Geo-Forest LiDAR Intelligence Demo

GroundLow vegetationMedium vegetationHigh vegetationBuildingWaterCanopy gap

Internal performance

86.49%Best internal mIoU-run OA

82.60%Best internal macro-F1

72.63%Best internal mean IoU

86.71%Robust model internal OA

Fresh external holdout

83.27%Overall accuracy

61.80%Macro-F1

50.98%Mean IoU

+10.35mIoU pp vs strong baseline

The holdout is geographically separate and tile-disjoint. Water support is only 0.064%, so absolute external water performance requires a more representative dataset.

Robust internal class IoU

Ground

76.49

Low veg.

43.91

Medium veg.

52.71

High veg.

94.53

Building

69.51

Water

97.86

External improvement by class

Class	Baseline	Robust	Delta
Ground	73.43	79.97	+6.54
Low veg.	21.23	27.70	+6.47
Medium veg.	47.28	56.46	+9.18
High veg.	81.93	88.47	+6.54
Building	17.80	50.50	+32.70
Water*	2.09	2.77	+0.68

Why companies buy it

Turn existing LiDAR into decision-ready geographic information without building an in-house AI team.

Reduce manual mapping: begin with an automatic classification instead of a blank layer.
Prioritize inspections: focus GIS and field teams on unusual or high-value areas.
Deliver faster: export classified LAZ, maps, grids and reports into existing workflows.
Reduce deployment risk: start with a bounded pilot and measurable acceptance criteria.

Start with a focused pilotOne area, one operational objective and a clear scale-up decision.

Land-cover classification