Study area and site selection

Iowa’s landscape is dominated by agriculture consisting of seven distinct landform regions: Northwest Iowa Plains, Missouri Alluvial Plain, Western Loess Hills, Des Moines Lobe, Southern Iowa Drift Plain, Iowan Surface, Paleozoic Plateau, and Mississippi Alluvial Plain [49]. Historically, these landform regions produced a variety of habitats including tallgrass prairie, savannas, and pothole wetlands that comprised most of the Iowa landscape. Although some of these habitats remain, 63% of Iowa is now agricultural land and 2.5% is developed [50].

As part of the Iowa Multiple Species Inventory and Monitoring (MSIM) program [51] we conducted our study on 295 public properties greater than 97 ha (Fig 1, S1 Table). A small number of private properties were included in our study to document SGCN on lands enrolled in the Landowner Incentive Program (see http://wsfrprograms.fws.gov/subpages/grantprograms/lip/lip.htm). We classified all properties according to 19 different habitat types (S1 Table) listed in the 2005 Iowa Wildlife Action Plan [50] using aerial imagery, 2002 Iowa landcover data in ArcGIS (ver. 10.1) [52], and knowledge of the local land manager. We classified properties by one or more habitat types depending on the degree of habitat diversity on the property. We then divided Iowa into four equal management districts for property selection to allow for equal representation of all habitat types across the state. We utilized a stratified random sampling technique to select properties which involved selecting properties of a specified habitat type (primary stratum) within each management district (secondary stratum). Once a property was selected for a particular habitat type, it was excluded from selection for a different habitat type. We then defined the “core” habitat type as the habitat type for which the property was selected. We repeated this procedure from 2006 to 2014 to obtain approximately 75 properties to survey each year. Beginning in 2017, we selected at least two properties each year to be surveyed annually. In this way, we increased the number of annual properties each year, resulting in 27 properties surveyed for three or more years, 18 surveyed for four or more years, and 10 surveyed for five or more years (S1 Table). The procedure for selecting annual properties was similar to that mentioned above, although we limited our selection to those properties owned and managed by the Iowa Department of Natural Resources Wildlife Bureau for ease of logistics. We obtained Special Use Permits from the U.S. Fish and Wildlife Service for data collection on National Wildlife Refuges (Permit # 2013–016), and a Scientific Collector’s Permit from the Iowa Department of Natural Resources (Permit # SC872). All field methods for this study were reviewed and approved by the Iowa State University Institutional Animal Care and Use Committee (IACUC; Protocol #3-12-7326-Z).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Sites surveyed for herptiles (black dots) in Iowa, 2006–2014.

Basemap data provided by [65].

https://doi.org/10.1371/journal.pone.0306655.g001

Herptile surveys

We used three different methods for surveying herptiles. Occurrence data collected using all methods were compiled for all years to establish a single occurrence history for each species on each property. Properties were also clipped to the known ranges of each species to increase accuracy for predictive models, resulting in unequal sample sizes for each species.

We conducted standardized visual encounter surveys for herptiles each year from April to October in the period 2006–2014. This survey method involved a timed search for herptiles in suitable habitats identified and mapped on each property at the beginning of the survey year. We divided each year into three survey seasons to minimize seasonal variation in detection probability of different species. Those seasons were spring (15 April–15 June), summer (16 June–15 August), and fall (16 August–15 October). We surveyed each property for four person-hours twice during each of the three survey seasons for a maximum of six visits (24 person-hours) per year. We surveyed properties only during the year in which they were selected except for those properties that were selected for annual surveys. Surveys were conducted two weeks apart on average to increase independence among visits. During the spring and fall, we conducted surveys during warmer hours of the day, which typically fell between 10:00 hours and 18:00 hours, to maximize detection probabilities. We did not perform nocturnal vocalization surveys for technician safety purposes. Two species, Cope’s gray treefrog (Hyla chrysoscelis) and eastern gray treefrog (Hyla versicolor), can only be distinguished via auditory detections. Given that we could not confidently differentiate between the two species, we combined their detection histories and modeled both species in aggregate as the gray treefrog complex.

We also placed coverboards at both systematic and random locations on the property to increase detection probabilities for herptiles. Six coverboards were placed systematically in a hexagonal arrangement 200 m apart in the core habitat type on the property and an additional nine coverboards were placed in locations that seemed most likely to capture individuals, for a total of 15 coverboards per property. Coverboards are frequently used by many species of herpetofauna as habitats for thermoregulation and cover, but may be a more efficient method for capturing reptiles than amphibians [53–56]. Each coverboard was marked with a global positioning system (GPS) unit and checked during each visual encounter survey. All herptiles observed during the survey were identified to species [57]. Surveys were not conducted on cool days (< 10°C) or during periods of rain.

Lastly, we used aquatic traps to target amphibians and turtles. We set a variety of aquatic traps, which typically consisted of three hoop nets, three box traps, three fish traps, and up to six minnow traps, for two trap nights once per each of three survey seasons described above for a total of six trap nights per property. We identified water bodies on each property, which included ponds, wetlands, lakes, rivers, and streams. Aquatic traps were placed within waterbodies at locations deemed most likely for captures. We baited larger traps with fish (e.g., sardines, dead grass carp [Ctenopharyngodon idella]) and checked traps once daily. We replaced bait as needed when traps were checked.

Habitat variables

We used ArcGIS (ver. 10.1) [46] to measure landscape-level habitat variables within 200-m, 500-m, and 1000-m radii of each property. Buffers at each scale were placed around each survey mid-point for each property using the buffer tool in ArcGIS toolbox (Analysis Tools, Proximity, Buffer) [52]. We then clipped a 2009 Iowa Landcover file, developed by the Iowa Department of Natural Resources, to all buffers at all properties using the “clipraster” command in the Geospatial Modeling Environment (GME) package [58]. The 2009 Iowa Landcover was developed using satellite imagery and includes 15 classifications such as coniferous and deciduous forests, corn and soybean crops, and anthropogenic structures, among others, at a 3-m resolution. We then aggregated the classifications into six landcover types: water, agriculture, development, wetland, woodland, and grassland. These landcover types were selected due to their potential influence on our focal species and their interest to land managers. We opted to use landcover data for Iowa from 2009 because it is centered in our time series of surveys (2006–2014) and because the file is higher resolution (3m) than most other landcover products.

We used FRAGSTATS (ver. 3.4) [59] to summarize landcover metrics at each buffer. FRAGSTATS is a computer program that analyzes spatial pattern based on categorical maps and allows the user to pick from a variety of metrics to assess landscape configuration [59]. We used the percentage of landscape (PLND), largest patch index (LPI), patch density (PD), and edge density (ED) for all analyses. Percentage of landscape is the area of a land-use classification divided by the total area of the landscape. Largest patch index is the area of the largest patch of a land-use classification divided by the total area of the landscape. Patch density is the count of patches corresponding to a land-use classification divided by the landscape area. Edge density is the amount of linear edge on the landscape for a land-use classification divided by landscape area. We extracted each FRAGSTATS metric for each of the six landcover types within three different spatial scales (200, 500, and 1000-m radii) to be included as prediction covariates in our models.

Predictive models

We used a robust design occupancy model framework [47] in Program Mark [60] to estimate occupancy and colonization probabilities for 15 herptile species in Iowa (Table 1). Robust design occupancy models account for imperfect detection across multiple survey years [41], allowing for the estimation of probability of initial occupancy (ψ), probability of local colonization (γ), probability of local extinction (ε), and detection probability (p) [47] as response variables. Unlike the single-season occupancy model where sites are closed to changes in occupancy state during the primary sampling season [1], the robust design occupancy model assumes sites are closed to changes in occupancy state between secondary sampling intervals (e.g., sampling occasions within a year) but are open to changes in occupancy state between primary sampling intervals (e.g., years) [47]. This allows for the evaluation of metapopulation dynamics through the process of determining the probability a site will remain occupied or become colonized. We used real parameter estimates for occupancy, which represent the initial occupancy probability at the first year of survey. The first year of survey varies by property, as properties were continually added throughout the survey period. We did not estimate local extinction probabilities for two reasons. First, we did not have sufficient data for some species for the full model to converge, so we chose to focus inferences on occupancy and colonization. Second, we wanted a numerical estimate of occupancy as opposed to one derived recursively from colonization and extinction probabilities [47]. One possible source of bias is that these models do not account for spatial autocorrelation. However, all properties were separated by a minimum distance of 530 m and were on average 10.2 km away from the nearest property, which we deemed adequate to meet the assumption of spatial independence.

Landscape-level habitat variables (S2 Table, S3 Table) and detection covariates were included in occupancy model sets as predictor variables along with intercept-only null models. The primary sampling intervals were the survey years (2006–2012, 2014) and the secondary sampling intervals were the survey occasions (days) within each sampling year (April-October). Data from 2013 were excluded from model building and later used for testing the accuracy of predictive models. We chose 2013 for model testing because it was one of the final years of surveys and more properties were surveyed in 2013 (n = 74) than in 2014 (n = 27). We estimated probability of occupancy and colonization on primary sampling intervals, and detection probability on secondary sampling intervals. Models were constructed in a hierarchical fashion separately for each species [61] and evaluated using Akaike’s Information Criterion adjusted for small sample sizes (AICc) [62]. Models with ΔAICc ≤ 2 were considered to have strong support [62]. We first modeled environmental covariates (e.g., temperature [˚F], percent cloud cover) on detection probability while keeping all other parameters constant. Using the best model from this step, we then modeled landscape-level covariates on probability of occupancy. Our goal was to identify the most important landscape-level factor that predicts occupancy and colonization probabilities for each species. Therefore, we compared each predictor in separate models to identify the variable with the most explanatory power, except for two species (northern redbelly snake and gray tree frog complex) where life-history characteristics were a priori expected to depend on environmental interactions (as determined by LeClere [57], S4 Table). We then selected the best model with effects on detection probability and probability of occupancy to be used in modeling effects on probability of colonization. Among the top models, we compared model deviance as an indicator of goodness-of-fit. We ultimately determined the top ranked model to be the best for predictions for every species, given that model averaging does not increase prediction accuracy, even when the top models have similar AICc weights [63]. We used the top model as a basis for inference to build our predictive maps.

To predict probability of occupancy across Iowa for each species, we first established a 1000-m point grid across the entire state using the tools extension package NPS AlaskaPak [64]. We used these points to extract landscape-level habitat characteristics of interest across Iowa. We used the same process described above for our sampled properties to estimate landscape-level habitat characteristics for each of the four land-use classifications, where the 2009 Iowa Landcover file was clipped to three buffers (200 m, 500 m, and 1000 m radius) around each point in the grid. We then developed predictive models for each parameter for each species using the linear coefficients from the top occupancy model. We calculated a probability of occupancy and colonization for each point in the point grid by taking the logit transformation of the product of the linear coefficient of the covariate in the top model (corresponding to the parameter of interest) and the value for the covariate at the respective point. The probabilities for all points in the grid were mapped using the raster conversion tool (Conversion Tools, To Raster, Point to Raster) [52] to create a raster corresponding to the predicted probability.

Using data from 2013, we tested the validity of our models using the area under the receiver operating characteristic curve (AUC). An AUC score of 0.5 suggests that the model prediction did not successfully separate true predictions from false predictions, compared to a score of 1.0, where all occurrence points were predicted correctly [65]. We considered models useful if AUC was ≥ 0.70 [65].

Lastly, we combined the predictive maps for all species using the Raster Calculator tool to sum the predictive occupancy rasters for all species with an AUC ≥ 0.70. The resulting raster provides a SGCN herptile hotspot map for Iowa.

Amphibians

Wetland density has been found to increase amphibian occupancy due to species’ relatively small core ranges around a wetland base [68–70]. A high density of wetlands can increase variability of wetland characteristics, such as hydroperiods and emergent vegetation that have been found to affect occupancy of some amphibian species [71–74]. In our study, this was true for the southern leopard frog, where occupancy probability was best explained by water patch density at the 1000-m scale. We also found that tiger salamander colonization probability increased with wetland patch density. In Illinois, wetland connectivity and proximity has been shown to be an important metric to tiger salamander occupancy and colonization [75, 76] and patch density can be a useful indicator of habitat connectivity in a metapopulation [77]. Our predictions on tiger salamander colonization probability were low across Iowa, but we found high probabilities within specific regions of the state, such as the Prairie Pothole Region (PPR) in north central Iowa. However, due to agriculture practices and changing weather patterns, the wetter regions of the state, such as the PPR, are under threat [78, 79]. Increasing wetland variability has been shown to alter metapopulation dynamics of amphibians that rely on wetlands to breed, where spatiotemporal variation in wetland hydrology can alter structural and functional connectivity of wetland patches [79]. To combat these stressors on amphibians and support occupancy and colonization, increasing wetland density and wetland size could provide the variation in wetland characteristics to support occupancy and colonization of new areas [80]. Since the 1990s, funds from the Wetland Reserve Program (WRP) has increased the area of wetlands in Iowa by approximately 5,000 ha, which had been converted into agricultural development [16]. These restored wetlands have been shown to have occupancy probabilities ranging from 81–100% for frogs and toads, where wetland colonization was faster when restored wetland were close to source wetlands [81].

Although wetlands are often considered the most important habitat factor for the occupancy of amphibian species, upland habitat can be just as critical [82–84]. Agricultural development may limit the movement between populations, thereby negatively affecting occupancy of northern leopard frog and Blanchard’s cricket frog. For example, Swanson et al. [24] found fewer telemetry locations of northern leopard frog in agricultural patches compared to adjacent grasslands and wetlands. However, the influence of agricultural practices and types of agriculture on occupancy and habitat use may be conditional on agricultural intensity [85, 86]. Lower intensity practices, such as cattle grazing or cover crops, have shown to improve occupancy as compared to high-intensity row cropping practices [85, 86]. While some species, such as the American toad, can use row crops in the late growing season [87], we found that the top model for four amphibian species included a negative effect of agriculture. Consequently, broad-scale conversion of native uplands into agriculture has likely limited the dispersal ability of several amphibian species of greatest conservation need statewide. Our findings suggest that conserving remaining grasslands and wetlands or restoring such areas can support the colonization of species like the Blanchard’s cricket frog. Additionally, grasslands can provide cover that can aid in reducing desiccation and predation, thereby allowing amphibians to disperse more easily [88].

Reptiles

Similar to amphibians, agriculture was the most important landscape feature affecting the occupancy of reptiles. This relationship was negative for four of these species, whereas Blanding’s turtle exhibited a positive relationship with agriculture. Many turtles, including Blanding’s, nest during May and June [58] when corn and soybean fields often have little to no vegetative growth. Turtles are often attracted to cultivated land for nesting because they seek out open-canopied habitats that allow subterranean nests adequate warmth to facilitate embryonic development [89–91]. However, nesting in agriculture fields can be an ecological trap because corn and soybeans often grow to heights that shade out nests, making the areas too cool for adequate development [91, 92]. Additionally, given the relatively long lifespan of Blanding’s turtles [58], occupancy in agricultural areas may result from remnant populations in new development. Whether Blanding’s turtle occupancy is actually higher in large patches of agriculture warrants further study. As for the four species of reptiles that exhibited a negative association with agriculture, this likely stems from a lack of diversity of microhabitats, flora, and fauna, and consequently, food resources in agricultural landscapes [93–95]. Additionally, agricultural chemicals can harm developing young [92] and kill invertebrates that lizards and juvenile snakes feed upon, crops can block visual cues that reptiles need to orient [95], and machinery likely kills reptiles during tilling, planting, chemical application, and harvesting [96]. Two species, northern watersnake and common snapping turtle, showed negative effects of agriculture on colonization probabilities. Although common snapping turtle are considered relatively tolerant of agriculture due to their common nesting forays into agriculture fields [92], our results suggest such presence is probably not indicative of preference.

Woodlands had a negative effect on common snapping turtle occupancy and were included in a negative interaction with wetlands in the top model for northern redbelly snake colonization. Like Blanding’s turtle, common snapping turtles construct subterranean nests to oviposit eggs, and the temperature of the nest influences both somatic and sexual development [91, 97]. A lack of open-canopied habitats that are preferred for nesting could contribute to our finding that forested areas negatively affect their occupancies. However, for some woodland specialists, such as northern redbelly snake, woodland edges along rivers and streams can act as travel corridors in Iowa’s agricultural matrix. Woodlands in Iowa have decreased since the mid 1800’s at the state scale [98]; however, forest cover has increased in Iowa’s urban areas [99]. We found occupancy probabilities of common snapping turtle decreased in urban areas. This is most apparent in the spatial predictions of common snapping turtle occupancy probability, where areas of lowest probabilities are in Des Moines and Cedar Rapids regions. However, the predicted colonization probabilities in urban areas were comparatively high for common snapping turtles, likely due to limited agricultural development in these areas. This contradiction illustrates the difficulty in interpreting the effect of urbanization on species, given that urbanization often correlates with other landcover types.

Water bodies and wetlands had species-specific effects on occupancy or colonization for seven species of reptiles coinciding with their unique ecologies. For example, western fox snake, which is commonly found near wetland edges [58], had a higher occupancy probability in areas with higher percentages of water; whereas plains garter snake, an upland specialist, had decreased occupancy probabilities in areas with high water LPI. Colonization probabilities were negatively affected by water and wetlands for two reptile species. Such negative relationships on colonization may represent tradeoff with preferred habitat. For example, wetland patch density decreased colonization of plains garter snake and water area decreased colonization of prairie ringneck snake. Plains garter snakes prefer upland habitat and prairie ringneck snakes prefer rocky terrestrial areas [58], which are both precluded by wetlands and water. In contrast, positive landscape associations likely represent greater connectivity with preferred habitat, such as spiny softshell colonization increasing with water patch density and western fox snake colonization increasing with water LPI.

Threats to wetlands in the PPR and native grasslands in the midwestern United States pose increased risk to the viability of species that rely on those habitats [99–102]. Changing weather patterns are predicted to reduce wetland productivity throughout the PPR [103], which will likely decrease habitat for wetland-dependent species. This is particularly true for wetlands in the northern and western portion of the PPR, which are at greater risk of drying and instability due to drought [103]. The eastern border of the PPR, including Iowa’s Des Moines Lobe ecoregion, may be a potential climatic refuge for wetlands [103] and wetlands edges. However, this refuge potential may be limited due to the widespread conversion of wetlands and grasslands to agriculture in Iowa [103]. Since the mid-1800s, almost 75% of Iowa’s grasslands have been converted into agriculture statewide and the majority of wetlands in the Des Moines lobe ecoregion have been drained [98], thereby restricting large patches of wetlands and grasslands. Grasslands in Iowa face an additional threat of woody encroachment [104], which poses a strong threat to the current abundances of reptiles but may increase some amphibian abundances [102]. For example, in northwest Indiana, a gradient of open grasslands to forest corresponded to a decrease in reptile abundances but increases in amphibian abundances [102]. Consequently, weather and broad-scale conversion of wetlands and grasslands likely pose the greatest risks to habitat for many SGCN herptiles, potentially resulting in community composition shifts. Future habitat management and protection efforts in Iowa should prioritize areas with the highest probability of species’ occupancy and colonization probabilities to increase the efficiency of efforts. Our analyses of cumulative occupancy probabilities indicate that these regions primarily include large water bodies, such as the Iowa River, Mississippi River, and Lake Red Rock, and their adjacent uplands.