Watershed hydrological processes driven by topography and landform are crucial to landscape patterns of soil properties, plant distribution, and disturbances. Topographic wetness index (TWI) is a widely used topographic variable in landscape ecology studies for representing such drainage influences. Different TWI algorithms have been developed, yet their differences in spatial autocorrelation structure remain elusive. Spatial autocorrelation is important in determining landscape heterogeneity and processes. Understanding the spatial structure derived from different algorithms can better inform landscape ecologists’ use of TWI. In this study we evaluated the spatial autocorrelation structure of three commonly used TWI algorithms: D8, Multiple Flow Direction (MFD), and D-infinity. The study area is a mountainous, forested watershed in the Cumberland Plateau of eastern Kentucky. GIS and a five-foot resolution DEM derived from LiDAR were used to generate sample points every two meters along six transects and six contours. The semivariograms and their corresponding parameters (nugget, partial sill, and range) of a spherical model were computed for each transect and contour. ANOVA analyses were conducted to test if there are any differences in general variability and spatial autocorrelation among the three algorithms.
Results/Conclusions
The mean, coefficient of variance (CV), and ratio of partial sill over total sill are significantly different among the three TWI algorithms, although the semivariogram ranges are not significantly different. The D8 algorithm has the greatest variance, while MFD has the lowest. Additionally, the D8 algorithm has the lowest partial sill: total sill ratio compared to MFD and D-infinity. The results suggest D8 has the lowest magnitude of spatial dependency when compared to MFD and D-infinity. The difference in spatial autocorrelation structure of these algorithms is important to consider when applying this hydrological model at the landscape scale.