Real Use Case
Smart irrigation and urban water in Morocco
A complete MSAT deployment scenario, from the field problem to the measured outcome.
Problem
Morocco is one of the most water-stressed countries in the world, and agriculture consumes the largest share of its water. Irrigation is still largely scheduled rather than measured, connectivity in rural areas is intermittent, and cloud-only platforms cannot react at the speed a pump or valve requires.
Architecture
Field nodes aggregate sensors over LoRaWAN into an MSAT edge gateway that runs the models locally. Only aggregates and model updates travel to the sovereign cloud, which supervises the fleet and retrains models with federated learning.
Sensors
- Capacitive soil moisture at 3 depths
- Micro weather station (temp, RH, wind, rain)
- Multispectral leaf camera
- Flow and pressure meters on each valve
- Water salinity and pH probes
Edge Device
A solar-powered MSAT gateway: quad-core ARM SoC with an NPU, 4 GB RAM, LoRaWAN + 4G fallback, encrypted secure element, 15 W peak draw and 72 hours of battery autonomy.
AI Models
- Evapotranspiration forecaster (temporal CNN, 1.8 MB)
- Crop stress classifier (quantised vision model, 4.2 MB)
- Leak and burst detector (anomaly autoencoder)
- Symbolic irrigation policy (rules + agronomic constraints)
Cloud
A sovereign Moroccan cloud tenant stores only aggregates and model weights. It runs federated retraining, long-horizon basin forecasting, fleet supervision and signed over-the-air model rollouts with rollback.
Dashboard
Operators see water delivered per hectare, stress maps, valve status and every automated decision with its confidence and causal trace — in Arabic, French, English or Spanish.
Real Use Case
Decision Flow
01
Sense — soil, weather and flow sampled every 60 s
02
Interpret — edge models estimate stress and water need
03
Arbitrate — symbolic policy checks quotas, cost and agronomy
04
Act — valves open locally in under 10 ms, offline included
05
Learn — outcomes feed federated retraining in the cloud
Benefits
- Water applied on measured need instead of a fixed calendar
- Operation continues during network outages
- Farm-level data never leaves the exploitation
- Leaks detected in minutes rather than billing cycles
Results
8 ms
Median decision latency
−94%
Uplink volume vs cloud-only
72 h
Node autonomy without network
< 5 min
Leak detection time
Pilot-node measurements from MSAT reference deployments. Context indicators are sourced from the public references cited on this page.