The Tree That Wants to Be Struck by Lightning

The Bolt That Did Nothing

Start in the rain.

Barro Colorado is an island because a dam made it one. Until 1914 it was a hill in the Panamanian jungle, and then the Chagres River backed up behind Gatún Dam and the valleys around it filled. Fifteen square kilometres of old-growth lowland rainforest. Hardly any patch of forest on Earth has been studied this hard for this long: on part of the island, every stem thicker than your thumb has been counted, measured, mapped and re-measured on a five-year cycle since 1980. Somebody has worn out a lot of tape measure.

Lightning hits the place constantly. Central Panama sits under some of the most electrically active weather in the Americas. A storm rolls in off the lake in the late afternoon; somewhere out in that forest, several times a season, a stepped leader finds a crown and a return stroke rips upward through thirty-odd metres of wet wood.

Here is what normally happens next. Bark blows off in strips. Sap in the conducting tissue flashes to steam faster than it can escape, the wood splits, and the cambium, the single living layer that wraps the trunk just inside the bark, cooks. The tree may stand green for six months and then, in the dry season, discover it cannot move water any more and die all at once. Lightning kills an estimated 40.5% of large trees over 60 cm diameter in this forest [2], and nothing else kills more of the biggest individuals in the stand. Nobody knew that until somebody built an array of electric-field sensors to find out.

Now the part that should not happen.

September 2019. A bolt hit a Dipteryx oleifera, a tonka bean tree, one of those emergent giants that shoves its crown clear above the canopy like a head above a crowd. The researchers walked to it. Scorched in a couple of places, otherwise completely fine, and two years of monitoring later, still fine. Meanwhile 115 trees around it had been damaged, about half of them fatally, and every liana on it, every woody vine that had spent years climbing all over that crown, was dead [1].

The tree did not survive the lightning. The tree used it.

Ten direct strikes on Dipteryx sit in the dataset. All ten left the tree essentially unharmed. Struck the same way in the same forest, trees of other species died 64% of the time and lost 5.7× more of their leaves when they did survive [1]. Call that a categorical difference in outcome, well clear of anything noise could produce, and it demands a mechanism.

The Standard Story, and Where It Cracks

To see why this is startling you need the textbook version. It has been stable since the 1960s. A good story. It is about to spring a leak.

A tree is a tall wet stick standing in a field of atmospheric potential. When a thunderstorm charges the cloud base to tens of megavolts relative to the ground, the electric field at the tip of the tallest nearby object gets enhanced enormously. Franklin's rod works on exactly that principle: pointed conductors concentrate field lines. A tree does the same thing without meaning to, and being tall is being a target; the relationship is not subtle. Gora and colleagues built a risk model from thousands of mapped stems, and strike probability in it scales steeply with height and with crown area, because a wide crown intercepts more of the ground-level field and offers more attachment points [7].

Once the discharge attaches, the tree becomes a conductor whether it likes it or not. From there the analysis is first-year physics. Current flows. The tissue has resistance. Resistive heating deposits energy. Joule's law, integrated over the messy time profile of a real flash, says it:

and the quantity \(\int I^{2}\,\mathrm{d}t\), the action integral, is the number lightning engineers reach for, because it predicts how much metal melts and how much wood splits [11]. Gora and colleagues [3] ran this calculation on real trees with real measured resistivities, and the classical conclusion held: more current means more heating, and so does more resistance. More tissue to spread the heat over means less heating per gram.

So the received wisdom becomes: lightning is a tax on being tall. Grow above the canopy, get more light, get more strikes, and eventually one of them kills you. Alan R. Taylor put the ecological version of this in print in 1969, cataloguing lightning-killed forest patches and treating them straightforwardly as disturbance [14], and fifty-six years of literature followed the same instinct. A 2025 global vegetation model estimates lightning kills something like 320 million trees a year worldwide and destroys 0.21–0.30 gigatonnes of carbon in dead biomass, and it frames all of that, reasonably enough, as loss [15].

Here is the crack. The tax metaphor assumes the tree is the only thing being charged. Wrong assumption. A struck tree sits at the centre of a very local and very violent mass-casualty event involving everything electrically touching it, and in a tropical forest a great deal is electrically touching it.

How You Find a Struck Tree in Fifteen Square Kilometres of Jungle

Follow the procedure. The finding is only as good as the ability to answer one question: which tree, exactly? For most of the history of forest ecology nobody could answer it at all.

Step one: hear the flash before you see the damage. Commercial lightning networks locate cloud-to-ground strikes to within a few hundred metres. Useless here. A few hundred metres of Barro Colorado contains thousands of trees. So the team installed an array of Marx meters, fast electric-field-change sensors, each one sitting under an upturned metal bowl to keep the rain off, sampling from roughly 1 Hz up to 400 kHz [10]. Several sensors, precisely time-synchronised, triangulate the source of the field change, and cameras on 40 m towers back them up. Out comes a strike location good to tens of metres. Call that the difference between somewhere in the forest and that tree.

Step two: go and look, fast. A field crew walks to the located point within days and searches a radius for the signature of a strike, which is harder than it sounds, because a lightning-struck tree in the tropics often shows nothing at all from the ground for weeks. The diagnostic suite the group developed [9] includes flash marks and bark stripping. It also includes, and this is the part that matters, arboreal survey (somebody climbs into the crown) and later drone imagery from above, because the most reliable early signal is crown damage you simply cannot see from underneath.

Step three: come back. And back. Lightning mortality is slow; a tree with a cooked cambium can hold green leaves for a year before it fails. The 2025 paper's central comparison, 64% of struck non-Dipteryx dead against 0 of 9 Dipteryx dead, only exists because the crews re-surveyed for two years after each strike.

Step four: count the neighbours, and count the vines. Here a survival story turns into an ecology story. For each strike the team scored the struck tree and also the ring of trees around it: how many damaged, how many dead, how much biomass. They scored the liana load on the struck crown too, before and after. A strike on a Dipteryx killed an average of 9.2 neighbouring trees and about 2.1 megagrams of competitor biomass, and removed 78% of the vines [1, 17].

Step five: the comparison doing the real work. Because the 50-hectare plot has decades of census data, the authors could also ask whether trees near a Dipteryx die faster in general, over decades, and not only in the weeks after a strike. They do: neighbours of Dipteryx were about 48% more likely to die [1]. That long-run signal elevates the claim from "one dramatic bolt" to "a recurring feature of this tree's neighbourhood."

Bench Notes: The Afternoon Our Model Cancelled Itself

What follows is ours, and not the paper's. Gora and colleagues measured a forest. We sat in a classroom with a whiteboard and asked whether the physics forbids what they measured. Everything in this section and the next three is the club's own simplified model. Every number came out of analysis/lightning-tree.py, which you can download and break.

We drew Schematic 1. A 40 m tonka bean tree, trunk 1 m across, of which we let 60% of the basal area actually conduct: outer sapwood plus the moist cambial zone, heartwood written off as dead weight. A liana 4 cm thick, wandering 45 m from crown to soil. Wood as a plain ohmic cylinder:

Trunk resistivity we set at 120 Ω·m, deliberately at the low end, because the whole hypothesis is that this species conducts unusually well; the liana we put at 180 Ω·m. That choice is an act of aggression against the literature. We come back to it in §8.

Trunk: 0.4712 m² of conducting cross-section, R = 10.7 kΩ, and 17.8 tonnes of wood in the current path; liana: 1.257 × 10⁻³ m², R = 6.446 MΩ, and 39.6 kg. The vine is 603 times more resistive and 450 times lighter. Obvious, we thought. The resistive thing with no thermal mass will cook.

Then we did it properly, because in a genuine parallel pair current divides by conductance, not resistance. The liana's share:

Which is 0.166% of the flash. Take the standard waveform: 200 A of continuing current, 115 ms [3]. The vine gets 0.331 A, and its power is 707 kW, which sounds enormous until you divide by 39.6 kg and get 17,873 W/kg. The trunk, hogging 199.67 A, dissipates 426 MW over 17.8 tonnes. Call it 23,937 W/kg.

The vine heats slower than the trunk. Ratio 0.75. Over the whole 115 ms the trunk warms 1.38 K, the vine 0.69 K. Nothing in this model is damaged at all. Not the vine. Not the tree. Not the neighbours. Not so much as a scorched leaf.

We assumed we had a bug. We did not. Watch what happens when you write power-per-kilogram out in full, for a path holding voltage \(V\) across its ends:

The cross-sectional area \(A\) appears in numerator and denominator and cancels exactly. In a shared-voltage divider, heating per kilogram depends only on the material and the path length: being thin buys you less current and less mass to spread it over, in precisely equal measure. To this model, a 4 cm vine and a 1 m trunk are the same object. We swept liana diameter from 1 cm to 16 cm, a 256-fold change in cross-section, and the heating rate crawled from 17,929 to 17,017 W/kg, under 6%, and in the wrong direction.

Our model cannot tell a vine from a tree trunk. So it cannot explain why one of them explodes.

So we broke the assumption instead of the arithmetic. Suppose the flash does not politely divide at a shared node, but attaches to the liana directly, or bridges a gap where no thick alternative sits beside it. Then the channel sets the vine's current upstream. Its own resistance gets no say. Push the full 200 A through it: 257.8 GW, or 6.51 GW/kg. Against the trunk's 24,017 W/kg, the ratio becomes

and now \(A\) is squared and does not cancel at all. The two limits differ by a factor of 363,221. That entire span is our ignorance about one topological question. Does the vine share the tree's voltage, or the tree's current?

Somebody pointed out that we were arguing about watts when tissue responds to joules, so we forgot currents for a moment and counted energy.

To kill living cambium you have to get it to roughly 60 °C. Start at a Panamanian ambient 25 °C, take wet wood's specific heat at around 2,000 J·kg⁻¹·K⁻¹, and the bill is 70,000 J/kg for trunk tissue and 105,000 J/kg for the wetter, higher-heat-capacity liana. Going further, actually flashing the tissue water to steam, which is what splits a trunk open, means paying the latent heat of vaporisation too: 2.26 MJ for every kilogram of water. At 45% water by mass in the trunk and 60% in the vine, the bills come to 1.17 MJ/kg and 1.58 MJ/kg.

Those thresholds are the dashed lines in Figure 1. Divide them by the heating rates and you get times.

A cleaner way to put it exists. Instead of asking how hot something gets, ask what current destroys it; rearranging \(E = I^{2}R\,\Delta t\) for the current that just delivers a lethal energy budget to a mass \(m\) in the 115 ms available:

The liana needs 9.2 A to boil dry. Only 2.4 A kills its cambium. Those are 4.6% and 1.2% of the continuing current. The trunk needs 4,111 A, twenty-one times what the flash carries. In current, the trunk's safety margin is 447×. In energy, which goes as the square, 200,188×.

So the vine does not need to capture the flash. It needs about one twentieth of it. And here is the claim our model actually supports. The gap between "annihilates a vine" and "warms a trunk by one degree" is so enormous that almost any plausible current-sharing arrangement lands somewhere inside it. Invulnerability is a stronger word than we have earned.

The Ledger

Every quantity we computed, in one place. No adjectives.

Every row above is the club's model output except the two marked as the paper's. Seed: 20250326.

Questions We Kept Asking Each Other

Three Hundred Years, Twenty Thousand Trees

The circuit tells you a vine can die, but not whether that matters over the life of a tree. So we built the cartoon.

Twenty thousand simulated trees per arm. One-year timesteps, three centuries, no realism beyond what the argument needs. Each year a tree may be struck; hazard scales as height squared times crown area, normalised so that our tall wide-crowned morphology gets hit once per 56 years, matching the paper. A strike knocks the liana load down to 22% of what it was, the paper's 78% kill, and opens a canopy gap that decays over about 22 years. Then, depending on which arm the tree is in, the strike either kills it or does not. Between strikes lianas regrow logistically, dragging down both seed output and survival. Reproduction starts at year 60 and ramps to full capacity over 40 years.

We ran two contrasts, because they answer two different questions, and conflating them is the easiest mistake available here.

D1, tolerance alone. Two identical trees. Same height, same crown, same strike rate, same everything. One survives strikes. The other dies 64% of the time. The paper's 14× refers to this contrast. Hold everything else fixed. What is survival alone worth?

D2, the whole strategy. A tall tolerant tree against a shorter, narrower, intolerant one. The short tree is struck once per 89 years instead of once per 56, a 58% difference in strike rate, which brackets the paper's reported 49–68% elevation for Dipteryx allometry [1]. But it also intercepts less light and carries a smaller crown.

The D1 answer is 7.23×. Mean lifetime seed output 45.12 arbitrary units against 6.24. Fraction still alive at 300 years, 7.1% against 0.2%; median age at death, 89 years against 39.

D2 comes out at 6.64×. Smaller, and correctly so. The avoider genuinely buys something with its shorter stature. Growing tall is not free, and a model that made the trade-off version larger would be broken.

Neither is 14.

We then took the model apart to find where the advantage lives. Switch off the liana penalty on seed output and the ratio falls from 6.82 to 5.12; dropping the canopy-gap bonus instead gives 6.42, and dropping the liana contribution to mortality gives 6.59. Kill all three and what remains is 4.58×. Pure survival, the advantage of simply not dying. So roughly two-thirds of our effect is bare survivorship. The other third is the competitive housekeeping the paper emphasises.

Steelmanning the Skeptics

Suppose you wanted to argue that none of this means what it appears to mean. Here is the strongest case we can build, put in the skeptic's own voice. We think most of it is serious.

"You have nine trees." Correct. The whole headline survival result rests on ten directly struck Dipteryx. Ten. Suppose the per-strike death probability for this species were genuinely 20%, which is meaningfully worse than immune; the chance of then observing zero deaths in ten trials is 0.8¹⁰ ≈ 0.11, and nobody would call that a small probability. The dataset cannot distinguish "essentially invulnerable" from "moderately tough" with any confidence, and the forest ecologist Gabriel Arellano made effectively this point in coverage of the paper, noting that sample sizes for most individual species remain too small for firm conclusions [19].

"Nobody has shown the mechanism." Also correct. The paper proposes high conductivity: less resistance, less internal heating. Stated, not demonstrated. As reported at the time, how D. oleifera survives remains unclear, and the competing hypothesis is purely geometric, that the crown architecture shunts current out through branches into neighbouring trees rather than down the trunk [18]. Those two hypotheses make different predictions, and the data do not yet separate them. Our own model is guilty here too, because we had to assume Dipteryx is the conductive outlier, reversing the general tree-versus-vine pattern Gora and Yanoviak measured [4]. If that assumption is wrong, so are §4 and §5. Specifically and correctably wrong. §10 tells you how to check.

"A benefit is not an adaptation." The objection with the most teeth, and we have no good answer to it. Everything in the paper is consistent with Dipteryx being a huge dense hardwood that lives for centuries, tolerates current well as a side effect of its wood anatomy, and is tall for the boring reason that tall is how you win at light. The lightning benefit would then be a lucky by-product rather than a selected function. Telling those apart needs heritability data and a comparative phylogenetic test across the genus. Ideally it also needs a species that is tall and conductive without being long-lived. None of that exists yet.

"Even the specialists cannot agree on the circuit." They cannot. The 2017 modelling paper concluded lianas protect host trees, cutting heating by up to 87% [3]; the 2023 field paper concluded lianas make lightning disturbance more severe, by bridging current into extra trees [5]. Both come from the same group. Both are defensible. They describe different aspects of one system. Anyone claiming the electrical ecology of a liana-laden crown is well understood is overselling. As Bianca Zoletto put it, the interaction between trees and lightning is genuinely difficult to pin down and sits more on the physics side than the ecological side [18].

And then the objection we would raise against ourselves, which is that our model is cruder than theirs in one specific and embarrassing way. We have no flashover. Real lightning largely travels along the wet outer bark surface rather than through tissue; that is the single largest omission in our arithmetic, and it is the reason a median return stroke's full action integral of 55,000 A²·s, if we let the trunk conduct all of it, deposits 588 MJ and warms 17.8 tonnes by only 16.5 K. We also treat wood as ohmic, homogeneous, isotropic and temperature-independent, and it is none of those [16]. Resistivity falls steeply with moisture content and temperature, then rises catastrophically once sap boils, which makes the real problem nonlinear in a way a resistor network cannot capture.

The result is probably right. The mechanism is a hypothesis, and the adaptation claim is a hypothesis resting on that one.

A Short History of Watching Trees Get Hit

This question is old. For a very long time it was also completely stuck.

In 1752 Franklin put a pointed iron rod on a roof and demonstrated that a good conductor, well grounded, could take a discharge and route it harmlessly to earth. The physics of lightning protection has been essentially settled ever since. For buildings. Nobody could apply it to forests, because nobody could say which tree was hit.

By the mid-twentieth century foresters had accumulated a substantial folklore. Oaks were "lightning-prone". So were poplars and elms. Beech, famously, was said to be rarely struck. Mostly this was inference from scars, and scars are a biased sample: you only find the trees that survived long enough to be scarred, in stands people walked through. Alan R. Taylor's 1969 survey [14] is the serious version. Even it is an inventory of aftermath.

The methodological problem is stark. To study lightning ecology you need to know, for a specific tree, that it was struck on a specific day. Then you need to come back for years. Commercial lightning networks, built for utilities, locate a flash to within a few hundred metres, which in a tropical forest means thousands of candidate stems, any one of which might be the one that burned. The whole field was blocked on a positioning problem.

What unblocked it was a physics instrument built for an unrelated reason: arrays of fast electric-field-change sensors of the kind characterised for the Huntsville Alabama Marx Meter Array [10]. Deploy several of those around a well-mapped forest, synchronise them, add tower cameras, and the location problem collapses. The first fruits arrived quickly: a 2017 methods paper on identifying lightning damage [9], then the 2020 result that lightning kills 40.5% of the largest trees in this forest [2], then a 2021 accounting of its contribution to biomass turnover and gap formation [6], a 2022 demonstration that species differ enormously in their lightning mortality [8], the 2023 liana-severity result [5], and now the 2025 Dipteryx paper [1].

Six major results in eight years, on a question that sat immobile for fifty. A sensor did that.

How Would We Test This?

Four experiments. Design them yourself before reading ours; you will learn more that way than by agreeing with us.

Start with the cheap one, which also happens to decide whether our §4 rests on sand. Drive electrodes into standing trunks of Dipteryx oleifera and into matched-size individuals of the species that die, using the in-situ resistivity method already run on temperate trees and vines [4] and on tropical species in [3]. Prediction: if the conductivity hypothesis holds, Dipteryx sapwood resistivity comes out measurably lower than that of comparable emergents. If not, the conductivity story is dead and the crown-geometry story inherits the field.

Then the expensive one. That whole 363,221-fold span of ours is a single question about where the current goes, and hardware can answer it. Put Rogowski coils or magnetic-field sensors on the major branches and on individual lianas of a canopy tree at a high-strike-rate site. Then wait. On the lucky day they record how the current actually divided. Expensive, and at the mercy of a storm that may not come for years. It would also settle physics that ecological inference currently has to guess at.

The third one you could almost run with a machete and patience. Liana removal is a mature technique with a large literature behind it [13]. Take matched Dipteryx individuals. Strip the vines off half of them by hand. Follow seed output for a decade. If the lightning benefit is mostly vine removal, hand-removal should reproduce most of it, and lightning stops looking like a weapon and starts looking like a free gardener with very poor impulse control. If hand-removal recovers only a fraction, neighbour-killing carries more weight than we thought.

And last, the one the adaptation claim actually needs. The genus Dipteryx has several species scattered across the Neotropics. They sit under wildly different lightning climatologies. Map the tolerance traits onto the phylogeny. If tolerance tracks strike frequency across the genus, call it selection rather than coincidence. If every Dipteryx turns out equally tough regardless of how much lightning it ever sees, the trait is ancestral baggage and the adaptive story weakens considerably.

What We Actually Believe

Here is where we ended up, stated as plainly as we can manage.

We believe the observation. Ten strikes is a small number. But the effect size is enormous, the neighbour-mortality signal is corroborated independently by decades of plot census, and the instrumentation is the best in the world for this specific question. When a categorical difference shows up in a small sample and a separate, much larger dataset points the same way, you are not looking at a fluke.

The physics permits it, and that part we worked out for ourselves. Our model spans a factor of 363,221 between its two limiting cases. Sounds like failure. Actually it is the finding. Within that span, the threshold for destroying a 40 kg vine and the threshold for destroying a 17.8 tonne trunk are separated by a factor of 200,188 in energy. A bolt has room to be lethal to one and trivial to the other. We could not build a version where vine and trunk are in comparable danger.

We do not believe the 14× is a stable number. Neither, we suspect, do its authors. The figure is a modelled projection over a lifespan that is itself only estimated. Our own version gives 7.23×, our sensitivity sweep across 27 parameter combinations spans 4.58× to 13.49×, and 14 sits just outside the top of that range. Reaching it would need the intolerant tree to die from about 78% of strikes rather than 64%, plus the strongest liana and gap effects in our grid. Those values are not absurd. They are simply not what we chose.

What we do not believe, yet, is the word evolved, taken in the strong sense of a trait selected for this function. That requires a comparative test nobody has run.

But here is what stays with us. For a century the lightning literature counted trees as casualties. What changed the picture was a handful of field-change sensors and a crew willing to walk into the forest after every storm for years, with no new theory required at all. The tonka bean tree had been doing this the whole time, and we simply had no way to notice.

Stand under a 40-metre crown in a thunderstorm. It looks like a tree taking a terrible risk. As far as anyone can tell, it is a tree collecting a debt.

References