Evaluating the ecological relevance of habitat maps for wild herbivores

Informed management of large herbivores depends largely on how well habitat availability and suitability are understood. The aims of the study were to quantify and map the distribution of sour and mixed grasslands in the 48,000 ha Songimvelo Game Reserve, Mpumalanga, South Africa. Mixed grassland retains its forage quality and hence its ability to sustain animal production for longer in the year than sour grassland. An unsupervised classification technique was applied to a LANDSAT 5 TM image acquired in 1993. The probability that each resulting cluster represented either sour or mixed grassland was calculated based on the proportional allocation of 428 sample plots. The 2 resulting probability maps were combined into a single image by selecting the class image that contained the maximum posterior probability and assigning that class to the output pixel. The accuracy of the vegetation map was assessed by ground-truthing with an independent set of 85 plots. This yielded a correct classification of 84.8% for the sour and 76.9% for the mixed plots. The mixed grasslands covered only 31.0% of the area but accounted for 66.1% of the game stocking. Water is widely distributed and is not a limiting factor to habitat selection. Based on a GIS analysis, the qualitative difference between mixed and sour grasslands overrides quantitative differences in forage availability, fire history and human disturbance in influencing herbivore distribution. The integration of field data and satellite imagery into a GIS system thus offers a powerful tool for the objective quantification and mapping of available habitat.