A Probabilistic Co‐Occurrence Approach for Estimating Likelihood of Spatial Overlap Between Listed Species Distribution and Pesticide Use Patterns

Abstract

Characterizing potential spatial overlap between federally threatened and endangered (“listed”) species distributions and registered pesticide use patterns is important for accurate risk assessment of threatened and endangered species. Because accurate range information for such rare species is often limited and agricultural pesticide use patterns are dynamic, simple spatial co‐occurrence methods may overestimate or underestimate overlap and result in decisions that benefit neither listed species nor the regulatory process. Here, we demonstrate a new method of co‐occurrence analysis that employs probability theory to estimate spatial distribution of rare species populations and areas of pesticide use to determine the likelihood of potential exposure. Specifically, we 1) describe a probabilistic method to estimate pesticide use based on crop production patterns; 2) construct species distribution models for 2 listed insect species whose ranges were previously incompletely described, the rusty‐patched bumble bee (Bombus affinis) and the Poweshiek skipperling (Oarisma poweshiek); and 3) develop a probabilistic co‐occurrence methodology and assessment framework. Using the principles of the Bayes' theorem, we constructed probabilistic spatial models of pesticide use areas by integrating information from land‐cover spatial data, agriculture statistics, and remote‐sensing data. We used maximum entropy methods to build species distribution models for 2 listed insects based on species collection and observation records and predictor variables relevant to the species' biogeography and natural history. We further developed novel methods for refinement of these models at spatial scales relevant to US Fish and Wildlife Service (FWS) regulatory priorities (e.g., critical habitat areas). Integrating both probabilistic assessments and focusing on USFWS priority management areas, we demonstrate that spatial overlap (i.e., potential for exposure) is not deterministic but instead a function of both species distribution and land use patterns. Our work serves as a framework to enhance the accuracy and efficiency of threatened and endangered species assessments using a data‐driven likelihood analysis of species co‐occurrence. Integr Environ Assess Manag 2019;00:1–12. © 2019 SETAC

A data accuracy evaluation strategy to improve the representation of potential pesticide use areas for endangered species assessments

“…Consideration of probabilistic approaches similar to Budreski et al (2016) and Richardson et al (2019) are ideal for better representation of specific crop class occurrence within a spatial footprint in a specific number of years of interest and their likelihood of spatial overlap to the listed species.…”

A critical review of effect modeling for ecological risk assessment of plant protection products

“…For this reason, we chose to gather in a spécifie category ail the mechanistic modeling studies, when any element of which falls within a landscape framework. Our literature searches also revealed the existence of a few PPP studies at the land scape level that are based on spatial statistical approaches (species distribution models, Szabo et al 2009;Richardson et al 2019, pressure-impact relationships, Kattwinkel et al 2011. These studies, while not based on dynamic mechanistic models, do incorporate various éléments of spatially explicit mod eling related to PPP uses and environmental fate, or ecological determinisms of non-target population exposure.…”

Probabilistic co‐occurrence assessment for suites of listed species

“…The method also adjusts probabilities of crop occurrence on a state‐by‐state basis using measures of CDL accuracy and crop production (US Department of Agriculture, 2021), and produces a probability raster where each pixel has a value scaled 0–1 representing the likelihood that a particular crop (and attendant pesticide use pattern) will occur in any given year. This method has been enhanced by using satellite imagery to detect field borders and correct misclassifications within them (Richardson et al, 2019). By contrast, the deterministic BE footprint model assumes that any areas that have been cropped in any number of recent years are cropped and will be cropped indefinitely, regardless of the frequency of cropping or the classification of neighboring pixels.…”

“…The resulting probabilistic model quantitatively evaluates each raster cell for habitat suitability individually, as opposed to deterministic models that are qualitative classifications of large geographical areas as suitable or unsuitable. We refine each of these predictions through novel incorporation of an array of remote sensing data products (Richardson et al, 2019). The product of these two grids reveals the combined likelihood of spatial co‐occurrence of species and pesticide use in each pixel in the area of interest.…”