CO₂ Interpolation Tool

The CO₂ Interpolation Tool generates a spatial concentration map by fitting a Gaussian process model to the active CO₂ sensor network. The result is visualized as either a smooth heatmap or contour polygons on the Cesium globe.

Prerequisites

• CO₂ Sensors must be enabled and loaded before running interpolation (see CO₂ Sensors)
• At least 3 sensors must pass the validity checks

Opening the Tool Panel

Enable CO₂ Sensors from the sidebar. A CO₂ Interpolation Tool panel appears as a floating panel above the bounding box area. It contains:

• Run Interpolation button
• Smooth / Contours mode toggle
• Opacity slider (0–0.95)
• Stats area (populated after a successful run)
• Target Mode button

Sensor Validity Checks

When Run Interpolation is clicked, the app screens each sensor:

Check	Criterion
Has coordinates	lat and lon (or long) must be present and numeric
Valid CO₂ range	CO₂ must be between 0 and 1500 ppm
Within Central FL bounds	Latitude: 27.5–29.5 °N, Longitude: 80.5–82.5 °W

Sensors that fail are excluded. A rejection popup lists each excluded sensor and the reason. If fewer than 3 sensors pass, interpolation is aborted with an error notification.

The Interpolation Request

Valid sensors are sent to the UCF JetStream Cloud backend:

POST https://ucf-urbangeolens-backend.cis251126.projects.jetstream-cloud.org/co2/interpolate-co2
Content-Type: application/json

{
  "coordinates": [[lat1, lon1], [lat2, lon2], ...],
  "values": [co2_1, co2_2, ...]
}

Response

{
  "status": "success",
  "grid_points": [[lat, lon], ...],
  "predictions": [co2_value, ...],
  "contours": { GeoJSON FeatureCollection },
  "l": 0.042,
  "area_m2": 1234567.8,
  "bbox_area_m2": 2000000.0
}

Field	Description
grid_points	Regular grid of coordinates covering the sensor convex hull
predictions	Predicted CO₂ value at each grid point
contours	GeoJSON polygons representing iso-concentration contours
l	Gaussian process length scale parameter (km)
area_m2	Area of the convex hull of input sensors (m²)
bbox_area_m2	Area of the bounding box around the sensors (m²)

Visualization Modes

Smooth Heatmap

A textured rectangle entity is rendered on the Cesium globe, sized to the sensor bounding box. The texture is generated by:

1. Mapping each prediction value to an RGBA color (blue → green → yellow → red)
2. Painting the colors onto a <canvas> as a grid
3. Converting the canvas to a Cesium.Material using MaterialProperty

Contour Mode

GeoJSON contour polygons from the API response are rendered as Cesium polygon entities. Each contour is filled with a semi-transparent color matching the CO₂ concentration range it represents.

Toggle between modes using the Smooth / Contours buttons in the tool panel.

Opacity Control

The opacity slider adjusts the transparency of both the heatmap and contour layers in real time (range 0–0.95).

Statistics Panel

After a successful run, animated stat cards display:

Stat	Description
Mean	Average predicted CO₂ across the grid (ppm)
Median	Median predicted CO₂ (ppm)
Std Dev	Standard deviation of predictions
Min / Max	Minimum and maximum predicted values
Length Scale (l)	Gaussian process characteristic length scale — represents the spatial correlation distance in km
Hull Area	Area of the convex hull of input sensors (km²)
Bbox Area	Area of the bounding box around sensors (km²)

Copy buttons allow copying each stat to the clipboard.

Target Query Mode

Click Target Mode to enable an on-map query cursor. Click any point on the globe to estimate the CO₂ concentration at that location using the fitted model. A pin is placed at the clicked location and the interpolated value is displayed.

Click Target Mode again to disable it.

Color Scale

The CO₂ color mapping runs from blue (low) through green and yellow to red (high), scaled to the min/max of the current prediction set.

Limitations

• Interpolation is limited to the Central Florida region (sensor deployment area)
• The JetStream Cloud backend must be reachable from the server
• Results are a statistical estimate, not a direct measurement — sensor density and placement affect accuracy
• The Gaussian process assumes spatial stationarity; it works best when sensors are distributed across the area of interest