Quantification of the cascading tipping probability from the AMOC to the Amazon rainforest

TL;DR

The paper tackles the problem of quantifying cascading tipping from the Atlantic Meridional Overturning Circulation (AMOC) to the Amazon rainforest using a coupled, CESM-tuned conceptual model. It introduces a rare-event sampling approach, Time Adaptive Multilevel Splitting (TAMS), to efficiently sample and analyze transitions that are otherwise extremely unlikely within a centennial horizon. By tracking degradation levels, MFPTs, AMOC-strength distributions, and cascading probabilities across two NW Brazilian regions, the study finds region-specific dynamics: in the northwest (region 1) AMOC collapse appears nearly necessary for large degradation, while in the coastal region (region 2 wildfires dominate degradation with minimal AMOC change). These results demonstrate how TAMS coupled to process-based yet simplified models can yield mechanistic insight into tipping cascades and inform risk assessment of climate–tipping interactions, while highlighting limitations of current CESM-derived hydrology and the need for more comprehensive coupling in future work.

Abstract

The Amazon rainforest and the AMOC are considered to be tipping elements: they are important components of the Earth system, but may collapse under climate change. Moreover, an AMOC collapse may favor the transition of the rainforest to a degraded forest by influencing the precipitation patterns over the Amazon. This phenomenon is known as tipping cascade and better understanding it is key to anticipating the impact of tipping events. Here, we investigate in a coupled conceptual AMOC-Amazon model the probability that an AMOC weakening affects tree cover loss in two regions of the rainforest. To get more insight into the mechanisms behind the tipping cascade, we also analyze the dynamics of both systems and their evolution during the Amazon transition. Namely, we track the transition probability and the transition time of the Amazon, and reconstruct the distribution of AMOC strength at every stage of this transition. These tasks require a large ensemble simulation, containing in particular a large number of transitions. Since such events may be too rare to be sampled by direct numerical simulation, the collapse of both systems is studied using TAMS, a "rare-event" algorithm designed to efficiently sample rare transitions. We find that, in the northwest of Brazil, a transition of the Amazon rainforest to a degraded forest within 200 years is very unlikely. However, in this region, such transition can only occur after an AMOC collapse, which would have a large drying effect that favors the development of extreme wildfires.

Figures (17)

• Figure 1: The AMOC conceptual model (see Sect. \ref{['sec:amoc_model']}) is represented on the right. Arrows represent volume transport between boxes. Solid red arrows correspond to fluxes that are always present. The dashed red arrow represents the downwelling in the northern Atlantic, and its shutdown characterizes an AMOC collapse. The on-state of the AMOC conceptual model is tuned to that of CESM. CESM is used to derive a relationship between the AMOC strength and MAP and MCWD (see Sect. \ref{['sec:amazon_model']}). The dynamics of the Amazon conceptual model are governed by the empirical potentials of MAP and MCWD, and by the stochastic process representing wildfires. The dynamics of the potentials and the fire intensity depend, in turn, on the tree cover.
• Figure 2: Potentials obtained from the data from Flores2024 after application of Gaussian KDE, with the bandwidth detailed in Sec. \ref{['sec:potentials']}. In all panels, a darker color indicates a smaller potential value. The left panel shows the potential corresponding to the Mean Annual Precipitation (MAP), while the middle panel shows the potential corresponding to the Maximum Cumulative Water Deficit (MCWD). The right panel presents the combined potential driving the Amazon model, where the $y$-axis of the Mean Annual Precipitation (MAP) has been rescaled from mm/year to mm/month to match that of MCWD. The white dashed line represents the target value used here as rainforest/low tree cover state threshold.
• Figure 3: Fire intensity $f$ (Eq. \ref{['eq:fire_intensity']}) as a function of tree cover and Mean Annual Precipitation (MAP).
• Figure 4: a) MAP difference in CESM over the Amazon between the AMOC collapsed state and the AMOC on-state. The grid represents CESM grid cells. Red areas indicate a decrease in precipitation due to the AMOC collapse. Blue areas indicate an increase in precipitation due to the AMOC collapse. The two studied regions are enclosed in solid lines and numbered. b) High vegetation cover from ERA5, averaged over the period $2014$-$2024$. The darker shades of green indicate a denser forest. The same regions as in panel a are enclosed in white lines. c) Zonally averaged tree cover from panel b in region 1 (solid line) and region 2 (dashed line). These curves are used as initial condition of the partial differential equation in Eq. \ref{['eq:amazon_model']}.
• Figure 5: Zonally averaged MAP and MCWD from CESM, plotted against the AMOC strength for each grid cell of the two regions highlighted in Fig. \ref{['fig:regions']}a. The upper row corresponds to MAP, the lower to MCWD. The left column corresponds to region 1, the right to region 2. Colored lines represent MAP and MCWD computed in CESM and smoothed by applying a $100$-year rolling mean, plotted against the AMOC strength from CESM, smoothed in the same way. The darker the line color, the larger the mean latitude of the corresponding grid cell in region 1 or 2 (see Fig. \ref{['fig:regions']}a). Dashed lines indicate the least square fit of the corresponding curve by a degree three polynomial.