eNose surface plot – manual

This app shows eNose measurements as 3D surfaces on a map, combined with the wind. The goal: not just “there was a peak”, but “there was a peak of this kind, from that direction, probably from that source”.

The idea

An eNose has four sensors (S1–S4) that each respond differently to substances in the air. A time series of the sum of those sensors tells you when something happened, but not where it came from or what kind of odour it was.

This app links every measurement to the wind at that moment. For each eNose it builds a wind rose: for every wind direction and wind speed it shows how strongly each sensor responded. Placed on the map, the plots of several eNoses point together towards the likely source – a form of cross-bearing.

Reading a surface plot

• Direction: a hill on the south-west side means: when the wind comes from the south-west, this eNose smells something. The source is therefore usually in that direction.
• Wind speed: a hill close to the centre (low wind) points to a nearby, low source where odour accumulates. A hill towards the outer ring points to a high stack or a more distant source that only arrives with more wind.
• Sensors and colours: each sensor has its own surface: S1 green, S2 red, S3 blue, S4 orange. Which sensor dominates is a hint about the type of odour.
• One scale: height and colour use one scale (in dB) for all sensors and all eNoses, so you can compare them directly. The current scale is shown under Display.

Getting started

1. Open the app. If you are not logged in you are sent to the Comon Invent login page automatically.
2. Project – click the field and type part of a name or a project ID (for example ICL or 582). Choose with the mouse, or with ↑/↓ and Enter. The app remembers your last project.
3. Virtuals – the eNose locations of the project. Click the field to open the list; use All / None or tick individual eNoses. Unticked eNoses disappear from the map immediately. For some projects (ICL, Coenhaven) a default set is ticked.
4. Period – type a start and end, for example 28-05-2026 08:00. Short forms work too: 28-5, 28.5 8, 28-05-26. A missing year becomes the current year, a missing time 00:00. The 📅 button opens a calendar. Default: the last 24 hours.
5. Press Load. The button is grey when everything for this period is already loaded; it becomes active again when you change the period or need data that is not loaded yet.

Wind model and wind source

The plots are only as good as the wind that is linked to the measurements.

Wind is linked to each measurement using the wind reading closest in time (at most 60 seconds apart). Measurements without wind are left out.

Aggregation per cell

Each wind rose is divided into cells: 10° of wind direction × 1 m/s of wind speed. All measurements in a cell are combined:

• Average – the typical response for that wind. A direction that occurs only now and then weighs as much as a common one. Best for finding where odour comes from.
• Sum – response × how often that wind occurred. Shows where most of the odour came from over the whole period, but common wind directions naturally come out higher.

Zero correction

An eNose that is not zeroed well has a baseline above or below 0, and that baseline can drift. With Baseline per eNose to 0 switched on, the app estimates the baseline per eNose and per sensor (20th percentile over a sliding 3-hour window) and subtracts it. A wrong zero or drift then no longer looks like a response; real peaks remain.

Complaints

Odour complaints of the project in the loaded period are shown as white circles with a red !. Hover a complaint for the time, odour, intensity and text (no personal data is shown). With the timeline open, only the complaints that were active in the time window are shown.

Which complaints you see depends on the rights of your account for the project (imports, app reports, DCMR).

Source analysis

Two separate switches, both off by default.

Source likelihood map

An automatic cross-bearing over all selected eNoses:

1. For every eNose the app determines when it clearly responded: a selected sensor more than 3 times its normal spread above its own median (robust z-score). This also works with a wrong zero.
2. For every wind direction it calculates how often the eNose responded when the wind came from that direction. Common wind directions therefore do not automatically weigh more, and “nothing smelled when the wind came from there” counts as evidence against that place.
3. Those directions are projected upwind onto the map. For every place the app averages what the eNoses within Reach saw in that direction (nearer eNoses count more). Where the raised directions of several eNoses meet, the map turns yellow → red → purple.

Places seen by fewer than 2 eNoses stay empty. Only what stands out from the typical level of the map is coloured.

Known sources

The sources registered for the project are shown as purple diamonds with their name. With the source likelihood map on, each source gets a score (the value of the map at its location) and the panel ranks them. Below the ranking the app tells you whether the strongest spot is at a known source, or how far it is from the nearest one – a spot far from every known source may be an unknown source.

Sensors and display

• Sensors – click the field to show or hide S1–S4. The source likelihood map only uses the selected sensors.
• Colour transition – from which part of the scale the colour becomes visible; lower values stay transparent.
• Height scale – how high the highest value is drawn (in metres on the map).
• Radius – size of the wind roses on the map; the outer ring is the maximum wind speed (12 m/s).

Timeline

Click ▸ Timeline at the bottom to play through the loaded period. The map then only shows the measurements in a window that ends at the chosen moment.

• ▶ / ❚❚ play and pause; at the end it starts again from the beginning.
• Slider – drag through time yourself. The window being shown is displayed next to it.
• Window – 15 minutes to 6 hours. Speed – 5 minutes to 3 hours of data per second.
• Height and colour keep the scale of the whole period, so frames can be compared. Empty cells are not filled in, so you only see what was actually measured in the window.
• Complaints and the source likelihood map follow the window as well.

Data reliability

Below the Load button a coloured dot summarises the data: the share of usable measurements (with sensor data and wind) compared with one measurement per minute. ● green from 90%, ● orange from 60%, ● red below. Hover the ⓘ to see per eNose how complete the data is, and which eNoses were unusable and why (no sensor data, no wind, loading failed).

Map controls

• + / − zoom; the compass resets the map to north.
• 2D / 3D switches between a top view and a tilted view; your own tilt is remembered.
• Tilt and rotate with the right mouse button, or Ctrl + drag.
• Hover an eNose dot for its name and number of measurements; hover a complaint or source for details.

A typical investigation

1. Choose the project and load a period around the event, preferably a few days for variety in wind direction.
2. Check the reliability line: are the important eNoses usable? If not, try another wind source.
3. Look at the surface plots: in which direction do the hills point, and which sensor dominates?
4. Switch on the source likelihood map and known sources: where do the eNoses meet, and which known source scores highest?
5. Open the timeline and play the event: when do the hills grow, does the hotspot move, and do complaints appear at the same time?
6. Switch sensors individually to see whether the responses are one odour or several.

Troubleshooting

Glossary