A 3D Tiles dataset consists of multiple nodes (data groups) organized into a 3D spatial tree structure. The root node of this tree defines the top level entry point for visualizing the tile dataset. When zooming in, the root node is checked to see if it needs to split to increase the level of detail shown on screen. This check is delicate: split a node too late and you see low quality data on screen. But split a node too soon and you spend too much time loading data that has no visual impact. It is a classic speed versus quality problem.

The parameter that controls the splitting of a node is called the “geometric error” and each node in the tree has a value for it. If the geometric error is large the node splits early. The most detailled nodes (leafs) of the tree have a very small geometric error, the root node has the largest one.

Each vender of software platforms that create 3D Tile datasets has their own way to calculate the geometric error. Some have rather large values, some rather low. To adjust for this variation of values the 3D Tile renderer can scale the geometric error. Finding the right scale factor requires some experience. One way to avoid loading too much data is starting with a low scale factor that avoids splitting and increasing it to add more quality.