Considerations for Inferring Remote Sources of Inoculum, Vectors, or Pests in Agroecosystems Using Spatial Data, when Dispersal Kernels Are Uncertain and Landscapes Are Heterogeneous

Spatial heterogeneity influences processes in agriculture and can be configured optimally for production and disease management. Landscape composition can be analyzed to infer sources of inoculum or vectors influencing points of interest such as crop fields, and to estimate risk to improve management. Substantial research in plant ecology and disease epidemiology has addressed the importance of the dispersal kernel to spatial epidemic dynamics, but a methodological knowledge gap remains for situations in which the magnitude of dispersal from different landscape types varies. This knowledge gap is important to the study of emerging pathosystems in which inoculum sources or reservoirs may not be well-characterized, and to vectored-disease systems in which transmitting arthropods are polyphagous with unknown host preference. Using simulated data, we describe issues that arise from the practice of summarizing spaces as concentric rings to infer dispersal kernels or identify influential landscape classes, and we demonstrate utility of instead analyzing variation at a point of interest as a sum of simultaneously distance- and class-weighted remote sources. Of particular interest are scenarios where the form of the true dispersal kernel is unknown and the goal is to rank landscape types depending on their magnitude of influence. Results emphasize that nonlinear regression methods are necessary for simultaneously fitting distance weighting functions and estimating relative influence of landscape types, and show that ring-based descriptions of landscapes for analysis can lead to misapprehensions about dispersal processes. Methodological development is needed in plant disease epidemiology research where dispersal kernels may vary depending on biological considerations.