threejs-rain-snow-and-wet-surfaces

Rain, Snow, and Wet Surfaces

Couple appearance to causes: one time source and wind field drive airborne precipitation, while one owner per receiver integrates deposited rain or snow. Visual particle count samples the weather; it never sets the deposited mass.

Use Three.js r185 WebGPURenderer, TSL, storage nodes, node materials, and the node post stack. Initialize the renderer before allocating compute or storage:

await renderer.init();

if (renderer.backend.isWebGPUBackend !== true) {
  throw new Error('This weather path requires the WebGPU backend.');
}

For cross-system precipitation or receiver state, declare its units, frame, time interval, authority/version, support, validity, and reset semantics; invoke $threejs-choose-skills when ownership spans skills.

Build sequence

1. Name the owners and units

Declare:

• one monotonically sampled time source and update interval in seconds;
• one air-velocity field in world metres per second, including its frame, support, cadence, and validity;
• temperature in kelvin and a named humidity convention when phase, melt, or evaporation depends on them;
• precipitation forcing as liquid/ice mass-area flux in kg m^-2 s^-1 over a physical receiver support;
• one receiver-state owner for each liquid or snow inventory;
• one owner for HDR presentation, tone mapping, and output conversion.

Treat cloud appearance and causal precipitation as separate branches. An appearance-only cloud may coordinate art direction. A causal cloud source publishes a mass flux or airborne inventory with a fall-delay/transport model; rain transports it to receivers on a later ordered stage.

This step is complete when every cause and persistent state has exactly one owner, every exchanged quantity has units and a frame, and every producer is sampled within its stated validity.

2. Select motion before allocating state

Use immutable seeds and analytic vertex motion when position is an exact function of seed, time, and integrated wind. Use recurrent GPU-resident state when turbulence, collisions, feedback, or path history affects the next state. Authored time-varying wind remains analytic only when its displacement integral is available; multiplying the current wind by total elapsed time makes all particles jump when wind changes.