Approach
How we build a monitoring system.
Forest Monitoring, Transmission Line Monitoring, Mine Monitoring and the GIS Surveying suite each have their own models, their own data and their own deployment. What they have in common is a method: get the imagery, find the change, give it an owner, send someone to look, keep the record.
The method
The six-stage method.
Change detection identifies where something has altered. It does not establish who is accountable for that area, whether it was inspected, what was found, or whether the matter remains open. Each system carries that process through to closure, using components tuned to its own domain.
Acquire
The sensor is selected against the question the system must answer. A forest division needs wide coverage every fortnight; a transmission corridor needs to resolve a single tree. Each product ingests what its own problem requires.
- Sentinel-2 optical and Sentinel-1 SAR
- Sub-metre commercial imagery
- Drone photogrammetry
- Overview pyramids built on ingest
Detect
A land-cover classifier for a forest division and an individual-crown detector for a transmission corridor are different models with different failure modes. Each product carries its own, tuned and scored against reviewed ground truth from its own domain.
- Multi-date change detection
- Object detection for individual features
- Surface-model differencing
- Scored against human-reviewed labels
Give it an owner
A detection becomes an alert when it has a jurisdiction, a priority and a named person. Each product models the hierarchy its own users work in — beat and range for a forest division, span and tower for a corridor, lease for a mine.
- Jurisdiction resolution
- Priority and risk scoring
- Assignment, reassignment, escalation
- Queue with full status history
Verify on the ground
Receive the alert with coordinates, navigate, capture geo-tagged evidence, submit a structured outcome. Built for low-end Android devices and for beats with no signal — the field client is a determining component of the system.
- Turn-by-turn navigation to the alert
- Geo-tagged photographs and voice notes
- Offline capture with background sync
- Structured outcomes, not free text alone
Keep the record
Imagery source, detection date, assigned officer, field position, submitted media and every status change stay with the case for its whole life. A closure challenged two years later can be reconstructed from the record rather than from memory.
- Immutable case history
- Provenance on every derived number
- Exportable for statutory reporting
- Retention aligned to the client's policy
Report from the same record
Coverage, response time, closure rate, area affected, volume moved — the numbers organisations already have to file come off the case record rather than being assembled by hand at the end of a quarter.
- Per-officer and per-unit performance
- Period-on-period change summaries
- Statutory report formats
- Data export in open formats
How we build
Design principles.
The imagery stays the source of truth
Analysts work over the real scene at full resolution, served live from the source rasters. Dispatch decisions are taken against the source imagery at full resolution.
Full resolutionHuman review is part of the model
Corrections made by reviewers are retained as training and evaluation data, so a detector improves on the client's own imagery rather than remaining at its initial accuracy.
Reviewed truthOpen delivery formats
GeoTIFF, standard vector layers, point clouds, CSV. Data is delivered in open formats and is not held exclusively within our software.
No lock-inDeployed where the data must live
Managed cloud or the department's own infrastructure, with role-scoped access down to the smallest administrative unit. Sovereignty requirements are a deployment decision, not a blocker.
Your infrastructureOffline operation is assumed
Field work happens in places with no signal. Every mobile surface caches what it needs and queues what it produces.
Offline-firstUncertainty is reported
A measurement the imagery could not support is flagged as such rather than replaced with an estimated value. A measurement the imagery cannot support is flagged rather than estimated.
Uncertainty reportedData
Data sources.
Resolution, revisit and cost pull in different directions. Each system is built around the sources that can answer its question, at the revisit that question needs.
| Sentinel-2 | Optical, wide-area change detection and vegetation health |
|---|---|
| Sentinel-1 | SAR — validation and confirmation through cloud |
| Cartosat-3 | Sub-metre multispectral for individual-feature detection |
| Daily optical | MODIS, VIIRS, PlanetScope for rapid large-event detection |
| Drone | Photogrammetric capture where centimetre accuracy is required |
| Ground | Field app submissions — geo-tagged imagery, audio, structured outcomes |
Built this way