Critical transition theory tells us that, as exogenous parameters drive the system towards a bifurcation and the emergence of a new equilibrium, we can see evidence of the upcoming transition through decreasing resistance to each oscillation peak15, like a ball rolling around in a cup whose walls are becoming less and less steep1,6. We find evidence of this critical slowing down within the year-to-year deceleration of rust growth rates just prior to their annual peak, or λ, our formal slowing down measure. This is coupled with a delay in the initiation month (takeoff) of the oscillation itself. The oscillations collapse in 2019 as the system apparently switches to a more benign non-epidemic state.
In the mid-elevation zone of southwestern Mexico where we conducted our study, the general expectation from climate change, in addition to increased temperature, is reduced rainfall19,20, a pattern broadly in evidence for the past five years (Fig. S2). With regard to the coffee rust disease, this precipitation trend is most closely associated with the change in λ, and reflects the obligate range of associated moisture and temperature conditions required for the rust to flourish9,17. On the other hand, rust-resistant replanting has been occurring throughout coffee farms in much of Mesoamerica since the rust outbreak in 2012–201321,22. This could limit plant-to-plant spread by reducing opportunities for direct contact with an infected plant, as well as decrease the environmental spore load by reducing the contributing pool of infected plants in the broader region. On our site, the slowing down patterns of both λ and the rust takeoff point were associated with more rust-resistant replanting on a local level, while the significant linear delay in the year-to-year rust takeoff over the study period may reflect larger-scale effects emerging from increased resistance across the farm.
Our preliminary treatment of April as the annual rust season initiation point corresponded generally with a seasonal pattern of increasing rainfall. This approach might be assumed by a manager experiencing the system in real time without knowledge of how the rust season will play out that year. Such an assumption works well in the first two seasons of our data, where the rust increase and rainy season both appear to begin in April. However, this association decouples in subsequent years. Evidently, the critical slowing down we initially estimated in Fig. 1B is partitioned into two components: one that imposes a lag on the initiation of the disease, and another that imposes a decelerating approach to the peak rust intensity, necessitating a joint model of the initiation month and the critical slowing down warning signal (λ). That the takeoff had a significant delaying trend over the study period suggests that the initiation of the rust season can also be seen as a measure of critical slowing down, in that a slower approach to the equilibrium point (the seasonal rust peak) will likely be reflected in a failure to even recognize the increase when it is still very small. Thus, the conspicuous change in takeoff point over time (Fig. 2A, B) itself suggests a critical slowing down and the negative values of λ in Fig. 1B may be mainly a reflection of a lag in takeoff time. The overall increasing lag in rust takeoff each year could mean that the transmission factor itself may be exhibiting a critical slowing down as the critical transition is approached, due to the progressive influence of parameters outside of the present analysis. Given the basic biology of the disease10, this trend may stem from a year-to-year secular decline in the environmental spore load causing initiation of the disease season to be deterred by a small amount each year.
The evident relationship between critical slowing down and reduced rainfall is suggestive of a connection to climate change, while a slow secular increase in the proportion of rust-resistant varieties could have inhibited local or regional spread, leading to an effective “herd immunity” to the disease. The joint operation of reduced rainfall and fewer susceptible plants raises the possibility of dual bifurcations and hysteretic zones in the dynamic landscape, which we illustrate in a qualitative fashion (generalized to climate and management) in Fig. 4a. The initial outbreak, which was unexpected in light of the historically low, but persistent, levels of rust in the region7,8, was likely a critical transition precipitated by interactions between local and regional processes14. As foregrounded in Fig. 4b, we propose that a possible parameter driving the system to this initial bifurcation was recent increases in precipitation, as evidenced in local rainfall records (Fig. S2) and regional trends19. This could have propelled the system past a hysteretic phase space to where seasonal conditions dictated that the system jump to a high rust intensity equilibrium, represented conceptually in Fig. 4a by the dotted trajectory leading to the upper surface of the landscape that corresponds to an epidemic state of the rust. Likewise, in the years following the outbreak, our findings suggest that, while the system still tracked precipitation, progressive replanting of resistant varieties emerged as another parameter axis (management) that drove the system through a second critical transition back to a low rust equilibrium (Fig. 4c). Although the dynamical landscape in Fig. 4 is a qualitative representation, the trajectory along the upper surface helps to visualize how two exogenous forces, operating separately, both contributed to the critical transition we observed in our data. The trajectory we propose brings attention to the interesting possibility that the main driver of slowing down shifted from climate (precipitation) to management (replanting), leading to the second critical transition. Indeed, though the average yearly replanting rate remained roughly the same year-to-year (Fig. S1), we note that much of the cutting prior to the collapse in 2019 seemed to be concentrated in April that year (Fig. 1D), accompanied presumably by a similarly timed replanting campaign.
Envisioning the combination of climate and management effects in a joint hysteretic framing, stemming from gradual change in both forcing parameters. (A) Relative positions of rust intensity for each year are illustrated in their approximate positions with red arrows (other trajectories could be imagined based on the data presented herein). Inset plots provide a conceptualized view of the dynamics between the two forcing variables on rust intensity: (B) management (in our case, resistant variety replanting) and (C) climate (precipitation). In the inset plots, arrows indicate directions of change; solid and dashed lines indicate stable and unstable equilibria, respectively.
Recent work on critical transitions suggests that perturbances driving a system to transition are more realistically not distinct or isolated, and that the stochastic and deterministic elements of the system can therefore be entangled or even interdependent4. Likewise, we find that the variability in the environmental covariates of monthly rainfall and resistant variety replanting better explained patterns in λ than a linear trend leading up to the transition, as represented by the year variable. The correlation between the quadrat grouping offset estimates from the λ and takeoff components of the multivariate model also suggest that slowing down and delayed takeoff were associated at the individual quadrat level (Fig. 3C). Accounting for this spatial effect, these two components do not appear to be correlated by year (Fig. 3D). This suggests that the shared variability between these two indicators reflects variability in spatial environment within the plot rather than idiosyncratic effects of unique years. Besides the direct effect of resistant varieties, local stochasticity and spread dynamics may also play a role. Local growing conditions, such as variability in shade from overstory trees, can affect dispersal through rainfall splash and wind23,24. Additional management factors may also play a role, such as the vegetation structure and the presence of paths25, as well as the physical relationship between coffee plants26.
Our observations of the rust dynamics themselves allow us to detect the general signals anticipating a critical transition, though the drivers may emerge from a complex system of dialectical interactions that must be considered in their whole7,27. The concept of critical slowing down thus may lend itself to application across coffee-growing regions, where predicted effects of climate change and other geographic conditions may differ9,19. Since the emergence of the rust outbreak, recommendations and protocols have been published for monitoring rust levels, potentially providing managers with regular data in changes in rust intensity for many areas9,28. As the resilience of a system can be interpreted through measuring critical slowing down prior to catastrophe2, as well as, in our case, the “exit time” from an undesirable regime4, we demonstrate that such concepts may be applied to this monitoring data to gain some insight into the system’s status. Future studies could explore signs of critical slowing down across coffee-growing regions and management systems to see how these signals predict significant changes and respond to local drivers, potentially adding to the vocabulary of agroecological management.
In sum, it is clear that both a lag in takeoff point for the seasonal oscillation and the rate of approach to the peak each year seem to conspire to produce a critical slowing down, strong evidence that the decline in the disease in 2019–2020 is indeed a critical transition, regardless of the underlying mechanism. While our model suggests that two exogenous forces, rainfall and resistant variety replanting, may be driving the slowing down in our case, the underlying dynamical landscape is likely not unique to our site. More generally, the phenomenon of multivariate bifurcations leading to subsequent critical transitions (e.g., Fig. 4) is perhaps more common than thought29. Examinations of critical transitions should therefore consider the larger dynamical landscape for the possibility of subsequent transitions.
Source: Ecology - nature.com
