Land Cover Change in Northern Botswana: The Influence of Climate, Fire, and Elephants on Semi-Arid Savanna Woodlands

Complex couplings and feedback among climate, fire, and herbivory drive short- and long-term patterns of land cover change (LCC) in savanna ecosystems. However, understanding of spatial and temporal LCC patterns in these environments is limited, particularly for semi-arid regions transitional between arid and more mesic climates. Here, we use post-classification analysis of Landsat TM (1990), ETM+ (2003), and OLI (2013) satellite imagery to classify and assess net and gross LCC for the Chobe District, a 21,000 km 2 area encompassing urban, peri-urban, rural, communally-managed (Chobe Enclave), and protected land (Chobe National Park, CNP, and six protected forest reserves). We then evaluate spatiotemporal patterns of LCC in relation to precipitation, fire detections (MCD14M, 2001–2013) from the Moderate Resolution Imaging Spectroradiometer (MODIS), and dry season elephant ( Loxodonta africana ) aerial survey data (2003, 2006, 2012, 2013). Woodland cover declined over the study period by 1514 km 2 (16.2% of initial class total), accompanied by expansion of shrubland (1305 km 2 , 15.7%) and grassland (265 km 2 , 20.3%). Net LCC differed importantly in protected areas, with higher woodland losses observed in forest reserves compared to the CNP. Loss of woodland was also higher in communally-managed land for the study period, despite gains from 2003–2013. Gross (class) changes were characterized by extensive exchange between woodland and shrubland during both time steps, and a large expansion of shrubland into grassland and bare ground from 2003–2013. MODIS active fire detections were highly variable from year to year and among the different protected areas, ranging from 1.8 fires*year −1 /km 2 in the Chobe Forest Reserve to 7.1 fires*year −1 /km 2 in the Kasane Forest Reserve Extension. Clustering and timing of dry season fires suggests that ignitions were predominately from anthropogenic sources. Annual fire count was significantly related to total annual rainfall ( p = 0.009, adj. R 2 = 0.50), with a 41% increase in average fire occurrence in years when rainfall exceeded long-term mean annual precipitation (MAP). Loss of woodland was significantly associated with fire in locations experiencing 15 or more ignitions during the period 2001–2013 ( p = 0.024). Although elephant-mediated damage is often cited as a major cause of woodland degradation in northern Botswana, we observed little evidence of unsustainable pressure on woodlands from growing elephant populations. Our data indicate broad-scale LCC processes in semi-arid savannas in Southern Africa are strongly coupled to environmental and anthropogenic forcings. Increased seasonal variability is likely to have important effects on the distribution of savanna plant communities due to climate-fire feedbacks. Long-term monitoring of LCC in these ecosystems is essential to improving land use planning and management strategies that protect biodiversity, as well as traditional cultures and livelihoods under future climate change scenarios for Southern Africa.