How it works
Four steps, running continuously.
Connect
Secure links to your sondes, flow meters and telemetry, plus open Environment Agency, Defra and Met Office data.
Clean
Automated checks for drift, fouling, gaps and implausible values, so sensor artefacts are not mistaken for pollution.
Infer and forecast
Pollutant estimates and 24–72 hour risk forecasts for every point on the network, updated hourly.
Alert and report
Alerts by email, SMS or API before a threshold is breached; audit-ready reports on demand.
Inputs and outputs
What goes in, and what comes out.
The engine works with whichever parameters are available at a site. More inputs generally mean tighter estimates; each deployment’s calibration report states what was used.
| Signal | Typical unit | Role | Source |
|---|---|---|---|
| Dissolved oxygen | mg/L, % sat | Input | Section 82 monitors, sondes |
| Temperature | °C | Input | Section 82 monitors, sondes |
| pH | pH units | Input | Section 82 monitors, sondes |
| Turbidity | NTU / FNU | Input | Section 82 monitors, sondes |
| Ammonium / ammonia | mg/L NH₄-N | Input | Section 82 monitors, sondes |
| Conductivity | µS/cm at 25 °C | Input | Sondes (where fitted) |
| River flow and level; rainfall | m³/s, m; mm | Input | Environment Agency and Met Office open data |
| Nitrate | mg/L NO₃-N | Estimated · first target | Proxy Inference Engine |
| Orthophosphate | mg/L PO₄-P | Estimated · in development | Proxy Inference Engine |
| Dissolved zinc and copper | µg/L | Estimated · in development | Proxy Inference Engine |
| Microplastic and pharmaceutical risk markers | risk index | Research stage | R&D programme |
Catchment to coast
Pollution moves. The forecast follows it.
Rivers are usually monitored point by point, by different organisations with different systems. EcoFlow Predict models the river as one connected network.
- Travel time and dilution: a signal at a headwater sonde becomes a timed forecast for the town, bathing water or estuary downstream.
- Tributaries and rainfall: runoff after rain and inflows from side streams are built into each forecast.
- Calibrated per catchment: each deployment is fitted to its own geology, land use and discharges, and keeps learning.
Illustrative long-section
Architecture
Cloud-native, and built to be audited.
Your data, plus open data
Client instruments over MQTT, HTTP/REST, SFTP or CSV; Environment Agency, Defra and Met Office feeds; partner networks. No proprietary hardware.
Validation and time-series storage
Drift and fouling detection, gap handling, unit and timestamp harmonisation, and a time-series store holding every raw and cleaned value.
Proxy Inference Engine
Machine-learning models calibrated per catchment, a propagation model for downstream forecasting, anomaly detection and risk scoring.
Dashboards, alerts, reports and API
Predictive heatmaps, an alert centre, compliance and ESG reports, and a REST API for integration with your own systems.
Security and data governance
Your data stays yours.
Clients own their raw data. EcoFlow Predict processes it under a data-sharing agreement and uses it only to deliver and improve the service.
| Hosting | Enterprise cloud with UK data residency |
| Encryption | TLS 1.3 in transit, AES-256 at rest |
| Access | Role-based access with strict separation between clients |
| Audit trail | Every reading, prediction and alert logged and traceable |
| Compliance | UK GDPR; Cyber Essentials planned for Year 1; ISO 27001-aligned controls |
| Availability | 99.5% target with automated backup and recovery |