Leading indicators of grassland to woodland transitions along a spatial gradient in a savanna ecosystem

Background/Question/Methods

Many ecological systems are susceptible to abrupt and potentially irreversible regime shifts. Such dramatic changes could potentially lead to the loss of critical ecosystem services, thus disturbing not only the ecology of the system, but also the human societies that depend on these ecosystems. Mathematical models have shown that various spatial metrics may serve as leading indicators of such transitions. Furthermore, these studies also show that spatial information may provide leading indicators without the need for long-term temporal data. However, due to the lack of large-scale spatiotemporal data and difficulties in experimentally manipulating ecosystems in the field, testing for leading indicators in spatial systems is difficult. To address this problem we took a ‘space-for-time-substitution’ approach by computing leading indicators of transitions from grassland to woodland states in the Serengeti-Mara ecosystem in Tanzania and Kenya.  The existence of alternative stable states and the possibility for abrupt transitions from grassland to woodland states together with the ecological and socio-economic importance of savannas make them ideal ecosystems in which to test for early warning signals of critical transitions.

Results/Conclusions

We found that spatial variance, spatial skewness, spatial correlation and spatial discrete Fourier transform at low frequencies increased prior to a sharp decrease in grass cover. We showed that the spatial indicators change substantially more than spatial mean along transects which exhibit transitions from grassland to woodland. Thus monitoring of spatial indicators to predict regime shifts would be better than the monitoring of spatial means alone. These results are largely in accordance with predictions from mathematical modeling of ecological systems that exhibit abrupt regime shifts. However, we found that the spatial metrics were most effective as early warning signals during the initial decline in grass cover. Our results suggest that it may be possible to determine early warning signals of regime shifts using spatial data.