comparison of analysis units for associating Lyme disease with forest-edge habitat

• Published in: Community ecology Vol. 7; no. 2; pp. 189 - 197
• Main Authors: Jackson, L. E, J. F. Levine, E. D. Hilborn
• Format: Journal Article
• Language: English
• Published: Akadémiai Kiadà 01.12.2006; Akadémiai Kiadó
• Subjects: autocorrelation; Borrelia; deer; disease incidence; Disease models; Disease risk; Forest cover; Forest habitats; Fragmentation; humans; Land cover; Landscape; Landscapes; Lyme disease; Maryland; Modifiable areal unit problem; monitoring; Negative binomial model; planning; Polygons; Roads; Scale; Spatial autocorrelation; spatial data; Spatial models; Wildlife habitats; Maryland
• ISSN: 1585-8553; 1588-2756
• DOI: 10.1556/ComEc.7.2006.2.6

The rational definition of spatial analysis units is critical to modeling and understanding large-scale ecological processes. This study assessed the relationship between forest-edge habitat pattern and Lyme disease incidence rate when modeled under three designs for spatial data aggregation. Incidence rates were calculated from passive surveillance data reported for 12 counties in the U.S. State of Maryland during 1996–2000. A design using road-bounded polygons that varied in size from 0.002 km 2 to 368 km 2 (n = 415) was compared with designs that used grid cells of 10 km 2 (n = 823) and 36 km 2 (n = 230). Major roads were chosen to approximate bounded populations of deer and humans engaged in outdoor activity around the home (peridomestic activity). While cell boundaries were arbitrary, cell sizes were chosen to eliminate outliers observed in small polygons, and to standardize the presumed zone of exposure. The single variable that explained the most variation in incidence rate across all study designs was percent of herbaceous edge adjacent to forest. The multi-variable model with the strongest explanatory power (R 2 = 0.87) resulted from the road-bounded design. Furthermore, this design controlled for spatial autocorrelation (p = 0.064), which was highly significant in both grid designs (p = 0.002). Findings demonstrate the utility of roads to delimit distinct zones of human-environment interaction, including development intensity and peridomestic contact with wildlife habitat. This study emphasizes the importance of judicious boundary selection to spatial models with the potential for real-world applications in landscape planning and design.