Raster and vector data: the foundation of geospatial AI

A map can show a road, a building and a temperature layer together, but those datasets represent the world in different ways. Roads are usually lines, buildings are often polygons and a temperature surface is commonly a grid of pixel values.

Geospatial AI needs to preserve those differences. LGM brings raster and vector services into a common research workflow while keeping the meaning, scale and provenance of each input visible.

Raster layers describe continuous surfaces

A raster divides an area into cells. Each cell stores one or more values, which might represent elevation, land-cover class, soil properties or an observation derived from imagery.

Resolution matters because it determines the scale of the information. A broad terrain layer can help screen a region but may be inadequate for assessing a small construction site. Resampling a coarse layer into smaller pixels does not create new observations.

LGM’s raster capabilities support the use of these surfaces in territorial analysis. A useful request specifies the area, the indicator and the statistic needed for the decision.

Vector layers describe geographic objects

Vector data represents points, lines and polygons with associated properties. A point might identify a transport stop, a line a street and a polygon a building or administrative district.

LGM uses vector services to retrieve and analyze geographic objects. Operations such as selecting objects inside a boundary or grouping them by area allow a model to move beyond a textual description of a place.

Object counts need interpretation. Building footprints do not directly measure occupancy, and a point labeled as a facility does not necessarily describe its capacity or operating status.

Why the storage and exchange formats differ

The Cloud Optimized GeoTIFF format organizes raster data so compatible software can access relevant portions using HTTP range requests. This helps workflows avoid downloading an entire large raster when only part of it is needed.

GeoParquet adds geospatial conventions to the Parquet format for vector datasets. It supports analytical data workflows that combine geometry with tabular attributes.

GeoJSON is a practical exchange format for geometries and properties. STAC describes and organizes spatiotemporal assets so software can discover relevant data.

These formats serve different roles. A catalog helps find an asset; a storage format organizes it; an exchange format delivers information to another application. LGM’s data architecture uses that distinction to connect discovery, analysis and delivery.

Combine layers around a specific question

Consider identifying candidate sites for a small market. Vector layers can describe streets, buildings and existing businesses. Raster or zonal indicators can add population context, terrain or environmental conditions.

The layers need compatible geography and a sensible analytical scale. A citywide indicator should not be treated as a parcel measurement. A dataset from one year should not silently stand in for current conditions.

Before ranking locations, the investigation should establish which inputs are available and which are only proxies for the desired measure.

Add private data without losing access boundaries

Organizations often hold the most decision-relevant layers themselves. Asset inventories, site plans and operating records can make a broad public-data investigation much more specific.

LGM Collections support uploaded and shared layers alongside public data. Its GeoAI deployments request only the layers available to the project account, including permitted public Overture and Australia datasets.

The next useful step is often a data inventory: what exists, what is missing and which sources can be used for the intended purpose. Explore LGM’s data foundation or regional geospatial data layers to plan that work.