Addition of finer scale data and uncertainty analysis increases precision of geospatial suitability model for non-native plants in the US

Abstract

“Proto3” is a geospatial model used by the United States Department of Agriculture (USDA) Plant Protection and Quarantine to predict the potential distribution of non-native weed species in the continental U.S. as part of routine weed risk assessments (WRA). While performing as well as other methods, this tool has the benefit of being simple to produce, expanding accessibility and reproducibility. However, it has the tendency to overestimate potential distributions. To address this shortcoming, this paper introduces the “Proto4” model and compares it with the established and mechanistically similar “Proto3” model currently used. Both models overlay Plant Hardiness Zones, precipitation, and Köppen-Geiger climate classes with global distribution of a plant species and rely on semi-qualitative assessments of a plant's affinity for each of the climate categories. However, Proto4 uses more detailed layers of the Plant Hardiness Zones and Köppen-Geiger climate classes, adds elevation as a fourth predictive variable to increase the precision of predictive maps. Additionally, we incorporate uncertainty to spatially distinguish regions of different potential suitability. We compared the performance of both models by estimating the predicted distributions of 30 broadly distributed, invasive plants in the U.S. with Proto3 and Proto4. We found that on average, the Proto4 model produces predicted distributions that are nearly 780,000 square kilometers (an area larger than the state of Texas) smaller than the Proto3, while only failing to capture a median of fewer than 0.5% more georeferenced points. Furthermore, the inclusion of uncertainty classes adds to the utility of Proto4 by distinguishing areas with greater and lesser degrees of evidence that a particular area is suitable for an invasive species, providing more information to help select invasive species prevention and management prioritization strategies.