ABSTRACT: Precision agriculture (PA) is the application of management decisions based on identifying, quantifying, and responding to space-time variability. However, knowledge of crop pest responses to within-field environmental variability, and the spatial distribution of their natural enemies, is limited. Quantitative methods providing insights on how pest-predator relationships vary within fields are potentially important tools. In this study, phloem feeders and their natural enemies, were observed over two years across 81 locations within a field of the perennial feedstock grass in Georgia, USA. Geographically weighted regression (GWR) was used to spatially correlate their abundance with environmental factors. Variables included distance to forest edge, Normalized Difference of Vegetation Index (NDVI), slope, aspect, elevation, soil particle size distribution, and weather values. GWR methods were compared with generalized linear regression methods that do not account for spatial information. Non-spatial models indicated positive relationships between phloem-feeder abundance and wind speed, but negative relationships between elevation, proportions of silt and sand, and NDVI. With data partitioned into three seasonal groups, terrain and soil variables remained significant, and natural enemies and spiders became relevant. Results from GWR indicated that magnitudes and directions of responses varied within the field, and that relationships differed among seasons. Strong negative relationships between response and explanatory factors occurred: with NDVI during mid-season; with percent silt, during mid-, and late seasons; and with spider abundance during early and late seasons. In GWR models, slope, elevation, and aspect were mostly positive indicating further that associations with elevation depended on whether models incorporated spatial information or not. By using spatially explicit models, the analysis provided a complex, nuanced understanding of within-field relationships between phloem feeders and environmental covariates. This approach provides an opportunity to learn about the variability within agricultural fields and, with further analysis, has potential to inform and improve PA and habitat management decisions.
