Delphi was built in a climate war room, and geospatial data is its native language. Maps are a first-class visualization type in every command center, and the agent has the tooling to compose sophisticated map layers directly from the data you already have — no GIS pipeline required.
Map layer kinds
A Delphi map isn’t a single visualization — it’s a stack of layers, each with its own kind and style. The full set of shipped layer kinds includes:
- geojson — shaded regions from a boundary source (choropleth). Good for “which state has the most of X.”
- scatter — points on the map, sized or coloured by a field. Good for events, incidents, or stations.
- heatmap — density surface from point data. Good for “where is the concentration.”
- hexagon / grid / screengrid — binned aggregations for dense point data.
- arc / line / great-circle — origin-to-destination flows between pairs of points.
- contour — isolines for continuous surfaces like elevation or concentration.
- path / trips — continuous routes, optionally animated over time.
- column — extruded 3D bars anchored to geographic points.
- text / icon — labels and markers for curated overlays.
- tile / mvt / tile-3d / wms — raster, vector, 3D tile, and WMS basemaps from external sources.
- point-cloud / scenegraph — 3D point clouds and anchored 3D objects for terrain and scene layers.
Most maps use one or two kinds; complex decision surfaces might layer four or five. Delphi picks the layer kinds based on the shape of your data and your stated goal.
Boundary overlays
For choropleth and region-based views, Delphi ships with locally-hosted GeoJSON boundary sets so you don’t have to hunt down and validate public feeds for every decision. The boundaries that ship today cover:
- Ethiopia — first and second-level administrative divisions.
- Kenya — first and second-level administrative divisions.
- United Kingdom — countries and administrative regions.
- Canada — provinces and territories.
- Australia — states and territories.
- United States — states, counties, and congressional districts.
- India — states and union territories.
- Antigua and Barbuda — parishes.
- World — country-level boundaries for global choropleths.
Local boundaries matter for sovereign and regulated deployments — customers in those contexts often can’t reach out to public feeds, and they need the boundary files under the same residency regime as their operational data. See Data residency for the governance side of that story.
For geographies outside this set, Delphi can join to any public GeoJSON URL, ArcGIS feature service, MVT tile endpoint, or WMS service — you just need a public URL and a join key. The agent can also pull boundary metadata from a curated registry of vetted sources before composing the map, so you don’t have to hand-type URLs.
Building a map via chat
The fastest way to get a map is to ask. The agent picks the basemap, layer kinds, theme, and camera based on what you describe:
Show me a choropleth of active FEMA disaster declarations by US state over the last 12 months.
Add a heatmap of earthquake events over M4.5 in the Pacific Northwest for the last 30 days.
Build a map showing vessel traffic through the English Channel this week — point layer with ship direction arrows.
Compare reservoir capacity across California over the water year as a graduated circle map, with this year and last year side by side.
Delphi parses your request into a goal (distribution, density, flow, routes, labels, or comparison), a geographic granularity (global through city), an audience (executive, analyst, public), and optional styling preferences — basemap (dark, light, satellite-lite, terrain-lite), theme (editorial, atlas, analytic), and view type (map, globe, orbit, orthographic). You can override any of those in the prompt if you have opinions, or let the agent pick.
Connecting operational data to a map
Any dataset with spatial fields becomes map-ready automatically. Delphi reads latitude and longitude for point observations, target_latitude and target_longitude for arc origin/destination pairs, and Data Commons place IDs for join-by-region views. If your dataset has a state name or country code and you want a choropleth, the agent will join it to the right boundary source.
The work happens in the chat tool layer: the agent knows what boundary sources are available, what layer kinds fit your goal, and how to style the result for readability at the target audience level. You describe the outcome; Delphi picks the layers.
Located stats as their own layer
Sometimes the data you want on a map isn’t a connector or a curated visualization — it’s just a set of points that each carry a few measurements. Soil samples across a farm. Water-quality readings across a river system. Vendor sites across a region. Delphi treats these as geo-data layers: ask the agent to put the readings on the map and they become a first-class layer with their own legend, tooltip, and metric picker. Multiple readings at the same point (a soil-test sensor and a soil-moisture sensor on the same plot, say) merge into a single point with all the metrics available behind a switcher, including first-to-last trend deltas where the data supports it. The layer turns on and off from the map’s layer panel like any other.
Weather overlays
Switchable weather layers — precipitation, temperature, clouds, and wind — sit alongside your operational layers on the same map. Flip one on from the layer panel and current conditions render on top of the existing view, with a colour scale legend so the numbers are readable at a glance. The weather feed is live, so the overlay reflects current conditions every time the map renders.
Rewinding the map
When you rewind a command center to an earlier moment (any view that exposes a time control — the dashboard’s context bar or a scenario replay), the map rewinds with it. Satellite imagery anchors to the growing season completed on or before the rewound date. Data-backed map values resolve to their state at that moment rather than their current state. Weather overlays switch to historical tiles for the rewound date. You see the map in its rewound state alongside the headline KPIs and charts in the same historical view.
Legends on raster layers
The raster overlays — NDVI, carbon, land cover, the weather layers — render their own legends in the map’s legend card so a reader can read the colour scale without leaving the canvas. Continuous overlays carry a gradient with units; categorical ones (like land cover) carry labelled swatches. The legend follows the focused layer, so swapping which raster is on top swaps the legend with it.
When maps aren’t the right answer
Maps are powerful but not universal. If your question is “how has X changed over time” and geography isn’t part of the answer, a line chart will land better. If you want a single number for a region, a KPI card with a filter is usually clearer than a map with one labeled cell. Reach for a map when the spatial distribution is the story.