We modeled corridors and tested three categories of corridor validation methods, categorized by data intensity, using a common corridor model. Our results highlight the subjectivity in corridor modeling and the importance of adequate corridor validation. Each of the three resistance surface transformations resulted in different spatial distributions of corridors, leaving the user to match management objectives with corridor distribution. Our results underscore the need for careful project planning to allow for corridor validation, and establish a novel method for more efficient single-species corridor validation.
We demonstrate with sample data from Florida black bears that corridor selection varies with resistance transformation and validation category. Despite the high correlation among the three continuous flow outputs, little of the top 10% of the corridors across the resistance surfaces overlapped, indicating relatively low spatial concordance of the highest flow areas across resistance transformation outputs (Fig. 1). Though we did not have validation data available for the entire state of Florida, the variability seen in the Highlands-Glades population validation likely reflects broader patterns. To create corridors statewide, we recommend validating the entire state with independent data. Across all validation results, the C2 or C025 black bear models may be considered best in Florida. With category 1 validation, we found that the top 10% flow areas of the C2 surface were the most specific, efficient use of space, including the greatest number of bear locations in the least amount of area (bears per km2). The C8 corridors identified areas used by individual dispersers but likely missed important areas not used by the entire population (Fig. 1). Those corridors may provide insights for managers when they must prioritize unique genetic movements or conservation planning that aims to preserve connectivity rather than core habitat.
In category 2 validation, we found no distinguishing differences using the point comparison or buffered point comparison methods; in all cases, bear locations had significantly higher flow values than random locations, indicating all resistance surfaces resulted in high flow areas in bear usage areas. However, the GLMM results indicated the C025 resistance model (approximately a linear transformation of the habitat suitability model to resistance) resulted in a stronger association of bear locations and higher flow areas, while the C8 transformation resulted in the weakest association. This was expected because the C025 transformation resulted in higher suitability values corresponding to higher resistance values, whereas higher c value transformations (such as the C8 model) result in higher suitability areas having lower resistance values. If we were to use the category 2 validation method in isolation, we may determine that the C025 model had the best results, but when taken in context with the category 1 validation results, the choice is not as clear – highest flow areas do not necessarily equate to more bears/km2.
Category 3 validation methods were inconsistent in identifying a top corridor model between C025 and C2. The novel use of a step-selection function to determine whether bears are selecting higher flow areas compared to available locations indicated the C2 model best represented bear corridors. However, when compared to the null model, C025 had the highest average, indicating higher flow than expected by an isolation-by-distance null model. This model was the only model where the confidence interval did not overlap zero. The C8 model in both of the category 3 validation methods performed the worst and had lower flow than expected compared to the null model. This was likely due to the transformation including more moderate suitability areas as low resistance areas. However, management objectives still must be considered given the unique output of C8.
The multispecies FEGN corridors or FWC, as expected, were the least efficient in terms of solely identifying bear locations per km2. From a broad ecological perspective, using FWC as conservation guidance can help Florida meet multiple conservation objectives, but does not provide conservation goals tailored specifically to black bears and enable tracking of the progress of this species. When interpreting our three models together, compared with FWC, and considering the behavior of the underlying transformation, we suggest the C2 model performed the best, as seen in the category 1 and 3 validation results. Still, the selection of the final model should be informed by management and conservation goals and by category 4 validation methods such as mark recapture/resight or gene flow analyses.
Validation of corridor work has increased in recent years, but validation of outputs using independent data is still quite rare (Riordan-Short et al. 2023). Testing the sensitivity of differing modeling decisions and using multiple categories of validation should be considered a best practice of corridor modeling. As shown here, using a single transformation on the resistance surface can result in inefficient corridors. The use of category 1 and category 3 validation methods revealed different ‘best’ corridors in our example population. If both differing resistance surfaces and multiple validation techniques are not used, the modeler may come to the conclusion that one resistance surface has high efficiency and propose corridors that capture only movement patterns of a subset of individuals and likely miss out on greatest possible gene flow. On the other hand, identifying wide multispecies corridors, which include the greatest number of species locations, may be wise ecologically, but without dedicated funding, these types of corridors may not be realistically conserved in multi-use, developing landscapes. More specifically, underfunded multi-species corridors could end up with scattered parcels that serve none of the species it was designed to support.
By testing multiple resistance surfaces and employing multiple validation techniques, the modeler can better understand the impact of parameter choices on corridors and better match proposed corridors to the population’s movement as well as to management goals. At a minimum, we suggest that researchers should evaluate the percentage of animal locations within corridors (category 1 validation) and should incorporate multiple validation methods into their modeling workflow. As one works through the validation framework from category 1 to category 4 validation, costs and time increase. Ensuring or restoring gene flow across a region is often the motivating factor underlying connectivity research, so if category 4 validation is available, other categories of validation may not be necessary. However, our results suggest the use of categories 1–3 in concert will likely lead to more informed decisions and possibly more effective corridors. We recognize that this proposed framework for validation is not an exhaustive list, but we present this framework as a first step in advancing and standardizing corridor validation. We encourage others to expand on this framework and to identify other robust validation methods, such as our novel usage of SSE.
The need for multiple validation methods springs from the variety of ways in which corridors can be modeled and the various decisions that are necessary along the modeling process. We selected some of the more common methods, and we recommend modelers carefully consider all options and choose the best option for their data, landscape, and management and/or conservation objectives. There is mounting evidence that linear inverse transformations may not accurately reflect the response of some species to human modification (Trainor et al. 2013; Keeley et al. 2016; Belote et al. 2022). As such, more care should be taken to go beyond empirical habitat suitability surfaces and match resistance surfaces with dispersal processes on the ground. Keeley et al. (2016) identified a method with which empirical corridors can be evaluated against random corridors to test which cost surface transformation performs the best. Because this is a relational comparison for this category of corridor modeling and there is no standard transformation, we deemed this comparison more relevant than comparisons between corridor modeling methods or differing corridor width tests, which are important considerations but have larger evidence bases from which to guide decisions (Poor et al. 2012; LaPoint et al. 2013; Cushman et al. 2013; McClure et al. 2016; Zeller et al. 2018; Lalechere and Berges 2021; Kumar and Cushman 2022). Methods choices may have an outsized impact on model results and the implications of these choices should be better understood and documented by modelers and communicated to landscape and species managers to ensure robust, functional corridors based on their specific management objectives, financial resources, and timelines.