• African savanna;
  • fragstats;
  • fuzzy classification;
  • fuzzy convolution;
  • physiognomic vegetation map;
  • SEM;
  • vegetation heterogeneity


Aim  The aim of this study is to introduce a structural vegetation map of the Serengeti ecosystem and, based on the map, to test the relative influences of landscape factors on the spatial heterogeneity of vegetation in the ecosystem.

Location  This study was conducted in the Serengeti–Maasai Mara ecosystem in northern Tanzania and southern Kenya, between 34° and 36° E longitude, and 1° and 2° S latitude.

Methods  The vegetation map was produced from satellite imagery using data from over 800 ground-truthing points. Spatial characteristics of the vegetation were analysed in the resulting map using the fragstats software package. Average patch area and nearest neighbour distance (NND) were determined for grassland, shrubland and woodland vegetation types. The heterogeneity of vegetation types was estimated with Simpson’s diversity index (D). Structural equation modelling (SEM) was used to explore the relationships between the spatial characteristics of vegetation and three predictor variables: annual rainfall, coefficient of variation (CV) in annual rainfall, and topographic moisture index (TMI).

Results  A vegetation map is presented along with a detailed summary of the distribution of land-cover classes and spatial heterogeneity in the ecosystem. Significant relationships were found between vegetation diversity (D) and TMI, and also between D and average rainfall. The average area of grassland patches showed significant relationships with average rainfall, with rainfall CV and with TMI. Grassland NND was positively associated with average rainfall. Woodland patch area showed a unimodal response to average rainfall and a negative linear association with TMI. Woodland NND showed a U-shaped association with annual rainfall and a weaker positive linear association with TMI. An acceptable model that explained variation in shrubland patch characteristics could not be identified.

Main conclusions  The vegetation map and analysis thereof resulted in three significant causal explanatory models that demonstrate that both rainfall and topography are important contributors to the distribution of woodlands and grasslands in the Serengeti. These findings further indicate that changes in patch characteristics have a complex interaction with rainfall and with topography. Our results are concordant with recent studies suggesting that percent woody cover in African savannas receiving less than c. 650 mm year−1 is bounded by average annual rainfall.