Fig. 3
From: Climate-human interaction associated with southeast Australian megafauna extinction patterns
Most likely drivers of megafauna extirpation timings and patterns. Relative importance of predictor variables for the top-ranked generalised least-squares model assuming a Gaussian spatial autocorrelation structure best decribing a the spatial gradient (bearings) of megafauna extirpation (Extb) in human-megafauna coexistence areas, b the timing of megafauna extirpation (Extt) in human-megafauna non-coexistence areas and c the bearings of megafauna extirpation (Extb) in human-megafauna non-coexistence areas. Human-megafauna coexistence and non-coexistence areas are described in Fig. 2a and in the top panel of each barplot in Fig. 3 along with red arrows indicating the bearing of megafauna extinction gradients. The relative importance of each predictor variable is calculated as the change in the full model likelihood when one of its predictor variables is removed. Climate predictors were from LOVECLIM19,39 for the time period in each grid cell corresponding to the estimated timing of megafauna extirpation and its confidence interval. Predictor variables of the model describing Extt are mean annual temperature anomaly (T), mean annual precipitation anomaly (P), mean annual freshwater availability anomaly (EminP), mean annual net primary production anomaly (NPP), and the fraction of desert anomaly within the grid cell (DF). Climate anomalies are calculated relative to the 50–30 ka mean time period (see Methods). Predictor variables subscripted b (Tb, EminPb) indicate that we used the directional bearing of these climate variables, including the directional vectors for the timing of initial human arrival (Hb), to build the model describing the spatial pattern bearing of megafauna extirpation (Extb). For clarity, we did not present the results describing Extt for the areas with human and megafauna coexistence because we did not have any relevant model (i.e., percentage of variance explained by the models ~ 0%) in those areas. For each variable, error bars represent the standard deviation of the relative importance of predictor variables for the top-ranked generalised least-squares model accounting for the temporal lag by regressing extirpation against climate from 0 to 5000 years (at a 1000-year time step = 5 temporal-lag scenarios, i.e., 6 values of relative importance in total) per grid cell. Source data are provided as a Source Data file.
