The interview data in our current study support this interpretation and produce the same finding in an additional legacy production region. Our approach of incorporating social or cultural data into ecological modeling is not unique to cannabis production, and is becoming more common in contexts as varied as deforestation , marine conservation , and human-wildlife conflict . One strength of incorporating qualitative data into quantitative models is the ability to capture nuances that may be left out or simplified in traditional modeling efforts. For example, while we did not identify any economic covariates functioning at the parcel level for our models, the interview data helped us recognize that broader economic changes are likely to influence changes in regional cannabis production over time. Another example was our use of local cannabis density as a proxy for supportive local attitudes towards cannabis farming. The interview data allows us to simplify a much larger concept of connection to community with this variable, while recognizing that in doing so, we may lose some local nuances – such as locations where there is a high neighborhood cannabis density but also strong negative community attitudes towards cannabis production. Some of the drivers identified in our study raise concerns that farmers may be actively selecting parcels that are in areas of greatest environmental sensitivity. For example, as farmers seek out more rural parcels, these are also likely to be ones with greater terrestrial wildlife habitat—in fact, as the interviews indicate,vertical farming supplies this faunal biodiversity is often something farmers appreciate and seek on the land in which they live and farm. Similarly, the preference for parcels closer to rivers and streams may result in negative impacts on freshwater systems.

Previous research has illustrated a potential overlap of cannabis agriculture in Josephine County with terrestrial and aquatic biodiversity , and our findings here suggest that this overlap is not incidental. It is possible that the ecological overlap observed in other rural cannabis-producing regions could be influenced by similar social/cultural drivers. The significance of ruralness and distance to freshwater in the model of new farm development further raises concerns that this proximity could increase over time. The emergent theme of connection to community, and the strength of its associated drivers for cannabis distribution illustrated the network reliance of cannabis farmers, which further suggests that development over time is likely to occur in areas that are current cannabis hotspots. The context provided by the interview data suggests that some of the same motivations leading farmers to grow in rural areas may also provide opportunities to mitigate potential environmental harm. While our sample of farmer perspectives is relatively narrow, they all expressed strong environmental stewardship values. Similarly, other studies from California have identified commitments to environmental practices among outdoor cannabis farmers . These values alone do not mean that private land cannabis farming has a low environmental footprint — the farmers themselves even expressed concerns over the impacts of the industry. Rather, environmental stewardship values, combined with farmer concerns about the lack of education on best management practices for cannabis, implies that there is a research, education, and outreach gap for sustainable cannabis farming. This gap is one that researchers have repeatedly noted . Moreover, in their connection to community, farmers explained that they rely heavily on learning from other farmers’ practices. Thus, there may also be opportunities to enforce conservation-minded practices via cultural dissemination to receptive farming communities. Our land use models illustrate a rapidly expanding cannabis farming industry, with a 116% increase in parcels with cannabis, and a 227% increase in plant count over 2-3 years from pre- to post-recreational legalization county-wide. Despite this rapid increase in cannabis production, most interviewed farmers were not optimistic about the future of the industry, with frequent comparisons to other “boom-bust” natural resource trajectories.

Moreover, many farmers also described an industry that was currently unpredictable, difficult to navigate , and unlikely to result in long term financial stability. This disconnect between the farmers’ perceptions of the industry compared with its rapid expansion could mean that the specific type of producers we interviewed were not benefitting from the industry increase that accompanied legalization. Other research on small scale cannabis producers from northern California supports this interpretation . It is also possible that landscape-scale industry change does not translate to the scale of an individual farm. If this is the case, it might help explain why the model of change in plant count had the fewest significant predictors—rather than being a more simplified process, it might instead be that the drivers for farms that existed before legalization are highly individualized or localized.Despite the uncertainty surrounding the trajectory of legacy cannabis farms, the models for new cannabis development provide insights into predicting the growth of the industry. While we did not project our predictions into the future, due in part to large policy changes that were not explicitly addressed in our interviews or models , our results do provide a baseline and contextualized understanding that could be used for future predictions. For example, based on farmer descriptions for why they may seek out large and rural parcels, it is unlikely that the strength of those drivers would decrease over time. On the other hand, farmers’ stated preference for farm-zoned parcels, which by contrast ended up as a significant driver in the opposite direction for new farm development, might be more likely to change over time as a potential driver due to shifts in regulation, enforcement, or social pressures for those renting/selling farm zoned parcels. While our results are broadly useful for understanding cannabis landscapes in southern Oregon, there are many levels of complexity that are not captured by the models. For example, we treat cannabis agriculture as a single entity for these models, while in reality it contains a diversity of production styles and regulatory statuses. It is entirely likely that a large-scale licensed hemp farmer and a small-scale unlicensed cannabis farmer will reveal different drivers of their land use. Similarly, whether a farmer owns their own land or rents it, or whether a farmer lives on site or off, could also change the relationship with potential drivers. While we did not have detailed information on each cannabis producer at the county level to classify or group production styles, this would be an important avenue for future research. Future research would also benefit from added time points, particularly after the 2018 federal hemp legalization. In addition, this study was largely confined to a small number of small-scale farmers, and thus an expanded interview or focus group data collection process might reveal new drivers that would be relevant for other production styles.

The relatively low pseudo r-squared values for our models suggests that there may be additional drivers functioning in this system, which extended interviews could help uncover. Our study focused on private land production, but it is important to remember that public land production also occurs in this area and influences not only the local environment, but the public perceptions of cannabis in the region. Incorporating the links between public and private industries might strengthen our understanding of these systems. Similarly, linking different scales of drivers would be a valuable next step. The interview data indicates that the southern Oregon industry is tied to regional and national markets , and that much of the economic decisions are either very fine scale at the level of the farm, or broader scale at the level of the state. Within the scale of Josephine County, the significant effect of mapped year implies that there may also be different dynamics in the two halves of the county that were mapped at different time points . Although it did not directly emerge in the interviews, while living in Josephine County, PPS observed different local approaches to integrating cannabis indoor greenhouse farmers into the community in Williams as opposed to the Illinois Valley. This is an example of a secondary way in which the observations that occur during the interview process can assist with model interpretation. Further research on differences in local policies, community standards, or other regional differences might elucidate this pattern. Capturing interrelated dynamics such as local to county-wide processes would require a complex modeling approach but might lend insights into multi-scalar drivers.Understanding wildlife response to disturbance across landscape gradients is a complex endeavor. Individual animals can respond to anthropogenic disturbance with a variety of different behavioral changes , but these responses are all context dependent . For example, in some studies, coyotes demonstrate a space use preference for agricultural areas , while in others, they avoid farmland ; similarly, at times they are labeled as urban exploiters , and at times avoiders . These differences are often tied to context-dependent responses and differences in landscape configurations . At a wildlife community level, the complexity of responses increases even more. Disturbance may affect some species more than others, or in opposite directions, leading to broader contractions or expansions in species assemblages and interactions . Changes in species interactions, especially if they involve keystone species, can have cascading effects on ecosystem function . The context-dependence of these shifts means that consistently predicting how wildlife communities will respond to rapid land use change at a local level is very difficult and requires understanding multiple interacting mechanisms . Nevertheless, wildlife community responses to disturbance matter because the context-dependent consequences in turn can affect ecosystem health , effectiveness of wildlife management strategies , and human-wildlife conflict . Thus, there is a continuing need to examine the effects of disturbance on wildlife in order to develop strategies to mitigate the negative effects of land use change. Understanding wildlife response to disturbance is particularly important in areas where land use change is occurring rapidly. Spaces of rapid development for agriculture are called frontiers, and are often spurred by the growth of a new industry, while accompanied by the movement or growth of human populations, and transportation structure improvements.

Frontiers are naturally spaces of rapid land use change, and often sites where different approaches to land use planning and conservation clash . While frontiers present a novel disturbance scenario, most studies of wildlife response to agricultural land use have been concentrated in Asia, South America, and Europe , and often in areas that have long been dominated by agriculture. Such studies may miss some of the immediate responses of wildlife to development that occur over shorter spatial and temporal scales . Recreational cannabis agriculture represents an ideal opportunity to study wildlife community response to disturbance generated by a currently expanding land use frontier. In the US, state level legalization of recreational cannabis has initiated a rapid land use frontier for outdoor cannabis production . This frontier is particularly noticeable in rural areas of the western US. Influenced by its illicit history, outdoor cannabis is often grown in remote, bio-diverse regions with minimal other non-timber agriculture . Regardless of individual legal status, private land cannabis farms are typically smaller than those of other commercial crops, and are clustered in space, creating a unique land use pattern of small points of development surrounded by less developed land . This pattern of development locates the cannabis frontier directly at the wilderness boundary—a somewhat rare characteristic for agriculture in the United States .At a broad scale, cannabis development in rural areas overlaps with regions that may be important habitat for wildlife , yet it is unclear whether, where, and to what extent this broad scale spatial overlap actually results in negative impacts on animals at a local scale. There have been studies suggesting that cannabis production may lead to habitat destruction or modification , and wildlife death due to toxicant use and poaching . However, most studies on direct impacts of cannabis farming have largely been conducted on illegal public land production sites , as opposed to private land sites. The research conducted to date on private land has not encompassed a full landscape gradient around cannabis farms. Not only have private land sites likely seen the largest production increases due to legalization in recent years , they are also often characterized by very different production practices than public sites. For example, on many private land farms, indirect sources of disturbance to wildlife such as noise and light are more common than direct causes of mortality.