Human societies, and their well-being, depend to a significant extent on the state of the ecosystems that surround them. These ecosystems are changing rapidly usually in response to anthropogenic changes in the environment. To determine the likely impact of environmental change on ecosystems and the best ways to manage them, it would be desirable to be able to predict their future states. We present a proposal to develop the paradigm of predictive systems ecology, explicitly to understand and predict the properties and behaviour of ecological systems. We discuss the necessary and desirable features of predictive systems ecology models. There are places where predictive systems ecology is already being practised and we summarize a range of terrestrial and marine examples. Significant challenges remain but we suggest that ecology would benefit both as a scientific discipline and increase its impact in society if it were to embrace the need to become more predictive.

Savanna ecosystems are dominated by two distinct plant life forms, grasses and trees, but the interactions between them are poorly understood. Here, we quantified the effects of isolated savanna trees on grass biomass as a function of distance from the base of the tree and tree height, across a precipitation gradient in the Kruger National Park, South Africa. Our results suggest that mean annual precipitation (MAP) mediates the nature of tree-grass interactions in these ecosystems, with the impact of trees on grass biomass shifting qualitatively between 550 and 737 mm MAP. Tree effects on grass biomass were facilitative in drier sites (MAP≤550 mm), with higher grass biomass observed beneath tree canopies than outside. In contrast, at the wettest site (MAP = 737 mm), grass biomass did not differ significantly beneath and outside tree canopies. Within this overall precipitation-driven pattern, tree height had positive effect on sub-canopy grass biomass at some sites, but these effects were weak and not consistent across the rainfall gradient. For a more synthetic understanding of tree-grass interactions in savannas, future studies should focus on isolating the different mechanisms by which trees influence grass biomass, both positively and negatively, and elucidate how their relative strengths change over broad environmental gradients.

1 Patch dynamics is a new, potentially unifying mechanism for the explanation of tree‐grass coexistence in savannas. In this scale‐explicit paradigm, savannas consist of patches in which a cyclical succession between woody and grassy dominance proceeds spatially asynchronously. The growing ecological and economic problem of shrub encroachment is a natural transient phase in this cycle. 2 An important step towards understanding patterns at the landscape scale is achieved by investigating mechanisms at a smaller scale. We developed the spatially explicit individual‐based simulation model SATCHMO to test the null hypothesis that cyclical succession cannot emerge from a realistic patch scale simulation model of the population dynamics of savanna woody species. 3 We calculated the partial temporal autocorrelation coefficient for 100 simulated time series of shrub cover over 500 years for time lags of up to 200 years to establish the existence and duration of successional cycles. We found a significant positive autocorrelation indicating the existence of cycles with a typical duration of about 33 years. 4 The shrub size frequency distributions over the course of a cycle showed shifts from dominance of small shrub sizes towards larger sizes during the increasing phase of a cycle and the reverse in the declining phase. This supports the three phase explanation as follows: (i) an initial phase when spatially and temporally overlapping favourable conditions lead to mass recruitment of shrubs; (ii) a build‐up phase when the shrub cohort grows; and (iii) a break‐down phase when increased competition due to crowding and unfavourable conditions lead to the break‐down of the shrub cohort. The frequency distribution of shrub age at death over 10 simulations was also in agreement with this explanation. 5 We investigated the relationship between shrub cover, annual precipitation and time‐lagged shrub cover to identify the driver of the cyclical successions. More than 90% of the variation in shrub cover was explained by shrub cover of the previous year, precipitation, and their interaction. 6 With the demonstration of precipitation‐driven cyclical succession at the patch scale, we show that the mechanistic, temporal component of patch dynamics can be used to explain tree‐grass coexistence in semi‐arid savannas.