courseshub.world · Mycorrhizal Networks · Module 05

Resilience and the Mother-Tree Effect

Scale-free networks are robust to random failure but fragile to targeted attack — quantifying the conservation implication.

Random attack — robustTargeted attack — fragile4 random trees removed → network largely intactmother tree removed → catastrophic fragmentation

A forest's common mycorrhizal network is a complex adaptive system, and its long-term function depends on its resilience to disturbance. Two qualitatively different threats are particularly relevant: random failure (a single tree dies of natural causes; a small patch of soil is compacted by a falling branch) and targeted attack (selective logging that removes the largest, most valuable trees first — precisely the ones with the most connections). Network science has shown definitively that scale-free networks respond very differently to these two scenarios.

The Cohen–Erez–Ben-Avraham–Havlin result

Cohen et al. (2000, 2001) showed that for scale-free networks with degree distribution$P(k) \propto k^{-\gamma}$ with γ < 3:

  • Under random failure, the percolation threshold $f_c$ (the fraction of nodes that must be removed to disconnect the network) is very close to 1 — the network is essentially indestructible by random damage.
  • Under targeted attack on highest-degree nodes, $f_c$ drops to less than 5% — removing just a handful of hubs catastrophically fragments the network.

This is the mathematical foundation of the "mother tree effect" in forest ecology: protect the top ~5% of hub trees, and the network can absorb extensive random damage. Remove those few hubs, and connectivity collapses regardless of which other trees survive.

Quantifying it in the simulation

The backend's resilience_analysis function compares two attack strategies on the forest graph:

  1. Random removal of a fraction f of nodes (averaged over many realisations).
  2. Targeted removal of the top-f highest-degree nodes.

For each strategy, several metrics are tracked as f increases from 0 to 1:

  • Size of the largest connected component (LCC): the fraction of trees still mutually reachable through the network.
  • Algebraic connectivity $\lambda_2$: how fast information still propagates through the surviving graph.
  • Mean shortest-path length: average number of hops between trees in the LCC.
  • T2 saturation ratio: equilibrium reach — how close to optimal transport the surviving network is.

The phase transition

Plotting LCC vs f reveals a sharp phase transition for the targeted-attack case: above a critical fraction fc ≈ 0.04 – 0.07 (depending on the BA parameter m), the LCC collapses from ~95% of the original size to near zero in a narrow window. Random attack shows no such transition for f < 0.5; the LCC degrades smoothly and gradually.

Empirical evidence

Simard (1997) and Klein, Ammoneit & Hannemann (2016) used radioactive carbon and DNA-fingerprinting techniques to quantify carbon transfer through real CMNs before and after selective logging treatments. Their findings:

  • Removing the largest 10% of trees reduced inter-tree carbon transfer by 60–80%.
  • Removing 30% of trees at random reduced transfer by only 25–35%.
  • Recovery time to pre-logging connectivity after mother-tree removal was >15 years for the studied stands.

What is the cheapest way to protect a forest?

Naïvely, conservation effort should be spread evenly across all trees. The Cohen–Havlin result says otherwise: a much higher per-tree marginal benefit accrues from protecting the very few highest-degree individuals. In practice this means:

  • Identify the largest, oldest trees in a stand (typically <5% of stems but accounting for >30% of total network connectivity).
  • Exclude these trees from harvesting plans.
  • Retain a buffer of mid-aged trees around each mother to preserve local hyphal continuity.
  • Avoid clearing large continuous areas — corridors of intact canopy let the network bridge across.
Conservation rule of thumb
Protect the top ~5% of hub trees and you preserve forest connectivity even under heavy random disturbance. Remove them and the network collapses regardless of which other trees survive. This result is not a metaphor: it follows directly from the scale-free degree distribution and the spectral properties of the graph Laplacian.

Caveats and open questions

  • Real mycorrhizal networks may not be strictly scale-free at all spatial scales; small forests can look exponential, large landscapes scale-free. The transition is itself an active research topic.
  • Trees do not simply propagate nutrients passively. The active dynamics of carbon allocation, partner selection, and seasonal reversals (Simard 1997) means that "connectivity" is more than topological.
  • Climate-change-driven mortality is not random: it preferentially affects drought-stressed trees, which may be the same individuals that are most connected (large canopies → high water demand). The attack distribution is somewhere between random and targeted.

Where this connects to broader theory

The phase transition between connected and fragmented networks is a textbook example of bond percolation on a complex graph. Cohen et al.'s analytical results use the generating-function formalism of Newman, Strogatz & Watts (2001). The Fiedler-value collapse accompanying targeted attack is a special case of spectral bottleneck emergence, also studied in computer-science contexts (graph clustering, expander graphs, social-network polarisation).

Share:XRedditLinkedIn