FINAL REPORT Title: Social-ecological resilience to shifts in fire regime in the Sonoran

Desert

JFSP PROJECT ID: 15-1-03-35

SEPTEMBER 2019

Clare Aslan Northern Arizona University

Brett Dickson Conservation Science Partners

Dave Theobald Conservation Science Partners

Leah Samberg Rainforest Alliance

The views and conclusions contained in this document are those of the authors and should not be interpreted as representing the opinions or policies of the U.S. Government. Mention of trade names or commercial products does not constitute their endorsement by the U.S. Government.

i

TABLE OF CONTENTS Table of Contents ....................................................................................................................... i List of Tables ........................................................................................................................... iii List of Figures .......................................................................................................................... iv List of Abbreviations/Acronyms ................................................................................................ v Keywords ................................................................................................................................. vi Acknowledgments ................................................................................................................... vii Abstract ..................................................................................................................................... 1 Objectives ................................................................................................................................. 1 Background ............................................................................................................................... 2 Materials and Methods .............................................................................................................. 6 Prediction of future fuels and fire regimes across the Sonoran Desert Ecoregion…………...6 Stakeholder engagement: implications of these changes for management programs...……...8 Integration of social and ecological data to develop spatial models and maps of fire resilience across the landscape…………………………………………………………………………….10 Spatial data processing……………………………………………………………………...11 Ecological and social risk…………………………………………………………………..12 Social risk statistical methods………………………………………………………………12 Ecological resistance and adaptive capacity………………………………………………..13 Social resistance and adaptive capacity…………………………………………………….14 Social resistance and adaptive capacity statistical methods – Interview…………………...15 Social resistance and adaptive capacity statistical methods – Survey……………………...15 Creating socioecological risk and resilience maps…………………………………………16 Analysis of non-spatial patterns of management objectives and activities………………...16

ii

Development and analysis of adaptive capacity survey……………………………………..17 Results and Discussion ............................................................................................................ 17 Outcomes of stakeholder engagement……………………………………………………….17 Integration of social and ecological data to develop spatial models and maps of fire resilience across the landscape……………………………………………………………………………..18 Analysis of non-spatial patterns of management objectives and activities………………….22 Survey and analysis……………………………………………………………………….…25 Spatial predictions of resilience in the Sonoran and their implications……………………..26 Implications of non-spatial findings regarding objectives and activities……………..……..27 Alternative activities and their promise……………………………………………………...27 Conclusions (Key Findings) and Implications for Management/Policy and Future Research .... 29 Literature Cited………………………………………………………………………………….31 Appendix A: Contact Information for Key Project Personnel………………………………….A1 Appendix B: List of Completed/Planned Scientific/Technical Publications/Science Delivery Products …………………………………………………………………………..…………….A2

iii

LIST OF TABLES

Table 1. Management objective and current management practice categories and descriptions..10 Table 2. Indicators used to compile an ecological resistance map of the Sonoran Desert.……..13 Table 3. Indicators used to compile an ecological adaptive capacity map of the Sonoran Desert…………………………………………………………………………………………….14 Table 4. Comparison of manager-reported confidence levels that objectives can be achieved and activities continued into the future under environmental change. ……………………………....25

iv

LIST OF FIGURES

Figure 1. Synthesis conceptual model, displaying the influence of fire regime change and management response on ecosystem resilience………………………………………………..5 Figure 2. Maps of study area showing the 3 different fire risk models in panels a, b, c……...9 Figure 3. Manager-reported objectives, activities, and alternative activities for their jurisdictions…………………………………………………………………………………..18 Figure 4. Social fire risk in the Sonoran Desert, as assessed by interview questions about fire suppression and fire management and survey questions about jurisdiction resources………19 Figure 5. Ecological resistance in the Sonoran Desert, as determined by topographic diversity, geophysical diversity, vegetation diversity, and water availability………………………….19 Figure 6. Ecological adaptive capacity as determined by habitat connectivity, species richness, and human modification……………………………………………………………………..19 Figure 7. Social resistance determined by manager responses to questions about meeting management objectives and preparing for surprises in ecosystem behavior………………...20 Figure 8. Social adaptive capacity determined by manager responses to questions about adopting new management practices in the future and in the past. …………………………………...20 Figure 9. Socioecological fire risk in the Sonoran Desert. …………………………………21 Figure 10. Socioecological fire resistance in the Sonoran Desert. …………………………21 Figure 11. Socioecological adaptive capacity in the Sonoran Desert. ……...………………21 Figure 12. Vulnerability in the Sonoran Desert. ……….…………………………………...22 Figure 13. The proportion of total objectives across all points comprised by the 10 identified management objectives. ……………………………………………………………………..22 Figure 14. The proportion of total objectives identified by each agency comprised by a given objective. ……………………………………………………..……………………………..23 Figure 15. Activities to meet the objective of fire suppression by management agencies that reported a fire suppression objective (Federal, Local, and Tribal agencies). …….………...24 Figure 16. Activities to meet the objective of infrastructure protection by management agencies that reported that objective. …………………………………………………….…………..24

v

LIST OF ABBREVIATIONS/ACRONYMS

ATP - Adaptation Turning Point AZ - Arizona DC - District of Columbia DEM - Digital Elevation Models SERDP - Department of Defense Strategic Environmental Research and Development Program GIS - Geographic Information System GM - General model MCMC - Markov Chain Monte Carlo Analyses NEPA - National Environmental Policy Act NDVI - Normalized Difference Vegetation Index OHV - Off-Highway Vehicles ORV - Off-Road Vehicles PSA - Public Service Announcement RCP - Representative Concentration Pathway SWEI - Shannon-Weaver Equitability Index TES - Threatened, Endangered. and Sensitive US - United States USGS – United States Geological Survey

vi

KEYWORDS

Adaptive capacity; adaptive tipping points; adaptive turning points; fire regime change; management activities; management objectives; resilience; resistance; socioecological systems; Sonoran Desert Ecoregion

vii

ACKNOWLEDGEMENTS

We are indebted to the many land managers and stakeholders who devoted time and energy to meeting with us during the initial needs assessment, attending one or both workshops, completing the survey and one-on-one interviews, and participating in the results webinar. We are grateful to the Southwest Fire Science Consortium for hosting our results webinar, and to the Sonoran Joint Venture for hosting our results workshop. Special thanks to M. Lata for additional consulting and feedback throughout this process.

1

ABSTRACT

Although the Sonoran Desert is considered a non-fire-adapted ecosystem, emergence of a novel fire regime is underway, driven by increasing strength and duration of drought as well as invasion of non-native annual plant communities. Increasingly larger and more frequent fires can diminish the long-term recovery potential of native Sonoran Desert plant communities and place desert systems at risk of ecosystem state change. To estimate social-ecological resilience across the Sonoran Desert Ecoregion, we used ecological modeling, manager interviews, and social system surveys to determine the likelihood that regional ecological and social systems will persist in their current state into the future. Our aim was to identify locations across the Sonoran where post-fire resilience is likely to be high and thus interventions such as restoration of low priority, as well as sites where post-fire resilience is likely to be low and thus active restoration more important. Following data collection, we performed a spatial overlay of social and ecological factors to assess total social-ecological fire risk, social-ecological resistance to state change, and social-ecological adaptive capacity across the study region. Highest total fire risk occurred in scattered locations, with greatest concentration in the eastern portion of the ecoregion. The southeastern ecoregion exhibited the lowest social-ecological resistance as well as lowest social-ecological adaptive capacity. These areas are thus vulnerable to fire-driven state change. Interviewees reported the likelihood that specific activities and objectives would continue under future fire regimes. Social and cultural resource protection and conservation were considered least likely to be achieved over the long term, whereas economic and military activities and objectives were most likely to be achieved over the long term. By considering the joint influences of ecological and social factors, this study enables an assessment of future responses to fire regime change across the management landscape of the Sonoran Desert.

OBJECTIVES

Our proposed work comprehensively addressed JFSP FON Task Statement 3: Implications of Changing Fuels and Fire Regimes. We predicted probable climate-driven shifts in vegetation and fuels across the Sonoran Desert of Arizona, projecting 20 and 40 years into the future. We used these predictions to guide an interdisciplinary process for assessing how predicted fire regime shifts will influence management and what parts of the study area are likely to be more or less resilient to fire events. We accomplished all objectives (below) of this project. Our first objective was to predict likely changes over the next 20 to 40 years in fuels and fire regimes across the study region. We accomplished this by deriving regional projections of current and future vegetation attribute patterns, based on available downscaled future climate models, and predicted fire regime shifts (changes in fire behavior, spread, risk, and connectivity [likelihood]). These outputs undergirded the following objectives and syntheses. Our second objective was to forecast the potential implications of these changes for management programs. We employed an initial needs assessment workshop and follow-up survey, through which we worked with managers to: (a) identify points along the fuels regime change scenarios at which current fire management methods will be unable to meet current objectives; (b) explore and define a range of alternative management strategies; and (c) quantitatively assess the ability of management units to adopt new approaches. Our third objective was to synthesize and integrate the results from objectives 1 and 2 to develop spatial models and maps of fire resilience across our study area. This enabled us to

2

distinguish areas where active management might be necessary to prevent changed fire regimes from sending ecosystems past their tipping points. We engaged managers and other stakeholders in a final workshop, wherein we worked with managers to jointly interpret model and map-based results, and to develop practical management recommendations in light of fire regime change. To guide this work, we hypothesized that: (1) flammable fuels and fire parameters (e.g., behavior, risk, likelihood) will increase over the next 20-40 years as a result of vegetation shifts driven by climate change and plant invasion (addressed the FON “changing fuels and fire regimes” questions); and (2) changing fire conditions on the landscape will lead to changes in the predicted effectiveness of management interventions, and thus changes in management decisions (addressed the FON “implications for management programs” questions).

BACKGROUND

Changing fire regimes are among the most pervasive drivers of rapid environmental change worldwide (Nolan et al. 2018). Shifting rainfall patterns and increasing temperatures result in strengthening drought in many locations (Schlaepfer et al. 2017). Increasingly powerful weather systems lead to fluctuations in moisture that are outside the historic range of variability, resulting in pulses of biomass production followed by rapid drying of fuels (Keyser & Westerling 2017; Pendergrass et al. 2017; Abatzoglou et al. 2018). Spread of invasive species alters the quantity, seasonality, and continuity of fuels in many systems (D’Antonio 2000; Brooks et al. 2004; Gaertner et al. 2014). Compounding all of these ecological changes, human activities imposing fire suppression and exclusion result in fuels buildup in some systems and increase accidental ignition rates, altering fire return intervals, severity, and extent (Bowman et al. 2011).

The Sonoran Desert in the southwestern United States is generally considered non-fire-adapted, with native vegetation growth patterns that result in non-continuous fuels and few native fine fuels (Alford et al. 2005; Fuentes-Ramirez et al. 2016). Today, however, the Sonoran Desert faces extreme drought as a result of steadily increasing temperatures and increasingly patchy rainfall (Seager et al. 2007; Abatzoglou & Kolden 2011). Additionally, widespread invasion of non-native annual grasses, introduced for cattle forage, creates a novel layer of continuous fine fuels that can link native vegetation patches and spread fires ignited by lightning or human activity (McDonald & McPherson 2013; Moloney et al. 2019). Although the Sonoran Desert is the most biodiverse desert in North America, home to numerous threatened and endangered species and the continental center of biodiversity for key biotic groups such as native bees (CBD 2014; Comus et al. 2015; citation), its sensitive, native vegetation communities are not adapted to short fire return intervals, and its native plants lack fire recovery mechanisms (Weiss & Overpeck 2005). Emergence of a novel fire regime in the Sonoran Desert, with increasingly larger and more frequent fires, can diminish the long-term recovery potential of native plant communities with limited inherent resilience to disturbance and place desert systems at risk of ecosystem state change (Abella 2009; Brooks & Chambers 2011). Resilience, discussed further below, is defined as the quantity of perturbation a system can absorb without permanently transitioning to a novel state (Gunderson 2000). To effectively support vegetation community resilience, and thus retain functioning ecosystems and native Sonoran Desert biodiversity on the landscape, management and restoration professionals in the Sonoran must confront the combined effects of plant invasion, drought, and fire (McCarty 2001).

3

In the Southwest, average temperatures are on the rise but observed and predicted precipitation displays more variability (Jardine et al. 2013). Specifically, rainfall is becoming patchier, both spatially and temporally: on any given year, most sites receive less rain than the historical average, but in rare years rainfall far exceeds that average (Coe et al. 2012). As an example of the rare precipitation burst, in the winter of 2004-2005, southeastern Arizona received an unusually high amount of precipitation, and as a result the lower Sonoran Desert accumulated a much greater load of standing non-native and native annual grass biomass than normal (Gray et al. 2014). After the winter rains, high temperatures during an unusually hot and lengthy pre-monsoon summer dried out this biomass, which had formed a continuous layer connecting native vegetation patches. This process fueled an unprecedented number of fire ignitions and acres burned in the arid spring and summer of 2005. Across the Lower Sonoran Desert Ecoregion, the fires of 2005 accounted for 89% of the total area burned between 1989 and 2010 (Gray et al. 2014). In interviews conducted for this study, managers of burned sites reported very poor recovery, with some locations remaining barren nearly 14 years after the fires.

Although the winter of 2004-2005 was exceptional, this pattern is likely to arise in the Sonoran with greater frequency as occasional wet winters are followed by ever-lengthening hot, dry pre-monsoon summers (Jardine et al. 2013; Gray et al. 2014). [Indeed, the pattern appeared once more in the winter of 2018/2019, near the end of the project period for the research described here, and in early summer 2019 the region experienced the Woodbury Fire, unofficially considered the fifth largest fire in state history, and the largest recorded fire to burn in non-fire-adapted Sonoran Desert vegetation (https://ktar.com/story/2629443/woodbury-fire-passes-willow-to-become-states-5th-largest-fire/).] Even if the wet winters arise only once every 20 years, the emergence of this novel fire regime may result in significant damage across the sensitive Sonoran Desert region.

In a study funded by the Department of Defense Strategic Environmental Research and Development Program (SERDP) and conducted between 2010 and 2015, members of our team explored the relationship between environmental factors and the occurrence of recorded fires across the Sonoran Desert Ecoregion between 1989 and 2010 (Dickson et al. 2016). The study mapped habitat requirements and predicted distributions of non-native species contributing fine fuels across the study region (Dickson et al. 2016). Additionally, researchers modeled occurrence of fires with respect to landscape-scale and remotely sensed variables and found that large (>20 ha) fires were significantly associated with maximum Normalized Difference Vegetation Index (NDVI) for the year of fire as well as with low values of elevation and road density (Gray et al. 2014). NDVI is a Landsat remote sensing imagery-derived variable that acts as a proxy for quantity of primary production, and can be thought of as vegetation greenup as recorded by satellite. This study thus identified the combination of year-to-year climate factors, environmental conditions, and social factors (distance from roads) as determinants of large fire occurrence. Our research followed from this previous study to examine the combined effects of social and ecological factors on fire response and recovery. Definition and components of resilience in the social and ecological spheres

Key to this research is the concept of resilience, a multi-faceted system characteristic that has been defined in various ways within different streams of literature. The central concept in resilience is that of the stable state, the set of characteristics that describe a given focal system prior to perturbation. When a perturbation occurs, the quantity of disturbance that can be absorbed by the system before it transitions to a new state is a metric of its resistance (Carpenter

4

et al. 2001; Lake 2013). If a system does transition to a new state, the quantity of disturbance after which the system can recover to its original state, without management assistance, is a metric of its resilience (Walker et al. 2004; Lake 2013). A system that demonstrates low resilience in its pre-disturbance state may require active restoration in order to regain its historical characteristics (Lake 2013). The disturbance event establishes a threshold of change that the system must overcome in order to recover, and the height of this threshold varies by disturbance type, intensity, and frequency, as well as by inherent system characteristics and starting point (Suding et al. 2004). Although resilience is typically considered beneficial, the definition is not synonymous with healthy. It is possible, for example, for a system in its historic state to exhibit low resilience, with a low change threshold, and thus to be easily perturbed from that state; an example might be a sensitive native grassland perturbed by fire followed by biological invasion. The same system may exhibit high resilience in its novel state, as an invaded grassland with a high threshold for change, that tends to retain its new invaded community indefinitely and tenaciously (Standish et al. 2014).

The concept of resilience frames our understanding and of ecological system change and recovery as well as the components (such as high biodiversity) that make a system likely to recover after disturbance. It guides management aimed at retaining native species or sensitive habitats and assists in prioritization of restoration resources. (An inherently resilient system, for example, will recover on its own from many disturbances and may not require large investment of restoration funds or effort.) Managing “for resilience” is a mandate of many public land management agencies in the US (Aslan et al. 2018). However, ecological resilience does not exist in a vacuum but is rather fundamentally influenced by and influences the social realm, as well. Social systems may affect resilience by dictating management values and definitions of success and hence activities and priorities, defining desired ecological state characteristics, imposing stressors and disturbance events, enabling or preventing management interventions of certain types and certain scales (Aslan et al. 2018). Additionally, social systems exhibit their own degree of resilience, retaining or losing certain characteristics in the face of disturbance. Once again, in social systems resilience is not always positive: a system that retains social inequalities, for example, may be highly resilient and difficult to change (Adger 2000).

Studies of ecological resilience examine factors that influence the amount of change required for an ecosystem to reach a point of no return (a.k.a. “tipping point”) at which it transitions to a different type of system (for example, a desert grassland to a mesquite-dominated woodland) (Ruesink et al. 2006). Deserts such as the Sonoran may be at risk of such changes, which in these systems manifest as loss of perennial vegetation, for example due to fire, leading to increased desertification and dominance by invasive plant communities which are, in turn, more fire-prone (Rietkerk et al. 1997). By understanding the invasion-fire cycle in the Sonoran and modeling future vegetation and fuels occurrence patterns across the landscape, we can identify locations likely to experience tipping points and. This information will be crucial for managers if future fire regimes will make it hard for them to achieve their management objectives (Allen et al. 2002).

It remains unclear how resilient both ecological and management systems will be to anticipated environmental changes (e.g., Elmqvist et al. 2003; Millar et al. 2007; Nelson et al. 2007). Current and projected patterns in the Sonoran include significant changes in both abiotic conditions (e.g., temperature and moisture availability) and in biotic conditions via species range shifts, single-species extinctions, and phenology shifts (Parmesan 2006; Araújo & Luoto 2007). In much of the arid West, warming temperatures foster invasive species spread, elevate

5

flammability of biomass, and reduce recruitment of native species (Shafer et al. 2001; Westerling et al. 2006; Jardine et al. 2013), exacerbating the conditions that led to the high fire year in 2005. In the face of these changes, key ecosystems (and associated dominant vegetation) may transition to alternative ecosystem types. Management can help boost resilience and prevent these changes by restoring sites that have been disturbed (Perrings & Walker 1997). Evaluating and anticipating ecological resilience across a landscape is a critical first step in facilitating management that can contribute effectively in this way.

Importance of interdisciplinary approaches to resilience research

Resilience as a concept is inherently interdisciplinary. Ecological factors of importance in many systems include temperature and precipitation shifts, biological invasions, genetic diversity, environmental heterogeneity, and traits of native species and communities. Social factors include flexibility of management tools and responses, availability of resources, constraints such as regulations and social pressures, and ongoing disturbances such as fire

ignitions, road construction, ORV use, etc. Whether a site is resilient to disturbance such as fire will fundamentally depend on both the starting-point ecological conditions and the social conditions at that location. For example, locations with high habitat connectivity, genetic diversity, topographic heterogeneity, flexibility of management choices, and abundant management resources will demonstrate higher resilience than locations lacking these factors. If a site is high in ecological resilience factors but not social resilience factors, it may return to its former state after disturbance without social intervention. However, if a site

is low in ecological resilience factors but high in social resilience factors, social interventions might help the system revert to its previous state following perturbation.

In spite of the inherent interdisciplinarity of resilience studies in these social-ecological systems, studies measuring and predicting resilience at a landscape scale rarely incorporate both social and ecological factors (Folke 2006). Methods of data collection, quantification, and scales vary between social and ecological science, making it difficult to match data from the two realms and thus arrive at truly interdisciplinary assessment of resilience. We here describe a study focused on ecosystem management as a key area of intersection between social and ecological realms, and an area in which both are essential to the trajectory of a system. Our methods were intended to distinguish socio-ecological turning points vs. tipping points in the Sonoran Desert Ecoregion (Fig. 1). The term “tipping point” refers to the point in the trajectory of a system at

Figure 1. Synthesis conceptual model, displaying the influence of fire regime change and management response on ecosystem resilience. Red squares = alternative management strategies for implementation at ATPs, or adaptive turning points. Under medium fire regime change, alternative strategies may boost ecosystem resilience under various levels of jurisdictional adaptability. Under high fire regime change, few alternative strategies may be available, and high jurisdictional adaptability may be crucial.

6

which the state of a social-ecological variable changes, or, from a management perspective, the point at which management objectives have become unachievable. The term “turning point” refers to the set of conditions under which management must change for objectives to be achieved (after Lenton et al. 2008; Werners 2013). Thus, a site may reach a turning point, via perturbation, prior to a tipping point; a shift in management activities to achieve current management objectives may be required at a turning point, but those current objectives can be achieved if that shift in activities occurs. After reaching a tipping point, however, it will not be possible to achieve original objectives under any activity. Our research approach distinguished activities from objectives and aimed to thus evaluate separately where sites exhibit resistance to state change, due to the present of socio-ecological alternatives that enable existing objectives to be reached; and where sites exhibit resilience, due to the ability for social management systems to adapt to fundamental changes in state as defined by the set of management objectives for a system.

In the socio-ecological literature, an adaptation turning point (ATP) describes the amount or threshold of environmental change that renders current management tactics unable able to meet objectives (Kwadijk et al. 2010; Werners 2013). By examining scenarios of environmental change based on quantitative forecasts, managers and decision-makers can determine whether they would be likely to alter their tactics to meet objectives under each scenario. By recognizing anticipated ATPs, managers and (by extension) policy-makers can better prepare for the likelihood that they will have to adjust their fire management strategies in order to meet objectives. Preparing for likely future turning points requires identifying those management objectives that are currently in use in a given location, and understanding likely response to environmental change requires identifying adaptation options available to the stakeholder (Berkhout et al. 2014). These options may include measures that have been attempted by a stakeholder in the past, or may include measures that have never been attempted or feasible. Stakeholder and decision-maker preferences may change over time, as a result of or independently of environmental change (Tompkins et al. 2008). The first step to using an ATP framework is to present likely environmental change scenarios to managers, with reference to their needs and concerns. Thus, bidirectional engagement is necessary, with researchers presenting likely environmental change scenarios and stakeholders considering and discussing the likely impact of each scenario on their management objectives and activities. Following that social system data collection process, stakeholders can evaluate which scenarios may require responses across jurisdictions and at a regional scale (Shackley & Deanwood 2003).

In our study of social-ecological resilience across the Sonoran Desert Ecoregion, we used a combination of ecological modeling and social system data collection to identify ecological and social sources of resilience. In this study, a social-ecological state refers to the set of ecosystem management objectives for that system. Our aim was to identify locations across the Sonoran where post-fire resilience is likely to be high and thus interventions such as restoration of low priority, as well as sites where post-fire resilience is likely to be low and thus active restoration might be more important. This work followed a study of fire regime change and its drivers in the Sonoran, performed by several members of our team, and together the two studies produce maps of fire risk and fire resilience across the region.

MATERIALS AND METHODS

Prediction of future fuels and fire regimes across the Sonoran Desert Ecoregion

7

We forecast vegetation abundance, including both native and non-native fuels, across two 20-year averages, 2015-2034 and 2035-2054, and under a “business as usual” emissions scenario, representing the effects of greenhouse gas emissions by their equivalent radiative forcings of 8.5 W/m2 in the year 2100 (RCP 8.5). We based these models on multiple-year Normalized Difference Vegetation Index (NDVI) projections. The annual maximum NDVI is derived from satellite imagery at 30-m resolution and is a good indicator of the available fuel in any given year (e.g., Burgan et al. 1998). In addition to the maximum NDVI of the current year, the maximum NDVI from the year prior can also give an indication of fuel that may have dried out but is still available to burn in the current year. Amount and location of fuels can be highly variable from year to year. In previous fire models for the study region, we created map-based ‘scenarios’ of fire risk that have been observed in the recent past. Fire risk was defined as the probability of a large fire (>20 ha) if an ignition were to occur, from either natural or human sources. Fire risk changed from year to year based on the maximum annual NDVI of the current and prior year, but also incorporated other landscape characteristics, such as topography and density of roads, that influence fire spread and the speed and success of fire suppression efforts and remained fixed from year to year. The scenario of high fire risk was derived from maximum NDVI following the high moisture period of 2004 and 2005, as well as landscape characteristics. As part of our previous work in the region, we also derived predictive models of invasive plant species. In the Sonoran Desert region, unlike in fire-adapted systems elsewhere, wildfire potential is not limited by temperature or humidity, meaning that summers are usually hot and dry enough for wildfires to burn. Rather, wildfire potential is limited by the amount and continuity of fuel, which depends largely on the growth of annual grasses and forbs. Invasive species represent a novel annual fuel source in the region and also contribute to the interannual variability in fuel abundance. While temperatures limit the overall range of invasives in the region, results from previous work showed that precipitation is the strongest influence on invader abundance (Gray et al. 2014; Wang et al. 2014; Gray & Dickson 2015). Climate projections for this area show increases in warm and cool season temperatures (increasing the range of some species in the region), but decreases in the abundance and frequency of cool season precipitation. At the same time, climate projections for this area also show an increase in cool season precipitation variability, so extreme precipitation events could become more extreme. It only takes one anomalous year of high precipitation and fuel growth, such as 2005, to introduce a lot of fire on the landscape, weaken native vegetation, and give a competitive advantage for invasives to spread. Here, we defined fire risk as the probability of large fire if an ignition were to occur, and used the current and prior year maximum NDVI as predictors of annual fire risk. For this effort, we also included climate variables (total winter precipitation, winter mean daily minimum temperature, and fire season mean daily wind speed, maximum temperature, and humidity) to predict past and future fire risk. These were dynamic model variables that change from year to year. The remaining predictor variables in the model were assumed to remain fixed from year to year and included road density, distance to urban development, surface heat load index, topographic roughness, elevation, aspect, and slope. To forecast fire risk into two future time periods, it was first necessary to first forecast the dynamic predictor variables. We used data at 4-km resolution, downscaled from 12 Global Climate Models, to forecast the meteorological predictors. To forecast NDVI into the future, we used 32 years of historical precipitation and

8

NDVI data to statistically relate cool season precipitation to the subsequent maximum annual NDVI. Predicted NDVI values were plotted against observed NDVI values and averaged over the whole study area. With the results of that statistical relationship and using available forecasted precipitation data, we were able to forecast annual maximum NDVI into the future. To build models of fire risk, we used points that either burned historically in a large fire or points that burned but did not become a large fire and related these to the predictor variables. We included the entire US Sonoran Desert ecoregion, so that we had many historical fires to draw from, and created 10 of these independent, random datasets from which to build 10 independent models. We used those 10 independent datasets, and accompanying predictor variables, to build 10 independent models that predict the probability of a large fire. The annualized estimate of fire risk was averaged over 10 datasets and 12 GMs, and includes the variability resulting from independent fire risk models and variability resulting from future climate projections. This effort focused on the maximum fire risk (i.e. an extreme rather than the mean) in each of three time periods. This decision was based on the concern that even a single, extreme climate and fire year over a 20-year period may cause changes in land cover, and can help focus adaptation planning. While some areas may not show significant change in maximum risk from the past to the future, even a slight increase would indicate, for instance, that risk in a future period may surpass what was experienced in 2005. While this may not represent a large change from the past, it may indeed represent a large change over the conditions a given manager has experienced in their tenure (Fig. 2). Stakeholder engagement: implications of these changes for management programs

We performed social system data collection via a three-step approach. First, we performed a needs assessment with a subset of jurisdictions in the Sonoran Desert Ecoregion study area. We met with stakeholder managers from US National Forest, Army, wildlife refuges, National Park Service, and Bureau of Land Management jurisdictions. During each of these meetings, we performed open-ended interviews, asking managers about fire concerns and management objectives and also asking them to recommend additional managers in the region that should be included in the study. Using the insights we obtained from this needs assessment, we next designed a spatial participatory mapping exercise for use in Workshop 1 data collection. For this workshop, we assembled stakeholders at the Sonoran Desert Conference Center in Ajo, AZ, in March 2016. We presented to these stakeholders background information about fire in the Sonoran Desert, drivers and consequences of fire regime change, and social-ecological resilience. We then distributed to them the maps of projected large fire risk we had developed. Stakeholders were provided with study area-wide maps for context as well as zoomed-in maps of their own jurisdictions. Each of the jurisdiction-scale maps was marked with 10 randomly-generated points, obtained via the ‘Create Random Points’ tool in ArcGIS. Participants interacted with the jurisdiction-scale maps in four ways: First, for management objectives directly related to fire (e.g., fuels removal or invasive plant suppression), participants examined each random point in turn and described the current management objectives relevant to that point. They then described current management activities occurring at that point and stemming from each objective. They recorded these objectives and activities on a pre-developed table. The table included a Likert-scale assessment of, first, the likelihood that current activities will continue to meet current objectives (scale 1-5, very unlikely to very likely), and, second, the likelihood that alternative activities can be selected to meet current objectives, as needed (also scale 1-5). The participants additionally identified alternative objectives that might be adopted in place of

9

current objectives, if environmental change drives the system to a tipping point and current objectives can no longer be achieved under any activity. Second, the participants were asked to repeat this process for management objectives not directly related to fire (e.g., grazing or recreation), since such objectives can still be affected by fire regime change. Third, participants were asked to outline additional areas on their maps for which projected fire regime change indicated some need to alter fire-related objectives or management activities. These areas differed from the random points in that they were participant-driven and indicative of locations of particular interest or concern to the participants themselves and thus certainly biased toward areas of high change and high value. The participants completed the same table of information for these locations and also gave the research team the hand-drawn maps, so the outline of indicated areas could be transferred as a polygon into the project GIS. Fourth, the participants completed this same participatory mapping process to designate areas of concern for objectives and activities not directly related to fire.

Following the

workshop, we coded all management activity and objective information into the GIS, drawing into the digital platform the polygons of areas identified by participants as important and attributing both random points and polygons with the objectives and activities relevant to those locations as well as the participants’ reported level of optimism that activities could continue to achieve objectives, new activities could be employed, or new objectives could be adopted. We were then able to use these data to develop a spatial model of projected occurrence of thresholds in fire management, given the fuels scenarios presented at the workshop, that are likely to require a change in management objectives. Furthermore, since participants identified management changes (if any) that they would likely implement to continue to

Figure 2. Maps of study area showing the 3 different fire risk models in panels a, b, c.

a.

b.

c.

10

meet fire management objectives, we could model projected spatial distribution of these turning points.

Finally, the participants were asked to complete a paper survey regarding their experience with past environmental change, fire, adaptation turning points, and adaptation tipping points. Survey questions were designed to evaluate the adaptive capacity of each jurisdiction, by assessing whether and how readily the jurisdiction has demonstrated adaptability in management previously, as well as what constraints might limit a jurisdiction’s ability to respond to changing conditions. In all, we obtained ATP and adaptive capacity information from 25 jurisdictions, accounting for 79% of the study area. All participant jurisdictions were governmental entities (Tribal, federal, state, or county lands), since private landowners represent an extremely small proportion of the study area (<8 percent) and, in the study area, are known to be exceptionally difficult to obtain contact information for. Following analysis of these social data, we reconvened stakeholders for two results engagement events: an online webinar in May 2019, supported by the JFSP Fire Exchange Network and Southwest Fire Science Consortium, in which we presented results and fielded questions from nearly 30 participants, and an in-person June 2019. Workshop 2 with a small focus group (6 participants) with which we discussed priority follow-up work and next-step research questions. Integration of social and ecological data to develop spatial models and maps of fire resilience across the landscape Interviewee-reported management objectives and current management practices were grouped into 9 (objectives) and 12 (practices) broader categories, each with a definition (Table 1). Objective and Management Practice categories). Statements from interviews and workshop materials were then coded into these categories. In order to assess inter-coder reliability, two coders first piloted the codes together for several randomly chosen jurisdictions. The coders then independently coded all statements for remaining jurisdictions. We calculated percent agreement for each code. Where differences existed between coders, the project team discussed the differences and came to consensus on the final code for each statement. Table 1. Management objective and current management practice categories and descriptions. The descriptions listed are groups of keywords that managers used in their interview responses.

Management Objective Description Conservation habitat/species protection, ecological integrity, conservation,

wilderness management, restoration, resource managed fires Resource Extraction mining Grazing grazing Agriculture agriculture General Fire Suppression fire suppression, roadside fire reduction, lightning strike monitoring Human/Infrastructure Protection infrastructure protection (private and commercial plus roads,

powerlines, visitor centers, and campgrounds), human safety (nearby communities), air quality, communications, water management (flood control)

Military military (training, etc.) and border activities (patrol, enforcing border laws, controlling illegal border activity, border patrol cooperation)

Recreation hunting, viewshed protection, infrastructure, camping, scenic driving, general

11

Cultural Resource Protection sites and infrastructure, historical sites and buildings, saguaro harvest, traditional uses of area and plants

None manager lists no management objective for that area or lists a historical (but not current) objective

Current Management Practice Description

Fire management general suppression, fire breaks, fuel breaks, access/pre-positioning for fire crews, let burn/monitor, prescribed burns/management fires/fire regime, fuel load monitoring, road and trail closures, shooting restrictions, burning/fires restrictions/bans, restricting access to culturally important sites, increasing patrols, hazard mitigation

Preventative vegetation/invasives removal for fire (Category is for preventative veg/invasives removal for fire.) general thinning, site-specific fuel removal (around infrastructure and roads), herbicide or manual removal invasives/grasses removal, mastication, tamarisk removal, general invasive grasses removal, buffer creation, mowing, grazing for fuel management, grazing for invasives/grasses removal

Grazing leasing for grazing, permits for grazing Outreach and education education, PSAs Native Species Management monitoring for conservation management (invasive/exotic species

including feral horses, native vegetation long-term, wildlife populations (like pronghorn and yellow-billed cuckoo), endangered species, birds (eagles, etc.), rare species, riparian habitat), restoration (native spp. seeding, of general spp./habitat, of spp. for cultural use), maintenance of vegetation and native plants (cultural uses, medicinal)

Crops leasing - crops, irrigation - crops Resources from partnerships funding, partnerships/agreements, general, partnering with local fire

departments Leasing and Permits other (not crops or grazing) leasing and permits, "improving

stipulations"/guidelines/rules for mining Infrastructure development general irrigation (not for crops), improved fencing, improved

wells/other water infrastructure, firewise campgrounds - infrastructure and training

Soil and watershed protection erosion prevention/retention for flood control or other reasons

Cultural and social resource protection maintenance of recreational sites, cultural resources, cultural sites, native plants (for cultural uses, medicinal reasons), preventative maintenance

Recreation hiking, camping, hunting/shooting, OHV/ORV; if 'Recreation' was listed as a management objective, this category was used in current management.

None manager lists no current management practices for that area

Spatial data processing For the interview data, we created polygons in ArcGIS of the same size, shape, dimensions, and location as the ones land managers hand drew on paper maps and paired them with interview responses, including Likert scale responses (1-5, very unlikely to very likely). We also paired random points with interview responses. Because land managers likely could not identify the precise location of each randomly selected point, we gave all points a 1 km radius buffer. We created a single layer for all polygon and point responses. For the fire suppression management objective, we selected only the polygons and points for which land managers identified this management objective. We then rasterized the selected polygons and points, coding areas with the management objective as a “1” and areas without the management objective as a “0”. We did the same for the fire management category of current management practices. Some geographical areas were covered by multiple organizations. In areas of overlap, areas with higher levels of constraint on fire management actions (such as due to policy, land use

12

classification, or agency mandate) took precedent. We then imported these rasters as a grid of points into R. We assigned survey responses at the jurisdiction level because managers were answering questions regarding perceptions of their entire organization, not specific geography within the organization’s land base. We coded Likert scale responses as 1-7 (strongly disagree to strongly agree). Ecological and social risk We defined ecological fire risk as the predicted risk of large fire within the period 2015-2035, as defined in the Methods, above. We defined social fire risk as a combination of four factors: whether or not a given location (as defined at the point or polygon level) reported 1) fire suppression as a management objective and 2) fire management as a current management practice, and 3) and 4) the survey responses (strongly disagree to strongly agree) to two statements about organizational resources. The survey statements were worded as follows: “My organization has sufficient resources & personnel to manage fire and fuels on a day-to-day basis” and “My organization has sufficient resources & personnel to manage fire and fuels during fire incidents.” The responses to each of these questions were scaled from 0 to 1 and summed. Social risk statistical methods Fire management and fire suppression were each scored 1 (for fire management or fire suppression happening at that grid cell) or 0 (for fire management or fire suppression not happening at that grid cell). Because the central points of grid cells did not capture all management or suppression happening within the full grid cell, we used a search radius of the 5000 m (the size of the grid cell), and included all management or suppression within that grid cell. This means that one grid cell could have three or four entries, each with a 0 or 1 management or suppression ‘state’ possible. Using multiple states per grid cell, where relevant, had the advantage of providing a more accurate average relationship between average fire management or suppression and our predictor variables. No state or military jurisdictions have fire suppression, which translates to no 1’s being fed into the Bernoulli regression which in turn causes converge issues. To overcome these convergence issues, we randomly selected 2 of 32 state grid cells (<10%) and 60 of 635 military grid cells (<10%) in which we reassigned the 0’s to be 1’s. The local jurisdictions only had fire suppression at 2 of the 62 grid cells, so we randomly selected 4 (for a total of 6, <10%) in which we also reassigned the 0’s to be 1’s. These 1’s keep the average fire suppression across all state, military, and local grid cells near 0 but make it possible to achieve convergence within the model. Fire management and fire suppression were modeled as Bernoulli-distributed. Our regression included a logit link and related distance to roads, distance to nearest privately-held land, elevation, latitude, and longitude to average fire management or suppression by jurisdiction type (as a random effect). All intercepts (the by-jurisdiction type random effect) were drawn from a common normal distribution. This common normal distribution pulls individual random effects closer to the common mean and reduce variance (O’Hara and Sillanpaa 2009). Roads were listed within interview responses as a major source of ignitions and as such a place where fire suppression and management were sometimes focused. Protection of nearby communities was listed within interview responses as a reason for fire management and fire suppression. We used distance to nearest privately-held land as a proxy for nearby communities. Lower elevations

13

within the Sonoran Desert ecoregion are at higher risk of large fires, following precipitation-driven pulses of biomass growth, due to more rapid drying (and on average lower precipitation) at lower elevations relative to higher elevations (Gray et al. 2014). Furthermore, precipitation increases within the ecoregion along a gradient from west to east and from south to north, resulting in increased vegetation cover (Sponseller et al. 2012). The relationship between distance to road, distance to nearby communities, and elevation and fire management or suppression were fit across all jurisdiction types because we first ran the model with a slope term for each jurisdiction type for all of these predictor variables but found no evidence that the relationship differed across jurisdiction types. We modeled the relationship between latitude and longitude and fire management or suppression by jurisdiction type because some different jurisdiction types displayed a relationship between latitude or longitude and fire suppression or management and some did not. We fit both analyses in JAGS (Plummer 2003) via the R packages ‘rjags’ version 4-8 (Plummer et al. 2018) and ‘R2jags’ (Su and Yajima 2015). We used uninformative priors for all stochastic nodes. We ran three MCMC chains for each analyses, for 500,000 iterations, and we used the Gelman-Rubin convergence statistic to check for convergence between and within all chains by ensuring that its values were ≥ 1 and <1.1 (Gelman and Rubin 1992). Using the fit linear regression and the values of the predictor variables for each cell, we predicted average fire management and suppression values for each cell across the study area. For both the fire management and suppression prediction rasters, we replaced predicted values with the known ‘states’ for each cell in which we had reported fire management or suppression ‘states’. The final fire management and suppression rasters were composed of predicted mean values (0-1) from our linear regression for cells which had no information reported by jurisdiction managers and known ‘states’ (0 or 1) for cells for which jurisdiction managers had provided information. To match the survey raster, we rescaled the resulting 5000 m x 5000 m grid cell size to 1000 m x 1000 m. Ecological resistance and adaptive capacity We separated ecological resistance and adaptive capacity into categories based on the ability to continue in the face of disturbance or generally persist over time (resistance) or the ability or capacity of a system to modify or change its characteristics or behavior so as to cope better with existing or anticipated external stresses (e.g. the system may be a different state, but it maintains key/desired ecosystem processes; adaptive capacity). To create an ecological resistance raster, we created metrics for topographic diversity, geophysical diversity, vegetation diversity, and water availability (Table 2). To create an ecological adaptive capacity raster, we created metrics for habitat connectivity (sensu Theobold 2013), species richness, and human modification (sensu Theobold et al. 2012) (Table 3). To create ecological rasters that could later be combined with social rasters, we summed these indicators within a single location for ecological resistance and ecological adaptive capacity, respectively, across all locations in the Sonoran Desert. Table 2. Indicators used to compile an ecological resistance map of the Sonoran Desert. Each indicator is paired with a description of how the metric was calculated and the scale used. Indicators were scaled from 0-1, with 0 being the lowest concern for managers and 1 being the highest concern for managers.

14

Indicator Metric Scale

Topographic diversity

Standard deviation of slope from DEM (min = 0, max = 1) using moving window (focal statistics) from 30m base map to 270m windows.

High standard deviation= 0 Low standard deviation = 1

Geophysical diversity

Shannon-Weaver Equitability Index (SWEI) at multiple spatial scales, equivalent to average sizes (1.2 – 115.8 km radii) of HUC 4-16 watersheds. Indices derived at multiple scales were then combined to produce a single multi-scale index at 30-m resolution.

High SWEI = 0 Low SWEI = 1

Vegetation diversity

Count of unique threatened, endangered. and sensitive (TES) classes Low SWEI = 0 High SWEI = 1

Water availability

Path distance from surface water features (springs, seeps, perennial rivers, perennial lakes and ponds from the National Hydrography Dataset (1:24,000; USGS 2009), with the path surface defined by the DEM layer.

Low Distance = 0 High Distance = 1

Table 3. Indicators used to compile an ecological adaptive capacity map of the Sonoran Desert. Each indicator is paired with a description of how the metric was calculated and the scale used. Indicators were scaled from 0-1, with 0 being the lowest concern for managers and 1 being the highest concern for managers.

Indicator Metric Scale*

Habitat connectivity

Sensu Theobald et al. (2012): resistance values for least-cost calculations based on the inverse of a landscape “naturalness” value.

High HC = 0 Low HC = 1

Species richness Mean of rarity-weighted species richness High SR = 0 Low SR = 1

Human modification

Sensu Theobald (2013): a metric that incorporates development, agriculture. energy production and mining, transportation and service corridors, biological resource use, human disturbance, natural system modification, invasive species, and pollution.

Low HM = 0 High HM = 1

Social resistance and adaptive capacity We defined social resistance as the ability for management objectives to continue to be met using current management practices. We pulled data from both the interviews and the survey that addressed social resistance as per our definition. From the interviews, we used the answers, very unlikely to very likely, to the question “[Given predicted future fire regimes] How likely are you to continue to meet current management objectives?” From the survey, we used the responses, strongly disagree to strongly agree) to the statement “In my organization, we expect and try to prepare for ‘surprises’ in ecosystem behaviors.” We defined social adaptive capacity as the ability for management objectives to continue to be met by changing current management practices. Like for social resistance, we pulled data from both the interviews and the survey that addressed social adaptive capacity as per our definition. From the interviews, we used the answers, very unlikely to very likely, to the question

15

“[Given predicted future fire regimes] How likely are you to change current management practices in order to be able to meet current management objectives?” From the survey, we used the responses, strongly disagree to strongly agree) to the statement “I have seen my organization successfully change management strategies when necessary.” Social resistance and adaptive capacity statistical methods - Interview Likert responses to “In this location, how likely is it that current activities will achieve current management objectives under future fire regimes?” and “If there are management alternatives, how likely is it that they would be adopted?” were scored from 1-5 (where 1=very unlikely, 5=very likely). Because the central points of grid cells did not capture all responses within the full grid cell, nor did it capture multiple overlapping responses in space, we used a search radius of the 5000 m (the size of the grid cell), and included all Likert responses within that grid cell. This means that one grid cell could have multiple entries, each with a 1-5 Likert response possible. Using multiple responses per grid cell, where relevant, had the advantage of providing a more accurate average relationship between average Likert response and our predictor variables. For both questions we modeled Likert responses with a multinomial distribution, using stepwise logistic regression (sensu Field et al. 2017). Our stepwise logistic regression included an intercept for each Likert response category (1-5), a term for each jurisdiction type, and projected future fire at a given grid cell as a predictor variable. We included projected future fire as a predictor variable because managers were asked to look at the projected future fire regime map while responding to the interview questions. For the question about the likelihood of current activities achieving current management objectives, we combined state and local jurisdiction types because 1) neither had responses within many grid cells (making each potentially unidentifiable separately) and 2) state and local responses were similar. We fit both analyses in JAGS as described above, with the same number of chains and iterations, and we assessed convergence in the same manner (Plummer 2003, Plummer et al. 2018, Su and Yajima 2015, Gelman and Rubin 1992). Using the fit linear regression and the values of the predictor variables for each cell, we predicted average Likert response values for each cell across the study area, for both questions. Unlike the fire suppression and fire management rasters, we did not replace predicted values with known values. Multiple areas across the landscape had overlapping multiple responses and this predicted value served as an average response in those areas. To match the survey raster, we rescaled the resulting 5000 m x 5000 m grid cell size to 1000 m x 1000 m. Social resistance and adaptive capacity statistical methods - Survey Because 1) the survey responses were at the level of the whole jurisdiction (no variation across the landscape within jurisdiction) and 2) we only received 23 survey responses distributed among four jurisdiction types, we could not do a landscape-level analysis nor could we regress predictor variables against survey responses by jurisdiction type. To be able to assign survey response scores to non-respondant jurisdictions, based on jurisdiction type, we averaged Likert responses (where 1 = Strongly Disagree to 7 – Strongly Agree) across each jurisdiction type (federal, local, military, and tribal). We spatially assigned respondant jurisdictions their response Likert score to their entire jurisdiction and non-respondant jurisdictions the average Likert score for their jurisdiction type to their entire jurisdiction. Because we analyzed state and local responses together for the interview responses, we used the average local responses for any state

16

jurisdictions. We rasterized this shapefile to a 1000 m grid, using the score in the center of the grid cell as the assigned grid cell score. Creating socioecological risk and resilience maps We scaled all ecological and social measures on a 0-1 scale, with 0 being the lowest concern for managers and 1 being the highest concern (Dressel et al. 2018). Because our social layers were scaled in the opposite way by virtue of how we converted yes/no responses to management objectives and practices as well as Likert responses, we inverted all of the social layers before combining them with the ecological layers. To get total risk across the Sonoran Desert, we summed the ecological and social risk layers, and rescaled them to 1, giving equal weight to each. We did the same for total resistance and total adaptive capacity. When rasters were at different resolutions, we resampled them to make all rasters equivalent in resolution to the social rasters. To get a measure of total resilience across the Sonoran Desert, where ‘resilience’ is defined as the combination of resistance and adaptive capacity, we summed the rasters for total resistance and total adaptive capacity and scaled the subsequent raster from 0-1. We performed a coarse check of sensitivity to the social component within the total resilience by multiplying by four (the number of rasters that comprise total resilience) and subtracting out the social resistance and adaptive capacity rasters, which resulted in an ecological resilience raster. We then re-summed and re-scaled the total resilience raster by up-weighting or down-weighting the social resilience (summed social resistance and adaptive capacity) raster. We compared the resultant, scaled total resilience values for the socially up-weighted, socially down-weighted, and equal weights rasters. Analysis of non-spatial patterns of management objectives and activities

In addition to landscape-scale social-ecological resilience assessment, our data collection techniques enabled exploration of management objectives and activity patterns disassociated with spatial patterns across the study area. We analyzed non-spatial patterns in participant responses using interview data associated with randomly generated locations across the study area. At each location, land managers reported their management objectives, activities to meet those objectives, and the likelihood of meeting objectives. For nominal response variables, we scored each spatial point as a trial with a binary outcome (i.e., ‘1’ if the objective or activity was reported for that area, and ‘0’ if not). We compared associations between land management agencies, objectives, and activities using Generalized Linear Models (R-package ‘glm’) fitted with a logit link function to specify a binomial error distribution. We performed pairwise post-hoc comparisons with a Bonferroni correction to account for multiple comparisons (R-package ‘emmeans’).

We examined manager-reported likelihood that current activities will continue to meet current objectives into the future, and also that alternative activities or alternative objectives will be able to be adopted. From these data, we were able to summarize the total diversity of objectives and activities reported by each jurisdictional type and compare levels of manager-reported optimism about the success and continuity of each activity or objective. We were also able to examine proposed activities that managers had not yet attempted but suggested changing conditions might inspire; we reviewed published literature for the use of these proposed methods in arid systems in order to determine whether such literature indicated that the proposed methods were likely to be successful.

17

We coded jurisdictions into the following hierarchical bins, in order to facilitate statistical analysis and comparison among jurisdiction types: Tribal (3 jurisdictions), federal (14 jurisdictions), state (2 jurisdictions), county (3 jurisdictions), and military (3 jurisdictions). We collapsed reported objectives into 9 categories and activities into 13 categories to provide sample sizes of each sufficient to enable comparison across jurisdiction types (Fig. 3).

We compared manager-reported optimism levels for both objectives and activities to determine whether managers felt more confident in their ability to achieve some objectives or to continue to carry out certain activities than others. Data did not meet assumptions of normality (based on quantile-quantile plots, used to evaluate normality due to limitations in sample size; Wood 2010), so we used nonparametric Kruskal-Wallis tests to examine differences among objectives and activities and employed Dunn’s multiple comparisons posthoc test to determine which pairs of objectives or activities differed significantly. Additionally, since federal jurisdictions (Bureau of Reclamation, Bureau of Land Management, National Park Service, National Wildlife Refuges, and military) were so abundant in our dataset compared with other hierarchical categories, we performed simple Student’s t tests to determine whether federal vs. non-federal jurisdictions varied in reported optimism that objectives would be achievable and activities successful. Development and analysis of adaptive capacity survey We developed a survey to explore manager perceptions of organizational adaptive capacity (Appendix 3). Survey items focused on organizational innovation and uncertainty, adaptive management practices, and organizational constraints to change. Scaled items were adapted from Lockwood et al. (2014) and quantified on a 7-point Likert scale. Additional questions were shaped by the Department of the Interior Adaptive Management Guide (Williams et al. 2009). The survey was piloted with federal fire managers and refined. The survey was provided to representatives of 93 organizations within the study area. Participants received the survey at workshop 1, and we followed up by emailing non-attendees a link to an electronic version of the survey, with three follow-up reminders. Survey responses were kept confidential and identifying characteristics were removed. Likert scale responses were entered into ArcGIS attribute tables for each jurisdiction and included in spatial modeling. Survey responses were summarized in SPSS v. 25 (I/BM Corp. 2019). In order to calculate summaries, values were assigned to categorical responses sequentially such that Strongly Disagree was scored as 1 and Strongly Agree as 7. Open ended responses were analyzed inductively, first grouped using open coding, and then systematically themed and categorized (Bernard 2011; Corbin and Strauss 1997).

RESULTS AND DISCUSSION

Outcomes of stakeholder engagement Participant stakeholders were affiliated with government at various levels—Tribal, federal, state, and county. These land managers provided spatially-explicit objectives and activities at both random points and manager-generated polygons across their jurisdictions. Furthermore, they assessed their optimism that activities would continue to meet objectives or that new activities or objectives could be adopted. This allowed evaluation of patterns of optimism across activity types and manager categories. On average, Likert-scale (options ranging from 1-5) optimism levels were highest for agriculture and general fire suppression as

18

objectives (that is, managers reported that those objectives had the highest likelihood of being achieved) (Fig. 3a). By contrast, cultural resource protection was considered least likely to be achieved under changing environmental conditions (Fig. 3a). Among activities performed to meet objectives, managers demonstrated optimism that direct fire management, native species management, and crops could continue to be employed into the future, but both recreation and cultural and social resource protection were considered less likely to continue (Fig. 3b). If alternative management activities are selected, direct fire management and outreach/education were considered the most likely to be adopted, whereas doing nothing was considered least likely (Fig. 3c). Figure 3. Manager-reported objectives and activities. Managers reported confidence levels on a Likert 1-5 scale, such that a 1 indicates that an objective or activity is “very unlikely” to succeed and a 5 indicates that an objective or activity is “very likely” to succeed. A 3 signifies “unknown” and therefore became a default answer for uncertainty. Managers considered each spatial location and reported, in turn: a) current objectives at that location and b) current activities at that location; and c) alternative activities possible for that location. a.

b.

Integration of social and ecological data to develop spatial models and maps of fire resilience across the landscape Social risk was generally higher in the southern and eastern parts of the Sonoran Desert, due to fewer jurisdictions that reported fire suppression objectives or fire management practices (generally southern) and fewer jurisdictions that agreed that they have sufficient resources to manage fire and fuels both on a day-to-day basis and during fire incidents (generally eastern) (Fig. 4). While responses about resources were highly correlated (r2 = 0.93), fire suppression and fire management were less correlated (r2 = 0.64). Both fire suppression and fire management had more influence than resources on the variation in social fire risk across the Sonoran Desert.

19

Ecological resistance to fire was highest in the southern, central part of the Sonoran Desert, and lowest in the most northern reaches, where Sonoran Desert transitions to other ecosystem types (Fig. 5). Geophysical and vegetation diversity were the main drivers of ecological resistance. Ecological adaptive capacity, on the other hand, was strongly driven by proximity to roads and other types of human landscape modification, with the lowest adaptive capacity near human modified landscapes, particularly roads, and the highest adaptive capacity the furthest away from them (Fig. 6). Higher ecological adaptive capacity was also exclusively found in the northwestern portion of the Sonoran Desert.

Figure 6. Ecological adaptive capacity as determined by habitat connectivity, species richness, and human modification. Darker turquoise colors indicate lower ecological adaptive capacity, and lighter turquoise colors indicate higher ecological adaptive capacity.

Figure 4. Social fire risk in the Sonoran Desert, as assessed by interview questions about fire suppression and fire management and survey questions about jurisdiction resources. Redder colors indicate higher risk and yellower colors indicate lower risk.

Figure 5. Ecological resistance in the Sonoran Desert, as determined by topographic diversity, geophysical diversity, vegetation diversity, and water availability. Darker turquoise colors indicate lower ecological resistance, and lighter turquoise colors indicate higher ecological resistance.

20

Social resistance was lowest in the southeastern Sonoran Desert as well as in a few scattered jurisdictions in the northern Sonoran, and it was highest in the southwestern Sonoran (Fig. 7). Social resistance was largely driven by how likely managers thought it was that management objectives would continue to be met under predicted future fire regimes. Social adaptive capacity was also lowest in the southwestern Sonoran Desert as well as in the same northern jurisdictions that also had low social resistance (Fig. 8). It was highest in the southwestern Sonoran Desert, within some of the same and some different jurisdictions than social resistance. Social adaptive capacity was driven more by whether or not an organization had changed management strategies in the past. Figure 7. Social resistance determined by manager responses to questions about meeting management objectives and preparing for surprises in ecosystem behavior. Redder colors indicate lower social resistance and yellower colors indicate higher social resistance. Gray areas are private lands that did not respond to interviews and were not surveyed.

Figure 8. Social adaptive capacity determined by manager responses to questions about adopting new management practices in the future and in the past. Redder colors indicate lower social resistance and yellower colors indicate higher social resistance. Gray areas are private lands that did not respond to interviews and were not surveyed.

The areas of lowest total risk were in the western part of the Sonoran Desert (Fig. 9). These areas of lowest total risk generally corresponded to some of the areas of the highest total resistance (southwestern Sonoran; Fig. 10) as well as the areas of highest adaptive capacity (western Sonoran; Fig. 11). Some areas of the southwestern Sonoran had some of the highest total risk, but also had some of the highest areas of both total resistance and total adaptive capacity. Many high total risk areas were in the eastern Sonoran, and the eastern Sonoran Desert is largely covered by medium to high total risk areas. Within these areas of high risk, the northeast Sonoran had some of the areas of lowest resistance. The southeast Sonoran also had low total resistance. The eastern Sonoran also had areas of low total adaptive capacity, covering a wider band around major highways than in the western Sonoran.

21

Figure 9. Socioecological fire risk in the Sonoran Desert. Warmer colors are higher risk, cooler colors are lower risk.

Figure 10. Socioecological fire resistance in the Sonoran Desert. Warmer colors are lower resistance, and cooler colors are higher resistance.

Figure 11. Socioecological adaptive capacity in the Sonoran Desert. Warmer colors are lower adaptive capacity and cooler colors are higher adaptive capacity.

We plotted total risk against total resilience to determine vulnerability across the Sonoran Desert (sensu Comer et al. 2012; Fig. 12A). We additionally translated these vulnerability scores into a map so as to be able to connect areas of low, medium, high, and very high vulnerability to particular areas within the Sonoran Desert (Fig. 12B). Very few locations fall into the low vulnerability category (low risk and high resilience), and those that do are all in the western Sonoran Desert. These areas largely align with areas of lowest total risk. Most areas of the Sonoran Desert fall into the medium or high vulnerability categories, with more of the medium vulnerability areas in the western half of the Sonoran Desert. The majority of the very high vulnerability areas fall within the northeastern or eastern part of the Sonoran Desert, although some are also found in the northwestern areas.

22

Figure 12. Vulnerability in the Sonoran Desert, shown both as the relationship between risk (Sensitivity) and resilience as well as the spatial arrangement of low, moderate, high, and very high vulnerability. Colors on the graph and map correspond to low (green), moderate (yellow), high (red), and very high (maroon) vulnerability. Vulnerability definitions and plotting sensu Comer et al. 2012.

Analysis of non-spatial patterns of management objectives and activities We performed non-spatial analyses to compare the frequency with which objectives and activities were reported and also their manager-reported likelihood of being achieved (optimism levels). While scoring interview data, we identified 10 management objective categories, including general fire suppression, conservation, grazing, agricultural, resource extraction, military, recreation, human and infrastructure protection, cultural resource protection, and no objective. Respondents cited certain management objectives more frequently than others (÷2=380.07, p < 0.0001; Fig. 13). Overall, objectives related to ecological conservation, fire suppression, and recreation were most commonly identified, while agriculture, military, grazing, and resource extraction objectives were least frequently mentioned (Fig. 13). Agencies differed in focal management objectives (Fig. 14; statistical results presented within the figure). Non-military federal agencies reported a diversity of land management objectives, including agriculture, conservation, cultural resource protection, general fire suppression, grazing, human and infrastructure protection, recreation, and resource extraction. Relative to other management entities, the non-military federal government more frequently mentioned cultural resource protection and general fire suppression objectives; state entities expressed agriculture-related objectives; and tribal institutions frequently reported cultural resource and human infrastructure protection objectives. All agency types reported objectives related to conservation and recreation, while resource extraction was rarely cited as an objective across agencies.

Figure 13. The proportion of total objectives across all points comprised by the 10 identified management objectives. Bars connected by the letter are not significantly different.

A B

23

Figure 14. The proportion of total objectives identified by each agency comprised by a given objective. Test statistics and corresponding p-values for each comparison are indicated in the upper left corner of each panel. Bars connected by the same letter are not statistically different.

Management activities to meet objectives also differed by agency. For the objective of fire

suppression, most agencies described a focus on active fire management following ignition. However, the non-military federal government also identified activities related to fire prevention, such as grazing to reduce fine fuel loads, native and non-native vegetation removal, and outreach and education (Fig. 15). Similarly, activities to prevent loss of infrastructure focused on response to active fires, but federal, local and military agencies also included preventative activities, most commonly vegetation removal, to protect structures from loss or damage due to fire (Fig. 16). To meet conservation objectives, federal, local, military, and tribal agencies frequently cited native species management activities, which included monitoring of plant and wildlife populations of valuable or imperiled species and general habitat restoration. The state agencies, by contrast, met conservation objectives primarily through invasive species removal efforts. Again, federal agencies took a more diversified approach to conservation, mentioning fire management, native species management, outreach and education, preventative vegetation removal, and leveraging resources from partnerships as activities to protect wildlife and habitat. Most agency types considered fire management a necessary activity to manage for recreation, in addition to direct maintenance of recreational areas (e.g., campgrounds, hiking trails) and recreational permitting. Only federal and tribal agency types managed land for cultural resources, though these entities adopted different approaches. Tribal governments focused on cultural and social resource protection, which included identifying and maintaining areas of cultural importance, and native species management of culturally important plants. Nonmilitary federal government agencies, on the other hand, emphasized active fire management and preventative vegetation removal around structures or areas of cultural importance. Federal and Native American agencies identified active fire management as a critical activity to meet agricultural objectives, while state agencies focused on managing agricultural crops alone. The activities to meet the objectives related to grazing, military, and resource extraction did not differ among agency types.

24

Figure 15. Activities to meet the objective of fire suppression by management agencies that reported a fire suppression objective (Federal, Local, and Tribal agencies). The test statistic and corresponding p-value for the overall test of differences in management activities among agencies is indicated in the upper left corner of the initial panel. Bars connected by the same letter are not statistically different.

Figure 16. Activities to meet the objective of infrastructure protection by management agencies that reported that objective. The test statistic and corresponding p-value for the overall test of differences in management activities among agencies is indicated in the upper left corner of the initial panel. Bars connected by the same letter are not statistically different.

For objectives and activities, manager-reported optimism levels hovered around the “unknown” value of 3, indicating uncertainty on the part of managers. However, a Kruskal-Wallis test detected significant differences in the likelihood that objectives would continue into the future (Kruskal-Wallis C2 = 67.84; p < 0.0001), with both military and economic objectives significantly more likely to be achieved than conservation, recreation, or cultural resource protection objectives (Dunn’s test results; Table 4). Likewise, a Kruskal-Wallis test detected significant differences in the likelihood that management activities will achieve objectives (Kruskal-Wallis C2 = 34.67; p < 0.0001), with economic activities significantly more likely to succeed than all other activities and direct fire-related activities significantly more likely to succeed than activities related to conservation or human resource protection (Dunn’s test results; Table 4). Across all activities, federal jurisdictions (N of unique locations = 592) reported higher likelihood of achieving results than did non-federal jurisdictions (N of unique locations = 196) (mean Likert-scale likelihood 3.42 vs. 3.13) (t = 4.58; p < 0.0001). Federal jurisdictions (N of unique locations = 405) also reported higher likelihood of adopting new activities than did non-federal jurisdictions (N of unique locations = 198) (mean Likert-scale likelihood 3.89 vs. 3.51) (t = 5.23; p < 0.0001).

Outreach and education Preventative vegetation removal Resources from partnerships

Fire management Grazing Native Species Management

Fede

ral

Loca

l

Nativ

e Am

erica

n

Fede

ral

Loca

l

Nativ

e Am

erica

n

Fede

ral

Loca

l

Nativ

e Am

erica

n

0.00

0.25

0.50

0.75

1.00

0.00

0.25

0.50

0.75

1.00

Management agency classification

Prop

ortio

n of

tota

l man

agem

ent a

ctiv

ities

χ2 = 30.06,p < 0.01*

A

B B χ2 = 1.40, p = 0.50 χ2 = 0.40, p = 0.82

χ2 = 2.65, p = 0.27 χ2 = 0.96, p = 0.62 χ2 = 1.25, p = 0.54

Outreach and education Preventative vegetation removal Resources from partnerships

Fire management Infrastructure development None

Fede

ral

Loca

l

Milit

ary

Nativ

e Am

erica

n

Fede

ral

Loca

l

Milit

ary

Nativ

e Am

erica

n

Fede

ral

Loca

l

Milit

ary

Nativ

e Am

erica

n

0.0

0.2

0.4

0.6

0.8

0.0

0.2

0.4

0.6

0.8

Management agency classification

Prop

ortio

n of

tota

l man

agem

ent a

ctiv

ities

χ2 = 20.76, p < 0.01* χ2 = 6.30, p = 0.09 χ2 = 3.33, p = 0.34

χ2 = 6.30, p = 0.09 χ2 = 4.101, p = 0.25χ2 = 11.65, p < 0.01*

A

A

B

A

A

B

A

C

25

Table 4. Comparison of manager-reported confidence levels that objectives can be achieved and activities continued into the future under environmental change. Comparisons were performed using Dunn’s tests for multiple comparisons. Cell values provide Z statistic and p-values.

Management objective categories

Con

serv

atio

n

Econ

omic

Fire

Hum

an re

sour

ces

Mili

tary

Rec

reat

ion

Conservation Z = -5.09; p < 0.0001

Z = -6.24; p < 0.0001

Z = -0.77; p = 0.47

Z = -4.50; p < 0.0001

Z = -3.38; p = 0.0015

Economic Z = -5.09; p < 0.0001

Z = 0.83; p = 0.51

Z = 4.47; p < 0.0001

Z = -0.05; p = 0.96

Z = 2.80; p = 0.0085

Fire Z = -6.24; p < 0.0001

Z = 0.83; p = 0.51

Z = 5.27; p < 0.0001

Z = -0.78; p = 0.50

Z = 2.93; p = 0.0063

Human resources Z = -0.77; p = 0.47

Z = 4.47; p < 0.0001

Z = 5.27; p < 0.0001

Z = -3.97; p = 0.0002

Z = -2.50; p = 0.019

Military Z = -4.50; p < 0.0001

Z = -0.05; p = 0.96

Z = -0.78; p = 0.50

Z = -3.97; p = 0.0002

Z = 2.49; p = 0.017

Recreation Z = -3.38; p = 0.0015

Z = 2.80; p = 0.0085

Z = 2.93; p = 0.0063

Z = -2.50; p = 0.019

Z = 2.49; p = 0.017

Management activity categories

Conservation Economic Fire Human resources Recreation Conservation Z = -4.06;

p = 0.0002 Z = -1.95; p = 0.072

Z = 1.78; p = 0.095

Z = -0.68; p = 0.49

Economic Z = -4.06; p = 0.0002

Z = 3.25; p = 0.0023

Z = 5.44; p < 0.0001

Z = 3.72; p = 0.0005

Fire Z = -1.95; p = 0.072

Z = 3.25; p = 0.0023

Z = 4.05; p = 0.0002

Z = 1.30; p = 0.22

Human resources Z = 1.78; p = 0.095

Z = 5.44; p < 0.0001

Z = 4.05; p = 0.0002

Z = -2.56; p = 0.018

Recreation Z = -0.68; p = 0.49

Z = 3.72; p = 0.0005

Z = 1.30; p = 0.22

Z = -2.56; p = 0.018

Survey and analysis The survey received 32 responses from respondents across 11 jurisdictions including state, county, federal, and tribal lands. Respondents described multiple drivers of change in objectives and strategies when it had occurred for their jurisdictions, which we grouped into ten categories: Change in response to a specific event, such as fire or litigation, change in direction from the regional or national level, changes due to updated planning documents, changes in strategy due to interagency collaboration, changes due to shifting priorities, changes informed by monitoring, changes due to funding increases or reductions, changes due to shifts in organizational leadership, changes in response to public opinion, and change as a continuous process. Changes due to increases or reductions in funding availability was the most frequently cited theme (n=5), followed by changes in response to specific events such as fire or litigation, changes in agency direction, changes due to new information, and changes due to leadership changes, (all n=3). Respondents also identified multiple barriers to changing management strategies or objectives when needed, which we grouped into eight categories: Inadequate time, inadequate funding, inadequate staffing, inadequate information, organizational norms, lack of coordination, political or policy context, and public opposition to change. The most cited theme was

26

inadequate funding (n=11), followed by inadequate time (n=6), inadequate staffing (n=5), and the political or policy context, such as NEPA constraints or other policies (n=5). A majority of respondents agreed with statements indicating their organization’s ability to innovate and manage in spite of uncertainty, with 75% to 100% of respondents selecting Somewhat, Agree, or Strongly Agree in response to each statement. For statements regarding perceptions of organization’s ability to conduct adaptive management, a majority of respondents also selected Somewhat, Agree, or Strongly Agree, with 71%-89% of responses falling in this range. Fewer respondents agreed with the statement “In my organization, we test or experiment with alternative management approaches” than any other statement assessing adaptive management. We also explored perceptions of organization’s constraints to making change, including time, finances, public opinion, neighboring land managers, and the financial leeway to make mistakes. A majority of respondents did not believe that their organizations have enough resources to manage fire and fuels either on a day-to-day basis or during fire incidents. A large majority, 89%, also agreed that while their organization may be open to trying different approaches, funding availability is a constraint to making changes. Responses varied more widely as to the perceived role of public opinion and neighboring land management units in constraining needed change, as well as the role of time and the financial flexibility to try different options or make mistakes, with percent agreement with these statements ranging from 57-64%. Spatial predictions of resilience in the Sonoran and their implications We found that areas of the eastern Sonoran Desert generally had the highest fire risk, assessed ecologically, socially, and socioecologically. These areas also had generally lower ecological, social, and socioecological resistance and adaptive capacity. As a result, the eastern Sonoran Desert generally is highly or very highly vulnerable to fire. Ecologically this vulnerability is driven by high NDVI, low elevation, low geophysical and vegetation diversity, and proximity to human modified landscapes, particularly roads. Social vulnerability is driven by a lack of fire suppression, fire management, not thinking it likely that management objectives can continue to be met, and not having changed management strategies in the past. Although some areas of the southwestern Sonoran Desert had equally or higher fire risk than the eastern Sonoran Desert, these areas also had some of the highest resistance and adaptive capacity. Because high risk drives the vulnerability so much, these areas still fall into the highly vulnerable category. The high risk in these areas is more socially than ecologically driven. The higher resistance and adaptive capacity of these southwestern areas means that if risk can be reduced somewhat, by for example setting fire suppression objectives or managing for fire, these areas would be far less vulnerable. Some areas of the western Sonoran Desert had the lowest fire risk, as well as the highest resistance and adaptive capacity. As such, these areas also had low vulnerability. Additionally, much of the northwestern Sonoran Desert had medium vulnerability, with low to medium risk and high to medium resistance and adaptive capacity. These areas of the Sonoran Desert obviously require less attention than those with higher vulnerability. This lower vulnerability is driven by low NDVI, high plant diversity, fire suppression and management, organizational flexibility, and the possession of more resources. Down-weighting social resilience in reference to ecological resilience made the Sonoran Desert less resilient overall than when ecological and social resilience were given equal weight. Likewise, up-weighting social resilience in reference to ecological resilience made the Sonoran

27

Desert more resilient overall, with far more areas of land, particularly in the southwest portion of the state, classified as low vulnerability. This indicates that despite the severity of the current and predicted future fire regime, managers remain more optimistic about the ability of their organization to handle the disturbances to the ecological landscape that these fires bring than an ecological assessment alone would determine. Implications of non-spatial findings regarding objectives and activities Unsurprisingly, some objectives and activities carry greater promise of success, from the point of view of managers, than others. Military and economic objectives are particularly likely to be achieved. Based on interviews, we attribute this to greater access to resources, in the form of funding, materials, personnel, and time, to support these objectives. By contrast, conservation and human resource protection objectives are the least likely to be achieved. Conservation objectives include protection of sensitive species, native species assemblages, and ecosystem integrity. Since fire regime change in the Sonoran represents a novel disturbance regime to which native species are ill-adapted, approaches to support conservation of those species become difficult to plan and select; traditional management measures are unlikely to continue to work under new, no-analogue conditions (Seastedt et al. 2008). Our interviews and surveys were limited to government-managed properties and thus not representative of the complete landscape. However, private lands are a distinct minority of the landscape, by total area, so our methods captured most of the topographical and elevational diversity across the focal region. Federal land managers exhibited higher levels of optimism that objectives would be achieved and activities could continue, relative to other levels of government, although the differences between the two groups were not large. Again, this is likely due to differences in resource availability. Federal agencies in the region, although they are not equally resourced, are generally better provisioned than state, local or tribal entities and thus better able to respond to changing conditions and management challenges. Arizona receives a higher than equal return on its federal tax dollars; that is, for every dollar paid to the federal government, the state receives more than a dollar back in federal investments (New York State Comptroller Report; https://www.osc.state.ny.us/reports/budget/2018/federal-budget-fiscal-year-2017.pdf). Thus, the resources available at the state level and below are lower than the resources available at the federal level. Alternative activities and their promise Interviewees were asked what alternative activities they might attempt in order to meet current management objectives, along with the likelihood of adopting those activities. On average, managers reported that they were unlikely to do nothing if current management activities fail, suggesting that attempting to meet objectives will remain important for these jurisdictions. Suggested alternative activities fell into the following broad categories: grazing for fuels management, development of permitting processes, outreach and education, fuels removal for fire prevention, and seeking resources from partnerships. Some of these activities are already utilized by managers across the study area, but at least some respondents who do not currently use them reported that they would likely explore their use if future needs arise. Grazing for fuels management is a means of removing significant amounts of both native and non-native biomass and thus reducing fuel loads. Effectiveness of the practice in fire management has been evaluated in a number of systems, with mixed results (Barry et al. 2015; Davy et al. 2015). Grazing reduced annual fuel loads, including from cheatgrass (Bromus

28

tectorum), and resulted in diminished flame length and fire spread rates in high-elevation sites in the Intermountain West (Strand et al. n.d.; Diamond et al. 2010; Schmelzer et al. 2014). Similarly, with careful calibration of grazing intensities, fine fuel loads in southern Arizona rangelands were successfully diminished enough to reduce flame lengths and thus the cost of fighting fires (Bruegger et al. 2016). Grazing has also been employed experimentally along with native plant restoration efforts to reduce buffelgrass (Cenchrus ciliaris) coverage in lowland arid regions; in some contexts, these efforts led to measurable reductions in buffelgrass coverage (Melzer 2015), but not in others (Fensham et al. 2014, 2015). As these studies illustrate, grazing for fuels reduction should be carefully planned and monitored to evaluate its effectiveness over time, particularly under changing climatic conditions and increased incidence of drought. Permitting processes can be implemented to manage controlled and managed burns as well as access to fire-prone regions. Restrictions and permits are not always within the purview of land management agencies, and any form of increased regulation can lead to objections on the part of the public. Nevertheless, communities that have experienced fire directly have been shown to approve of access and use restrictions to diminish fire risk (Brown et al. 2008), although in such cases damage may already be extensive. Establishment of permitting processes entails delineation of permit requirements and the conditions under which certain activities will be permitted; this in turn increases the capacity of permitting agencies to plan and predict behaviors and also to respond to changing conditions (Riebau & Fox 2001). For example, understanding risks of smoke to human health under various wind conditions, and balancing those against ecological needs, can enable alignment of human health and ecological objectives in planning (Long et al. 2018). At the same time, permitting processes and programs can be onerous for land managers and may make them less likely to perform prescribed or managed burns (van Wilgen et al. 2012), as can increased manager liability associated with formal permitting (Wonkka et al. 2015). Furthermore, the layers of bureaucracy and processing of permitting programs can make fire management less flexible in response to changing ecological conditions, preventing managers from responding to changing fuels and fire risk needs (Kolden & Brown 2010; van Wilgen et al. 2012). As fire risk grows, outreach and education may diminish risky behaviors on the part of the public, reducing the likelihood of accidental ignitions (Prestemon et al. 2010; Crow et al. 2015). If they reduce ignition incidence, the investment of monetary resources in outreach programs carries high rate of economic return by saving fire-fighting resources (Prestemon et al. 2010). However, the extent to which outreach and education activities penetrate into a given community is heavily influenced by social cohesion and connectedness to place (McCaffrey 2015). Social capital can increase the amount of uptake of information and engagement with outreach efforts (Bihari & Ryan 2012). Importantly, socially vulnerable populations in Arizona, including lower-income and non-English speaking communities, were found to be less likely to be reached by wildfire education and outreach efforts (Ojerio et al. 2011). In general, outreach and education efforts are particularly effective when a fire has been experienced by a community, but also when particular individuals act as champions of the issue, external funding is supporting outreach efforts, and dedicated personnel are engaged in the process (Mockrin et al. 2018). Fuels removal for fire prevention can include the creation of fuel breaks in native or non-native vegetation. In our study area, according to our interviewees, most such efforts are focused on removal of non-native grasses to disrupt the continuity of fuels in upland desert or removal of non-native tamarisk (Tamarix sp.) woody debris to disrupt fuels continuity in desert riparian

29

areas. Fuel breaks are common to reduce wildfire risk in many habitat types, but entail expenditure of fire preparation resources as well as disturbance to native vegetation and soils; thus, careful placement and planning of efforts is needed to maximize their effectiveness (Rytwinski & Crowe 2010). In Arizona, experimental elimination of buffelgrass fuels required intensive herbicide application and dead material remained flammable until three years post-removal effort (McDonald & McPherson 2013). In southern California forests, fuel breaks were found to reduce fire extent largely by providing access points for fire-fighting personnel (Syphard et al. 2011). Individual fuel breaks were found to have low return on investment in a scenario planning exercise, but a network of fuel breaks across a landscape reduced overall landscape susceptibility to fire outbreaks (Oliveira et al. 2016). Such landscape-scale approaches typically involve multiple jurisdictions, requiring communication and collaboration among neighboring landowners for maximal effectiveness (Fischer & Charnley 2012; Oliveira et al. 2016). A perception of fire risk is often important for motivating landowners to participate in such cross-boundary collaboration, reiterating the need for effective wildfire education (Fischer & Charnley 2012). Resources from partnerships are increasingly essential in response to fire events, where agencies at multiple levels of government pool personnel and fire-fighting resources to coordinate and carry out fire response (Seekamp et al. 2013). Although increasingly large and severe fires often require this type of coordination, there are costs to the coordination itself, in terms of time investment, and a survey of agency personnel revealed that many do not feel that these costs are adequately accounted for in planning and compensation (Seekamp et al. 2013). Additionally, misalignments in mission can make it difficult for such partnerships to work effectively toward objectives that are shared among participants (Fleming et al. 2015). Other partnerships can include less tangible resource sharing. Public agencies have partnered with schools to boost wildfire and fire ecology education directed toward youth, for example (Monroe et al. 2016). Science-management partnerships have aimed to obtain scientific information from academic and non-academic scientists to inform on-the-ground management being carried out by agency personnel (Littell et al. 2012). Information-sharing among agencies may assist with planning and response, but at least one study found that information flows are not consistent across relevant entities, with, for example, fire response personnel communicating within separate circles from fire ecologists (Fischer & Jasny 2017). Public-private partnerships may help to leverage and extend fire resources, but such partnerships can be hampered by contractual difficulties or regulatory impediments (Schulz & Lueck 2014). Streamlining these resource exchanges may be necessary to make partnerships effective as alternative activities that can be pursued to meet objectives.

CONCLUSIONS AND IMPLICATIONS FOR MANAGEMENT AND FUTURE RESEARCH

Sources of environmental change can threaten biodiversity, ecological functions, and ecosystem services. Fire regime change, driven by altered temperatures and precipitation coupled with invasion of fire-adapted plants, is particularly concerning in the Sonoran Desert. This non-fire-adapted system is the most biodiverse desert in North America and sustains unique ecological interactions while providing ecosystem services to more than 6 million residents (www.census.gov). Occurrence of fire reduces both plant and associated animal diversity at local scales (McCaffrey 2015). A feedback loop promoting fire-adapted non-native species threatens

30

to turn burned Sonoran Desert ecological communities into grasslands dominated by non-natives (McDonald & McPherson 2013). Resilience at its heart is about state change (or lack thereof) and recovery, but relevant components of states are matters of perception. That is, a state change will only be identified as such if characteristics of value or viewed as intrinsic to a system are lost. Determining whether changes warrant alarm or management response entails determining what desirable states and components of systems can be and whether those desirable states are threatened by environmental change. Success in resilience management is likewise a feature of perception; what one manager sees as successful is not necessarily the same as what another manager sees as successful (Aslan et al. 2018). For this reason, we asked managers to define the spatially-explicit objectives that direct their management decisions on their landscapes, such that success entails meeting those objectives via any activity, and changes in activities or objectives entail varying levels of adaptability. Integration of social and ecological definitions and elements of resilience is likewise challenging in its semantics and conceptual discontinuities (Timpane-Padgham et al. 2017; Aslan et al. 2018; Petersen et al. 2018). Ecological state changes are defined as crossing a threshold into an alternative set of characteristics, and the resilience of a system is represented in the height of that threshold. A resilience system demonstrates a high threshold, such that a strong perturbation is required to cross the threshold (the strength of such a perturbation is one metric of resilience) (Holling 1973). In social systems, resilience may imply the ability of certain institutions or functions to persist following disturbance (e.g., government, economic sectors, etc.), and resilience is not always positive; indeed, it can indicate continuation of conditions of inequality or inefficiency in spite of “perturbations,” or efforts to alter those conditions (MacKinnon & Derickson 2013). Our approach did not aim to identify “healthy” or “positive” states, but rather the likelihood that the current state of the system, in whatever form, could continue in light of future projected fire regime shifts. Responses to vast, wholesale environmental change fall into two categories: adaptation and mitigation (Blicharska et al. 2017). Mitigation entails preventing change by, for example, removing halting the spread of invasive species (Watkiss et al. 2015). Adaptation entails reducing the vulnerability of systems to change by, for example, assisting the reproduction and relocation of threatened native species (Watkiss et al. 2015). The fundamental drivers of fire regime change in the Sonoran – drought and non-native species – are difficult (or perhaps impossible) to prevent, so understanding where they will be most impactful is necessary to boost adaptive capacity or reduced vulnerability to their effects. The spatially-explicit analyses we have performed aim to assist managers in identifying sites of most concern in order to support such adaptation efforts. Going forward, the maps and assessments produced here may serve managers across the region by pinpointing locations of vulnerability to state change resulting from fire. Separately considering the effects of management flexibility and resources vs. underlying ecological conditions may help managers and collaboratives identify areas of intervention to support protection of native species and ecosystem services. Preparing for change in advance of perturbations may enable management entities to select and screen alternative management approaches before the need for them arises.

31

LITERATURE CITED

Abatzoglou, J.T. and C.A. Kolden. 2011. Climate Change in Western US Deserts: Potential for Increased Wildfire and Invasive Annual Grasses. Rangeland Ecology and Management 64:471-478.

Abatzoglou, J.T., Williams, A.P., Boschetti, L., Zubkova, M. and Kolden, C.A., 2018. Global patterns of interannual climate–fire relationships. Global change biology, 24(11), pp.5164-5175.

Abella, S.R., 2009. Post-fire plant recovery in the Mojave and Sonoran Deserts of western North America. Journal of Arid Environments, 73(8), pp.699-707.

Abramson DM, Grattan LM, Mayer B, Colten CE, Arosemena FA, Bedimo-Rung A, Lichtveld M. 2015. The Resilience Activation Framework: a Conceptual Model of How Access to Social Resources Promotes Adaptation and Rapid Recovery in Post-disaster Settings. The Journal of Behavioral Health Services & Research 42:42–57.

Adger, W.N., 2000. Social and ecological resilience: are they related? Progress in human geography, 24(3), pp.347-364.

Aldrich DP, Meyer MA. 2015. Social Capital and Community Resilience. American Behavioral Scientist 59:254–269.

Alford, E. J., J. H. Brock, and G. J. Gottfried. 2005. The effects of fire on Sonoran Desert plant communities. USDA Forest Service Proceedings RMRS-P-36.

Allen, C. D., M. Savage, D. A. Falk, K. F. Suckling, T. W. Swetnam, T. Schulke, P. B. Stacey, P. Morgan, M. Hoffman, and J. T. Klingel. 2002. Ecological restoration of southwestern ponderosa pine ecosystems: a broad perspective. Ecological Applications 12:1418–1433.

Araújo, M. B., and M. Luoto. 2007. The importance of biotic interactions for modelling species distributions under climate change. Global Ecology and Biogeography 16:743–753.

Araújo, M. B., and M. New. 2007. Ensemble forecasting of species distributions. Trends in Ecology & Evolution 22:42–47.

Aslan, C.E., Petersen, B., Shiels, A.B., Haines, W. and Liang, C.T., 2018. Operationalizing resilience for conservation objectives: the 4S's. Restoration Ecology, 26(6), pp.1032-1038.

Barry, S., Bush, L., Larson, S. and Ford, L.D., 2015. The Benefits of Grazing—Livestock Grazing: A Conservation Tool on California's Annual Grasslands. University of California Agriculture and Natural Resources Publication.

Berkhout, F., B. van den Hurk, J. Bessembinder, J. de Boer, B. Bregman, and M. van Drunen. 2014. Framing climate uncertainty: socio-economic and climate scenarios in vulnerability and adaptation assessments. Regional Environmental Change 14:879–893.

Bernard, H.R. (2011). Research Methods in Anthropology: Qualitative and Quantitative Approaches (5th ed., 666 pp.). AltaMira Press.

Bihari, M. and Ryan, R., 2012. Influence of social capital on community preparedness for wildfires. Landscape and Urban Planning, 106(3), pp.253-261.

Blicharska, M., R. J. Smithers, M. Kuchler, G. K. Agrawal, J. M. Gutierrez, et al. 2017. Steps to overcome the North-South divide in research relevant to climate change policy and practice. Nature Climate Change 7:21-27.

Bowman DMJS, Balch J, Artaxo P, Bond WJ, Cochrane MA, D’Antonio CM, DeFries R, Johnston FH, Keeley JE, Krawchuk MA, Kull CA, Mack M, Moritz MA, Pyne S, Roos

32

CI, Scott AC, Sodhi NS, Swetnam TW, Whittaker R. 2011. The human dimension of fire regimes on Earth. Journal of Biogeography 38:2223–2236.

Brooks, M. L., and J. C. Chambers. 2011. Resistance to invasion and resilience to fire in desert shrublands of North America. Rangeland Ecology & Management 64:431–438.

Brooks, M.L., D'antonio, C.M., Richardson, D.M., Grace, J.B., Keeley, J.E., DiTomaso, J.M., Hobbs, R.J., Pellant, M. and Pyke, D., 2004. Effects of invasive alien plants on fire regimes. BioScience, 54(7), pp.677-688.

Brown, P. M., E. K. Heyerdahl, S. G. Kitchen, and M. H. Weber. 2008. Climate effects on historical fires (1630-1900) in Utah. International Journal of Wildland Fire 17:28–39.

Bruegger, R.A., Varelas, L.A., Howery, L.D., Torell, L.A., Stephenson, M.B. and Bailey, D.W., 2016. Targeted grazing in southern Arizona: using cattle to reduce fine fuel loads. Rangeland ecology & management, 69(1), pp.43-51.

Burgan RE, Klaver RW, Klaver JM. 1998. Fuel Models and Fire Potential From Satellite and Surface Observations. International Journal of Wildland Fire 8:159–170.

Carpenter, S., Walker, B., Anderies, J.M. and Abel, N., 2001. From metaphor to measurement: resilience of what to what? Ecosystems, 4(8), pp.765-781.

Coe, S.J., Finch, D.M. and Friggens, M.M., 2012. An assessment of climate change and the vulnerability of wildlife in the Sky Islands of the Southwest. Gen. Tech. Rep. RMRS-GTR-273. Fort Collins, CO: US Department of Agriculture, Forest Service, Rocky Mountain Research Station. 208 p., 273.

Collier MJ. 2015. Novel ecosystems and social-ecological resilience. Landscape Ecology 30:1363–1369.

Comer, P. J., B. Young, K. Schulz, G. Kittel, B. Unnasch, D. Braun, G. Hammerson, L. Smart, H. Hamilton, S. Auer, R. Smyth, and J. Hak. 2012. Climate Change Vulnerability and Adaptation Strategies for Natural Communities: Piloting methods in the Mojave and Sonoran deserts. Report to the U.S. Fish and Wildlife Service. NatureServe, Arlington, VA.

Comus, P.W., Dimmitt, M.A., Phillips, S.J. and Brewer, L.M. eds., 2015. A natural history of the Sonoran Desert. Univ of California Press.

Coops, N. C., and R. H. Waring. 2011. A process-based approach to estimate lodgepole pine (Pinus contorta Dougl.) distribution in the Pacific Northwest under climate change. Climatic Change 105:313–328.

Corbin, J. and Strauss, A. (1997). Basics of qualitative research: techniques and procedures for developing grounded theory (3rd ed. 400 pp.). USA: Sage Publications, Inc.

Coops, N. C., and R. H. Waring. 2011. A process-based approach to estimate lodgepole pine (Pinus contorta Dougl.) distribution in the Pacific Northwest under climate change. Climatic Change 105:313–328.

D’Antonio, C.M., 2000. Fire, plant invasions, and global changes. Invasive species in a changing world, pp.65-93.

Davy, J., Roche, L., Robertson, A., Nay, D. and Tate, K., 2015. Introducing cattle grazing to a noxious weed-dominated rangeland shifts plant communities. California Agriculture, 69(4), pp.230-236.

Diamond, J.M., Call, C.A. and Devoe, N., 2010. Effects of targeted cattle grazing on fire behavior of cheatgrass-dominated rangeland in the northern Great Basin, USA. International Journal of Wildland Fire, 18(8), pp.944-950.

33

Dickson, B.G., Sisk, T.D., Sesnie, S.E., and Bradley, B.A. 2016. Integrated Spatial Models of Non-Native Plant Invasion, Fire Risk, and Wildlife Habitat to Support Conservation of Military and Adjacent Lands in the Arid Southwest. Final Report submitted to SERDP Program, Department of Defense.

Dressel, S., G. Ericsson, and C. Sandstrom. 2018. Mapping social-ecological systems to understand the challenges underlying wildlife management. Environmental Science and Policy 84:105-112.

Elmqvist, T., C. Folke, M. Nyström, G. Peterson, J. Bengtsson, B. Walker, and J. Norberg. 2003. Response diversity, ecosystem change, and resilience. Frontiers in Ecology and the Environment 1:488–494.

Fensham, R.J., Wang, J. and Kilgour, C., 2015. The relative impacts of grazing, fire and invasion by buffel grass (Cenchrus ciliaris) on the floristic composition of a rangeland savanna ecosystem. The Rangeland Journal, 37(3), pp.227-237.

Fensham, R.J., Silcock, J.L. and Firn, J., 2014. Managed livestock grazing is compatible with the maintenance of plant diversity in semidesert grasslands. Ecological Applications, 24(3), pp.503-517.

Field, C. R., A. A. Dayer, and C. S. Elphick. 2017. Landowner behavior can determine the success of conservation strategies for ecosystem migration under sea-level rise. Proceedings of the National Academy of Sciences 114:9134-9139.

Fischer, A.P. and Charnley, S., 2012. Risk and cooperation: managing hazardous fuel in mixed ownership landscapes. Environmental management, 49(6), pp.1192-1207.

Fischer, A.P. and Jasny, L., 2017. Capacity to adapt to environmental change: evidence from a network of organizations concerned with increasing wildfire risk. Ecology and Society, 22(1).

Fleming, C.J., McCartha, E.B. and Steelman, T.A., 2015. Conflict and collaboration in wildfire management: the role of mission alignment. Public Administration Review, 75(3), pp.445-454.

Folke, C., 2006. Resilience: The emergence of a perspective for social–ecological systems analyses. Global environmental change, 16(3), pp.253-267.

Fuentes-Ramirez, A., Veldman, J.W., Holzapfel, C. and Moloney, K.A., 2016. Spreaders, igniters, and burning shrubs: plant flammability explains novel fire dynamics in grass-invaded deserts. Ecological applications, 26(7), pp.2311-2322.

Gaertner, M., Biggs, R., Te Beest, M., Hui, C., Molofsky, J. and Richardson, D.M., 2014. Invasive plants as drivers of regime shifts: identifying high-priority invaders that alter feedback relationships. Diversity and Distributions, 20(7), pp.733-744.

Gelman, A., and D. B. Rubin. 1992. Inference from Iterative Simulation Using Multiple Sequences. Statistical Science 7:457–472.

Gray, M., Dickson, B. 2015. A new model of landscape-scale fire connectivity applied to resource and fire management in the Sonoran Desert, USA. Ecological Applications.

Gray, M.E., Dickson, B.G. and Zachmann, L.J., 2014. Modelling and mapping dynamic variability in large fire probability in the lower Sonoran Desert of south-western Arizona. International Journal of Wildland Fire, 23(8), pp.1108-1118.

Gunderson, L.H., 2000. Ecological resilience—in theory and application. Annual review of ecology and systematics, 31(1), pp.425-439.

Haase, D. 2013. Participatory modelling of vulnerability and adaptive capacity in flood risk management. Natural Hazards 67:77–97.

34

Holling, C. S. 1973. Resilience and stability of ecological systems. Annual Review of Ecology and Systematics 4:1–23.

Jardine, A., Merideth, R., Black, M. and LeRoy, S., 2013. Assessment of climate change in the southwest United States: A report prepared for the National Climate Assessment. Island press.

Keyser, A. and Westerling, A.L., 2017. Climate drives inter-annual variability in probability of high severity fire occurrence in the western United States. Environmental Research Letters, 12(6), p.065003.

Kolden, C.A. and Brown, T.J., 2010. Beyond wildfire: perspectives of climate, managed fire and policy in the USA. International Journal of Wildland Fire, 19(3), pp.364-373.

Kwadijk, J. C., M. Haasnoot, J. P. M. Mulder, M. M. C. Hoogvliet, A. B. M. Jeuken, et al. 2010. Using adaptation tipping points to prepare for climate change and sea level rise: a case study in the Netherlands. Wiley Interdisciplinary Reviews: Climate Change 1:729–740.

Lake, P.S., 2013. Resistance, resilience and restoration. Ecological management & restoration, 14(1), pp.20-24.

Lenton, T. M., H. Held, E. Kriegler, J. W. Hall, W. Lucht, S. Rahmstorf, and H. J. Schellnhuber. 2008. Tipping elements in the Earth’s climate system. Proceedings of the National Academy of Sciences 105:1786–1793.

LeQuire, E. 2013. Climate change tipping points: a point of no return? Fire Science Digest 15:1-13.

Li D, Wu S, Liu L, Zhang Y, Li S. 2018. Vulnerability of the global terrestrial ecosystems to climate change. Global Change Biology 24:4095–4106.

Littell, J.S., Peterson, D.L., Millar, C.I. and O’Halloran, K.A., 2012. US National Forests adapt to climate change through Science–Management partnerships. Climatic Change, 110(1-2), pp.269-296.

Lockwood, M., C.M. Raymond, E. Oczkowski and M. Morrison. 2014. Measuring the dimensions of adaptive capacity: a psychometric approach. Ecology and Society 20(1):37. http://dx.doi.org/10.5751/ES-07203-200137

Long, J.W., Tarnay, L.W. and North, M.P., 2017. Aligning smoke management with ecological and public health goals. Journal of Forestry, 116(1), pp.76-86.

Looy KV et al. 2019. The three Rs of river ecosystem resilience: Resources, recruitment, and refugia. River Research and Applications 35:107–120.

MacKinnon D, Derickson KD. 2013. From resilience to resourcefulness: A critique of resilience policy and activism. Progress in Human Geography 37:253-270

McCaffrey, S., 2015. Community wildfire preparedness: a global state-of-the-knowledge summary of social science research. Current Forestry Reports, 1(2), pp.81-90.

McCarty, J. P. 2001. Ecological consequences of recent climate change. Conservation Biology 15:320–331.

McDonald, C.J. and McPherson, G.R., 2013. Creating hotter fires in the Sonoran Desert: buffelgrass produces copious fuels and high fire temperatures. Fire Ecology, 9(2), pp.26-39.

Melzer, R.I., 2015. When is stock grazing an appropriate 'tool' for reducing 'Cenchrus ciliaris' (Buffel grass) on conservation reserves? Proceedings of the Royal Society of Queensland, The, 120, p.53.

Millar, C. I., N. L. Stephenson, and S. L. Stephens. 2007. Climate change and forests of the future: managing in the face of uncertainty. Ecological Applications 17:2145–2151.

35

Mockrin, M.H., Fishler, H.K. and Stewart, S.I., 2018. Does wildfire open a policy window? Local government and community adaptation after fire in the United States. Environmental management, 62(2), pp.210-228.

Moloney KA, Mudrak EL, Fuentes-Ramirez A, Parag H, Schat M, Holzapfel C. 2019. Increased fire risk in Mojave and Sonoran shrublands due to exotic species and extreme rainfall events. Ecosphere 10:e02592.

Monroe, M.C., Ballard, H.L., Oxarart, A., Sturtevant, V.E., Jakes, P.J. and Evans, E.R., 2016. Agencies, educators, communities and wildfire: partnerships to enhance environmental education for youth. Environmental Education Research, 22(8), pp.1098-1114.

Nelson, D. R., W. N. Adger, and K. Brown. 2007. Adaptation to environmental change:

contributions of a resilience framework. Annual Review of Environment and Resources 32:395.

Nolan, C., Overpeck, J.T., Allen, J.R., Anderson, P.M., Betancourt, J.L., Binney, H.A., Brewer, S., Bush, M.B., Chase, B.M., Cheddadi, R. and Djamali, M., 2018. Past and future global transformation of terrestrial ecosystems under climate change. Science, 361(6405), pp.920-923.

O’Hara, R. B. and M. J. Sillanpaa. 2009. A review of Bayesian variable selection methods: what, how, and which. Bayesian Analysis 4:85-118.

Ojerio, R., Moseley, C., Lynn, K. and Bania, N., 2010. Limited involvement of socially vulnerable populations in federal programs to mitigate wildfire risk in Arizona. Natural Hazards Review, 12(1), pp.28-36.

Oliver TH et al. 2015. Biodiversity and Resilience of Ecosystem Functions. Trends in Ecology & Evolution 30:673–684.

Oliveira, T.M., Barros, A.M., Ager, A.A. and Fernandes, P.M., 2016. Assessing the effect of a fuel break network to reduce burnt area and wildfire risk transmission. International Journal of Wildland Fire, 25(6), pp.619-632.

Parmesan, C. 2006. Ecological and evolutionary responses to recent climate change. Annual Review of Ecology, Evolution, and Systematics 37:637–669.

Pendergrass, A.G., Knutti, R., Lehner, F., Deser, C. and Sanderson, B.M., 2017. Precipitation variability increases in a warmer climate. Scientific reports, 7(1), p.17966.

Perrings, C., and B. Walker. 1997. Biodiversity, resilience and the control of ecological-economic systems: the case of fire-driven rangelands. Ecological Economics 22:73–83.

Petersen B, Aslan C, Stuart D, Beier P. 2018. Incorporating Social and Ecological Adaptive Capacity into Vulnerability Assessments and Management Decisions for Biodiversity Conservation. BioScience 68:371–380.

Plummer, M. 2003. JAGS: A program for analysis of Bayesian graphical models using Gibbs sampling. Proceedings of the 3rd International Workshop on Distributed Statistical Computing (DSC 2003) March, pp. 20–22.

Plummer, M., A. Stukalov, and M. Denwood. 2018. Rjags: Bayesian graphical models using MCMC. R package version 4-8. https://CRAN.R-project.org/package=rjags.

Prestemon, J.P., Butry, D.T., Abt, K.L. and Sutphen, R., 2010. Net benefits of wildfire prevention education efforts. Forest Science, 56(2), pp.181-192.

Quinlan AE, Berbés-Blázquez M, Haider LJ, Peterson GD. 2016. Measuring and assessing resilience: broadening understanding through multiple disciplinary perspectives. Journal of Applied Ecology 53:677–687.

36

Ramirez F. 2015. Flammable deserts: understanding the impacts of fire on southwestern desert ecosystems of USA. PhD Thesis. Dissertation, Iowa State University.

Rehfeldt, G. E., N. L. Crookston, C. Sáenz-Romero, and E. M. Campbell. 2012. North American vegetation model for land-use planning in a changing climate: a solution to large classification problems. Ecological Applications 22:119–141.

Riebau, A.R. and Fox, D., 2001. The new smoke management. International Journal of Wildland Fire, 10(4), pp.415-427.

Rietkerk, M., F. van den Bosch, and J. van de Koppel. 1997. Site-specific properties and irreversible vegetation changes in semi-arid grazing systems. Oikos 80:241–252.

Ruesink, J. L., B. E. Feist, C. J. Harvey, J. S. Hong, A. C. Trimble, L. M. Wisehart, and others. 2006. Changes in productivity associated with four introduced species: ecosystem transformation of a ‘pristine’ estuary. Marine Ecology Progress Series 311:203–215.

Rytwinski, A. and Crowe, K.A., 2010. A simulation-optimization model for selecting the location of fuel-breaks to minimize expected losses from forest fires. Forest ecology and management, 260(1), pp.1-11.

Schlaepfer, D.R., Bradford, J.B., Lauenroth, W.K., Munson, S.M., Tietjen, B., Hall, S.A., Wilson, S.D., Duniway, M.C., Jia, G., Pyke, D.A. and Lkhagva, A., 2017. Climate change reduces extent of temperate drylands and intensifies drought in deep soils. Nature Communications, 8, p.14196.

Seastedt TR, Hobbs RJ, Suding KN. 2008. Management of novel ecosystems: are novel approaches required? Frontiers in Ecology and the Environment 6:547–553.

Schmelzer, L., Perryman, B., Bruce, B., Schultz, B., McAdoo, K., McCuin, G., Swanson, S., Wilker, J. and Conley, K., 2014. Case Study: Reducing cheatgrass (Bromus tectorum L.) fuel loads using fall cattle grazing. The Professional Animal Scientist, 30(2), pp.270-278.

Schulz, K.B. and Lueck, D., 2014. Contracting for control of landscape-level resources. Iowa L. Rev., 100, p.2507.

Seager, R. et al. 2007. Model projections of an imminent transition to a more arid climate in Southwestern North America. Science 316:1181–1184.

Seekamp, E., Barrow, L.A. and Cerveny, L.K., 2013. The growing phenomenon of partnerships: A survey of personnel perceptions. Journal of Forestry, 111(6), pp.412-419.

Shackley, S., and R. Deanwood. 2003. Constructing social futures for climate-change impacts and response studies: building qualitative and quantitative scenarios with the participation of stakeholders. Climate Research 24:71–90.

Shafer, S. L., P. J. Bartlein, and R. S. Thompson. 2001. Potential changes in the distributions of western North America tree and shrub taxa under future climate scenarios. Ecosystems 4:200–215.

Sponseller, R. A., S. J. Hall, D. P. Huber, N. B. Grimm, J. P. Kaye, C. M. Clark, and S. L. Collins. 2012. Variation in monsoon precipitation drives spatial and temporal patterns of Larrea tridentata growth in the Sonoran Desert. Functional Ecology 26:750-758.

Standish, R.J., Hobbs, R.J., Mayfield, M.M., Bestelmeyer, B.T., Suding, K.N., Battaglia, L.L., Eviner, V., Hawkes, C.V., Temperton, V.M., Cramer, V.A. and Harris, J.A., 2014. Resilience in ecology: Abstraction, distraction, or where the action is?. Biological Conservation, 177, pp.43-51.

Sterk M, van de Leemput IA, Peeters ET. 2017. How to conceptualize and operationalize resilience in socio-ecological systems? Current Opinion in Environmental Sustainability 28:108–113.

37

Strand, E.K., Launchbaugh, K.L., Limb, R.F. and Torell, L.A., 2014. Livestock grazing effects on fuel loads for wildland fire in sagebrush dominated ecosystems. Journal of Rangeland Applications, 1, pp.35-57.

Su, Y.-S., and M. Yajima. 2015. R2jags: Using R to Run 'JAGS'. R package version 0.5-7. https://CRAN.R-project.org/package=R2jags.

Suding, K.N., Gross, K.L. and Houseman, G.R., 2004. Alternative states and positive feedbacks in restoration ecology. Trends in ecology & evolution, 19(1), pp.46-53.

Syphard, A.D., Keeley, J.E. and Brennan, T.J., 2011. Comparing the role of fuel breaks across southern California national forests. Forest Ecology and Management, 261(11), pp.2038-2048.

Theobald, D.M., S.E. Reed, K. Fields, and M. Soulé. 2012. Connecting natural landscapes using a landscape permeability model to prioritize conservation activities in the United States. Conservation Letters 5:123-133.

Theobald, D.M. 2013. A general model to quantify ecological integrity for landscape assessments and US application. Landscape Ecology 28:1859–1874.

Timpane-Padgham, B. L., T. Beechie, T. Klinger. 2017. A systematic review of ecological attributes that confer resilience to climate change in environmental restoration. PLoS ONE 12:e0173812.

Tompkins, E. L., R. Few, and K. Brown. 2008. Scenario-based stakeholder engagement: incorporating stakeholders preferences into coastal planning for climate change. Journal of Environmental Management 88:1580–1592.

van Wilgen, B.W., Forsyth, G.G. and Prins, P., 2012. The management of fire-adapted ecosystems in an urban setting: the case of Table Mountain National Park, South Africa. Ecology and Society, 17(1).

Walker, B., Holling, C.S., Carpenter, S. and Kinzig, A., 2004. Resilience, adaptability and transformability in social–ecological systems. Ecology and society, 9(2).

Wang O, Zachmann LJ, Sesnie SE, Olsson AD, Dickson BG. 2014. An Iterative and Targeted Sampling Design Informed by Habitat Suitability Models for Detecting Focal Plant Species over Extensive Areas. PLoS ONE 9:e101196.

Watkiss, P., M. Benzie, and R. J. Klein. 2015. The complementarity and comparability of climate change adaptation and mitigation. Wiley Interdisciplinary Reviews: Climate Change 6:541-557.

Watson JEM, Iwamura T, Butt N. 2013. Mapping vulnerability and conservation adaptation strategies under climate change. Nature Climate Change 3:989–994.

Werners, S. E. 2013. Adaptation Turning Points: decision support methods for adaptation, MEDIATION Project, Briefing Note 9.

Westerling, A. L., H. G. Hidalgo, D. R. Cayan, and T. W. Swetnam. 2006. Warming and earlier spring increase western US forest wildfire activity. Science 313:940–943.

Williams, B. K., and E. D. Brown. 2012. Adaptive management: The US Department of the Interior applications guide. US Department of the Interior, Adaptive Management Working Group.

Williams, B.K., R.C. Szaro, and C.D. Shapiro. 2009. Adaptive Management: The US Department of the Interior Technical Guide. Adaptive Management Working Group, US Department of the Interior, Washington, DC.

38

Wonkka, C.L., Rogers, W.E. and Kreuter, U.P., 2015. Legal barriers to effective ecosystem management: exploring linkages between liability, regulations, and prescribed fire. Ecological applications, 25(8), pp.2382-2393.

Wood, S. N. 2010. Statistical inference for noisy nonlinear ecological dynamic systems. Nature 466:1102-1107.

A1

APPENDICES

Appendix A. Contact information for key project personnel. 1) Clare Aslan, Principal Investigator Landscape Conservation Initiative Box 5694 Northern Arizona University Flagstaff, AZ 86011 clare.aslan@nau.edu 928-523-2487 2) Brett Dickson, Co-Principal Investigator Landscape Conservation Initiative Box 5694 Northern Arizona University Flagstaff, AZ 86011 brett.dickson@nau.edu 928-523-2487 3) Sasha Stortz, Research Associate Landscape Conservation Initiative Box 5694 Northern Arizona University Flagstaff, AZ 86011 sasha.stortz@nau.edu 928-523-2487 4) Martha Sample, Research Specialist Landscape Conservation Initiative Box 5694 Northern Arizona University Flagstaff, AZ 86011 martha.sample@nau.edu 928-523-2487

A2

Appendix B. List of Completed/Planned Scientific/Technical Publications/Science Delivery Products. Published articles

• Aslan, C. E., B. Petersen, A. Shiels, W. Haines, and C. Liang. 2018. Operationalizing resilience for conservation objectives: the 4 S’s. Restoration Ecology 26:1032-1038.

• Aslan, C. E., L. Samberg, B. G. Dickson, and M. E. Gray. 2018. Management thresholds

stemming from altered fire dynamics in present-day arid and semi-arid environments. Journal of Environmental Management 227:87-94.

Articles in preparation

• Aslan, C. E., B. G. Dickson, L. Zachmann, C. Levine, and L. Samberg. In prep. Our assumptions about non-native plants and fire are based on a tiny number of species. Fire Ecology Journal. Target submission date: September 2019.

• Aslan, C. E., S. Souther, M. Sample, M. Sandor, and S. Stortz. In prep. Jurisdictional

variation in management objectives and activities: relevance to altered fire regime. Journal of Environmental Management. Target submission date: October 2019.

• Sandor, M., C. Levine, C. E. Aslan, S. Souther, M. Sample, S. Stortz, and M. Gray. In

prep. Relationships between social and ecological resilience to fire in the Sonoran Desert. International Journal of Wildland Fire. Target submission date: October 2019.

Master’s thesis

• Perrone, M. 2018. Mapping Land Management Agency Resilience in the Sonoran Desert Ecoregion. MS Thesis. Environmental Science and Policy. Northern Arizona University.

Webinar

• JFSP Fire Exchange Network and Southwest Fire Science Consortium. May 2019. Stakeholder results webinar: FireAdapt: Examining the role of management thresholds in Sonoran Desert fire resilience.