Estimating social-ecological resilience: fire management futures in the Sonoran Desert
Data files
Oct 19, 2020 version files 77.10 KB
-
JFSP_All_Codedpoints_Translated.xlsx
Abstract
Methods
In this study, we used existing ecological layers and a stakeholder mapping approach to develop quantitative maps of social-ecological resilience to fire across the EPA Level III Sonoran Desert Ecoregion of Arizona. We used an overlay analysis to map coupled social-ecological resilience across the region. This analysis enabled us to identify areas of likely low and high resilience. Since resilience is meaningless unless examined in light of disturbance risk, we also modeled across the landscape both ecological and social aspects of fire risk. We then plotted resilience against modeled ecological and social fire risk to assess vulnerability across the study region.
Study system
The Sonoran Desert is the most biologically diverse desert in the US and also one of the regions of fastest human population growth, with more than 6 million residents according to the 2010 US Census (Dimmitt et al. 2015). Due to lack of fuels continuity in native plant communities, the Sonoran Desert is considered non-fire-adapted; occurrence of fire reduces both plant and associated animal diversity at local scales (McCaffrey 2015). However, fire risk has been increasing across the region in recent decades (Gray et al. 2014). Emergence of a novel, high-frequency fire regime in the Sonoran Desert is driven by increasingly variable precipitation in combination with the spread of invasive, fine fuels (Seager et al. 2007, Abatzoglou and Kolden 2011, McDonald and McPherson 2013, Moloney et al. 2019). These changes impact native plant communities with limited inherent resilience to disturbance, placing desert systems at risk of fundamental ecosystem state change (Abella 2009, Brooks and Chambers 2011). That is, a feedback loop promoting fire-adapted non-native species threatens to turn burned Sonoran Desert ecological communities into grasslands dominated by non-natives (McDonald & McPherson 2013). Resource managers and restoration professionals in the Sonoran confront biological invasions and aim to reduce fire risk in order to effectively retain functioning ecosystems and native Sonoran Desert biodiversity on the landscape (McCarty 2001). In the US, the Sonoran Desert Ecoregion consists largely of government-managed land, but is a patchwork of federal, state, and local jurisdictions, military bases, and tribal reservations. Management responses to fire across the region are varied and span a range of jurisdictional mandates as well as constraints and resources (Aslan et al. In revision).
Modeling ecological aspects of resilience
We gathered ecological spatial layers to (a) predict ecological large fire risk over the next 20 and 40 years across the study area, so that we could present risk maps to managers and discuss their responses, and (b) include ecological components of system resilience in our coupled systems GIS. We defined fire risk throughout the study as the probability of large fire (>1000 acres) if an ignition were to occur, and used the current and prior year maximum NDVI as predictors of annual fire risk (after Gray et al. 2014). We also included as dynamic model variables 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 (after Gray et al. 2014). Fixed variables in the model included road density, distance to urban development, surface heat load index, topographic roughness, elevation, aspect, and slope (Gray et al. 2014). To forecast fire risk into two future time periods, it was first necessary to forecast the dynamic predictor variables. We used data at 4-km resolution, downscaled from 12 Global Climate Models (GCMs), to forecast the meteorological predictors. To forecast NDVI into the future, we used 32 years of historical precipitation and 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 U.S. Sonoran Desert ecoregion, so that we had many historical fires to draw from, and from these burned/unburned points created 10 of these independent, random datasets, which were combined with the accompanying predictor variables described above to build 10 independent models that predict the probability of a large fire. The annualized estimate of fire risk was then averaged over these 10 datasets and over 12 GCMs. This approach thus incorporated the variability resulting from independent fire risk models as well as variability resulting from future climate projections, which heightened the robustness of final estimates. 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 can cause changes in land cover (Gray et al. 2014), and this worst-case scenario 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 a meaningful and challenging change over the conditions a given manager has experienced in their tenure.
To model ecological aspects of resilience, we mapped variables indicative of a system’s likelihood of retaining its characteristics and species in the face of disturbance, as indicated in ecological resilience literature. We thus created layers for topographic diversity, geophysical diversity, vegetation diversity, water availability, habitat connectivity (sensu Theobald et al. 2012), species richness, and human modification (Table 1). To create ecological rasters that could later be combined with social rasters, we summed these indicators to obtain a single value per location for ecological aspects of resilience across the Sonoran Desert.
Modeling social aspects of resilience
To collect data on social aspects of resilience, we designed a mapping exercise for land managers, who were the stakeholders for this study. We reached out to managers of all governmental jurisdictions in the study area (county, state, federal, and tribal) and invited them to participate in the study. To identify invitees, we examined land ownership maps of the region and used internet searches and our existing contacts to match management units to individual managers. In all, we obtained participation from managers of 25 jurisdictions, accounting for 79% of the study area (Table 2). We included only governmental jurisdictions, since private landowners represent an extremely small proportion of the study area (<8 percent). Throughout the below, we use the term “jurisdiction” to refer to a contiguous land area managed by a given government agency.
To all participating land managers, we presented background information about fire in the Sonoran Desert, drivers and consequences of fire regime change, and social-ecological resilience. We provided managers with maps of their jurisdictions displaying the projected large fire risk we had developed; maps included 10 randomly generated points established within the jurisdiction of each manager. We asked managers to identify on the map the locations of their various management objectives and activities, as well as the likelihood that these objectives and activities would be effective under projected future fire risk (Supplementary information). Managers provided this information for all of the random points and also for polygons they hand-drew on printed maps to indicate areas of particular concern (e.g., locations of cultural resources or endangered species management) in light of fire regime change. Following interviews, we transferred all management activity and objective information into a GIS. To do this, we transcribed into the digital platform the polygons of areas identified by participants as important. We attributed both random points and polygons with the objectives and activities relevant to those locations, as well as the manager-reported likelihood (see below) that activities could continue to achieve objectives, new activities could be employed, or new objectives could be adopted (Supplementary information). For analysis, we classified participant-reported management objectives and current management activities in the GIS into 9 objective categories and 12 activity categories (Table 3). We then imported rasters of each category, along with levels of likelihood as described above, as a grid of points into R version 2.14.1 (R Core Team, 2012).
After completing the mapping exercise, participants were asked to complete a paper survey reporting their experience with past environmental change, fire, and adaptive management. Responses to survey statements were recorded on Likert scales as 1-7 (strongly disagree to strongly agree) (Supplementary information). Survey questions were designed to assess jurisdictions’ past adaptability and constraints to change, and Likert-scale values were mapped onto the GIS at the scale of the jurisdiction. For institutional reasons or due to resource constraints, some agencies or individuals have difficulty changing their management methods, while others are more adaptive (LeQuire 2013, Haase 2013). Quantifying the flexibility of each jurisdiction can be critical to assessing the implications of changing conditions for management.
We defined the “state” of the social system in this study as the set of management objectives at a location, and the social element of resilience as the likelihood that current management objectives would continue to be met, either using current management practices or via the adoption of new practices. To quantify this, we used spatially-explicit data from both the interviews and the surveys. From the interviews, we used the answers (very unlikely to very likely) to the questions “[Given predicted future fire regimes] How likely are you to continue to meet current management objectives?” and “[Given predicted future fire regimes] How likely are you to change current management practices in order to be able to meet current management objectives?” Participants responded to these questions with respect to specific polygons or points using a Likert scale from 1-5 (where 1=very unlikely, 5=very likely), and those Likert values were mapped into the GIS as point and polygon attributes. From the survey, our estimate of social resilience was informed by the responses (Likert scale 1-7) to the statements “In my organization, we expect and try to prepare for ‘surprises’ in ecosystem behaviors.” and “I have seen my organization successfully change management strategies when necessary” (Supplementary information).
We defined social fire risk as a combination of four factors influencing whether an ignition becomes a large fire. From the interviews, we included 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. From the surveys, we included the responses to two statements about organizational resources: 3) “My organization has sufficient resources & personnel to manage fire and fuels on a day-to-day basis” and 4) “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 for use in the overlay analysis (below).
Although the indicators of ecological resilience that we included were measured objectively, these indicators of social resilience are clearly subjective, reflecting the experiences and opinions of the interviewed land managers. This is an important distinction to acknowledge, but we felt that our interview approach allowed us to capture the lived experiences of the managers and therefore incorporated the subtleties of fire response options that are difficult to adequately equate to more objective indicators of social management context such as parcel size, funding, staffing, etc.
Overlay analysis: coupled ecological and social estimate of resilience
We developed spatial layers of social-ecological fire risk as well as social-ecological resilience (Supplementary information). We scaled all ecological and social measures on a 0-1 scale, with 0 indicating lowest concern for managers and 1 indicating highest concern (Dressel et al. 2018). We used a simple additive model (i.e., a linear relationship) between ecological and social factors because, since research on quantitative social-ecological resilience is so limited, we are unable to defensibly select any nonlinear relationship that would mathematically inflate the contribution of either the social or ecological factors to the overall estimate of integrated resilience. To calculate total risk across the Sonoran Desert, then, we summed the ecological and social risk layers, and rescaled them to 1, giving equal weight to each. We did the same for total resilience. When rasters were at different resolutions, we resampled them to make all rasters equivalent in resolution to the social rasters. We performed a coarse check of sensitivity to the social component within the total resilience assessment by up-weighting or down-weighting the social resilience raster and comparing the resultant, scaled total resilience values for the socially up-weighted, socially down-weighted, and equal weights rasters.
Lastly, we plotted total risk against total resilience to determine vulnerability of locations to fire across the Sonoran Desert (sensu Comer et al. 2012). Vulnerability can be considered the intersection of system sensitivity (which in our study we equate to risk) and resilience, with the most vulnerable areas being those where high sensitivity and low resilience intersect (Comer et al. 2012). We translated these categorical vulnerability scores into a map to identify areas of low, medium, high, and very high vulnerability within the Sonoran Desert.