Dr Tigga Kingston

Assistant Professor

>

Fig 1. GIS created for 1 of the 5 study grids in Krau Wildlife Reserve, Malaysia. Clear squares are harp trap positions set in 2002, filled squares are harp traps in which one particular species, Kerivoula papillosa were captured. The challenge is test this observed distribution of captures for local departures from randomness – clusters of individuals that are not attributable to the underlying distribution of traps.

Fig 2. Hotspot fusion surface for the Kerivoula papillosa captures. This is a concordance surface based on 19 outputs from artificial neural networks, distance indices and G statistics tests for local spatial autocorrelation. Black areas indicate hotspots identified by all three methods, a level of concordance conferring the highest degree of confidence in the true presence of a statistically significant hotspot. Dark grey indicates two methods in agreement, light grey one and blank areas are those in which no hotspots were detected by any method.

BOX 1. A fusion approach to detect spatial clusters in ecological count data

Spatial variation in the distribution and density of individuals is a common species attribute (Fig. 1). Spatial autocorrelation of individuals (clustering) both influences and reflects the processes shaping species, community and ecosystem interactions. It can result from spatially dependent or autocorrelated responses to a structured environmental variable, or endogenous clustering processes that derive from the species’ biology.  Most tests of spatial autocorrelation assume stationarity of the causal process across the study area. As a consequence, these ‘global’ tests generally fail to detect local departures from randomness that are often of biological interest, and cannot localize or characterize their extent, area, or density. Existing local tests of spatial association have different strengths and weakness; each tends to capture different aspects of the data, a problem exacerbated when there is no a priori knowledge of the system to inform parameterization of sampling and analysis units.  To overcome these, I have used a fusion approach to characterize clusters in ecological data using three disparate, but commonly-used tests of local autocorrelation: a traditional spatial analytical approach (G statistics); a permutation method designed for ecological count data using distance indices (SADIE); and SOM (Self-Organizing Map) neural network. Agreement among method is analyzed using coincident analysis which provides explicit representation of confidence for individual (Fig. 2).