Mapping species of greatest conservation need and solar energy potential in the arid Southwest for future sustainable development

Study area

The study area encompasses the arid regions of the Southwest, spanning California, Nevada, Utah, Colorado, Arizona, New Mexico, and Texas (Fig. 1). It is rich in habitat and biodiversity, encompassing topographic extremes and wide climate diversity. Notably, it includes the country’s highest point, Mount Whitney, at 14,494 ft, and the lowest, Death Valley, at −282 ft (Southwest Climate Adaptation Science Center, 2023). Extreme topography variation can significantly influence various climatic parameters, such as temperature, precipitation, soil characteristics, and other ecological factors (Dillon et al., 2011). The Southwest is home to several ecosystems, including deserts, grasslands, woodlands, chaparral, tundra, wetlands, and various forested environments (Dahms & Geils, 1997). The region also supports extensive human activities and is home to some of the nation’s most productive agricultural land and urban areas, thus making this region an intriguing intersection of ecological complexity and human influence.

Species of greatest conservation need

The LeConte’s Thrasher (Toxostoma lecontei) and Bendire’s Thrasher (Toxostoma bendirei) are experiencing significant declines, making them among the fastest-declining species in the Southwest (Ammon et al., 2020). These birds are native to desert flats with sparse vegetation, such as saltbush, cholla cactus, and low shrubs (Sheppard, 2020; England & Laudenslayer Jr, 1993). Both species are particularly vulnerable to habitat alteration and loss due to their preference for specific vegetation types (Sheppard, 2020; England & Laudenslayer Jr, 1993). The risk is further exacerbated because SED projects often target landscapes with low-growing shrubs that can be easily removed during construction (Mierzwiak & Calka, 2017).

The Mojave Desert Tortoise (Gopherus agassizii) and the newly distinguished Sonoran Desert Tortoise (Gopherus morafkai) (Murphy et al., 2011) are biologically similar. Both species are characterized by their fossorial behavior; they construct burrows, creating microhabitats that provide shelter for themselves and many other desert inhabitants (Lovich & Ennen, 2011). The conservation status of G. agassizii is listed as critically endangered as it faces many threats, including those arising from renewable energy development (Berry et al., 2021).

The population of Burrowing Owl (Athene cunicularia) within the Southwest has steadily declined for many years, prompting conservation efforts to preserve this species (US Fish and Wildlife Service, 2023). The resident population within the Southwest is valuable to the long-term persistence of the species, considering its unique genetic diversity (Hughes, Daily & Ehrlich, 1998).

SED and SDM preparation

The combined approach to find areas of high habitat suitability for endangered species within areas of high SED potential included two major steps: (1) the use of abiotic variables to identify areas with the highest potential for SED and (2) the use of environmental variables and machine learning models to find suitable habitat of Species of Greatest Conservation Need.

SDM preparation and evaluation

Occurrence and environmental data.

We used occurrence data for the selected species obtained from GBIF (GBIF, 2023a; GBIF, 2023b; GBIF, 2023c; GBIF, 2023d), as well as solar and bioclimatic data from BioClim (WorldClim, 2020), and topography data from EarthEnv (Amatulli et al., 2018). To ensure high data quality and accuracy, the occurrence data underwent a thorough cleaning process, which included reducing spatial aggregation, removing duplicate occurrences and records with missing values, and incorporating pseudo-absences and background data. To mitigate the potential impacts of sampling bias and improve the quality control of the data, we first removed duplicate records and those with incomplete or inaccurate geographic positions (e.g., in the ocean). We then created a grid with the same resolution of the environmental variables to randomly select one per site (Hijmans et al., 2022).

Variable selection.

To minimize potential multicollinearity in the model, we used the VarSel function from the SDMTune package (Vignali et al., 2020) to remove highly correlated variables. This step helps to ensure that only variables with the most significant influence are included in the analysis, thereby improving the accuracy and reliability of the results. We generated 10,000 background locations using the dismo package in R (Hijmans et al., 2021). The data was then split into training and testing data sets, with a 20% allocation for testing. Then, a Maxtent model (Phillips, Dudik & Schapire, 2004) was employed, containing all variables (Elith et al., 2011). Maxent is successfully used to estimate species’ habitat suitability (Elith et al., 2011). The varSel function was applied to perform data-driven variable selection. Starting from the provided model, it iterates through all the variables, starting from the one with the highest contribution (permutation importance or maxent percent contribution). The method used for assessing variable correlation was Spearman’s rank correlation coefficient, and the threshold used was 0.7 (Vignali et al., 2020). The varSel function selects the least correlated variables based on the specified correlation threshold (Vignali et al., 2020). This process was performed for each target species.

Model evaluation & prediction.

First, we prepared presence and background locations and split the data into 80% for training and 20% for testing. Then, we used the subset of variables indicated by the variable selection process, the training dataset, and Maxent to estimate habitat suitability for each species. We restricted the produced habitat suitability to the known geographical ranges of the target species described by the IUCN (Figs. 2 and 3). Once we estimated habitat suitability, we applied the specificity/sensitivity threshold to transform suitability maps into binary (presence/background) (Liu et al., 2005). Lastly, we calculated the AUC (Area Under the Curve) and TSS (True Skill Statistics) values to evaluate each model. AUC and TSS are commonly used metrics for evaluating the performance of species distribution models, such as Maxent models (Elith et al., 2011). The AUC measures the model’s ability to distinguish between presences and absences. AUC scores range from 0 to 1, meaning higher AUC values indicate better model performance in determining suitable and unsuitable habitats (Elith et al., 2011). The TSS is another evaluation metric that combines specificity and sensitivity into a single statistic. TSS scores range from −1 to 1, meaning higher TSS values indicate better model performance in correctly identifying suitable and unsuitable habitats (Elith et al., 2011).

Identifying high-priority habitats

We calculated the overlap between the species presence produced by the SDM and the potential SED locations to identify overlapping areas. We considered areas where the potential SED locations and species presence had a value of 1 while assigning N.A. values to any cells with values other than 1. To express the overlap quantitatively, we calculated the percentage of overlap as the ratio of the sum of cells with overlapping values to the sum of cells with presence data (excluding N.A. values). The overlap percentage was reported as critical areas for SED development, i.e., areas where solar energy development could be pursued while considering the selected species’ habitat.

SED analysis

The sites identified through the site suitability analysis are depicted in Fig. 4. These selected locations exhibit consistent traits, including a slope of less than five degrees, proximity within five miles of an existing substation, classification within one of four land cover categories (barren land, shrubland, grassland, or cropland), and a prominent GHI value ranking within the top 30%. The geographic distribution of these sites indicates that the sites suitable for potential SED development are mainly distributed in Texas (all over the state), California, and Colorado (Fig. 4). A major portion of suitable sites was also observed outside of urban areas of Arizona and Colorado. The overlay between the map of sites suitable for SED development and the USPVDB map was moderate, since 48% of USPVDB facilities were included in our proposed solution.

SDM results

The SDM outcomes for each target species are presented in Figs. S.1 and S.2. These figures show the habitat suitability and presence patterns for each species. These findings provide valuable insights into the specific variables that strongly influence the distribution and suitability of habitat for each target species. The high AUC and TSS values emphasize the reliability and accuracy of the SDM results. Values ranged from 0.89−0.98 for AUC and 0.63−0.91 for TSS (Athene cunicularia: AUC = 0.87, TSS = 0.6, Toxostoma lecontei: AUC = 0.98, TSS = 0.89: Toxostoma bendirei, AUC = 0.98, TSS = 0.89: Gopherus agassizii, AUC = 0.98, TSS = 0.89, Gopherus morafkai: AUC = 0.98, TSS = 0.92).

The variable response curves for A. cunicularia (Fig. S.6), highlight the species’ sensitivity to precipitation and solar radiation, which are key drivers of its distribution across California, Colorado, New Mexico, and Texas. For G. agassizii (Fig. S.7) and T. lecontei (Fig. S.9) temperature and solar radiation significantly influence their suitability, with high suitability patterns evident in California. Lastly, the response curves for G. morafkai (Fig. S.8) and T. bendirei (Fig. S.10) indicate a strong association with temperature and solar radiation, leading to a high suitability concentration in Arizona.

High-priority habitats

The geographic patterns of areas with projected presences for endangered species within regions of high SED potential are illustrated in Figs. 5 and 6. For the Burrowing Owl (A. cunicularia), the most critical areas are distributed in Texas, California, and Arizona. For LeConte’s Thrasher (T. lecontei), Bendire’s Thrasher (T. bendirei), the Mojave Desert Tortoise (G. agassizii), and the Sonoran Desert Tortoise (G. morafkai), the most critical areas are observed in California and Arizona (Figs. 5 and 6).

Regarding the overlap between areas of projected presence and SED potential for non-migratory species, the highest values are observed for the Sonoran Desert Tortoise (G. morafkai, 46.36%) and LeConte’s Thrasher (T. lecontei, 35.81%), while the lowest is observed for the Mojave Desert Tortoise (G. agassizii, 16.69%).

For migratory species, the overlap of projected presences and SED potential for Bendire’s Thrasher (T. bendirei) is 31.44% for resident extant individuals who are present year-round and 6.38% for breeding individuals who migrate for a portion of the year. Similarly, for the Burrowing Owl (A. cunicularia), the overlap is 6.76% for resident extant individuals and 13.63% for breeding individuals. This distinction is important as it highlights the different conservation needs and potential impacts of SED projects on species with both migratory and resident populations.

SED suitability analysis

The site suitability analysis identifies potential regions for large-scale solar energy developments, including (a) California Central Valley: This region’s potential for future solar projects is primarily due to its relatively flat terrain and extensive agricultural land. The flat terrain facilitates easier installation and maintenance of solar panels. At the same time, agricultural land may offer large, open spaces suitable for SED (Doljak & Stanojević, 2017; Mierzwiak & Calka, 2017; Bukhary, Ahmad & Batista, 2018; Nebey, Taye & Workineh, 2020); (b) Greater Phoenix Valley, Arizona: The dense cluster of potential sites in this area is due to the combination of flat topography and high Global Horizontal Irradiance levels. High GHI levels indicate abundant solar radiation, which is necessary for maximizing the efficiency and energy output of solar panels (Gbémou et al., 2021; GlobalSolarAtlas, 2021; Doljak & Stanojević, 2017), (c) Texas: The analysis reveals suitable locations throughout Texas. The widespread potential in Texas is attributed to high GHI levels and ideal land cover, including shrublands, croplands, and grasslands. These land cover types are often well-suited for SED because they can provide large, relatively unobstructed areas for panel installation (Mierzwiak & Calka, 2017; Bukhary, Ahmad & Batista, 2018; Nebey, Taye & Workineh, 2020).

Our results, however, revealed a moderate overlap between areas of high suitability of SED and the US Large-Scale Solar Photovoltaic Database (USPVDB, Fujita et al., 2023). This moderate overlay could be linked to the lack of incorporation of datasets (i.e., residential, commercial and industrial PV facilities) representing the current distribution of operational solar facilities. We did not incorporate them, because these are not widely available at the scale of our study. Even with these limitations, our analysis offers a predictive framework, serving as a backdrop for conservation data. By identifying areas with SED potential the inclusion of biodiversity data can highlight conflict with sensitive habitats.

SDM results

Our findings indicate a significant correlation between habitat suitability for each of the five target species and climate and environmental variables, albeit with some variation. Notably, three key climatic and environmental variables exhibited the most substantial influence, consistently shared across all five target species. Precipitation emerged as a top influential factor for T. lecontei, T. bendirei, G. morafkai, and A. cunicularia. This correlation can be attributed to their dietary reliance on arthropods and small rodents, populations of which frequently correlate with precipitation levels and the overall health of vegetation (Desmond & Sutton, 2017; McDonald, Korfanta & Lantz, 2004). Temperature played a crucial role for T. lecontei, G. agassizii, and G. morafkai, with temperature shifts impacting activity levels, feeding patterns, and species survivorship (Meyer, 2008 Sheppard, 1970). Topographic variables like elevation and Topographic Roughness Index (TRI) significantly impacted T. bendirei and A. cunicularia, influencing habitat preferences, distribution, and adaptation to specific landscapes (Desmond & Sutton, 2017; Meyer, 2008).

The influence of these variables on the species’ habitat suitability emphasizes their critical importance in shaping the ecological conditions for each target species. Consequently, any significant or prolonged alterations in the environmental or climatic variables could directly affect the target species. For instance, precipitation’s impact on food source availability reveals concerns about these species’ ability to find sufficient sustenance, potentially hindering reproductive success and population health (Desmond & Sutton, 2017; McDonald, Korfanta & Lantz, 2004). Temperature’s influence on activity levels and survivorship highlights these species’ adaptations to cope with extreme heat conditions, with different responses seen in burrowing owls and desert tortoises (Meyer, 2008; Sheppard, 1970). The connection between topography and habitat preferences also showcases the importance of specific landscape characteristics for T. bendirei and A. cunicularia, influencing their presence in landscapes with features like exposed ground patches or open flat land (Desmond & Sutton, 2017; Meyer, 2008).

The Southwest has been identified as a climate change “hotspot,” with projected increases in air temperature, aridity, and seasonal variability (Gutzler & Robbins, 2011). Arid environments, such as deserts in the Southwest, are particularly vulnerable to the impacts of climate change, supported by abundant evidence (Archer & Predick, 2008; MacDonald, 2010; Garfin, 2013). For instance, the region is expected to experience fewer frost days, more frequent unusually high temperatures, increased water demand, and a higher frequency and intensity of extreme events such as droughts, heatwaves, and floods (Archer & Predick, 2008; MacDonald, 2010; Garfin, 2013). While changes in precipitation hold a higher degree of uncertainty, likely, both precipitation and temperature variations will directly impact vegetation and ecosystem processes throughout the Southwest (Archer & Predick, 2008). Nevertheless, it’s important to acknowledge that these findings are grounded in current data, and future changes in these variables due to factors such as climate change and future urban expansion may introduce additional complexities to these species’ habitat suitability and species welfare. Given the strong correlation between temperature, precipitation, and the overall fitness of target species, the suitable habitat is subject to change based on future environmental and climate changes.

Overlap results—implications for conservation

Our analysis considered habitat suitability and projected species presence to gain insights into the potential impacts of future SED locations. Species presence refers to documented occurrences of a particular species within a specific geographic area. At the same time, habitat suitability assesses the environmental conditions and resources necessary for a species to thrive and persist. It’s important to note that the percentages of species presence provided in our analysis were determined within the ranges specified by the IUCN, and they are approximations. Additionally, these percentages may change based on currently available data and the data used in our analysis is subject to change over time.

Limitations

While our study provides valuable insights into the overlap between high-priority habitats and suitable locations for SED, certain limitations must be acknowledged. First, the site suitability analysis relies on current substation data without accounting for future substation or transmission line availability. Additionally, ground truthing with existing solar data is needed to evaluate the areas of SED better, as commercial and residential solar data were excluded from the analysis due to a lack of open access. Including this data could offer important insights for future studies, and its absence may affect the long-term viability and feasibility of the identified SED sites. We acknowledge that the SED model can be further refined to enhance its predictive accuracy and better capture the complexities of SED suitability.

Secondly, the focus on the species’ range within the United States Southwest, while providing valuable information, may not fully capture the complete spatial distribution or habitat suitability, as the range of each target species extends beyond our study area.

Furthermore, it’s important to recognize that our spatial overlap analysis provides an approximation of the interaction between the potential SED sites and the target species’ habitat. The actual spatial dynamics at a finer scale may differ, and the species’ habitat suitability could change over time due to factors like climate change, impacting distribution and habitat availability. It is also important to note that the percentages of species presence and habitat suitability provided in our study were determined within the ranges specified by the IUCN, and they are approximations. Additionally, these percentages may be subject to change based on currently available data, and the data used in our analysis will be subject to change over time.

Although site suitability analyses are standard in research and planning, our inclusion of species-specific data provides added ecological insight by highlighting interactions between future SEDs and high-priority habitats. We envision that similar analyses expanding to a broader range of taxa could prove valuable in guiding future conservation decisions. By recognizing these limitations, we aim to inspire further studies and conservation efforts that actively address potential changes and uncertainties tied to substation availability and environmental conditions.