Measuring tropical rainforest resilience under non-Gaussian disturbances

Vitus Benson, Jonathan F. Donges, Niklas Boers, Marina Hirota, Andreas Morr, Arie Staal, Jürgen Vollmer, Nico Wunderling

TL;DR

Standard early warning signals for ecosystem "tipping points" assume that disturbances follow a predictable, bell-shaped Gaussian distribution. This paper proves that these indicators remain robust even under "heavy-tailed" non-Gaussian disturbances—like fires and extreme droughts—provided that researchers resolve small-scale temporal dynamics and apply specific statistical filters to reduce false alarms.

What Problem Does This Paper Address?

The Amazon rainforest is a "tipping element" of the Earth’s climate system. If it loses too much resilience, it could abruptly shift from a lush forest to a dry, savanna-like state. Scientists track this resilience using Critical Slowing Down (CSD): as a system nears a tipping point, it takes longer to recover from small shocks.

However, the mathematical foundation of CSD usually assumes "Gaussian white noise"—essentially, a constant stream of small, predictable nudges. In reality, rainforests face "pulse-like" disturbances: massive fires, sudden deforestation, or windthrows. These events follow power-law distributions (where extreme events are much more likely than a bell curve suggests) and can be modeled as $\alpha$-stable Lévy noise.

The core challenge: Do our existing early warning tools still work when the "noise" consists of discrete, violent jumps rather than smooth, random fluctuations?

Methods

The authors used a conceptual tipping element model (based on the van Nes et al. model) that describes tree cover as a system with three stable states: forest, savanna, and desert.

Key Innovations in the Approach: * Model Modification: The original model only resolved recovery over centuries. The authors added a "short-term resilience" term to simulate how a forest recovers from weather-scale fluctuations (months to years). This is critical because CSD indicators rely on observing these faster recovery rates. * Lévy Noise Simulation: They replaced standard Gaussian noise with $\alpha$-stable Lévy noise, varying the "anomaly index" ($\alpha$) to simulate different levels of "jumpiness" in the disturbances. * The Processing Pipeline: They tested several indicators, including the Interquartile Range (IQR)—a more robust version of standard deviation—and Lag-1 Autocorrelation (AC1). They then used the non-parametric Kendall-$\tau$ slope to determine if these indicators were significantly increasing over time.

Datasets

This study is primarily theoretical and uses simulated time series data generated from the modified differential equations.

  • Benchmarks: They compared their modified model against the original "van Nes" model to show why resolving small timescales is necessary.
  • Noise Configurations: They evaluated the indicators across a wide range of noise amplitudes and $\alpha$ values (from Gaussian $\alpha=2.0$ to highly impulsive $\alpha=0.5$).

Key Insights and Ideas

  • CSD is Robust to Jumps: Surprisingly, the "jumps" caused by Lévy noise do not break the CSD signal. The underlying "potential landscape" of the ecosystem still flattens as the tipping point approaches, and this change remains detectable even through the chaos of power-law disturbances.
  • The Necessity of Fast Dynamics: If a model or observation method doesn't capture fast-timescale responses (like leaf-out after a dry spell), it will almost entirely miss early warning signals. In their tests, models without these fast dynamics had a recall of only ~11%.
  • The "Adjusted" Kendall-$\tau$: One of the biggest risks in tipping point analysis is the "false alarm." The authors found that by filtering out signals where the absolute temporal autocorrelation was low (below 0.5), they could drastically reduce false positives without losing the real warnings.

Key Results

  • High Sensitivity: Using their optimized pipeline (IQR + Kendall-$\tau$ + Autocorrelation filtering), the researchers achieved a recall of 0.99. This means they correctly identified nearly all approaching tipping points, even under non-Gaussian noise.
  • Low False Alarms: The "Adjusted" processing chain reduced the False Positive Rate (FPR) from 10% down to just 2%.
  • Indicator Performance: While standard deviation and AC1 worked well, the Interquartile Range (IQR) slightly outperformed them because it is naturally more resistant to the extreme outliers (the "jumps") inherent in Lévy noise.

Limitations and Future Work

  • Dimensionality: The study uses a 1D conceptual model. Real-world forests are spatially extended, meaning future work needs to test if these findings hold up when accounting for "patchiness" and spatial feedbacks.
  • Transient States: The model assumes the system is in a quasi-equilibrium. However, the real Amazon is currently undergoing rapid, non-steady-state changes due to active deforestation, which might complicate the detection of CSD.
  • Bounded Variables: In the model, tree cover is bounded between 0% and 100%. Extremely large Lévy jumps can push the system beyond these bounds, which the authors handled by "clipping" the noise, but this remains a simplification of real-world ecological limits.

Summary updated

Abstract

The Amazon rainforest is considered one of the Earth's tipping elements and may lose stability under ongoing climate change. Recently a decrease in tropical rainforest resilience has been identified globally from remotely sensed vegetation data. However, the underlying theory assumes a Gaussian distribution of forest disturbances, which is different from most observed forest stressors such as fires, deforestation, or windthrow. Those stressors often occur in power-law-like distributions and can be approximated by $α$-stable Lévy noise. Here, we show that classical critical slowing down indicators to measure changes in forest resilience are robust under such power-law disturbances. To assess the robustness of critical slowing down indicators, we simulate pulse-like perturbations in an adapted and conceptual model of a tropical rainforest. We find few missed early warnings and few false alarms are achievable simultaneously if the following steps are carried out carefully: First, the model must be known to resolve the timescales of the perturbation. Second, perturbations need to be filtered according to their absolute temporal autocorrelation. Third, critical slowing down has to be assessed using the non-parametric Kendall-$τ$ slope. These prerequisites allow for an increase in the sensitivity of early warning signals. Hence, our findings imply improved reliability of the interpretation of empirically estimated rainforest resilience through critical slowing down indicators.
BibTeX
@inproceedings{benson2024measuring,
  author = {Vitus Benson and Jonathan F. Donges and Niklas Boers and Marina Hirota and Andreas Morr and Arie Staal and Jürgen Vollmer and Nico Wunderling},
  title = {{Measuring tropical rainforest resilience under non-Gaussian disturbances}},
  booktitle = {Environmental},
  year = {2024},
  doi = {10.1088/1748-9326/ad1e80},
}
Figure