Smart Irrigation System for Precision Agriculture

Introduction

The Edge AI Precision Irrigation System is a fully autonomous, on-device solution for dynamically adjusting water supply to nursery-stage crops (e.g., onion seedlings) in response to environmental and crop conditions. It integrates historical climate data and real-time soil moisture sensing into a machine learning inference engine running on a microcontroller, which computes the crop’s current water demand and controls pump actuation without cloud reliance.

Smart Irrigation System

In our implementation:

  • Weather inputs (Tmax, RH, Irradiation) via weather API feed an MLP-based model on a Raspberry Pi Pico W to estimate daily reference evapotranspiration (ET₀).
  • ET₀ is then used to derive actual crop evapotranspiration (ET₀ × Kc).
  • The computed water requirement is used by the Pico to set the irrigation volume.
  • An ESP32 node drives a water pump via MQTT for the calculated duration.

Conventional irrigation often over-waters or follows rigid schedules, wasting water and harming crops. Onion seedlings are especially vulnerable:

  • Early over-watering disrupts root aeration and yield.
  • Uneven watering or waterlogging leads to soil salinity and nutrient loss.

In water-scarce regions, these issues are severe.

This system:

  • Adjusts irrigation dynamically based on growth stage, weather, and soil moisture.
  • Predicts reference ET₀ and applies the onion-specific Kc to deliver precise water per growth stage.
  • Runs entirely on the edge, ensuring cloud independence.

Data Collection

Historical Climate Data

  • Source: NASA POWER API
  • Variables:
    • Min & Max Temperature (T2M_min, T2M_max)
    • Relative Humidity at 2m (RH2m)
    • Wind Speed (WS2m)
    • Solar Radiation (ALLSKY_SFC_SW_DWN)

Label Generation (ET₀)

  • Computed using the FAO-56 Penman–Monteith formula.
  • These ET₀ values serve as labels for supervised model training.

Real-Time Sensor Data

  • Analog soil moisture sensor installed per bed or pot.
  • Outputs voltage proportional to volumetric water content (VWC).
  • Converted to available water depth.
  • Feeds the controller for real-time irrigation decisions.

Preprocessing & Feature Engineering

  • Outlier Removal: Filtered extreme/erroneous values.
  • Scaling: Applied Z-score normalization using StandardScaler.
  • Feature Selection: Top 3 predictors identified via correlation analysis:
    • T2M_max (Max Temp)
    • RH2m (Relative Humidity)
    • ALLSKY_SFC_SW_DWN (Solar Radiation)

Model Design and Training

Model Selection

  • Evaluated: Linear Regression, SVR, Random Forest
  • Final model: Feedforward MLP Regressor
  • Accuracy: ~97% variance explained

Architecture

  • Two hidden layers: 16 and 8 neurons
  • Activations: ReLU
  • Optimizer: Adam

Training Configuration

  • Dataset: ~5,569 samples (80% training / 20% testing)
  • Early Stopping after 1500 iterations
  • Performance:
    • MSE ≈ 0.03
    • R² ≈ 0.971

Methodology

ET₀ Prediction

  • Pico gathers weather data via API: Tmax, RH, Solar Radiation
  • Inputs fed to MLP model on-device
  • Outputs ET₀ (mm/day)

Crop Water Demand (ETc)

  • ET₀ × Kc (crop coefficient for growth stage) = ETc
  • Represents daily crop water requirement

Soil Moisture Update

  • Sensor feeds real-time VWC
  • Converted to Available Water Depth
  • Used to adjust or skip watering if soil is already wet

Irrigation Volume Calculation

  • Rule-based logic or controller computes volume:
    • If soil moisture < target → deliver (ETc – depletion)
  • Ensures optimal water delivery

Pump Actuation

  • Pico sends MQTT command to ESP32
  • ESP32 activates relay/pump for the required time
  • Operates in a closed loop every 24 hours or on-demand

Hardware and Software Stack

Hardware Components

Component Specification Role/Use
Raspberry Pi Pico W RP2040 dual-core @ 133MHz, 264 KB RAM, Wi-Fi Hosts MLP model, sensor reading, volume calculation
ESP32 Dual-core 240MHz MCU with Wi-Fi/Bluetooth Receives MQTT, actuates pump
Soil Moisture Sensor Analog resistive sensor Measures real-time VWC
Relay Module 5V/12V controlled relay board Switches pump power via GPIO
Water Pump 12V DC pump (~X L/min flow rate) Delivers water to crop

Software Stack

Tool Use
Python (PC) Model training, preprocessing: pandas, numpy, scikit-learn
MicroPython Edge inference on Pico: matrix ops, manual activations
ESP32 FW MQTT listener, GPIO toggle for pump relay
Support Tools Thonny IDE, umqtt.simple, Jupyter for model export

Results

ET₀ Prediction Accuracy

  • R² ≈ 0.97
  • MSE ≈ 0.10 mm²/day²

Model Efficiency

  • Original model: ~18 KB
  • Pruned for deployment: ~12 KB
  • Inference time on Pico: ≤ 50 ms
  • Fits within Pico W’s RAM/flash constraints

Challenges and Mitigations

1. Integrating Multiple Sensors

  • Challenge: I/O and bandwidth limitations
  • Mitigation: Multiple ESP32s; Wi-Fi aggregation

2. Sensor Accuracy Drift

  • Challenge: Drift due to temperature or low-cost hardware
  • Mitigation:
    • Use temperature-compensated sensors
    • Periodic recalibration
    • Firmware-level smoothing (moving average)

3. Resource Constraints (Pico W)

  • Challenge: Limited flash/RAM
  • Mitigation:
    • Prune and quantize model
    • Manual implementation of inference logic

Conclusion

This project presents a complete Edge AI-based irrigation system for nursery crops:

  • Leverages long-term climate data and on-device MLP for accurate ET₀ prediction
  • Integrates soil feedback for optimal irrigation
  • Achieves ~97% accuracy against FAO-56 benchmarks
  • Fully cloud-independent and runs on low-cost hardware
  • Successfully combines hardware and AI into a working prototype

Future Directions:

  • Add user interface (LCD/app)
  • Extend to more crop types via ensemble models
  • Integrate additional environmental sensors
  • Modular, open-source design supports replication and customization

References

  • Allen, R. G., et al. (1998): FAO-56 Penman–Monteith Method
    Crop evapotranspiration: FAO Irrigation and Drainage Paper 56

  • NASA POWER API: NASA POWER Documentation

  • Pedregosa, F., et al. (2011): Scikit-learn: Machine Learning in Python
    JMLR 12, 2825–2830

  • Onion Crop Coefficients (Kc): Allen et al., 1989
    Agronomy Journal, 81(4), 650–662

  • MicroPython on Raspberry Pi Pico: Monk, S. (2019), The MagPi Magazine

  • ESP32 Datasheet: Espressif Systems (2021)

  • Irrigation Scheduling: Jones, H. G. (2004)
    Journal of Experimental Botany, 55(407), 2427–2436

  • Wireless Sensor Networks: Akyildiz, I. F., et al. (2002)
    Computer Networks, 38(4), 393–422

  • Project Repository:
    AYRUS06 / Edge AI Based Precision Irrigation (GitHub)