Ray-marches a scalar field baked once into a 3-D texture.

Collaboration diagram for Scalar field volume:

Visualization SimObject showing a whole region of a scalar field at once.

The field is evaluated onto a regular grid filling a NED axis-aligned box once, in RenderInit, and uploaded as a 3-D texture. A box proxy is rasterised, and each fragment marches one ray through the texture, compositing colour and opacity front to back, so the region reads as a semi-transparent body of water rather than as a surface. Nothing is evaluated per frame, so the cost is independent of Resolution once baked.

Opacity is corrected for step length against ReferenceLength, a fixed world-space length, so AlphaRange means the same thing here as on VisualScalarPlane and the apparent density does not change with Resolution or with view direction.

Only a time-invariant scalar field can be baked. A field that varies with time is refused unless VisualiseAsConstant is set.

Occlusion, and two accepted residuals

A ray-march is a single fragment, so the whole ray is depth-tested at the depth of whichever box face was rasterised. The front face is rasterised, so the ray is tested where it enters the box, and the seafloor mesh - drawn opaque, with depth writes, ahead of any transparent geometry - hides the volume behind it. Two artefacts follow from the one depth value, are accepted, and are not solvable without depth-buffer sampling, which is out of scope:

A ray passing through the seabed still accumulates below-seabed samples and blends them over the seafloor. For a ParametricScalarField with maskBelowSeabed those samples are masked and contribute nothing, but ConstantScalarField and DepthLayeredScalarField know nothing about the seabed, so for those the artefact is unmitigated and a below-seabed column tints the seafloor.
While the camera is inside the box the front faces are clipped by the near plane, so the back face is rasterised instead and the fragment is tested at the far box face. An opaque object between the camera and that far face - the seafloor above all
- then erases the volume across its whole footprint instead of occluding only the part behind it. The max(nearIntersection, 0) clamp in the shader does not remove this: it moves where the ray starts, not the depth the fragment is tested at.

Example Configuration (XML)

<Lib LibName="fhsim_environment" SimObject="VisualScalarVolume" Name="TemperatureVolume">
   <Parameter name="FieldName">Temperature</Parameter>
   <Parameter name="BoxCentre">0, 0, 100</Parameter>
   <Parameter name="BoxSize">2000, 2000, 200</Parameter>
</Lib>

Input ports

None.

Output ports

None.

Configuration Parameters

Name	Width	Description
FieldName	1	Name of the scalar field to show, as registered by the Environment. (Required)
BoxCentre	3	Centre of the sampled box in the NED frame [x, y, z] [m]. Ignored when CentreOnCamera is set: the first bake then centres on the camera. (Default: 0, 0, 100)
BoxSize	3	Extent of the sampled box along each NED axis [m]. (Default: 2000, 2000, 200)
Spacing	3	Grid spacing of the bake along each NED axis [x, y, z] [m] - the NED frame, not the local [u, v] frame VisualScalarPlane uses for the same parameter. Mutually exclusive with Resolution; configuring both is an error. A spacing coarser than BoxSize on an axis still gives at least two samples there, because the bake is always a 3-D texture. (Default: absent, a spacing near 4194304 cells is derived)
Resolution	3	Samples of the bake along each NED axis [x, y, z] - the NED frame, not the local [u, v] frame VisualScalarPlane uses for the same parameter. Mutually exclusive with Spacing; configuring both is an error. The third entry is raised to 2 when it is 1, because the bake is always a 3-D texture. (Default: absent, a spacing near 4194304 cells is derived)
ReferenceLength	1	Path length over which AlphaRange is the opacity [m]. A fixed world-space length, so halving or doubling Resolution leaves the apparent density unchanged. It combines with AlphaRange over the whole path: a ray crossing L metres of visible field at per-ReferenceLength opacity alpha emerges at 1 - (1 - alpha)^(L / ReferenceLength). Worked example: AlphaRange 0.02, 0.35 with ReferenceLength 10 leaves the faint edges nearly clear while the densest core reaches about 1 - 0.65^6 = 0.9 opacity through a 60 m core. The default of 1 m saturates any ray crossing more than a few metres of visible field - at the default AlphaRange 0.3, 0.6 a 15 m path is already opaque and a 200 m box is solid - so ReferenceLength should normally be set to the scale of the structure being shown, as the shipped examples do with 10 and 20. (Default: 1)
CentreOnCamera	1	Move the box centre to follow the camera, re-baking on a worker thread once it has moved far enough. The encoding range is fixed at the first bake and never revisited, so values at a later box position that fall outside it are silently clamped; a warning naming ColorRange and VisibleRange is logged when a re-bake clamps more than a few percent of its cells, and pinning those two parameters is the cure. (Default: false)
RebakeFraction	1	Fraction of BoxSize the camera must move on any axis before a CentreOnCamera re-bake starts. (Default: 0.25)
VisibleRange	2	Value span drawn at all [min, max]; values outside it are fully transparent. (Default: probed from the field and widened by one encoding code)
ColorRange	2	Value span mapped across the full colour ramp [min, max]. (Default: probed from the field)
AlphaRange	2	Opacity span applied across the visible range [min, max], stated over a path of ReferenceLength and clamped to [0, 1]: the lowest visible value gets min and the highest max. A ray accumulates it over its whole path as 1 - (1 - alpha)^(L / ReferenceLength), so it is the pair AlphaRange and ReferenceLength that decides how dense the result looks, never AlphaRange alone. It means the same thing on VisualScalarPlane, where a single fragment is one sample and no path correction applies. (Default: 0.3, 0.6)
ShowExtrapolated	1	Draw values the field flagged as inaccurate instead of masking them. Leaving it false has two surprising consequences. A DepthLayeredScalarField reports OUT_OF_RANGE_INACCURATE at every depth outside its profile, so a box below the deepest layer renders nothing. A ParametricScalarField returns the same status both for a position below the seabed when maskBelowSeabed is set and for a DepthProfile background clamped outside its layers, and the two cannot be told apart, so such a field masks everything outside that profile and not only what is below the seabed. (Default: false)
VisualiseAsConstant	1	Bake a field that is not time-invariant anyway, showing it as it is at time zero. (Default: false)

This SimObject is referred to as VisualScalarVolume