The research interface
Under /api/v1 there are five data endpoints: active fire detections, burnt areas, fire weather index, sensor stations and sensor readings. Each returns either JSON or CSV, and each accepts a time range and an area filter.
Access runs through a personal key sent in the request header. The description of the contract, parameters, status codes, examples in curl, Python and R, is readable without signing in: a student must know how to use a key before having one, and an examiner must be able to check the methodology statements of a thesis without ever holding an access key.
Reproducibility
Four commitments make work on this data traceable.
All timestamps are UTC, stored, computed and delivered. Display is in local time, but everything that goes through the interface is UTC without exception.
Simulated stations carry a field of their own and can therefore be excluded in any analysis. Work that accidentally computes on demonstration data would be worthless, and the error would be invisible from outside.
Readings are available as a raw series and as a corrected series, selectable by parameter. The raw series never changes; the corrected one carries the version of the correction model.
And the interface is versioned: a change of response format would receive a new version number in the path instead of silently breaking existing analyses.
Open datasets
One frozen, citable dataset per season is planned, published with its own identifier (DOI) under Creative Commons Attribution. That makes theses and publications citable to this network, and the network in turn becomes academically referenced.
The rule behind it is the same as for ground sensing: raw data is inviolable, corrections are versioned layers above it, and the methodology documentation accompanies every dataset.
Catalogue of thesis topics
Eight topics to begin with, updated annually with the supervising chairs. Every completed piece of work is listed in the annual report and linked on the platform.
| Topic | Core question | Data basis | Level |
|---|---|---|---|
| Calibrating low-cost particulate sensors against a reference | How good does a low-cost sensor become after co-location and correction? Which class of correction, linear, humidity-dependent, learned, is worth the effort? | Co-location campaigns, official reference station | Bachelor / Master |
| Inversion weather situations in a valley basin | How do valley inversions shape the particulate distribution across station heights and times of day? | Station network and weather data, winter season | Master |
| Early smoke signature versus satellite latency | By how many minutes or hours does the ground network pattern precede the first satellite detection with its roughly three hours of latency? | Sensor layer against the FIRMS archive | Master |
| Energy profile of a wireless sensor node | What is the real energy balance of the street-lamp architecture with night charging and deep sleep, compared with the design? | Telemetry of battery voltage and charge state | Bachelor |
| Humidity bias of optical particle counters | How large is the water uptake effect in the local climate, and how can it be corrected? | Co-location and climate series | Bachelor / Master |
| Ageing and drift of particulate sensors | How does a sensor drift over months of operation, and when is replacement cheaper than correction? | Rotation data and change log | Bachelor |
| Dispersion modelling and source location | Can the source region of a particulate event be narrowed down from the station pattern and the wind field? | Station network, weather, event log | Master |
| Water temperature and low water of the Bosna | What do time series from the water variant show about hot summers, low water and drought? | Water boxes along the river | Bachelor / Master |
Supervision and stewardship
The network is to be operated by a registered student association in the Federation of Bosnia and Herzegovina: maintenance, calibration rotation, data quality and the first point of contact for questions from authorities. samoLabs builds the product, installs it and carries product responsibility.
One supervising chair per university is envisaged as mentor, accompanying theses from the catalogue, the universities of Zenica and Sarajevo are the intended partners. Specific faculties and contacts are to be settled before the pilot starts; what stands here is the plan, not its execution.
The academic self-interest is at the same time the insurance against the usual fate of sensor pilot projects: a network that is cited and on which work is continuously produced has advocates beyond the funding period.
Access
A key for the research interface is issued on request. Anyone planning a piece of work should state the topic and the period, which endpoints and which periods are unlocked depends on that.