Relative Species Richness
Models of the relative species richness of trees/shrubs in burnt and unburnt plots had limited explanatory power (i.e., 5.78%), and model averaging revealed that the 95% confidence intervals of most parameter estimates overlapped zero (Table 2). Protection status did, however, influence relative species richness of trees/shrubs – with species richness increasing at burnt sites relative to unburnt controls in unprotected sites, whilst within protected sites species richness was more similar in burnt and unburnt sites (Table 2; Fig. 2a). Models of forb relative species richness in burnt and unburnt plots had much higher explanatory power (i.e., 41.74%) than models for other life forms (Table 2). Model averaging revealed interactions between biome type and type of fire. In flooded grasslands and savannahs forb species richness increased in plots experiencing non-prescribed burns relative to unburnt controls, with species richness being more similar in burnt and unburnt controls following prescribed burns (Table 2; Fig. 2b). In contrast, in moist broadleaved forest prescribed burns increased forb species richness relative to unburnt controls, whilst non-prescribed burns resulted in similar species richness in burnt and unburnt controls (Table 2; Fig. 2). Models of the relative species richness of climbers in burnt and unburnt sites also had relatively limited explanatory power (i.e., 5.17%) with model averaging revealing that prescribed burns resulted in climber richness increasing in burnt plots relative to unburnt ones, whilst non-prescribed burns resulted in species richness declining in burnt sites relative to unburnt controls (Table 2; Fig. 2c). The best model of graminoid relative species richness had a higher AICc (88.90) than that of the null model which lacked predictors (AICc 86.97).
Table 2:  Results from model averaging across multiple regression models of relative species richness in burnt sites relative to control (unburnt) sites for trees/shrubs, forbs, and climbers. Results for graminoids are not included as no model had a lower AICc than the null model (i.e., one that lacked predictors). Parameter estimates are provided with 95% confidence intervals in brackets. FGS (Flooded Grasslands and Savannas); TSCF (Tropical and Sub-tropical Coniferous Forests); TSDBF (Tropical and Sub-tropical Dry Broadleaf Forests); TSGSS (Tropical and Sub-tropical Grasslands, Shrublands and Savannas); and TSMBF (Tropical and Sub-tropical Moist Broadleaf Forests).