Inspiration
I make video games for a living. Looking around at the world, I often feel overwhelmed by all the problems we face, and it seems like I just make entertainment. Well, what if I could use the power of games to teach people something in a fun way that might, just maybe, change their mind?
This is my attempt to apply what I know about game dev to climate change. Starting from the basics: let's teach how the climate works, and at the same time answer the question, what does +2 °C look like?
What it does
Transform Earth puts a miniature landscape at arm's length and fills the air above it with a live 3D simulation of wind, heat, humidity, cloud and rain. Nothing is scripted: every cloud exists because air rose, cooled and condensed.
You play with your hands alone. Sweep the wind, pinch up a mountain, warm or wet the ground, or blow into the microphone to make a breeze. Then turn the time dial from weather to seasons to years and watch forests climb the wet slopes, deserts spread in the rain shadow and rivers find the coast.
A guide (voiced by me) teaches the basics on a test island, then the world opens up: Hawai'i's Big Island, the Southern Alps, the Atacama, San Francisco's fog, the Western Ghats monsoon and the Grand Canyon. Finally, Mars: a sealed crater where you warm the ground, pour the first lake and grow a settlement.
How we built it
The heart of the game is Jos Stam's stable fluids, solved on the Quest's GPU in compute shaders over a 128 × 96 × 16 grid in real units. Each 30 Hz tick solves, roughly:
$$ \frac{\partial \mathbf{u}}{\partial t} = -(\mathbf{u}\cdot\nabla)\,\mathbf{u} - \frac{1}{\rho}\nabla p + b\,\hat{\mathbf{z}} - f\,\hat{\mathbf{z}}\times\mathbf{u} + \mathbf{F}_{\text{hands}}, \qquad \nabla\cdot\mathbf{u} = 0 $$
with buoyancy b from temperature and moisture, and Coriolis f = 2Ω sin φ. A multigrid pressure solve, simplified Kessler microphysics (condensation, latent heat, rain and snow), a sun that heats land faster than sea, soil moisture and rivers complete it. Your hands are a force in the same equation.
Each land tile keeps rolling annual rainfall P and biotemperature T, and is classified with Holdridge life zones (PET ratio = 58.93 T / P) into 12 readable display biomes. Change the climate and the biomes follow.
Our acceptance test was the sea breeze: with no scripting, the coastal wind had to turn onshore by late morning and fade at night. When it did, we knew the sim was honest.
Built in Unity 6.6 with Meta XR SDK 207 and hand tracking throughout. I worked with a team of Claude Code agents: builders wrote each module, and independent critics judged it from their own captures in the Editor, the XR Simulator and on the Quest. About 1,970 commits in six weeks. And 25 Million output tokens.
Challenges we ran into
- A climate on a mobile GPU. The solver, clouds, rain and over 12,000 terrain tiles fit in about 9 ms of GPU time at 72 fps. Its insane when you think about it.
- NaNs NaNs NaNs When all your core code is shaders, every hardware is a new possibility for a small error an early return or an uninitialised variable or unintended optimisation to blow the whole thing up .
- Clouds. Raymarched clouds looked good on desktop, but its brutal on the quest GPU. Low-poly mesh clouds built from the sim's own cloud field read better and cost less.
- Simulation times. On desktop, we can simulate years in minutes, but on quest it's much harder. We can't just load a random level and calculate what it's climate will be.
- Breath as input. Making the mic hear breath but not your voice or the game's own sound took three rounds.
Accomplishments that we're proud of
- Real, emergent weather on a standalone headset: sea breezes, rain shadows, monsoons and San Francisco fog through the Golden Gate, none of it scripted.
- Hands-first from end to end, with no controller needed anywhere.
- A journey from a test island to real places to another planet, taught through play rather than text.
What we learned
- Honesty is a design principle. Faked clouds always looked worse than letting the physics run.
- Measure on the device, early. Only the headset settles comfort and GPU cost.
- Small simulations can feel huge. A 128-cell grid is tiny next to a weather model, but at arm's length it reads as a living world.
What's next for Transform Earth
- We have a sandbox, now to build the game! A city builder on mars that responds to the weather.