299 stations provide measurements of precipitation, 47,639 for snowmelt, and 10,768 for maximum and minimum temperature over the contiguous United States (CONUS). All measurements within a county are averaged for each day to the county level. If no measurements are available in a county for a given day, we find the nearest station to the county centroid. Relative humidity is estimated from dew point temperature obtained from PRISM 2004 using the method described in Lawrence (2005). There are 153,366 matched observations between ATUS responses and weather data by county and date.

We also estimate wet bulb temperature, or the air temperature at 100 percent relative humidity. Wet bulb temperature is often used to indicate potential for heat exhaustion, since it is about the temperature on the skin when it perspires. We estimate wet bulb temperature using the empirical approach described in Stull (2011) using maximum daily temperatures and relative humidity described above.

We use the same methodology employed by Neidell et al. (2021), described in more detail in Graff-Zivin and Neidell (2014), to identify sunset and sunrise times by county and date. These were adjusted to account for the 2007 daylight savings time extension.

2.3. Future climate data

We use the same set of climate projections used in the second phase of the CIRA2.0 project (EPA 2017). These future climate projections are a subset of those generated for the Intergovernmental Panel on Climate Change’s Fifth Assessment Report (AR5), namely from climate models CanESM2, CCSM4, GISS-E2-R, HadGEM2-ES, MIROC5, and GFDL-CM3. Under the Representative Concentration Pathways (RCP) 8.5. The projections were downscaled using the procedure described in (Pierce et al. 2014) at a 1/16 degree grid. The gridded climate data was spatially averaged to the county scale for this analysis. To estimate changes in recreation activity, we compare with the CIRA baseline climate for 1986-2005, also used in Pierce et al. (2014) for the downscaling procedure. We use these climate trajectories to present impacts by degree using the methods described in Sarofim et al. (2021) and EPA (2021). We use Representative Concentration Pathways (RCP) 8.5 because it provides projections for the full range of plausible twenty-first century temperatures, obviating the need to run multiple scenarios to address low, medium, and high impacts.

2.4. Valuation data

We value changes in outdoor recreation participation using welfare estimates per trip from the Oregon Statewide Comprehensive Outdoor Recreation Plan (SCORP), a recent application of the Oregon State University Recreation Use Valuation Database (RUVD) (Rosenberger 2018, Recreation Use Values Database 2016). The RUVD is a collection of outdoor recreation valuation studies coded for characteristics such as location, activity, valuation method, and year commonly used in benefits transfer studies. The SCORP reported average values per trip by activity from the RUDV after eliminating estimates for Canada and removing outlier estimates, which we match with the outdoor recreation activities reported in ATUS (see Table S14). The RUVD does not include values for the non-limited activities, with the exception of walking. Sports have not been a focus of the valuation literature in terms of values per trip or outing, although there have been some studies outside of the U.S. on willingness to pay for sports on an annual basis (see Wicker 2011; Johnson et al. 2007; and Downward and Rasciute 2011). We assign all non-limited activities, including sports, the value for walking, following the method employed in the SCORP.

3. Methods

We perform two analyses to explore the relationship between weather and time use choices. The first estimates the relationship between weather and outdoor recreation activity and the second examines the reallocation of time between outdoor recreation and other time uses across weather conditions. We then use the results of the first analysis to project changes in outdoor recreation participation under future climate and calculate the associated change in welfare value associated with the change in participation.

3.1. Analysis 1 model: Weather and outdoor recreation participation

We seek to characterize the relationship between weather and outdoor recreation activity. To this end, we use a partially non-parametric regression framework that is common in the climate impacts literature (see, e.g., Chan and Wichman 2022). We model recreation choice as follows:

$$R_{icrst}=\alpha +f(W_{crst}\mid \Theta )+\beta X_i+{\zeta }_{rs}+{\tau }_t+{\epsilon }_{icrst}.$$ (Eq. 1)

$R_{icrst}$ represents individual $i$’s binary choice of whether to participate in the specified recreational activity or group of activities. Subscripts $c$ and $r$ represent the respondent’s county and region (where regions are defined as the Fourth National Climate Assessment (NCA) regions), while $s$ and $t$ represent the season and year in which the activity took place. $X_i$ represents individual characteristics, specifically age, sex, race, ethnicity, education level, income, family status, and employment status. We define $R_{icrst}=1$ if the survey respondent spent any time on that activity during the diary day, and $R_{icrst}=0$ otherwise.

The primary explanatory variable of interest is $W_{crst}$, which represents the weather (maximum daily temperature and total daily precipitation) experienced by the respondent on the diary day. We use a standard non-parametric approach for modeling the relationship between weather and recreation, as represented by the function:

$$f(W_{crst}\mid \Theta )=\sum \begin{array}{c}A\hfill \\ a\hfill \end{array}{\gamma }_{a}1[T_{crst}\in a]+\mu 1[P_{crst}\ge 0.25].$$ (Eq. 2)

This function creates binary variables for each 5-degree (F) bin of maximum temperature (consistent with the approach in Dundas and von Haefen 2020) and four binary variables describing the level of precipitation in inches. Our omitted temperature category is the 70 to 75-degree (F) bin, so all coefficients on temperature bins can be interpreted as temperature impacts relative to a day in the 70 to 75-degree category. Likewise, our omitted precipitation category is less than 0.01 inches of daily rainfall. Alternative treatments of temperature and rainfall including using wet bulb temperatures as opposed to dry bulb temperatures (Table S8) and accounting for precipitation as a binary (0 inches or greater than 0 inches) (Table S9) yield similar results to the main specification.

Additional variables in the estimating equation are individual and day-specific attributes $(X_i)$, region-by-season fixed effects $({\zeta }_{rs})$, and year dummies $({\tau }_t)$. See Table S3 for a complete list of weather and control variables, with descriptive statistics. We estimate the relationship using logistic regression, weighted according to the ATUS survey weights. We also tested Analysis 1 without survey weights (Table S6) and including state-season instead of region-season fixed effects (Table S7) and found the main conclusions hold.

3.2. Analysis 2 model: Weather and time allocated across activities

The preceding regressions identify the effect of weather on a given outdoor recreation activity. However, we are also interested in understanding how individuals adapt and respond to weather more generally. To this end, we outline a regression framework that allows us to evaluate how individuals may reallocate time between outdoor recreation and other time uses in response to weather conditions.

In particular, we estimate a system of equations with multiple categories of activities as outcome variables: outdoor recreation, indoor recreation, other activities in the home, and other activities outside the home. Outdoor recreation captures all activities described above. Indoor recreation includes leisure pursuits that are generally undertaken in enclosed areas, such as bowling and gymnastics. Other activities encapsulate all other time uses captured in ATUS, and we code each as home or outside of the home based on assumptions by activity code (see Table 1 for more details on activity category definitions). The system of equations as follows:

$$\begin{gathered} outdoorre{c}_{icrst}=\alpha +f(W_{crst}\mid \Theta )+\beta X_i+{\zeta }_{rs}+{\tau }_t+{\epsilon }_{icrst} \\ indoorre{c}_{icrst}=\alpha +f(W_{crst}\mid \Theta )+\beta X_i+{\zeta }_{rs}+{\tau }_t+{\epsilon }_{icrst} \\ otherhom{e}_{icrst}=\alpha +f(W_{crst}\mid \Theta )+\beta X_i+{\zeta }_{rs}+{\tau }_t+{\epsilon }_{icrst} \\ otherwa{y}_{icrst}=\alpha +f(W_{crst}\mid \Theta )+\beta X_i+{\zeta }_{rs}+{\tau }_t+{\epsilon }_{icrst} \end{gathered}$$ (Eq. 3)

Similar to the prior regression equations, these equations relate each activity category to maximum daily temperature and total daily precipitation while controlling for a host of potential confounders. However, instead of a binary indicator of participation, $[timeuse{]}_{icrst}$ represents the minutes per day spent on the given time use category by individual $i$. The four equations will sum to 1,440 minutes (24 hours), allowing us to explore if changes in time allocated to outdoor recreation are generally substituted with other recreation and leisure activities or with other time uses.

As in Eq. 1, subscripts $c$ and $r$ represent the respondent’s county and climate region, while $s$ and $t$ represent the season and year in which the activity took place. The independent variables are also the same as those shown in Eq. 1. Standard errors are clustered at the county level. The results of this analysis are used to inform the discussion of the implications of changing participation in outdoor recreation beyond welfare measures.

3.3. Projection process

To project the future number of outdoor recreation trips linked to changes in temperature and precipitation levels throughout the 21^st century, we employ the CIRA framework. First, we utilize the output of climate models described in Section 2.3. Specifically, we count the number of days in each future year where the predicted maximum daily temperature falls into each temperature bin and above the precipitation threshold described in Section 3.1. Because we are interested in changes in outdoor recreation participation relative to the baseline period (1986-2005), we ultimately describe the difference in the number of days in each temperature and precipitation bin in the future. Given the climate data is observed at the county level, our results are county specific.

Second, we multiply the number of days in these bins per future year by the estimated coefficients from Eq. 1 to project changes in the annual number of future outdoor recreation trips at the individual level. Because Eq. 1 is estimated using a logistic model, we first convert these outputs to marginal effects estimated at the sample mean. We offer three scenarios of results that demonstrate the potential effects of adaptation: (1) all counties estimated using CONUS-averaged results, (2) all counties estimated using region-matched results, and (3) all counties estimated using results from the southern regions specifically. The third scenario is expected to mimic conditions if people in northern regions respond to changes in future weather more similarly to people in southern regions currently (similar methods for approximating weather-related adaptation have been used in other studies such as Mills et al. 2015).

Third, we aggregate to the county level by multiplying the annual change in number of recreation trips at the individual level by the total population age 15 and above in the matched county. For population data, we use 2010 population counts from the U.S. Census and hold population constant in our projections to isolate effects of climate change.

Finally, we synthesize results across climate models by future temperature change using the impacts by degree approach described in Sarofim et al. (2021) and EPA (2021). Overall, this approach identifies the arrival year of a given quantity of warming relative to the baseline, then averages across impacts in the 11 years around the arrival year (see the supplemental materials of Sarofim et al. 2021 for arrival years by GCM). We present our results for all of CONUS by degree increments for each of six degrees Celsius above baseline (1986-2005), specific to and averaged across climate models. We transition from degrees Fahrenheit used in the model to degrees Celsius for the projections for consistency with the CIRA framework (described in Section 1) and comparison to other climate impact studies that commonly refer to warming in degrees Celsius.

3.4. Valuation process

We estimate the total change in welfare value associated with the change in participation in all outdoor recreation and the limited set of recreation activities based on activity-specific values described in Section 2.4. To account for the variation in response to temperature by activity, we assign activity-specific values to activity-specific results by degree. For activities in the limited category, for which we have activity-specific values, we first calculate the total change in trips by activity (see Table S16), then multiply those totals by the activity-specific value from SCORPS (see Table S14). We then sum the welfare estimate across limited activities and divide by the total change in activity days to calculate a weighted average dollar per trip for each degree. Note that this process excludes boating trips, for which marginal effects could not be calculated due to limitations of the sample. The resulting values vary slightly by degree, from $36.74 at 1 degree to a maximum of $37.53 at 5 degrees, based on the relative mix of participation changes across activities. We then apply this weighted average value to the total limited set activity trips.

For activities in the non-limited set (which includes all sports, walking, rollerblading, and climbing, spelunking, or caving), we do not estimate activity-specific changes in total trips. Instead, we subtract the change in total limited activity trips from the change in all outdoor activities estimate to arrive at the total change in non-limited activities. We value the non-limited activity trips using the SCORP value for walking (a similar approach was used in the Oregon SCORP), equivalent to $13.63. Note that walking trips are projected to decrease under future climates (Table S16) while other non-limited trips are projected to increase under future climates. This valuation method may over- or under-estimate activity day values based on the duration of participation and quality of trip; however, the average values used provide a useful sense of the magnitude of welfare changes that could be experienced in the absence of activity-specific values for the non-limited set.

4. Results and Discussion

4.1. Analysis 1 model: Weather and outdoor recreation participation

We summarize results for Analysis 1 in in Table 3. We see that participation in outdoor recreation tends to increase on warmer days, with similar patterns for all outdoor recreation, non-sport, and our limited set. In particular, we see strong, negative, statistically significant coefficients for low temperature bins and null or positive statistically significant coefficients for high-temperature bins when participation is aggregated across outdoor recreation types. These general participation patterns are consistent with prior work that has found similar relationships between weather and temperature (Graff Zivin and Neidell, 2014; Chan and Wichman, 2020; Chan and Wichman 2022). The overall increase in participation at warmer temperatures is driven by increases in water sports, which show strong, statistically significant increases above 75 degrees (see Figure 1) and has the highest levels of baseline participation in the limited set. In fact, dropping water sports from the limited set reverses the sign of participation coefficients above 85 degrees to a reduction in participation, which is statistically significant above 90 degrees. We separately test removing walking from the all outdoor recreation model and find the marginal effects are more significant and of larger magnitude at higher temperatures suggesting walking, the most common outdoor recreation activity, is less sensitive to high temperatures than other activities. Finally, we remove snow and ice sports from the all outdoor recreation run and see larger decreases in activity at low temperatures, as expected (see Table S11 for results of runs removing specific activities).

Table 3:

Marginal effects of weather variables for model 1 (outdoor recreation participation, all CONUS)

	All	Non-sport	Limited	Biking	Fishing	Hiking	Hunting	Running	Snow and ice sports	Water sports
Temp <30 (°F)	−0.0426*** (0.0063)	−0.0363*** (0.0052)	−0.0110*** (0.0031)	−0.0018** (0.0009)	−0.0003 (0.0003)	−0.0006*** (0.0001)	0.0000 (0.0001)	−0.0058*** (0.0010)	0.0023 (0.0015)	−0.0032** (0.0016)
Temp 30-35	−0.0439*** (0.0063)	−0.0342*** (0.0056)	−0.0045 (0.0040)	−0.0031*** (0.0005)	−0.0001 (0.0003)	−0.0005*** (0.0001)	0.0002 (0.0002)	−0.0041*** (0.0012)	0.0023 (0.0015)	−0.0012 (0.0020)
Temp 35-40	−0.0362*** (0.0058)	−0.0279*** (0.0054)	−0.0070** (0.0035)	−0.0028*** (0.0006)	−0.0010*** (0.0001)	−0.0003 (0.0002)	0.0001 (0.0001)	−0.0019 (0.0015)	0.0010 (0.0007)	−0.0014 (0.0016)
Temp 40-45	−0.0265*** (0.0062)	−0.0258*** (0.0048)	−0.0064** (0.0030)	−0.0012 (0.0007)	−0.0004 (0.0003)	−0.0005*** (0.0001)	0.0001 (0.0001)	−0.0026** (0.0011)	0.0011 (0.0008)	−0.0019 (0.0015)
Temp 45-50	−0.0241*** (0.0061)	−0.0218*** (0.0050)	−0.0083*** (0.0027)	−0.0019*** (0.0005)	−0.0003 (0.0002)	−0.0004** (0.0002)	0.0000 (0.0001)	−0.0027** (0.0011)	0.0001 (0.0002)	0.0003 (0.0020)
Temp 50-55	−0.0182*** (0.0052)	−0.0153*** (0.0043)	−0.0053** (0.0025)	−0.0009 (0.0007)	−0.0006*** (0.0002)	−0.0002 (0.0002)	0.0001 (0.0001)	−0.0021** (0.0010)	0.0005 (0.0004)	−0.0006 (0.0014)
Temp 55-60	−0.0127*** (0.0047)	−0.0107** (0.0042)	−0.0048* (0.0026)	−0.0015** (0.0006)	−0.0002 (0.0002)	−0.0002 (0.0002)	0.0001 (0.0001)	−0.0012 (0.0011)	0.0005 (0.0004)	−0.0013 (0.0015)
Temp 60-65	−0.0113** (0.0048)	−0.0113*** (0.0040)	−0.0065*** (0.0024)	−0.0013** (0.0006)	−0.0002 (0.0002)	−0.0002 (0.0001)	0.0001 (0.0001)	−0.0022** (0.0010)	0.0002 (0.0002)	−0.0011 (0.0011)
Temp 65-70	−0.0033 (0.0048)	−0.0038 (0.0042)	−0.0017 (0.0026)	−0.0010* (0.0006)	0.0001 (0.0003)	0.0000 (0.0002)	0.0001 (0.0001)	−0.0007 (0.0011)	−0.0000 (0.0001)	0.0003 (0.0012)
Temp 75-80	−0.0008 (0.0045)	0.0006 (0.0043)	0.0021 (0.0025)	−0.0003 (0.0005)	0.0001 (0.0002)	0.0000 (0.0001)	0.0001 (0.0001)	−0.0001 (0.0011)	0.0000 (0.0001)	0.0030** (0.0012)
Temp 80-85	0.0093* (0.0048)	0.0091** (0.0041)	0.0074*** (0.0026)	0.0001 (0.0006)	0.0003 (0.0003)	−0.0000 (0.0001)	−0.0000 (0.0000)	0.0010 (0.0012)	−0.0001*** (0.0000)	0.0051*** (0.0014)
Temp 85-90	0.0089 (0.0055)	0.0110** (0.0048)	0.0133*** (0.0034)	−0.0004 (0.0006)	0.0000 (0.0002)	−0.0000 (0.0002)	0.0000 (0.0000)	0.0022 (0.0014)	0.0000 (0.0001)	0.0099*** (0.0022)
Temp 90-95	0.0062 (0.0060)	0.0079 (0.0058)	0.0137*** (0.0036)	0.0003 (0.0010)	−0.0002 (0.0002)	−0.0003* (0.0001)	0.0000 (0.0001)	−0.0011 (0.0012)		0.0134*** (0.0028)
Temp 95-100	0.0019 (0.0074)	0.0024 (0.0064)	0.0130*** (0.0045)	−0.0006 (0.0009)	−0.0003 (0.0002)	−0.0002 (0.0002)		0.0010 (0.0018)		0.0143*** (0.0034)
Temp >100	0.0031 (0.0090)	0.0065 (0.0079)	0.0202*** (0.0064)	−0.0011 (0.0009)	0.0008 (0.0011)	−0.0002 (0.0002)	0.0003 (0.0005)	0.0017 (0.0024)		0.0161*** (0.0042)
Prcp 0.01-0.25 (inch)	−0.0086*** (0.0026)	−0.0048** (0.0022)	−0.0023* (0.0013)	−0.0005 (0.0004)	−0.0001 (0.0001)	0.0001 (0.0001)	−0.0000 (0.0000)	−0.0004 (0.0007)	−0.0000 (0.0000)	−0.0007* (0.0004)
Prcp 0.25-1	−0.0163*** (0.0032)	−0.0121*** (0.0030)	−0.0034** (0.0016)	−0.0007 (0.0005)	−0.0004*** (0.0001)	−0.0000 (0.0003)	−0.0000** (0.0000)	−0.0001 (0.0009)	0.0001 (0.0000)	−0.0007 (0.0005)
Prcp >1	−0.0289*** (0.0060)	−0.0197*** (0.0056)	−0.0095*** (0.0030)	−0.0028*** (0.0004)	−0.0005** (0.0002)	−0.0006*** (0.0001)	−0.0001** (0.0000)	−0.0010 (0.0019)	0.0002 (0.0002)	−0.0019*** (0.0007)
Observations	166,539	166,539	166,539	166,247	165,969	165,677	159,059	166,539	121,328	165,955

This table presents the temperature and precipitation regression results for analysis model 1 described in Section 3.1. Each column describes a separate model run where the outcome variable describes a sub-set of outdoor recreation activities (see Table 1 for descriptions) or specific activity within the limited set. For each model variable, the top line presents the coefficient while the bottom line presents the standard error (in parenthesis).

^*denotes statistical significance at the 90^th confidence level

^**at the 95^th confidence level, and

^***at the 99^th confidence level. For the temperature variables (expressed in degrees Fahrenheit), all coefficients are relative to the 70-75 degree omitted category. For the precipitation variables (expressed in inches), all coefficients are relative to the <0.01inch omitted category.

Figure 1: Change in Participation by Temperature and Activity Set (model 1).

Percentage change in participation for activity groups and individual activities in the limited set (boating excluded due to unsolvable margins) relative to participation on days between 70 and 75 degrees Fahrenheit. Bars represent 95^th percent confidence interval. Estimates not available for snow and ice sports over 90 degrees due to limited observations. See note of Table 1 for a list of activities defined within each activity set.

Changes will be heterogenous across regions. For example, Figure 2 shows the projected changes in participation in biking and hiking by county. For biking, some areas in the south will see decreases in participation as temperatures reach extreme highs even though overall participation increases due to warming in historically cooler areas. Chan and Wichman (2020) also examine cycling activity using a different dataset from urban bikesharing programs; they find similar heterogeneity across the country, with more pronounced increases in cycling demand in the northeastern and western states.

Figure 2: Map of Annual Change in Participation in Biking and Hiking.

County-level projected change in annual trips per 1,000 people over 15 years old for biking at 1, 3, and 6 degrees Celsius of CONUS warming relative to the 1986-2005 baseline. Projections are calculated by applying the estimated coefficients from Eq. 1 to county-level temperature and precipitation projections.

A notable example of an activity with decreasing participation as temperatures warm is skiing, skating, and snowboarding. Here, we see positive coefficients for low-temperature bins, indicating reductions in participation as temperatures warm. Note that these relationships are all significant under 60 degrees when using 10-degree bins, which allows for more observations per bin, which can get thin when segmented by activity (see Table S5). A direct implication for climate change is that future warming will tend to depress participation in these cold-weather activities (e.g., Wobus et al. 2017; Steiger, Damm, Prettenhaler, and Proebstl-Haider 2021).

We also subset our data into two broad regions—northern and southern regions defined as aggregations of the regional delineations used in the National Climate Assessment (NCA) of the U.S. Global Change Research Program—and rerun Eq. 1 on each subset separately. This analysis reveals whether warmer regions (southern NCA) respond differently to temperature than colder regions (northern NCA), shedding some light on potential adaptation and acclimatization.

The results are reported in Table 4 and Figure 3. In both regions, we continue to see negative and significant effects of cold temperatures. However, the point estimate for this impact is larger in magnitude in southern NCA regions, which are less accustomed to cold weather, although the error bars overlap. For the hottest temperatures (Temp>100), there is a statistically significant negative effect on outdoor recreational activity in northern NCA regions, and conversely a statistically significant positive effect in southern NCA regions in the limited set. Thus, extreme heat appears to stimulate outdoor recreation in regions where such conditions are more common, whereas it reduces outdoor recreation where such events are rarer. Collectively, all of these results are consistent with adaptation and acclimatization, as cold-weather states are less averse to recreating in low temperatures and warm-weather states are more likely to increase outdoor recreation activity in extreme heat.

Table 4:

Marginal effects of weather variables for model 1, separate for north and south regions (outdoor recreation participation)

	Northern NCA regions			Southern NCA regions
	All Recreation	Non-sport	Limited	All Recreation	Non-sport	Limited
Temp <30 (°F)	−0.0383*** (0.0073)	−0.0308*** (0.0061)	−0.0083** (0.0039)	−0.0457*** (0.0136)	−0.0436*** (0.0119)	−0.0127** (0.0058)
Temp 30-35	−0.0402*** (0.0071)	−0.0293*** (0.0066)	−0.0017 (0.0048)	−0.0396** (0.0171)	−0.0374*** (0.0120)	−0.0074 (0.0081)
Temp 35-40	−0.0319*** (0.0076)	−0.0216*** (0.0072)	−0.0020 (0.0048)	−0.0362*** (0.0097)	−0.0371*** (0.0090)	−0.0177*** (0.0041)
Temp 40-45	−0.0265*** (0.0071)	−0.0249*** (0.0055)	−0.0066** (0.0031)	−0.0098 (0.0126)	−0.0147 (0.0098)	0.0004 (0.0059)
Temp 45-50	−0.0095 (0.0080)	−0.0086 (0.0065)	−0.0034 (0.0039)	−0.0495*** (0.0077)	−0.0450*** (0.0059)	−0.0153*** (0.0034)
Temp 50-55	−0.0137* (0.0079)	−0.0106 (0.0066)	−0.0020 (0.0036)	−0.0208*** (0.0067)	−0.0192*** (0.0058)	−0.0085** (0.0035)
Temp 55-60	−0.0108 (0.0075)	−0.0077 (0.0066)	−0.0029 (0.0036)	−0.0139** (0.0060)	−0.0138*** (0.0052)	−0.0068* (0.0036)
Temp 60-65	−0.0060 (0.0079)	−0.0085 (0.0065)	−0.0025 (0.0034)	−0.0158*** (0.0060)	−0.0139*** (0.0050)	−0.0103*** (0.0032)
Temp 65-70	0.0034 (0.0081)	0.0008 (0.0068)	0.0044 (0.0043)	−0.0091* (0.0055)	−0.0079 (0.0052)	−0.0071** (0.0029)
Temp 75-80	0.0079 (0.0070)	0.0094 (0.0067)	0.0063 (0.0039)	−0.0084 (0.0058)	−0.0070 (0.0054)	−0.0023 (0.0031)
Temp 80-85	0.0111 (0.0079)	0.0136** (0.0064)	0.0090** (0.0042)	0.0083 (0.0058)	0.0052 (0.0054)	0.0056* (0.0031)
Temp 85-90	0.0173* (0.0103)	0.0192** (0.0089)	0.0186*** (0.0058)	0.0030 (0.0058)	0.0042 (0.0055)	0.0084** (0.0039)
Temp 90-95	0.0122 (0.0122)	0.0184* (0.0110)	0.0189** (0.0075)	0.0019 (0.0066)	0.0011 (0.0069)	0.0094** (0.0039)
Temp 95-100	0.0130 (0.0168)	0.0119 (0.0152)	0.0147* (0.0086)	−0.0021 (0.0083)	−0.0027 (0.0074)	0.0099* (0.0052)
Temp >100	−0.0211 (0.0359)	−0.0534*** (0.0123)	−0.0164** (0.0066)	0.0011 (0.0096)	0.0047 (0.0085)	0.0190*** (0.0065)
Prcp 0.01-0.25 (inch)	−0.0129*** (0.0030)	−0.0085*** (0.0026)	−0.0036** (0.0015)	−0.0038 (0.0041)	−0.0004 (0.0036)	−0.0004 (0.0021)
Prcp 0.25-1	−0.0227*** (0.0040)	−0.0189*** (0.0034)	−0.0052*** (0.0016)	−0.0074 (0.0051)	−0.0022 (0.0052)	−0.0005 (0.0031)
Prcp >1	−0.0287*** (0.0073)	−0.0201*** (0.0069)	−0.0083** (0.0040)	−0.0281*** (0.0096)	−0.0181** (0.0089)	−0.0106*** (0.0040)
Observations	80,630	80,630	80,630	85,909	85,909	85,909

This table presents the temperature and precipitation regression results for analysis model 1 described in Section 3.1 when dividing the ATUS sample into observations in northern and southern NCA regions. The northern NCA regions include the Northeast, Northwest, Midwest, and Northern Great Plains. The southern NCA regions include the Southeast, Southwest, and Southern Great Plains. Each column describes a separate model run where the outcome variable describes a sub-set of outdoor recreation activities (see Table 1 for descriptions). For each model variable, the top line presents the coefficient while the bottom line presents the standard error (in parenthesis).

^*denotes statistical significance at the 90^th confidence level

^**at the 95^th confidence level, and

^***at the 99^th confidence level. For the temperature variables (expressed in degrees Fahrenheit), all coefficients are relative to the 70-75 degree omitted category. For the precipitation variables (expressed in inches), all coefficients are relative to the <0.01 inch omitted category.

Figure 3. Change in Participation by Temperature, Activity Set, and Region.

Percentage change in participation for activity groups relative to participation on days between 70 and 80 degrees Fahrenheit for counties in northern CONUS (blue) and southern CONUS (orange). Bars represent 95^th percent confidence interval. See note of Table 1 for a list of activities defined within each activity set.

4.2. Analysis 2 model: Weather and Time Allocated Across Activities

We present results for Analysis 2 in Table 5. Here, rather than examine binary participation decisions, we instead focus on duration (minutes) spent on four activity categories: outdoor recreation, indoor recreation, other activities in the home, and other activities outside the home. The outdoor recreation results here closely resemble those from Analysis 1 model, with less time spent on outdoor recreation in cold temperatures and increasing engagement as temperatures warm. Other activities outside the home demonstrate a similar pattern for cold temperatures, but also show statistically significant reductions at the two highest temperature bins.

Table 5:

Regression results for weather variables in model 2 (substitution)

	Outdoor recreation	Indoor recreation	Other activities, home	Other activities, outside of home
Temp <30 (°F)	−2.68***(0.86)	0.70 (0.81)	19.42*** (6.37)	−17.44*** (6.33)
Temp 30-35	−2.27** (1.05)	1.18 (1.35)	17.53*** (5.93)	−16.44*** (5.68)
Temp 35-40	−1.89** (0.95)	1.41* (0.76)	8.80 (5.49)	−8.33 (5.41)
Temp 40-45	−1.76* (0.97)	0.92 (0.68)	0.38 (4.84)	0.46 (4.83)
Temp 45-50	−1.64* (0.87)	0.26 (0.66)	4.46 (4.99)	−3.07 (4.78)
Temp 50-55	−2.05** (0.79)	−0.21 (0.60)	0.33 (4.23)	1.93 (4.04)
Temp 55-60	−0.59 (0.78)	−0.22 (0.57)	2.34 (4.10)	−1.53 (4.04)
Temp 60-65	−0.55 (0.77)	0.24 (0.60)	−1.93 (3.84)	2.24 (3.79)
Temp 65-70	0.21 (0.73)	−0.38 (0.45)	3.29 (3.85)	−3.12 (3.85)
Temp 75-80	0.51 (0.79)	0.31 (0.48)	−3.77 (3.43)	2.96 (3.48)
Temp 80-85	1.71** (0.79)	0.60 (0.44)	0.88 (3.92)	−3.19 (3.83)
Temp 85-90	1.35 (0.88)	0.31 (0.53)	−0.01 (4.81)	−1.64 (4.73)
Temp 90-95	2.02** (0.93)	−0.12 (0.52)	5.69 (4.17)	−7.59* (4.12)
Temp 95-100	1.78 (1.32)	−0.40 (0.59)	4.39 (5.86)	−5.77 (5.65)
Temp >100	1.09 (1.68)	1.03 (1.24)	11.49 (7.62)	−13.61* (7.98)
Prcp 0.01-0.25 (inch)	−1.46*** (0.39)	0.17 (0.25)	−0.44 (1.85)	1.73 (1.87)
Prcp 0.25-1	−2.68*** (0.46)	−0.22 (0.33)	11.80*** (2.22)	−8.90*** (2.23)
Prcp >1	−5.06*** (0.83)	−0.21 (0.62)	19.60*** (5.39)	−14.33*** (5.31)
Observations	166,539	166,539	166,539	166,539

This table presents the temperature and precipitation regression results for analysis model 2 described in Section 3.2. Each column describes a separate model run where the outcome variable is a time use category (see Table 1 for descriptions). For each model variable, the top line presents the coefficient while the bottom line presents the standard error (in parenthesis).

^*denotes statistical significance at the 90^th confidence level

^**at the 95^th confidence level, and

^***at the 99^th confidence level. For the temperature variables (expressed in degrees Fahrenheit), all coefficients are relative to the 70-75 degree omitted category. For the precipitation variables (expressed in inches), all coefficients are relative to the <0.01 inch omitted category.

Allocation of time in a day spent on indoor recreation in Table 5 appears rather unresponsive to temperature, with only one temperature bin with a statistically significant coefficient (at the 90^th confidence interval). A run of the Analysis 1 model indoor recreation (i.e., an estimation of the binary decision to spend any time recreating indoors) does show an increase in participation at lower temperatures (see Table S4) but because Analysis 2 results do not show a significant reallocation of time, away from outdoor recreation to indoor recreation, the aforementioned gains in outdoor recreation activity from warming temperatures appear to represent true increases in overall time spent on recreation rather than an offset from countervailing changes in indoor recreation activity. This has potential implications for health and surplus through net increases in recreational participation.

Rather, the most visible time reallocations appear to take place between other activities in the home and other activities outside the home. The response functions for these two categories demonstrate opposite patterns and comparable magnitudes. As outdoor activities become less attractive in cold temperatures and extreme heat, individuals tend to shift their time toward indoor activities instead.

The supplemental materials include several additional model runs that demonstrate the robustness of our Analysis 2 results to alternative specifications, including removing survey weights (Table S12) and including state-day average weather conditions for observations without county identification dropped from our main specification (Table S13).

4.3. Projections and Valuation

The results presented in Section 4.1 demonstrate that people alter their outdoor recreation behavior at different temperature thresholds, specific to each recreation type. This section explores how participation in outdoor recreation may change under a future climate with fewer cold days and more warm and hot days (Figure S1 presents the change in number of days in each temperature bin per year at different degrees of warming). Given individuals tend to participate in outdoor recreation more on warmer days than cooler days, this could lead to an increase in total outdoor recreation as the climate warms.

Table 6 presents the results of our projection process described in Section 3.3. Under Scenario 1, where future projections are calibrated using the CONUS-wide model results from Table 3, we anticipate an increase in 88 million outdoor recreation trips per year at 1 degree Celsius of warming in the CONUS average temperature relative to the 1986-2005 baseline and up to 400 million trips at 6 degrees Celsius of warming. When applying our valuation framework, these trips result in an additional $3.2 billion to $15.6 billion in consumer surplus each year (2018 dollars). Excluding all non-limited trips except walking (i.e., dropping all activities for which activity-specific values were not available) does not meaningfully change the results (values range from $3.1 billion to $15.7 billion each year). Table S16 shows the projected change in trips for all limited set activities and walking, the non-limited activity with the highest levels of baseline participation (and large projected decreases in participation under future climates). These results assume a constant population; the magnitude of results could be larger when considering the future population at the arrival time of each integer degree of warming.

Table 6:

Projections and valuation

Degree (C)	Projected number of trips (relative to baseline), millions						Value of trips (relative to baseline), billions (2018 USD)
	All outdoor recreation			Limited outdoor recreation			All outdoor recreation			Limited outdoor recreation
	Scenario 1	Scenario 2	Scenario 3	Scenario 1	Scenario 2	Scenario 3	Scenario 1	Scenario 2	Scenario 3	Scenario 1	Scenario 2	Scenario 3
1	87.8	88.1	85.4	84.8	82.5	81.4	$3.16	$3.12	$2.98	$3.12	$3.04	$2.92
2	165.5	161.6	157.2	153.9	143.2	150.5	$5.85	$5.57	$5.58	$5.69	$5.32	$5.49
3	237.1	230.8	224.9	225.5	203.2	224.5	$8.51	$7.96	$8.24	$8.35	$7.58	$8.23
4	305.7	290.5	288.4	293.2	250.7	295.2	$11.08	$9.93	$10.81	$10.91	$9.39	$10.90
5	372.0	334.0	350.7	357.8	279.4	365.7	$13.62	$11.26	$13.45	$13.43	$10.52	$13.66
6	401.0	330.6	385.2	432.0	297.8	449.3	$15.58	$11.50	$15.76	$16.00	$11.05	$16.64

This table presents the result of our projections and valuation process described in Section 3.3 and Section 3.4 respectively. “Scenario 1” is all counties projected using CONUS-wide estimate; “Scenario 2” is counties projected using matched north and south regional estimates; “Scenario 3” is all counties projected using south regional estimates. Definitions of “all outdoor recreation” and “limited outdoor recreation” can be found in Table 1. Results are disaggregated by future degree of warming (expressed in Celsius). The table includes mean values across GCMs included in each degree Celsius. See supplemental materials (Table S15) for details by GCM.

Despite differences in data sources and methods, these changes in recreation participation values are well-aligned with previous estimates from Chan and Wichman (2022) and Mendelsohn and Markowski (1999). Because our results in Analysis 2 demonstrate that individuals are not substituting between indoor recreation and outdoor recreation activities, these values represent a net gain in consumer surplus related to recreation.

Scenario 2 offers a projection where changes in number of trips are estimated separately for counties in the north and in the south using the regional econometric model results (from Table 4). In other words, individuals in northern counties continue to reduce their outdoor recreation participation on days over 100 degrees Fahrenheit whereas individuals in southern counties continue to increase their outdoor recreation participation on these days. Under Scenario 2, we anticipate an increase in 88 million trips per year at 1 degree Celsius of warming and up to 331 million trips at 6 degrees Celsius of warming. Scenario 2 results in fewer predicted trips than Scenario 1 at most degrees of warming, but still a considerable expansion relative to baseline. Again, the activities in our limited set continue to make up the vast majority of the increase in total outdoor recreation trips.

Finally, Scenario 3 presents how the number of trips would change if everyone in the continental United States responds to temperature and precipitation changes in the future like people in southern regions do in the baseline. In this scenario, individuals in northern counties would no longer reduce their outdoor recreation participation at days that reach 100 degrees Fahrenheit and would instead adapt to the new climate conditions by increasing their outdoor recreation participation like individuals in southern counties. Such adaptative behavior among current southern residents may include shifting activities to cooler times of the day, as is common in southern summers, but currently may be less common in northern regions (Miller et al. 2022 discuss the various mechanism recreators and land managers have at their disposal to adapt under a changing climate). Under Scenario 3, we anticipate an increase in 85 million trips per year at 1 degree Celsius of warming and up to 385 million trips at 6 degrees Celsius of warming. At 6 degrees Celsius of warming, Scenario 3 results in 17 percent more trips than Scenario 2 for all outdoor recreation and 51 percent more trips for the limited recreation set, representing $4.3 million and $5.6 million in potential adaptation benefits, respectively. This underscores the sensitivity of our results to how people may adapt their outdoor recreation behavior as the climate changes.

The results in Table 6 are net across all outdoor recreation activities and specific to the limited set, however our investigation by activity reveals which activities are projected to see increases and decreases in participation relative to the baseline. Table S16 of the supplemental materials presents projections by activity. Using changes in trips at 3 degrees Celsius of warming as an example, snow sports are expected to see the largest decline in trips (8 million), followed by hiking (200,000). The largest increase in trip volume will come from water sports (175 million). Other activities expecting to see an increase in participation are running (31 million), biking (8 million), fishing (2 million), and hunting (400,000). Therefore, while outdoor recreation will overall benefit from future temperature and precipitation changes, some activities will experience a net reduction, while the increase will be concentrated among activities involving water. Changes in sports activities, which are often group activities, perhaps with less flexibility to reschedule based on conditions than other recreation activities, do not contribute significantly to the change in trip volume. The significant contribution of water sports to the net increase in outdoor recreation participation assumes the supply of water sports is otherwise unchanged in the future. As discussed in the following section, other climate stressors may limit access to water sports (e.g., drought, harmful algal blooms, water-borne disease). Table S17 presents projected trips and valuations under the assumption that water sport participation remains at baseline levels. Excluding water sports from future projections decreases the consumer surplus gains relative to baseline by approximately 75 percent. Although a scenario in which no additional water sport trips will be available is unlikely, this analysis offers a useful bounding exercise.

4.4. Caveats and Limitations

This analysis provides a method and resulting estimates of the change in outdoor recreation participation and associated values due to climate change; however there are important caveats and limitations to our findings. First, we focus on the impacts of climate change on outdoor recreation as a function of temperature and precipitation, however there are a number of other climate stressors that are not captured in this analysis (e.g., shrinking beach widths under sea level rise, decreased air quality—particularly following wildfires which are projected to increase in frequency and intensity, decreased water quality and harmful algal blooms, and changing distributions of species that typically draw wildlife viewing and hunting). These stressors are generally expected to decrease outdoor recreation participation and are not directly incorporated in our model, although we may implicitly include some of these effects to the extent the baseline data captures response to any of these concerns.

Second, there is uncertainty in the mapping between time use and weather data. We use county of residence to match respondents and their time use choices to historical weather following the approach used in Neidell et al. 2021 and others, however it is not uncommon for recreators to travel outside of their county, or even state, to participate in some outdoor recreation activities, particularly winter sports or water activities. It is possible that the weather in the respondent’s county is correlated with the weather at their recreating location but the assigned weather may not accurately represent the conditions at the recreation site. This is a limitation of using the ATUS data which does not provide specific location information.

Third, there is uncertainty in future supply and demand of outdoor recreational opportunities independent of climate change. There are reasons to believe both determinants of the equilibrium could shift over time, however the direction and magnitude is unknown, therefore we hold both constant in our analysis. For example, on the supply side, a large portion of the increase in trips comes from participation in water sport on high heat days but we do not account for potential increases in drought conditions during the same periods that could reduce access to this type of recreation or increases in harmful algal blooms that could also limit access or cause illness (Chapra et al. 2017). On the demand side, we do not account for non-climate driven changes in societal preferences or availability of substitute activities. By the end of the century, society could place a higher or lower value on outdoor recreation, increasing or decreasing demand. Poor air quality conditions projected to worsen under climate change, particularly surrounding more frequent and intense wildfires, may also decrease future demand for and welfare derived from recreational trips (Gellman et al. 2022). Increasing prevalence of vector borne disease may also decrease demand for time spent outdoors under climate change (Belova et al. 2017). Due to the same uncertainties, we hold willingness to pay per trip constant. We also choose to hold population constant for this analysis given uncertainty in the arrival times of the degrees of warming explored. Incorporating future population growth would increase the magnitude of the results.

Fourth, our valuation metrics are limited by the availability of activity-specific information in the ATUS dataset and the SCORPS valuation dataset. For example, we can project a change in general fishing activity but cannot examine subsets of fishing (e.g., coldwater and warmwater fishing) which have varying associated values. Coldwater fishing for species such as trout is highly valuable and vulnerable to climate change while fishing for warmwater species is less valuable and may see an increase in participation with a warmer climate. The total projected change in fishing activity represents the net of these two subtypes of activities and we are unable to account for the difference in values associated with the two sub-activities.

Fifth, our estimates can only identify marginal effects for the temperature ranges that currently exist with enough frequency to model recreation behavior. Temperatures exceed 100 degrees Fahrenheit in only one percent of observations, and observations at the highest temperatures (i.e., over 110 degrees) are even more limited. It is possible that demand for recreation reaches a tipping point at these extreme temperatures that we are unable to capture in our estimates due to limited observations. This would result in an overestimate of the projected benefits as these extreme heat days are projected to occur more frequently.

Finally, our estimate only captures a portion of the total economic impacts of changing recreation behavior. We measure the economic impacts of climate change on outdoor recreation as a change in welfare, however there are additional costs and revenue associated with the expansion or contraction in outdoor recreation participation (e.g., spending related to tourism at outdoor recreation sites, increased potential for accidents, and health risks of physical exertion at high temperatures). Also, because the ATUS survey is only administered to people ages 15 years and older, we are unable to estimate the impacts of climate change on children’s outdoor recreation. This is a potential area of future research given the importance of outdoor recreation and sport for children.

5. Conclusion

This paper identifies the historical relationship between weather conditions and outdoor recreation participation in the United States then projects future changes in the number and value of trips attributable to changes in temperatures and rainfall conditions. It follows the CIRA framework developed by EPA to quantify the economic impacts of climate change in the United States, allowing comparability with other anticipated impacts. We find that participation in outdoor recreation tends to increase as weather warms and decrease at cooler temperatures, and that these changes are not the result of substitutions between indoor and outdoor recreation. If individuals continue to respond to temperature changes over the remainder of the 21st century the same way that they have in the recent past, then our model anticipates an increase in 88 million trips per year at 1 degree Celsius of warming and up to 400 million trips at 6 degrees Celsius of warming, valued between $3.2 billion and $1 5.6 billion in consumer surplus each year. If individuals partially adapt to higher temperature days, our model predicts an additional 17 percent increase in trips per year across all outdoor recreation activities and a 50 percent increase in trips for the limited set of activities at 6 degrees Celsius of warming, underscoring the sensitivity of our results to how people may choose to alter their outdoor recreation behavior as the climate changes. At lower temperatures, there is a slight decrease in trips for the all outdoor recreation set (less than 3 percent difference).

While these aggregate magnitudes suggest an overall increase in outdoor recreation participation, the analysis also demonstrates that changes in participation will vary by region and activity. For instance, we find that people in northern regions are more likely to avoid outdoor recreation on the hottest days while people in southern regions are more likely to continue participating at high temperatures. Additionally, the largest increase in trip volume will come from water sports, followed by running, biking, fishing, and hunting while snow sports are expected to see the largest decline in trip numbers, followed by hiking. The net increase in outdoor recreation trips suggests that the overall increase in time spent on water-related activities, in particular, dominates the reduction in activities associated with cooler temperatures.

This work quantifies one way in which climate change is expected to result in additional value to humans, through the time they spend pursing activities for which there is a demonstrated willingness to pay and that provide some health and other well-being benefits (e.g., through exercise and fresh air). However, there will be winners and losers across activities and geography, and the expected magnitude of the net benefit is far less than the overall economic costs associated with climate change across sectors.

SIGNIFICANCE STATEMENT.

We extend the extant literature in four key ways. First, we probe how impacts vary by region of the United States, revealing disparate impacts across geographies. Second, we analyze substitution patterns between alternative activity categories, which sheds light on the broader implications beyond participation in outdoor recreation. Third, we add depth to the future projection of climate impacts by considering how our projections might change under empirically informed scenarios of adaptation and acclimatization. Finally, our analysis showcases one way in which climate change results in net benefits.

Data Availability Statement.

All files are available in the EPA’s Environmental Dataset Gateway: XX (DOI: XX).