Run ONNX wind-field prediction end-to-end. Computes SDF, building height, Zrelative, U/Uref, direction features from geometry, assembles the 8-channel input tensor, runs ONNX inference, and outputs predicted wind speeds. Supports legacy 1ch (U), 2ch (U + k) and new 4ch (U + k + Uroof + kroof) models. ML screening prediction, not CFD: use it to compare options, then confirm anything that matters with a Wind Study (planning authorities do not accept surrogate results). Check a result with Prediction Confidence and Verify Prediction. Method: Kastner et al. (2026), SSRN preprint 6401886, doi:10.2139/ssrn.6401886.
Input
Name
Nickname
Description
Type
Points
List of 3D points. Each point's Z must be absolute elevation (m).
Point
Buildings
List of Brep or Mesh objects representing buildings.
Geometry
Model
Full file path to the ONNX model. Connect the FilePath output from the ML Model component.
Text
Boundary Conditions
BC
Boundary conditions from the ABL or Uniform Flow component. Uref, zRef, z0, wind directions, and EPW path are extracted automatically.
Generic Data
Pedestrian Level
PedLevel
Pedestrian mount height (m). Default = 1.8
Number
Filter Margin
FiltMargin
Margin (in meters) to mask out from the outer perimeter of the prediction plane due to unstable boundary effects. Default = 100.0
Number
Palette
Color palette for visualization.
Text
Legend Domain
Domain
Optional custom domain [min, max] to lock the color bounds. If empty, the colors scale dynamically to the data.
Domain
Interpolate
Interp
Visualization style. Flat (pixelated) vs Smooth (interpolated colors).
Text
Field
Field to visualize. Affects M, LM, LP, LV outputs.
Text
Level
Visualization level. Roof Level only applies when a 4-channel model is loaded.
Text
Selection Points
SelectPts
Optional locations to output. The model still runs on all input Points; each selection receives the result of its nearest input sensor, in selection order.
Point
Radius
Marker size (m): circle radius or square half-width.
Number
Legend Scale
LegScale
Uniform scale factor for the legend bar and label spacing.
Number
Mesh Type
MeshType
Prediction marker topology. Square uses one quad per point and is fastest; Circle uses a 16-sided n-gon.
Text
Engine
Eng
Who measures the SDF and building-height grid the model reads (not the inference provider — that is the right-click menu). Auto (the default): the OpenCL GPU when every building is a mesh or a Brep with only planar faces, where it measures exactly what Rhino does; the CPU otherwise (curved faces, or no GPU), and the Remark says which. GPU: the OpenCL GPU on every building, curved ones on their render mesh; a fallback to the CPU is reported, never silent. CPU: Rhino's Brep rule on every core, the one the model was trained on.
Text
Run
Sticky on/off switch for the prediction (a toggle, not a momentary button). On: every input change re-runs inference. Off: inference is skipped and the last result stays on the outputs, so the canvas can be rewired without paying for a solve. A wired boolean works the same as the inline switch.
Boolean
Output
Name
Nickname
Description
Type
Points
Pts
The location of each predicted value, in Values order, at the input points' own height: the input Points, or the Selection Points when those are wired.
Point
Values
V
Predicted field values at each valid input point. Outputs wind speed (m/s) or turbulent kinetic energy (m²/s²) depending on the Field input. Branches represent different wind directions.
Number
Grid Mesh
M
A fast-rendering contiguous coloured preview mesh of the predictions.
Mesh
Legend Mesh
LM
A colored mesh strip acting as a visual legend.
Mesh
Legend Points
LP
Locations for the legend text in the 3D viewport (Generic type to prevent red cross preview).
Generic Data
Legend Values
LV
Text values corresponding to the generated legend.
Predicted roof-level field values at each valid input point. Outputs wind speed or TKE depending on the Field input. Branches represent different wind directions. Empty unless a 4-channel model is used.