Our study area was situated within the Oregon portion of the Klamath-Siskiyou Ecoregion and consisted of farms spread across three sub-watersheds in Josephine County, southwestern Oregon . We set cameras at 1,240 m to 1,910 m above sea level. The study area included a mix of vegetation types, including open pasture, serpentine meadows, oak woodland, and mixed conifer forest. Rainfall in this region varies seasonally and by elevation, with an average of 82.7 cm annually . Mean temperatures ranged between 3.9-20.6°C in 2018–2019 . The Klamath-Siskiyou Ecoregion is one of the most biodiverse temperate forest regions on Earth, in an area that straddles the Oregon-California border and contains several regions identified as critical climate change refugia . Several species of concern are present in the county, including native salmonids, threatened Humboldt martens , Pacific fishers , and spotted owls , all of which are hypothesized to be directly or indirectly affected by cannabis agriculture . Southern Oregon, and Josephine County in particular, have a long history of illicit and medical cannabis cultivation, as well as an active presence in the growing legal industry in Oregon . Southern Oregon has become known as a prime destination for outdoor cannabis production, and Josephine County has the highest number of licensed producers relative to population size in the state . Production in the county accelerated after recreational legalization in 2014 , and takes a similar form to cultivation occurring across the border in northern California, with clusters of small farms surrounded by undeveloped or less developed rural land .Cannabis farms for this study included one licensed recreational production site, one medically licensed production site, and six unlicensed sites. All farms were producing cannabis for sale, though in different markets depending on their access to licensed markets. We selected these eight farms because they were representative of the size and style of cultivation predominant in Josephine County in the years immediately following recreational legalization in 2015 , were all established after recreational legalization except for the medical farm, did not replace other plant-based agriculture, and granted us permission to set up cameras on site.

Our sampled farms were small ,indoor grow rack had conducted some form of clearing for production space, and three had constructed some form of fence or barrier around their crop. Nonetheless, specific land use practices and production philosophies differed between farms . We cannot disclose farm locations, as per our research agreement for access.Monitored farms were clustered within each watershed: one farm in Slate Creek, five in Lower Deer Creek, and two in Lower East Fork Illinois River. We placed un-baited motion sensitive cameras on and surrounding cannabis farm clusters as well as in random locations up to 1.5 km from the farms. To guide the placement of cameras, we overlaid the area surrounding each cannabis farm cluster with a 50 x 50 m grid and then selected a random sample of at least one quarter of grid cells , stratified by vegetation openness and distance to cannabis farm. We rotated 15-20 cameras through the sampled grid cells, ensuring each camera was deployed for a minimum of two weeks. As a result of sampling across two years, we likely violated the model’s assumption of geographic and demographic closure , but given our interest was in space use associations and not estimates of occupancy, we believe this is a minimal issue. For this analysis, we restricted our data to a subset of cameras on cannabis farms and cameras in 500 m proximity to farms active during the same camera rotation . Because of rotations and field constraints, all cannabis sites were not monitored at the same time or for the same length of time . Each cannabis site had at least one, and up to three comparison cameras within 500 m during each of its active rounds. Because of farm clustering, some comparison cameras were within 500 m of more than one farm. Half the cameras on farms were monitored for more than one round, but the comparison camera were not always the same for all rounds due to rotations. We summarized species observations at cannabis farms and created detection histories using the package CamtrapR in program R . We used a 24-hr time interval because our focus was on estimating space use associations instead of occupancy, and a short interval reduced the likelihood of the same individual animal being detected on both the farm and comparison camera . We used the detection matrix to summarize detection rates per 100 operation nights for species found on cannabis sites and comparison sites.

We then modeled the occupancy probabilities of the three most commonly detected wild species, which included black-tailed deer, lagomorphs , and common gray foxes , using the UNMARKED package in Program R . We used single-species occupancy models to assess factors influencing the likelihood that a species used the area around each camera station and the probability that the species would be detected given they were present . In this case, detection can also be influenced by fine scale activity and/or habitat use patterns We hypothesized that cannabis cultivation, elevation, water access, and vegetation type would influence species’ spatial relationships, and therefore included them as predictors of occupancy in the model. We predicted that cannabis cultivation would have a negative influence on a species’ probability of using an area. We included a binary, categorical variable in the models to characterize whether detection occurred on a cannabis site or a nearby comparison site . This variable reflected and distilled the on-site practices that are common across farms, including increased human activity and fencing. We expected regional elevation to influence species’ vegetation use, and therefore used the average elevation within a 1 km buffer of each camera location, from the 30 m National Elevation Dataset . Water access is frequently an important predictor for wildlife occupancy , especially during dry periods such as during our study years, so we included distance to streams as a predictor of occupancy . To represent vegetation, we used the percent evergreen forest, as determined via the National Land Cover Database within a 1 km buffer of each camera site as a vegetation predictor variable. Finally, to distinguish general biogeographic variation between regions, we used watershed as a categorical predictor for occupancy . For modeling detection, we hypothesized that cannabis production sites would negatively influence the probability that a species was photographed given they were available in the general area, due to both physical barriers to wildlife accessing these sites, and to behavioral shifts, such as animals moving less or moving more cautiously around areas of higher human activity .

We used distance to road as a proxy for human activity separate from cannabis production that might also negatively influence detection probability. Although cannabis cultivation can be associated with the creation of new roads , the roads used in these analyses were not those created or used exclusively by cultivators. Finally, we included year as a categorical variable to account for potential inter-annual variation in detection ability. We standardized covariates to have a mean of zero and a standard deviation of one. We used Akaike’s Information Criterion  to compare model fits. We modeled all of the detection covariates first, and then kept our top ranked model for detection constant before modeling our occupancy covariates. We used our top ranked model to assess covariate relationships and determine which variables influenced species use and probabilities of being photographed.We analyzed over 5,000 animal detections over 957 operation nights . We found that the communities of wildlife present on cannabis farms were qualitatively different from the surrounding, uncultivated areas. Wildlife on cannabis farms were often smaller-bodied species, and co-occurred with higher human and domestic dog activity. There were 18 different species recorded on cannabis farms, and 24 on comparison cameras. Wild predators were predominantly detected on comparison cameras rather than cannabis farms. For example, gray foxes had 18.5 detections per 100 operation nights on cannabis sites compared to a detection rate of 31.6 on comparison sites,ebb and flow system while black bears had a detection rate of 2.5 on cannabis sites compared to 4.9 on comparison sites, and coyotes had a rate of 1.9 on cannabis sites and 6.1 on comparison sites. By contrast, domestic predators such as cats and dogs, had a detection rate twice as high on cannabis production sites than comparison sites . It is also worth noting detections of two rarer carnivores: we detected mountain lions seven times on a cannabis farm and once on a comparison site, and bobcats two times on each.For the single species occupancy models, detection variables varied by species. The top models for deer and gray foxes included a negative association with cannabis production for detection, while the top model for lagomorphs did not have similar associations . Distance to roads was retained in all models for detection, and was positively associated with detection for all species, such that detection increased with increasing distance from roads. For occupancy, here defined as use, cannabis production had a weak negative association with gray fox occupancy, and was not a top occupancy variable for any of the other species .

Because watershed and forest cover were correlated , we only used the variable with the highest univariate effect size for each species. For instance, watershed had a higher univariate effect size than forest cover for deer and gray fox occupancy, so we used watershed for candidate selection in those models, and forest cover for lagomorphs. No single variable was consistently selected as a predictor of occupancy across all species.This study represents a first step to quantify patterns of wildlife avoidance and coexistence on and surrounding active small-scale cannabis farms on private land. Our observational monitoring data suggest that wildlife species may be affected by these locations and may be altering their use of these environments. Specifically, our results suggest that 1) wildlife are consistently present on and around cannabis farms, 2) private land cannabis production may influence the local space use of some species more than others, and 3) cannabis farms may deter larger-bodied wildlife species in particular. Although limited by a small dataset, these results offer valuable insights into the ecological outcomes of the emerging cannabis industry. The assessment of wildlife detection rates suggest that many wildlife species are consistently present at cannabis production sites . Whereas some species detected on cannabis farms are ones that have been recorded in the western United States as more tolerant to agriculture or disturbance , others are species that tend to avoid human activity . While we did detect some relatively rare species , we did not detect others such as fishers or ring tails , and cannot assess whether this is due to true absence or simply short study duration. We infer detection of wildlife on cannabis farms implies a potential for these species to move through these areas. In addition, some photos revealed foraging or resting behavior , which may indicate that cannabis agriculture could maintain biodiversity as other small scale agricultural crops have in other systems . However, understanding long term impacts of cannabis production would require information on farm-level land use practices. For example, if animals on private land cannabis farms suffer fitness consequences similar to the toxicant exposure occurring on public land production, then coexistence on these sites may be detrimental in the long term . Modeled use and detection probability results indicate that despite a general wildlife presence at cannabis farms, some animals may be more affected by these areas than others. For detection, both deer and gray fox were influenced by cannabis farms . Distance to roads was positively associated with all species detection, suggesting that animals are consistently avoiding roads, but no other variable was consistent across all species for either detection or use. For occupancy , cannabis farms were not selected for deer or lagomorph models , but we suspect this could have been due to our close proximity of cannabis and comparison locations. It is possible that these species would move >500m within a 24-hour period, making it difficult to distinguish space use. Additionally, because we pooled lagomorph species, it is possible that either brush rabbits or black tailed jackrabbits individually might have responded differently to cannabis production. Nonetheless, cannabis farms influencing detection probabilities for deer and gray foxes may imply an influence on repeated visits over our time period, and potentially a behavioral adjustment near cannabis farms.