The threat of wildfire to cannabis agriculture in California

23 At the intersection of climate change and rural development, wildfire has emerged as a threat to 24 agriculture in the western United States. This nexus is particularly problematic for the rapidly developing 25 cannabis industry in California, which includes farms located outside of traditional agricultural zones and 26 within landscapes potentially more prone to wildfire. Using fire hazard severity metrics, current and 27 historical wildfire perimeter data, and future burn regime projections, we compared the location of 28 licensed cannabis farms in California to other agricultural types, to determine if cannabis is uniquely 29 vulnerable to wildfire. We found that cannabis farming was located closer to wildfire perimeters and 30 more often in high fire hazard severity zones than other agriculture. Over the last 50 years, the distance 31 between cannabis farm locations and fire perimeters decreased significantly, and projected burn 32 regimes for the remainder of the century place cannabis farms at greater risk than other agricultural 33 types. Our findings highlight cannabis’ particular vulnerability to wildfire in California. In light of the 34 sector’s growing importance in the state, and given potentially direct and indirect consequences (e.g., 35 human health risks, socioeconomic impacts), these risks should be considered for the development of 36 future cannabis and rural development policies. 37

Analysis functions). Areas likely to experience high levels of wildfire activity in both space and time were 118 identified and classified into several different hot and cold spot categories based on the spatial and 119 temporal progression of modeled wildfire activity (Table 1). The Space-Time Cube and Emerging Hot 120 Spot Analysis functions were used to analyze the data in 3D across both space and time by aggregating 121 the predicted number of hectares burned into space-time bins. Modeled wildfire data provided 122 estimates of the number of hectares burned for each year between 2020 and 2100 across California 123 (area divided into 10688 grid cells -one grid cell extending approximately 6 km 2 ). The initial wildfire 124 projection data was aggregated into space time bins so that each bin incorporated one grid cell and 125 contained modeled data for one time slice (temporal interval was set on a yearly basis to capture the 126 gradual progression of wildfire activity). In total, our analysis included 4,950,973 hectares of hot-spots 127 (76.90% of the study area) and 149,981 hectares of cold-spots (2.32% of the study area), which 128 represent predicted fire dynamics for the period analyzed (2020-2100). 129 130 on a statewide basis? 131 To understand if cannabis farms were on average located closer to wildfires than other 132 agricultural types across California, we compared distance to wildfires between agricultural types, using 133 aggregated fire perimeters dating back to 1950. We used distance to fire perimeters as our main metric 134 of comparison, because neither FHSZ data nor burn probability data (from Westerling, 2018) are 135 comprehensive statewide. For each agricultural type, the distance was calculated between each data 136 point and the nearest fire perimeter. Although the majority of agricultural data points (especially those 137 of cannabis) were not contemporary with many of these fires, the perimeters instead are used herein as 138 a proxy for measuring geospatial susceptibility to wildfire. 139

Does wildfire pose a greater threat to legal cannabis than to other forms of agriculture
We fit a multilevel model, using the lme4 package in R Statistical Computing Software ( 156 We conducted a similar analysis focusing only on cannabis producing counties, restricted to 157 those counties comprising at least 1% of all CDFA outdoor cultivation licenses statewide. These included: 158 Humboldt, Lake, Mendocino, Monterey, Nevada, San Luis Obispo, Santa Barbara, Santa Cruz, Sonoma, 159 Trinity, and Yolo Counties (Figure 1). In these counties, we compared the threat of wildfire to cannabis 160 against the remaining agricultural types (pasture, grapes, general crops) using Fire Hazard Severity Zone 161 (FHSZ) data. We also compared cannabis wildfire risk between counties using fire perimeter data. 162 In order to address whether the threat of wildfire to cannabis has changed over the preceding 163 decade, we measured the distance of licensed cannabis farms to historic perimeters of wildfires that 164 occurred between 1970 and 2020. We compared the proximity of cannabis farms to historic fire 165 patterns were summarized between agricultural types in cannabis producing counties, as well as 226 between cannabis producing counties, focusing exclusively on cannabis data points. Using a space-time 227 approach in analyzing modeled wildfire activity throughout the twenty-first century allowed us to take 228 into account not only the spatial, but also the temporal dynamic of predicted wildfire activity and helped 229 identify areas likely to experience different wildfire threats through time. These distinctions are 230 important in assessing not only the current wildfire risk of cannabis farms, but also evaluating how this 231 risk is predicted to evolve in the future. 232 As a final metric of the threat to cannabis posed by wildfire, we calculated the proportion of 233 cannabis farms within fire perimeters of the 2020 fire season. The proportion of cannabis farms within 234 fire perimeters was calculated for each cannabis producing county. Cannabis was also compared to the 235 remaining agricultural types, focused specifically within cannabis producing counties. 236 were all reliably positive, indicating larger distances than those of cannabis (Intercept MLE=1.46 , 264 SE=0.11 , OSE= 4.30 km). Additionally, the percentage of cannabis farms that were located within 265 historical fire perimeters (9.89%) was higher than grapes (2.91%), pasture (1.75%), and general crops 266 (0.89%). 267

How does the threat of wildfire vary among legal cannabis producing counties and is the
268 threat increasing? 269 The percentage of cannabis farming in high (43.88%) or very high FHSZs (35.25%) was also larger 270 than any other type of agriculture found in cannabis producing counties (Figure 3). Grapes were the next 271 most common type to occur in high (24.53%) or very high FHSZs (5.09%), followed by pasture (12.43%; 272 0.76%, respectively). General crops almost never occurred in high (6.83%) or very high FHSZs (0.45%).    California. Our spatial analysis of statewide wildfire risk in California suggests that cannabis agriculture is 364 uniquely vulnerable to wildfire impacts relative to other crops. At the statewide scale, cannabis farms 365 are on average located within 3km of a past wildfire, whereas pasture is located over twice as far, 366 grapes three times as far, and general crops are located over four times as far from wildfires. 367 Furthermore, although the statewide distribution of cannabis agriculture is largely confined to a handful 368 of relatively fire-prone counties, the distribution of cannabis farms within these counties is still closer to 369 historic wildfire perimeters, and more likely to be found in high fire hazard severity zones, than are all 370 other agricultural types. Our results further suggest an alarming trend of increasing fire risks to cannabis 371 in the future. The wildfire risk for cannabis increased markedly during the five year period from 2015-372 2020 compared to the preceding 45 years. Overall, we estimated that the distance between cannabis 373 farms and fire perimeters has shrunk by 36% during this time period. Likewise, using data on projected 374 burn regimes, we found that a disproportionate number of cannabis farms are located in wildfire 375 hotspots under future climate scenarios. 376 legal market, many remain in these remote areas, given that their farms are already established. While these small farms vastly outnumber their counterparts elsewhere in the state, the majority of legal 385 cannabis production has already shifted to the Central Coast, where farms are less numerous, but orders 386 of magnitude larger (Dillis et al., 2021). Unfortunately, the future wildfire outlook for the Central Coast 387 also poses a concern in that all three top cannabis producing counties in this region have more than half 388 of their farms in zones projected as persistent, new, or intensifying wildfire hotspots. In fact, over 95% 389 of its cannabis farms in Santa Barbara County, which is now the top cannabis producing county in the 390 state, are located in new or intensifying hot spot zones. 391

Geography of Cannabis in California May Exacerbate Threat from Wildfire
It is worth noting the counties that produce the vast majority of the state's irrigated agricultural 392 crops are located in the Central Valley and are generally considered to have very low wildfire risk. 393 However, aside from Yolo County, every county in the valley has continued to prohibit cannabis 394 agriculture through local ordinances. As a consequence, many areas suitable for cannabis cultivation 395 that have lower fire risk are currently inaccessible. Future changes in policy that allow for cultivation 396 within these counties may significantly lower the overall wildfire risk to cannabis in the state. Within 397 current cannabis producing counties, many land use policies have encouraged production on lands 398 already used for agriculture. To the extent that these lands have less exposure to wildfire, it is possible 399 that newly establishing farms may have lower fire risk. As an example, Monterey and Trinity Counties 400 have both experienced an exceptional amount of wildfire since 2015 (covering 16% and 33% of their 401 land areas, respectively), yet the latter has over 90% of its cannabis farms in very high fire hazard 402 severity zones, while the former has none. This is largely because cannabis farming in Monterey County 403 is new and confined to agricultural zones, while production in Trinity County is still located in remote 404 legacy areas and there is relatively little agricultural land throughout the county in general. 405 406 Although the number of cannabis farms that were directly damaged by wildfire in 2020 (i.e., 407 inside fire perimeters) is small (0.63%), a much larger proportion of farms were likely affected by their 408 close proximity to fire, and experienced impacts from smoke exposure or infrastructure damage (e.g., 409

Cascading impacts of wildfire on the cannabis industry
power and water systems). It is unknown what proportion of farms experienced crop damage or losses 410 from wildfire smoke. While the adverse effects of wildfire smoke on the chemical composition and 411 quality of wine grapes ("smoke taint") is well documented and known to cause significant economic 412 impacts (e.g, (e.g., Krstic et al., 2015), the effects of smoke on the quality of cannabis products is less 413 well understood (but see Kukura, 2020;Schiller, 2020). In 2020, tax revenues from legal cannabis sales in the state amounted to over $780 million (State of crop losses from wildfire are likely to have critical economic impacts, particularly in rural communities 434 with a higher direct social and economic dependence on cannabis agriculture (Kelly & Formosa 2020). 435 This could also disproportionately impact already marginalized small-scale cannabis farmers who may 436 not have resources to recover from wildfire-related losses. state should also pursue options for providing crop insurance to farmers that aren't eligible for federal 447 programs. Furthermore, given that the impacts of wildfire extend beyond fire perimeters, research on 448 smoke exposure risks for cannabis crops and farm workers is an urgent priority. Collectively, these steps 449 will help bolster the resilience of the developing regulated cannabis industry with respect to wildfire. 450 The impacts of wildfire on cannabis farming may be particularly severe, but serves more generally as an 451 example of the vulnerability of rural agriculture, and its dependent communities, in the face of climate 452 change and the consequent increase in natural disasters such as wildfire. 453 Table 1. Descriptions of projected burn patterns adapted from ESRI (2016). Aggregates of individual burn patterns are indicated by horizontal lines. Each description of a hot spot pattern also applies to an equivalent description of a cold spot pattern.

Burn Pattern Description
New Hot Spot A location that is a statistically significant hot spot for the final time step and has never been a statistically significant hot spot before.
Intensifying Hot Spot A location that has been a statistically significant hot spot for 90% of the time-step intervals, with the intensity of clustering increasing overall and that increase is statistically significant.
Historical Hot Spot The most recent time period is not hot, but at least 90% of the timestep intervals have been statistically significant hot spots.

Persistent Hot Spot
A location that has been a statistically significant hot spot for 90% of the time-step intervals with no discernible trend indicating an increase or decrease in the intensity of clustering over time.

Sporadic Hot Spot
A location that is an on-again then off-again hot spot. Less than 90% of the time-step intervals have been statistically significant hot spots and none of the time-step intervals have been statistically significant cold spots.
Oscillating Hot Spot A statistically significant hot spot for the final time-step interval that has a history of also being a statistically significant cold spot during a prior time step. Less than 90% of the time-step intervals have been statistically significant hot spots.
Diminishing Hot Spot A location that has been a statistically significant hot spot for 90% of the time-step intervals, with the intensity of clustering decreasing overall and that decrease is statistically significant.
No Pattern Detected Does not fall into any of the hot or cold spot patterns.