Smart Irrigation Prediction System

A time-series pipeline that ingests soil-moisture, temperature, and humidity telemetry over MQTT into InfluxDB, then trains Prophet and LSTM models to forecast soil-moisture trajectories and upcoming irrigation demand. Predicted deficits drive an automated irrigation schedule, shifting from reactive watering to proactive, weather-aware management.

Reactive irrigation either over-waters (wasting water and leaching nutrients) or under-waters (stressing crops), because it responds to current readings rather than anticipating need. Growers need a forecast of soil-moisture decline that accounts for weather.

Sensor telemetry streams over MQTT into InfluxDB. Prophet captures seasonality and trend while an LSTM models short-horizon non-linear dynamics; an ensemble forecasts soil-moisture trajectories and flags upcoming deficits. Predictions feed an automated scheduler that waters just before stress thresholds are crossed.

Planned — to be measured: forecast error against held-out sensor data, achievable water savings versus a fixed-schedule baseline, and crop-stress avoidance.

1. Phase 1

   Telemetry pipeline

   Stream soil and atmospheric sensors over MQTT into InfluxDB with a clean, queryable schema.

2. Phase 2

   Forecast models

   Train and compare Prophet and LSTM models on historical soil-moisture series.

3. Phase 3

   Scheduling

   Convert deficit forecasts into an automated, weather-aware irrigation schedule.

4. Phase 4

   Evaluation

   Benchmark water savings and forecast accuracy against a fixed-schedule baseline.