Partitioning of spatial heterogeneity in an object-oriented riparian boundaries classification system for a South African savanna

Riparian boundaries have structural characteristics that are distinguishable from upland vegetation. These riparian elements within a savanna matrix located in Kruger National Park, South Africa are the result of several agents of spatial heterogeneity including fire effects, flood events and geomorphic gradients. A riparian-savanna object-oriented classification system was developed for this study region by integrating information on a generalized canopy cover model adjusted for fire, floodplain hydrology and terrain unit data with ecologically relevant indices derived from a multispectral tassel cap analysis of Landsat data. Results from this work indicate that riparian boundaries account for 29% of the total study area with the majority located outside the 500 year floodplain. The tassel cap greenness and wetness indices were found on average to be greater in riparian boundaries compared to upland patch types. The inclusion of riparian habitats within a South African savanna mosaic increased overall patch shape complexity indicating an increase in fragmentation within the mosaic. There were several distinct structural differences found along the lateral gradients of the major rivers. The nearest-neighbor patch arrangement analysis illustrated that the average patch distance from stream centerline to uplands is longer in the basalt geology type compared to granite. The buffer area distance analysis revealed that the majority of granite block riparian patch type was evenly distributed across the landscape once one move away from the stream centerline. Lastly, the boundary between riparian and upland patch types was found to be almost three times greater on the granite geology type compared to basalt. This could imply that edge effects have a larger impact on ecosystem function located on the granite geology type compared to basalt. The majority of riparian---upland interface on granite geology occurred within the granite block class, and the tributary class on basalt. An accuracy assessment of this classification was conducted for each of the four scales of analysis using a confusion matrix. The Land System scale produced an over all accuracy of 99%, Land Element 85%, Land Type 81%, and Land Unit 64%.