Lucas pixel art
Published on

Creating a Map Visualization Using React-map-gl and


I'll take any excuse to interact with mapbox - I love their tech, easy to use, intuitive, great results. I recently joined a challenge to create a Point of Interest UI to simulate a mission command for autonomous recon drones.

The result

Skipping straight to what I built for a moment - I'm pretty happy with how it turned out. There's a panel on the left that gives you all the POIs in a list, and the map with a route and POIs available to classify. The code is here - and to play with the map in your browser, go here (but load it, wait ~3m, and try again. It's on a render PaaS free tier and the API has to wake up.)

The usual pan/zoom functions are supported -

The drawer is linked to the map such that hovering on the drawer centers the UI, and hovering on the map autoscrolls into view and indicated in the list view which is currently active.

Finally, there is a little modal that allows an end user to classify POIs -

Building the UI

react-map-gl seemed like the de-facto implementation for the web, but the visualizations were a little lackluster. provides quite a few really fancy annotations, including 3d and animations. Later I learned custom animations would be a little more difficult than simple css.

My front-end was created with create-react-app. Much of the UI work was pretty straightforward React code. I skipped any css or state libraries in the interest of time. There were a few snippets to call out -

This is how to connect to react-map-gl - a bit hard to find but I think this is something that recently changed with a newer version of the libraries. It seems like the strategy around how to keep both libraries synched in terms of viewState has been changing a little, but what this does is automatically pass viewState down (it's not called out in directly, but it's in there in the spread params) such that whenever that changes via the Map, it'll thread into the overlay and the overlay will get the update as well.

 * A thing we have to do to connect react-map-gl to deck-gl (for the newer version of map-gl & mapbox)
 * Detail here:
 * This ensures and map-gl are synchronized behind the scenes - both respond to updates in
 * viewState in unison.
function DeckGLOverlay(
  props: MapboxOverlayProps & {
    interleaved?: boolean;
) {
  const overlay = useControl<MapboxOverlay>(() => new MapboxOverlay(props));
  return null;

I used a geojson layer to display three different types of elements - the polys, the lines, and the annotation texts. The upside here was the simplicity, the downside was that this layer ended up needing a kitchen sink of fields for those different types of geo elements.

// the geojson layer handles displaying everything - drone lines, pois, and text
  const geojson = new GeoJsonLayer({
    id: 'geojson-layer',
    data: [...flightpaths, ...poiList, ...namesFromPois(poiList)],
    pickable: true,
    stroked: false,
    filled: true,
    extruded: true,
    lineWidthScale: 20,
    lineWidthMinPixels: 2,
    getLineWidth: 1,
    getLineColor: () => [255, 140, 0],
    getTextColor: [255, 255, 255, 200],
    getTextSize: 22,
    getTextPixelOffset: () => [-20, 0],
    getTextAnchor: 'end',
    pointType: 'text',
    getFillColor: (d) => getFillColorFor(d, highlightedPoiId, hoveredPoiId, 215),
    getElevation: (d) => {
      if (d.geometry.type !== 'Polygon') return 30;
    onHover: (d: any) => {
      if (d.featureType !== 'polygons') return;
    onClick: (d: any) => {
      if (d.featureType !== 'polygons') return;

Building the API

I used my turbo setup from my notes app project, which I'm really becoming quite comfortable with. The API format exposed GeoJSON - I made the choice to stick to the spec 100% due to simplicity. No misunderstandings around the API payload if you can just look up preexisting detailed man pages. You can play with the API using my autogenerated swagger spec here (again if the free tier is sleeping, just give it a minute and try again)

Turbo config

I'm starting to really grok the turbo dependencies! This was the most elaborate turbo setup so far.

API generate comes first, which makes the swagger.json. Then Client generate can come next, which makes the autogenerated client that the FE can use to talk to the API nicely. Client build needs to happen before the rest of the builds, then generally everything else can happen.

  "$schema": "",
  "pipeline": {
    "generate": {
      "dependsOn": ["^generate"],
      "outputs": ["client/**", "public/**", "geodata/**"]
    "client#generate": {
      "dependsOn": ["api#generate"],
      "outputs": ["client/**"]
    "client#build": {
      "dependsOn": ["generate"],
      "outputs": ["client/**"]
    "build": {
      "outputs": ["build/**", "dist/**"],
      "dependsOn": ["generate", "^build"]
    "test": {
      "outputs": ["coverage/**"],
      "dependsOn": []
    "lint": {},
    "dev": {
      "dependsOn": ["generate", "api#build", "client#build"],
      "cache": false,
      "persistent": true
    "clean": {
      "cache": false


If I ever get back to Runstrike, this would be a great setup for the web UI!