Geospatial intelligence and Geo AI

Lucenia provides geospatial intelligence capabilities unmatched by any other search engine. From GeoTIFF content extraction to spatial reprojection and spatially-aware image tiling, Lucenia brings together geospatial data processing and AI retrieval in a single platform.

Geospatial field types

Lucenia supports rich geospatial data types for indexing and querying location-aware content:

Field type	Description	Use case
geo_point	Latitude/longitude coordinates	Store and query point locations
geo_shape	Complex geometric shapes (polygons, lines, multi-polygons)	Boundary searches, region containment, spatial joins

GeoTIFF and raster imagery

Full geospatial GeoTIFF handling — coordinate reference system (CRS), geo-transform, band information, and internal tiling — is read by the image_tiling processor via its Apache SIS raster source, not by content_extract (which handles document text/images and reports image dimensions and MIME type, but not the geospatial referencing). Point image_tiling at a GeoTIFF and each georeferenced tile carries its own CRS and bounding box.

Spatial reprojection

The reproject ingest processor transforms geospatial coordinates between coordinate reference systems at ingest time — critical when data arrives from multiple sources in different projections and must be normalized to a common CRS before indexing.

{
  "reproject": {
    "field": "geometry",
    "source_crs": "EPSG:3857",
    "target_crs": "EPSG:4326"
  }
}

Parameter	Required	Default	Description
field	Yes	—	The field holding the geometry to reproject.
source_crs	Yes (ingest)	—	Source CRS (for example EPSG:3857).
target_crs	No	EPSG:4326	Target CRS.
tolerance	No	0.01	Densification tolerance for reprojected shapes.
shape_type	No	geo_shape	Output shape type.
ignore_missing	No	—	Skip documents where field is absent.

The search-side reprojection processor (for reprojecting query geometries) defaults both source_crs and target_crs to EPSG:4326.

Spatially-aware image tiling

The image_tiling processor splits large imagery (Cloud Optimized GeoTIFFs, aerial photography, satellite imagery) into a grid of tiles for multimodal search. Choose a profile preset or set the tiling parameters directly:

{
  "image_tiling": {
    "source_uri_field": "raster_uri",
    "profile": "geo_search"
  }
}

Parameter	Default	Description
profile	—	Preset bundling tile size/overlap/max tiles: rag (512/64/100), detail (256/32/400), overview (1024/128/25), geo_search (512/0/200).
tile_size	(from profile)	Tile edge in pixels.
overlap	(from profile)	Pixel overlap between adjacent tiles.
max_tiles	(from profile)	Cap on tiles produced per source.
tile_format	jpeg	Tile image format (jpeg or png).
tile_quality	0.85	JPEG quality.
block_types	["image","geo_raster"]	Which extracted block types to tile.
source_uri_field	—	Field holding the raster object URI.

Each tile chunk carries tile_id, tile_index, a pixel_window (x/y/width/height), image_data, and image_mime_type. For georeferenced rasters only, tiles also carry bbox (west/east/south/north) and crs — so a geographic-region search can retrieve exactly the tiles that cover it.

Geospatial queries

Lucenia supports a full range of geospatial queries:

Query type	Description
geo_bounding_box	Find documents within a rectangular geographic region
geo_distance	Find documents within a distance from a point
geo_polygon	Find documents within an arbitrary polygon
geo_shape	Find documents that intersect, contain, or are within a shape

Integration with AI retrieval

Geospatial capabilities are deeply integrated into the AI retrieval pipeline:

1. Content extraction: GeoTIFF files are ingested with full spatial metadata preservation
2. Image tiling: Large imagery is tiled with geographic bounding boxes for each tile
3. Embedding: Tiles and text chunks are embedded for vector search
4. Spatial filtering: Search results can be filtered by geographic region before or after vector search
5. Retrieval grounding: The retrieval grounding processor includes spatial context (coordinates, bounding boxes) in its output, enabling location-aware RAG applications

This end-to-end integration means you can build applications that answer questions like "Show me satellite imagery of flooding in the Gulf Coast from last week" — combining temporal metadata, geospatial filtering, and semantic search in a single query.