An Open Research Project

Trees may shape where rain falls — and that may matter for global heat transport.

A citizen scientist, two AIs, a laptop, and one question worth asking carefully.

Read the Story Read the Paper
Aerial view of Amazon rainforest with cumulus clouds building above the canopy
Where it started

A strange pattern on an iPhone.

One day, looking at the weather map on his phone, an engineer noticed something odd. The equatorial rainforests showed continuous rainfall while the adjacent oceans showed intermittent rainfall.

That seemed backwards. Oceans have unlimited water for evaporation. Why would rain be more persistent over land?

Tropical rainforest topsoil is shockingly thin — typically 9 to 15 inches. Far too shallow to store an abundance of water. Trees in this environment must receive nearly continuous rain or they die. Evolution would favor any mechanism that lets trees secure their own water supply.

"Have trees evolved to inject something into the air to grab water for themselves?"

A search of the literature revealed the answer: yes.

In 2012, researchers led by Pöhlker et al. at the Max Planck Institute for Chemistry in Mainz, Germany — with Lawrence Berkeley National Laboratory collaborators doing the electron microscopy — discovered that fungi living on Amazon rainforest trees eject microscopic potassium salt particles into the air. These salt crystals act as seeds for cloud formation, triggering local precipitation. The trees and their symbiotic fungi are performing continuous cloud seeding — the biological equivalent of the airplanes that seed clouds for agriculture.

Salt nanoparticles rising from rainforest canopy into forming clouds
The thermodynamic insight

Where the rain goes, the heat goes.

Every gram of water vapor that condenses into rain releases 2,500 joules of heat. The atmosphere is a heat pipe — water vapor carries latent energy from the surface to wherever it condenses, depositing that energy as warmth.

If salt traps rain locally at the equator, it traps heat at the equator.

If the forests are cut and the salt source is removed, that humidity is free to travel poleward, carrying its latent heat to higher latitudes and releasing it closer to the poles.

This is not a greenhouse gas mechanism. It does not change how much heat Earth traps from the sun. It is a redistribution of existing heat — changing where thermal energy is deposited.

Ice cubes in a glass melt slowly in still water. Stir the water and the ice melts dramatically faster — not because stirring adds energy, but because it thins the thermal boundary layer, letting heat conduct through faster. Removing the equatorial salt barrier is like stirring the atmosphere.

The stirring analogy: still water with slow-melting ice vs stirred water with fast-melting ice
Heat transport diagram: with forest salt vs without salt
The gap

No mainstream model appears to explicitly represent this mechanism.

Hygroscopicity and cloud condensation nuclei (CCN) observations derived from Pöhlker-era work have entered broader aerosol-climate modeling in various ways. What we have not found, however, is a mainstream global climate model that explicitly represents rainforest biogenic-salt warm-rain emissions from the tree-fungi pathway as a dedicated parameterized process. The aerosol fields in most major climate models are dominated by pollution, sea salt from ocean spray, and mineral dust — the biogenic-salt-specific pathway, coupled to vegetation type, is not where our search has turned up an explicit representation.

This mechanism sits at the intersection of four fields that rarely interact:

Tropical Biology
Biogenic salt aerosol emissions linked to rainforest ecosystems were highlighted in 2012 (Pöhlker et al.); we have not found evidence that this specific mechanism has been explicitly incorporated into mainstream global climate-model physics as a dedicated parameterization.
Cloud Microphysics
Giant salt particles enhance rain (opposite of pollution aerosols which suppress it). The distinction is critical.
Global Dynamics
Meridional energy transport carries 5 petawatts poleward. Even a 2% change is climatologically significant.
Deforestation Science
Deforestation studies have focused primarily on albedo and carbon sinks. We have not identified published climate-model studies that explicitly isolate the aerosol-transport consequences of losing this rainforest biogenic-salt source.
The researcher

Not a meteorologist. A rocket scientist who became a systems pattern-matcher.

The lead researcher trained for five years under Robert Truax — the inventor of the regeneratively cooled rocket engine and Jet Assisted Takeoff (JATO), president of the American Rocket Society, and advisor to President Eisenhower on establishing NASA. He earned two NSF grants studying fluid-structural combustion instabilities in liquid-fueled rockets at the UC San Diego Supercomputer Center, and completed a Master's at Stanford in Aeronautics and Astronautics.

That rocket work taught a lesson that kept reappearing: in complex systems, a chemistry problem in one place can travel through a medium and cause damage somewhere far away. Fuel sloshing in a tank feeding back through the turbopumps and into the combustion chamber — driving POGO oscillations and combustion instabilities that can destroy an engine or tear an airframe apart. Process chemistry in one chamber making defects elsewhere in a semiconductor. And — most relevant to the current work — his later microbiome research (patented) showed that vanishingly small amounts of invisible chemicals secreted by gut microorganisms, reaching the brain via the bloodstream and vagus nerve, can be lethal or cause serious disabling neurological disease. The same pattern holds in climate: nanometer-scale salt particles emitted by rainforest trees can dramatically reshape global rainfall, heat transport, and polar ice balance. In both cases the active species is invisible to the naked eye, present in quantities that look negligible by any bulk measure, yet controls the outcome of a system orders of magnitude larger than itself.

He is not a career meteorologist. But he has decades of training in exactly the physics this problem requires: fluid mechanics, heat transfer, two-phase flow, numerical simulation, and — most directly relevant here — plasma physics. Cloud microphysics and plasma physics are closely related kinds of problems mathematically: huge numbers of small particles colliding, activating, and growing, where the overall behavior depends on how often particles meet, how fast they are moving, and whether they cross an activation threshold. A K-salt particle nucleating a cloud droplet obeys very similar governing equations to an ion forming in a plasma — just slower, colder, and electrically neutral. He has worked on problems of this kind across multiple fields.

He recognized the same template in climate — with an inverted polarity. Rainforest trees produce salt that condenses rain and releases latent heat locally at the equator. When deforestation removes the salt source, moisture escapes poleward, redistributing latent heat — disrupting precipitation and heat-transport patterns that have persisted since the end of the last ice age, roughly 11,000 years ago. The common thread: when critical chemistry goes wrong in one place — either by producing the wrong thing or failing to produce the right thing — the consequences show up somewhere far away. Atmospheric scientists have studied each piece of this chain separately. The hypothesis here is just that the pieces connect — and that the connection has been hiding in plain sight, because continuous rain over rainforests has always been the way weather maps look.

30+
Patents across six fields
2
NSF research grants
5
Years under Robert Truax
MS
Stanford Aero/Astro
The toolkit

Government software. Consumer AI. A laptop.

Every tool used in this research is freely available. The atmospheric models were built with decades of US government and academic investment. The AI assistants are consumer products anyone can access. The hardware is a laptop you can buy at Best Buy.

🌍
MPAS-Atmosphere
Global atmospheric model from NCAR and LANL. 10,242 hexagonal cells wrapping the entire Earth. 55 vertical layers from surface to stratosphere. Free, open-source, funded by NSF.
🤖
Claude + ChatGPT
Claude (Anthropic) built the Docker containers, compiled the models, wrote the Fortran modifications, and designed the experiments. ChatGPT (OpenAI) cross-verified the equations and caught three errors.
💻
A Consumer Laptop
Intel i7-12700H, 64 GB RAM. Each 30-day global simulation runs in 1.5 hours at 240 km or ~9 hours at 120 km. Total cost of hardware: ~$1,500. No supercomputer required.
Consumer laptop running MPAS global atmospheric simulation

The WRF and MPAS models represent decades of taxpayer-funded investment by NCAR, NOAA, DOE, and university research labs worldwide. This software was built so that atmospheric science could advance. Consumer AI made it accessible to anyone with the right question.

The software sat unused on a powerful laptop for two years. Then AI arrived, and the rest is history.

Seven phases of discovery

We got it wrong. Then we got it less wrong. Then a second AI found bugs in our fix.

Every misstep taught us something. We share them all because transparency matters more than polish.

Phase 1
WRF Channel Domain
Failed. Boundary artifacts and domain too small. Lesson: regional models can't study global transport.
Phase 2
MPAS v7 — Autoconversion Hack
Forced more rain at the equator by boosting cloud-to-rain conversion. Signal appeared: Arctic cooled 2.5 K. But salt was applied over ocean too — 5x too broad.
Phase 3
MPAS v8 — Prognostic Aerosol Transport
Real aerosol tracking. Salt emitted only from rainforest cells. But Thompson microphysics treated salt as small CCN (Twomey effect) — wrong rain physics. Arctic warmed instead of cooled.
Phase 4
Giant CCN (GCCN) Tracer — Simplified
Dedicated salt tracer modeled as giant cloud condensation nuclei (GCCN — the larger, ≥1 μm subset of CCN that initiate warm rain via collision-coalescence) with coalescence physics. Strongest rain enhancement (+0.19 mm/day). But fixed drop parameters (E=0.6, V=0.25, R=50μm) gave wrong transport sign.
Phase 5
GCCN Tracer — Full Lifecycle
The breakthrough. Drops grow from 10μm through the coalescence gap. Size-dependent collision efficiency (Hall 1980), terminal velocity (Beard 1976), wet scavenging. Transport flipped to correct sign: −95 TW. Antarctica cooled −1.03 K.
Phase 6
Iterative Debugging Surfaced Three Bugs
Iterative cross-checking of our GCCN physics against Thompson conventions surfaced three errors: a drop-radius reset keeping collectors below 20 µm, a missing air-density factor, and rain-number routing through the wrong pathway. We fixed them, rebuilt, and re-ran April 240 km. The corrected results differ qualitatively from the buggy version — and the fact that implementation details flip signs is itself the central finding.
Phase 7
120 km January with Bug-fixed Physics — Completed
Seasonal mirror partially confirmed. Rain redistributes northward in January (+0.18 mm/day at 5°N, −0.21 at 5°S) — the flip of April's southward shift. 30°N transport reduced by −211 TW (NH winter-ward hemisphere), 30°S increased by +71 TW (SH summer-ward). Antarctic near-neutral at +0.04 K (consistent with SH summer). One puzzle: Arctic still warms (+0.88 K) despite reduced transport, pointing to either a dry-heat pathway or weather noise dominating a single realization.
Phase 8
Pöhlker Baseline-CCN Matrix — Eight-Run Completed 2026-04-22
K-salt's effect flips sign depending on the baseline pollution level, but at the observed particle size the local rainforest benefit is preserved while far-field effects are modest. A reviewer-anticipating concern: MPAS's default Amazon aerosol baseline (~4,400/cm³) is 10–20× higher than Pöhlker et al.'s (2012) direct pristine ATTO observations (~100–300/cm³). We ran an eight-simulation matrix (four K-salt pairs across baseline and activation-size states), switched K-salt representation from giant CCN to accumulation-mode CCN (per Pöhlker's actual observations, Dg = 150 nm, κ ~ 0.8), and explicitly disabled ngccn. Results: at the polluted baseline, K-salt suppresses Amazon rainfall (−17.1 mm/30-day); at the Pöhlker-anchored pristine baseline with the observation-matched particle size (Thompson l=4, 160 nm diameter), K-salt enhances Amazon rainfall (+5.4 mm/30-day) with modest global-circulation impact (+31 TW at 30°N, +0.12 K Arctic). An upper-bound-size sensitivity test (l=5, 320 nm) reverts the local effect toward zero, suggesting the larger size pushes the system past its precipitation optimum. All 8 runs passed independent sanity checks. Implication: evolution optimizes K-salt for local rainfall (the only fitness-accessible outcome); the observed size produces clean local enhancement with only modest, physically downstream global consequences. Modern anthropogenic CCN pollution may already have pushed the Amazon out of the regime where this mechanism works — regional air-quality protection matters alongside canopy preservation.
The numbers

Salt is a remarkably sensitive variable.

Ten different implementations of the same hypothesis. A swing of ~364 TW in the global heat transport response — and the sign of the rainfall effect over the Amazon itself flips between polluted and pristine baselines. At the Pöhlker-observed particle size, the local Amazon enhancement is preserved while far-field signals are modest — consistent with evolution selecting for local rainfall, not global climate.

Metricv7 Autoconvv8 TwomeyGCCN Simp.GCCN (buggy)Bug-fixApril 2026240 km mesh · 30-day
v8.3.1 GCCN · NH summer		Bug-fixJanuary 2025120 km mesh · 30-day
v8.3.1 GCCN · NH winter		Phase 7 · PöhlkerPolluted4,400 → 8,800 /cm³
80 nm · κ=0.4
ngccn=0		Phase 7 · PöhlkerPristine150 → 300 /cm³
80 nm · κ=0.4
ngccn=0		Phase 7 · Pöhlker-matchedPristine + Dg150 → 300 /cm³
160 nm · κ=0.8
Matches Pöhlker Dg=150 nm		Phase 7 · sensitivityPristine + upper-bound150 → 300 /cm³
320 nm · κ=0.8
Upper-bound size test
Equatorial rain (band avg)−0.10+0.05+0.19+0.05+0.17 mm/day−0.002 mm/day−0.13 mm/day−0.09 mm/day+0.15 mm/day−0.03 mm/day
Rain at 5°S zonaln/an/an/an/a+0.62−0.21−0.28−0.25+0.47−0.08
Rain at 5°N zonaln/an/an/an/a−0.31+0.18−0.62+0.50+0.24+0.21
Rain over Amazon rainforest
(SALT − NOSALT, 30-day total, averaged over evergreen broadleaf forest cells only)		n/an/an/an/a+0.5 mm+6.9 mm−17.1 mm+5.6 mm+5.4 mm−3.0 mm
30°N transport−153+66+42−95+153 TW−211 TW−109 TW−172 TW+31 TW+136 TW
30°S transportn/a−330n/a−18−61 TW+71 TW+210 TW+15 TW−125 TW−249 TW
Arctic temp−0.15+0.32+0.21+0.43+0.14 K+0.88 K+1.73 K+0.86 K+0.12 K+0.24 K
Arctic 10m windn/an/an/an/a+0.00 m/s+0.38 m/s−0.20 m/s−0.16 m/s+0.12 m/s+0.45 m/s
Antarctic temp+1.5−0.50−0.13−1.03−1.26 K+0.04 K0.00 K+0.03 K+0.03 K+0.12 K

Reading the table. Columns are grouped by mechanism. Columns 1–4 (grey): earlier implementations with known limitations. Columns 5–6 (green): the two “Bug-fix” runs that modeled K-salt as giant CCN (200 nm collision-coalescence, ngccn active) — April and January seasonal mirrors. Columns 7–10 (blue): four Pöhlker Phase 7 runs that model K-salt as accumulation-mode CCN through the nwfa field, with ngccn explicitly disabled. The column with the blue accent bar (column 9, “Pristine + Dg”) is the primary Pöhlker-matched configuration: 160 nm diameter, closest to Pöhlker’s reported accumulation-mode geometric mean diameter Dg = 150 nm. Column 10 (“Upper-bound”, 320 nm) is a sensitivity test rather than a Pöhlker-matched configuration.

The central sign-flip finding (columns 7–10). Starting from MPAS’s default polluted Amazon baseline (~4,400 nwfa/cm³, representative of modern Amazon air influenced by anthropogenic sources), adding K-salt suppresses local rainfall by 17.1 mm over 30 days — the air is already past its precipitation-efficient CCN concentration. Lowering the baseline to match Pöhlker’s direct clean-air observations (~150/cm³) and adding the same fractional K-salt contribution enhances Amazon rainfall by 5.4–5.6 mm (the sign-flip is robust to the activation-size choice). The sign of the rainforest’s self-watering response flips with the background pollution level. The upper-bound sensitivity (column 10, 320 nm) reverts the effect toward zero, consistent with the system being pushed past its precipitation optimum when the assumed particle size is too large relative to the observed Dg. The response depends on effective CCN activation capacity (number × size × hygroscopicity), not raw particle number alone.

At Pöhlker’s observed particle size (column 9), global circulation responses are modest. 30°N transport changes by only +31 TW (vs. −172 at default size and +136 at upper-bound), and Arctic temperature changes by +0.12 K — near the noise floor of single-realization runs. This is scientifically important: plants and fungi in the rainforest cannot plausibly have been selected by evolution for effects at the poles. Atmospheric transport timescales from the Amazon to the Arctic are too slow (months to years) for any polar consequence to feed back on individual-organism fitness. The observed K-salt size should therefore be expected to optimize for local rainfall with modest far-field consequences — which is exactly what column 9 shows. The larger transport signals in the other columns appear to be artifacts of size assumptions that do not match the observed biological emission.

Arctic wind still responds physically at the polluted and pristine baselines. Surface wind speed over the Arctic core (beyond 70°N) decreases with K-salt addition at both the polluted (−0.20 m/s) and pristine-default (−0.16 m/s) configurations. Since polar ice loss is dominated by convective heat transport (turbulent sensible heat flux, proportional to wind speed) rather than conductive warming from air temperature, a 3% wind reduction translates to ~1–2 W/m² reduction in heat delivery to sea ice — within the magnitude range of regional radiative perturbations of climatic interest. At the Pöhlker-matched size (column 9), Arctic wind response is smaller (+0.12 m/s), consistent with the view that the observed biology-selected size produces clean local effects with muted far-field consequences.

Columns 5–6 still show the seasonal-mirror pattern. For the GCCN bug-fixed April/January pair: rain direction, 30°N transport, 30°S transport, and Antarctic temperature all flip sign with the season — consistent with salt modulating whichever Hadley branch is currently active. Arctic temperature warms in both seasons.

The 30°N transport numbers range from −211 TW to +153 TW across ten implementations of the same salt hypothesis, with mean ~−31 TW and standard deviation ~126 TW — larger than the entire estimated effect of anthropogenic CO2 on poleward heat transport. Changing a single drop-radius constant from 10 µm to 25 µm flips sign. Changing the baseline CCN concentration from 4,400/cm³ to 150/cm³ changes magnitude and sign. Changing the aerosol size assumption across the Pöhlker-matched lookup options (160 nm vs 320 nm diameter) reverses direction again.

This is not noise. Each run is deterministic on the same initial conditions. The spread is the fingerprint of how strongly biogenic salt couples to the atmospheric circulation. The interpretation we favor: salt functions as a variance amplifier, not a monotonic forcing. Some weather regimes get large latent-heat pulses from GCCN-enhanced rain; others don't. The mean may be near zero. The amplitude of the fluctuations may be large. Testing this requires running paired ensembles — next up.

Within the most physically complete April 240 km bug-fixed configuration, three robust findings emerge: (1) salt redistributes equatorial rain southward — zonal-mean rainfall rises by +0.62 mm/day at 5°S and falls by −0.31 mm/day at 5°N, consistent with salt precipitating moisture on the winter-ward side of the equator before it can cross northward; (2) Antarctica cools by −1.26 K in the hemisphere entering its winter season; (3) southward moisture transport at 30°S is reduced by −61 TW. The Northern Hemisphere 30°N transport (+153 TW) is opposite in sign and is most likely an artifact of the convective parameterization at 240 km.

The January 120 km bug-fixed run partially confirms the seasonal mirror. Rain redistributes northward (+0.18 mm/day at 5°N, −0.21 at 5°S, opposite of April). 30°N transport is reduced by −211 TW (mirror of the April summer-hemisphere increase), 30°S transport is increased by +71 TW. Antarctic temperature is near-neutral at +0.04 K (SH now in summer), consistent with the framework. The unresolved puzzle: the Arctic warms by +0.88 K in January despite reduced moisture transport, pointing to either a dry-heat pathway or weather noise dominating a single-realization experiment.

We make no claim about specific directions or magnitudes here — single realizations cannot distinguish forced signal from weather noise. The community must help us test this properly with ensembles, convection-permitting resolution, and longer integrations.

−17 → +5 mm
Amazon rainfall sign flips with baseline CCN: polluted → suppression, pristine → enhancement (Phase 7)
±150 TW
Spread of 30°N transport response across implementations
−1.26 K
Antarctic cooling with salt (April 240km bug-fixed)
+0.62
mm/day
Rain rise at 5°S; falls by −0.31 mm/day at 5°N. Salt shifts equatorial rain toward the winter hemisphere.
Phase 7 baseline-CCN sensitivity matrix — 8-run summary

Phase 7 — Amazon rainfall response flips sign with baseline CCN, not with particle size. ΔP = Rain(salt) − Rain(no-salt) across four K-salt pairs. Top row: Amazon zoom. Bottom row: global Robinson. Shared color scale. Columns 1–2: baseline-CCN sensitivity — at MPAS's default polluted baseline K-salt suppresses Amazon rain by 17 mm/30-day (coherent red); at Pöhlker-anchored pristine conditions K-salt enhances it by 5.4 mm (coherent blue). Column 3: at Pöhlker's observed particle size (Dg = 150 nm, Thompson l=4, our primary Pöhlker-matched configuration) the local enhancement is preserved (+5.2 mm) while far-field responses are modest (+31 TW at 30°N, +0.12 K Arctic) — consistent with evolutionary selection for local rainfall rather than far-field climate effects. Column 4: at an upper-bound sensitivity size (Thompson l=5, 320 nm diameter, not observed by Pöhlker), the effect reverts toward zero, demonstrating that the response depends on effective CCN activation capacity (number × size × hygroscopicity) rather than raw particle count. The central finding: modern anthropogenic CCN pollution may have pushed the Amazon out of the regime where its evolved self-watering mechanism can work.

The lesson

One radius constant flipped the sign of global heat transport.

The coalescence gap: droplet growth stages from 10 to 80 micrometers with increasing collision efficiency

When a salt particle activates in a cloud, it creates a droplet about 10 micrometers across. At that size it cannot collect other droplets — the collision efficiency is effectively zero. This is the coalescence gap: a size range where condensation has become too slow to grow drops further, and drops are still too small for collisions with neighbors to work efficiently.

Our first GCCN implementation had a bug: the collector-drop radius reset to 10 µm every microphysics timestep. Over one timestep drops only grew to ~20 µm — right at the boundary where collision efficiency starts but is still minimal. The GCCN physics was essentially idling. That version gave us a −95 TW reduction in 30°N poleward heat transport.

Subsequent cross-checking of the Fortran code against Thompson's native aerosol-scavenging conventions caught the radius-reset bug, the missing air-density factor, and the rain-number routing error. We fixed them. Now collector drops start at 25 µm (an "already-activated embryo"), which gives them a realistic collision efficiency. The same hypothesis now gives us +153 TW at 30°N — opposite sign, larger magnitude.

A swing of 364 TW in the global heat transport response across nine implementations — and a sign flip in Amazon rainfall response between polluted and pristine baselines. The transport swing alone is larger than the estimated entire effect of anthropogenic CO2 on poleward heat transport. The Phase 7 sign flip (columns 7–9 of the table) adds a second axis of sensitivity: the atmospheric background the salt is acting in matters as much as the salt itself. We interpret this not as "we finally got the right answer" but as evidence that biogenic salt is sensitive enough that mainstream models not explicitly representing this specific rainforest-emission mechanism — or the baseline-CCN regime it acts in — are probably missing something that matters.

A note on accuracy

What consumer hardware can and cannot do.

MPAS Voronoi mesh with 10,242 hexagonal cells wrapping Earth, rainforest emission cells highlighted in green

Our simulations run at 240 km resolution — each grid cell covers roughly the area of West Virginia. Individual thunderstorms cannot be resolved at this scale. About 90% of tropical rainfall is handled by a convective parameterization (Kain-Fritsch), not the explicit microphysics where our salt code operates. Going to 120 km helps only marginally: it is still fully parameterized. True convection-permitting resolution is 30 km or finer — well beyond what a laptop can do for ensembles.

We also have only N = 1 ensemble member per experiment. The atmosphere is chaotic; weather variability at 30 days is large (~2.67 K mean global divergence between our CONTROL and NO-SALT runs). We cannot yet say whether any specific transport number is a real forced signal or one draw from a wide noise distribution. The range of values we see across different microphysical implementations (−153 to +153 TW at 30°N) is consistent with either interpretation and most likely reflects both — salt couples sensitively to microphysical implementation AND to weather regime.

What we can trust
The sensitivity of the global response to salt microphysics, plus a baseline-CCN regime dependence confirmed in Phase 7: a swing of ~364 TW across nine implementations, and an Amazon rainfall sign flip (negative under polluted baselines, positive under pristine Pöhlker-anchored baselines). Salt is a first-order variable whose sign depends on the atmospheric state it acts in. That finding does not require any specific number to be correct.
What we cannot yet trust
Any specific number in our tables. Our N = 1 experiments and 240 km parameterized convection mean our results are illustrative, not definitive. Ensembles and convection-permitting resolution are needed to make definitive statements.
What the community needs to do
(1) Paired ensembles at 240 km to test variance amplification (doable on a consumer laptop). (2) 30 km convection-permitting paired runs (requires HPC). (3) Different microphysics schemes. Our open-sourced code and Docker containers make all of these reproducible.

The strongest claim we can defend from our current data: biogenic salt from rainforests is a sensitive enough variable that mainstream models not explicitly representing this specific warm-rain mechanism are probably missing a meaningful process. The specific sign and magnitude of its effect on polar heat transport, and whether existing CCN/hygroscopicity parameterizations partially capture it, require community replication and expert review to determine.

An invitation

This hypothesis belongs to everyone.

We are sharing the code, configuration files, analysis scripts, Docker build recipes, and this entire paper. Raw simulation output (~200 GB of NetCDF) is available on request and planned for a public data archive. We believe this question is too important to sit behind a paywall or wait for a grant cycle.

HPC-enabled climate modelers (highest priority)
Run our GCCN code at 30 km or finer (convection-permitting). A single paired run would settle whether the 240 km Hadley intensification is real physics or parameterization artifact. Tell us what your MPAS or WRF setup can do — we'll help port.
Cluster or cloud-compute access
Run larger ensembles than we can (20+ members), or longer integrations (90+ days). These directly test the variance-amplification hypothesis and can be done at 240 km for much less compute than convection-permitting runs.
Cloud microphysicists
Review the GCCN lifecycle code in our repo. Try the same hypothesis with Morrison or P3 microphysics — which parts of the response are Thompson-specific?
Tropical ecologists
Which tree species emit the most salt? What are the emission rates by canopy density, soil moisture, diurnal cycle? Field measurements inform all future modeling.
Citizen scientists
Download our Docker container and reproduce the April 240 km experiment. Try different seasons. Three days of laptop time produces a publishable replication. AI can help you set it up.
Policy makers and conservation organizations
The magnitude of sensitivity we observe is sufficient reason to ensure rainforests are not destroyed faster than we can understand what we lose. Our results do not yet determine the sign of the polar effect — but neither does "it is probably zero" look defensible anymore.
The bigger picture

When budgets are slashed, citizens step up.

The tools that made this research possible — WRF, MPAS, GFS data, satellite observations — represent decades of investment by the US government, NCAR, NOAA, DOE, NASA, and academic institutions worldwide. They were built to advance science. They are free because taxpayers funded them.

Consumer AI (Claude, ChatGPT, Gemini) made these tools accessible to non-specialists. A rocket scientist with a laptop can now ask questions that previously required a team of meteorologists and a supercomputer. The accuracy needs professional validation, but the questions can come from anywhere.

With research budgets under pressure, this model — government tools + consumer AI + citizen scientists + professional oversight — may be the future of climate research. Not replacing institutional science, but accelerating it with a thousand curious minds running experiments that no single lab has the bandwidth to attempt.

Biogenic salt from rainforests appears to matter more than mainstream models explicitly capture through a dedicated warm-rain mechanism. The specific effect on polar ice is still being determined. That uncertainty is not a reason to wait — it is a reason to ensure forests are not destroyed faster than we can understand what we're losing.

Polar bear on Arctic ice connected by atmospheric flow to Amazon rainforest
For citizen scientists

You can run this experiment today.

1
Get Docker Desktop
Free download. Runs on Windows, Mac, or Linux. This is how we containerize the entire atmospheric model — no complex library installations.
2
Build the container
Our repo has the Dockerfile for MPAS v8.3.1 with the full GCCN tracer modification. docker build sets it up. Full step-by-step in reproducibility/REPRODUCE.md.
3
Ask AI to help
Use Claude, ChatGPT, or Gemini to modify the experiment. Try different seasons, different emission rates, different regions. The AI can write the namelist changes and analysis scripts.
4
Share your results
Post your findings. Challenge our conclusions. Improve our physics. Science advances fastest when everyone can participate.

Minimum hardware: Any computer with 16+ GB RAM, 4+ cores, and 100 GB free disk space. Runs on consumer hardware — no GPU required.