Glossary

A cave survey and the web page that shows it come from two worlds that speak different languages: centrelines, length and cave registers on one side; meshes, textures and WebGL on the other. Here are the words of both, explained for what they are. The first ones describe cave surveying in general, not what this application does.

The survey

Cave survey

The systematic measurement of a cave and its graphical rendering. It is the document that records where a cave runs, how long it is, how far it descends and what shape its chambers have: the basis of every study, every later exploration and every rescue operation.

Centreline

The broken line joining the survey stations one after another, and the backbone of a traditional survey. Each leg carries a distance, a bearing and an inclination: out of those three numbers, repeated over hundreds of shots, the shape of the cave emerges.

Survey station

A fixed point in the cave from which the next measurement is taken, often marked with a tag or a spot of paint so that it can be found again. The quality of a survey depends on how well the stations are chosen and on how findable they still are years later.

Shot (leg)

A single measurement between two stations: distance, direction and inclination. Shots are the bricks a survey is built from, and their number gives the measure of the work behind a documented cave.

Length

The total extent of a cave. Surveyed length adds up the legs as they lie in space; horizontal length adds them up after projecting them onto the horizontal plane. The same cave has two different numbers, and confusing them is the commonest mistake when caves are compared.

Vertical range

The difference in elevation between the highest and the lowest point of a cave. For a deep system it is the defining number, and it should always be read together with the length: a long cave is not necessarily a deep one.

Closure error

When a survey loop returns to a station already measured, it rarely lands exactly on it: the gap between the calculated point and the real one is the closure error. It is the most honest measure of a survey’s accuracy, and it is distributed around the loop to make the figures agree.

Georeferencing

Placing a survey on the surface of the Earth, usually by anchoring the entrance to a GPS-measured coordinate. Without it a survey describes a shape; with it, a shape in a place — and it becomes comparable with the geology, the hydrology and the other caves around it.

WGS84

The reference system used by GPS receivers and by almost every online map. It is what the latitude and longitude that place a cave on the map refer to.

Geographic coordinates

The pair of numbers — latitude and longitude — that pins a point on the Earth’s surface. In the 3D Cave Viewer catalogue they are the two fields that make a cave appear on the map, in the right place.

Laser rangefinder

The instrument that measures a distance with a laser pulse. The models used in caving also carry a compass and a clinometer, returning all three values of a shot at once and cutting down the time spent writing numbers in the cold.

Plan and elevation

The two projections a survey has traditionally been drawn as: the plan looks at the cave from above, the elevation cuts it along its length. They are still the standard language of caving documentation, and they are the limitation out of which the idea of a three-dimensional model grows.

Cave

Any natural or artificial underground void. The technical usage is broader than the everyday one: it takes in swallets, shafts, crawls and man-made underground spaces as well.

Cave statistics

The set of figures describing a cave: length, vertical range, entrance altitude, number of entrances. They are a cave’s identity card, the ones that appear in registers and publications.

Cave register

The record, usually regional, that gives every known cave a unique number and collects its position, its statistics and its survey. It is the reference by which a cave can be cited without ambiguity.

From the real cave to the model

LiDAR

A technique that measures distances with laser pulses and returns thousands of points per second, reconstructing the shape of a space. Since it arrived in some phones, surveying a chamber or a passage in three dimensions no longer needs professional equipment: that shift is what this project grew out of.

Photogrammetry

Reconstructing a three-dimensional object from many photographs taken from different positions. Underground it is demanding because of the lighting, but it needs no instrument beyond a camera.

Point cloud

The raw output of a scan: millions of points in space, each with its coordinates and often a colour. It describes a surface without yet defining one — between one point and the next there is nothing — and it has to become a mesh before it can be walked through.

Mesh

The three-dimensional shape proper: a net of triangles closing the surface between the points. It is what 3D Cave Viewer reads and displays, and what collision detection stops you from walking through.

Vertex

One of the corners of a triangle in the mesh, defined by its three coordinates. The vertex count is the measure of a model’s weight, and it is what the optimisation tools try to bring down without changing the shape.

Polygon

A single face of the mesh, almost always a triangle. A survey of a medium-sized cave can hold several million: how many a device can carry is the practical constraint that decides whether a model can be published as it stands.

Texture

The image wrapped onto the mesh that gives it the colour and the substance of rock. It often weighs more than the geometry itself, and it is the first thing to compress when a model has to be made light.

Material file (MTL)

The file that accompanies an OBJ model and says which images clothe which parts of the surface. It is why an OBJ is never a single file: without its MTL files and its textures it arrives grey.

CloudCompare

A widely used free program for working with point clouds and meshes: aligning scans, cleaning them up, reducing their weight, exporting them. It is often the last step before preparing a model for the web.

Up axis (Z-up and Y-up)

The convention that decides which of the three axes points upwards. Technical software and CAD packages usually use Z; the glTF format and web 3D engines use Y. It is why a perfectly good model can arrive lying on its side — and it is straightened by declaring a rotation, without reprocessing the file.

Model scale

The ratio between the file’s units and real-world measurements. Get it right and a distance measured in the viewer matches the one underground, and 1:1 scale in virtual reality gives back true proportions; get it wrong and everything stays consistent but no number means anything.

GLB

The recommended format for publishing a model: it gathers geometry, materials and textures into a single binary file. One file to upload, and no assets that get lost along the way.

glTF

The open standard for three-dimensional models on the web, designed to be transmitted and displayed quickly. GLB is its single-file packaging; in glTF form the textures stay as separate files.

OBJ

An old and universal format, read by practically every program. It compresses nothing and keeps textures outside, so for the web it is the least efficient of the three: fine as a starting point, less so as a publishing format.

Draco compression

A technique that sharply reduces the weight of a glTF model’s geometry, often by an order of magnitude, at the cost of a brief decompression on opening. It is the step that makes a survey downloadable over a mobile connection when it otherwise would not be.

Decimation

Reducing the number of triangles in a mesh while keeping its shape recognisable. It is needed when a survey is more detailed than a device can carry: you lose fine relief and gain the ability to look at it at all.

Looking at it in the browser

WebGL

The technology that lets a web page draw three-dimensional graphics using the graphics card, with no plugin. It is why a survey can be explored inside the browser rather than inside a program you have to install.

Three.js

The library the viewer is built on: it handles the scene, the lights, the camera and the materials, resting on WebGL underneath. It is the layer between the model file and what appears on screen.

WebXR

The standard that lets a web page talk to a virtual reality headset: head position, controllers, tracking. It is what makes it possible to enter a cave without installing any dedicated application.

Virtual reality

The mode in which you enter the cave at 1:1 scale with a headset: you move with the sticks, turn in 45-degree steps and reach a distant point with a teleport. The proportions are the real ones, because the model is a survey and not a stage set.

Progressive Web App

A site that behaves like an application: it installs from the browser, sits among the other apps and keeps working without a network. It goes through no store and takes up nothing like the space of a real program.

Offline use

Being able to reopen a cave you have already visited when there is no network, because the browser has kept a copy. In the mountains, where signal is often missing, it is the difference between having the survey and not having it.

Orbit mode

The classic way of looking at a model: you stay outside and turn it, zoom it, move it, as you would with a physical model on a table. It is the view you need to grasp the overall run of a cave.

First person (POV)

The mode in which the camera sits inside the cave at a person’s height and you move with the keyboard or a virtual joystick. It is the one that conveys the real size of the chambers, which is badly judged from outside.

Collision detection

The check that stops the camera from passing through the walls as you move inside the model. Without it you would walk through rock and the sense of the space would be lost.

BVH

A tree structure that groups the triangles of a mesh into ever smaller volumes, so that they need not all be tested at every step. It is what keeps navigation and collisions fluid even on surveys of millions of polygons.

Clipping plane

A plane that virtually cuts the model along one of the three axes and hides what lies on one side, showing the internal structure of the cave as in a tomography. It is moved with a slider and can be flipped.

Orthogonal view

A view aligned to one of the axes — from above, from below, from the front, from the side — without perspective. It is how a plan or an elevation is read, and it lets a model be compared with the traditional drawing.

Scale bar

The ruler overlaid on the view that says how much a given length on screen is worth. It adapts to the zoom level, so there is always a metric reference even inside a small chamber.

Vertex snapping

The automatic pull of a measurement point onto the nearest vertex of the mesh. It takes the imprecision out of the hand and makes two measurements taken on the same edge give the same number.

Comfort vignette

The gradual darkening of the edges of vision during faster movements in virtual reality. By narrowing the peripheral field it softens the conflict between what the eyes see and what the body feels, which is what causes headset motion sickness.

Teleport

Moving instantly to a point aimed at with the controller, rather than travelling the distance. In virtual reality it is the least tiring way of crossing a long space, and the one that causes the least discomfort.