Supplementary Materials
S1. Generation of spatially weighted pseudo-absences
To generate the spatially weighted pseudo-absences (PA), we developed a kernel density (KD) raster based on cumulative D.citri abundance for all the years between 2009 and 2013 at 1 km2 resolution. The search radius for estimating the KD was chosen such that the spatial autocorrelation of D.citri abundance is minimized, and this was done by fitting a semi-variogram to the abundance data (Figure 1a). The resulting KD raster depicts a heat map of D.citri invasion for urban LA, which was then classified into 9 distinct bands of varying invasion intensity (Figure 1b). The KD raster informs the density of points within each band; these band-specific densities metric were inversed and weighted by the product of average density of D.citri detections (i.e. 8.5 detections per sq.km) for the entire study area and band-specific polygon area, thus yielding spatially weighted pseudo-absences that are inversely related to the intensity ofD.citriinvasion.
PA band 1 = (KDband 1)-1 * Mean D.citri detections per sq.km * Polygon areaband 1
Generating pseudo-absences in this manner is ideal for estimating potential distribution(Lobo et al. 2010) of the invasive species,allowing meaningful comparison of establishment and impact risk.
Figure S1a
Empirical semivariogram showing strong spatial dependency in D.citri abundance (in 1 km2pixels) for a lag distance of approximately 18000 meters (range of the semivariogram). The estimated range value was used as the bandwidth for making a D.citri heat map using a kernel density function.
Figure S1b
A) D.citri heat map for the study area derived from a kernel density function overlaid with the simulated pseudo-absences used in D.citri distribution modeling. As described above, the spatially weighted pseudo-absences simulation results in hot spots (red) with high density of D.citri detections containing fewer pseudo-absences, while cold spots (green) with relatively lower densities of D.citri contain greater numbers pseudo-absences. The heat map was developed using the kernel density function with a bandwidth of 18000 meters as estimated by the semivariogram above (Figure 1A). B) location of the urban Los Angeles study area (bounded by black solid line) within Southern California with all D.citri detections (2008 – 2013) as grey dots.
S2: Point-pattern analysis to estimate neighborhood scale
Figure S2
Plot showing the strength of bivariate clustering for a cross-pair correlation function, corrected for inhomogeneity in data – ginhom(r), where the focal points were D.citri detections in 2013 and neighborhood points were all prior D.citri detections for the years 2010, 2011 and 2012. The large ginhom(r) values on the y-axis are associated with short lag distance radii on the x-axis indicating clustering of prior D. citri detections is most strong within the immediate neighborhood of 2013 D.citri detections. The black arrow points toward the distance threshold (500m) selected for estimating prior D.citri density as a measure of spatiotemporal lag.