DFUSIONET_RT

Irrigation Prediction Model

This repository contains a MATLAB script for predicting irrigation requirements using sensor and spatial data. The script preprocesses data, trains a neural network model, makes predictions, and generates irrigation prescriptions based on the predicted evapotranspiration (ET).

IMPORTANT: before running the real-time prediction main file (main_ET_pred_1step_RT.m), make sure the specific sites in the datafiles/Irrigation_rec/[sitename]/ directory has 3 folders named 'Figures', 'Final Output', and 'SpatialPredET'.

Table of Contents

• Usage
• User Inputs
• Script Workflow
  • Data Preprocessing
  • Model Training
  • Prediction
  • Irrigation Prescription
• Dependencies
• Data Files
• Results
• Contact

Usage

1. Clone the repository:

   git clone https://github.com/[]
   cd irrigation-prediction
   

2. Place the required data files in the 'datafiles' directory.

3. Open the MATLAB script and set the user inputs as needed.

4. Run the script in MATLAB.

User Inputs

• site_idx: Index to select which site to work with (1 for R5, 2 for R6).
• currentdate: Current irrigation date.
• last_irr_date: Last irrigation date.
• site_dir: Directory containing site data files.
• filename_2023: Filename of the 2023 data file.
• n_features: Number of features for the model.
• window_size: Sliding window size (in days) for the time series.
• a: RNG array for initializing neural network predictors.
• xval: Hidden layer sizes for the neural network.
• NN_param: Parameters for the neural network (batch size, learning rate, dropout probability, epochs).

Script Workflow

Data Preprocessing

• Load sensor (Arable) ETc data and spatial ET data for the selected site.
• Filter and normalize the data based on the required dates and features.
• Perform proportional-offset interpolation (POI) to align site-specific values.
• Partition the data into sliding windows for time series analysis.

Model Training

• Initialize the neural network with specified parameters.
• Train the neural network with multiple random initializations.
• Save the trained networks.

Prediction

• Load the trained networks.
• Generate predictions for the specified date range.
• Save the predictions to CSV files.

Irrigation Prescription

• Load the predicted spatial ET data.
• Calculate total ET from the last irrigation date to the current date.
• Save the irrigation prescription to a CSV file.

Dependencies

• MATLAB R2021b or later
• Neural Network Toolbox

Data Files

• datafiles/Irrigation_rec/: Directory containing site data files.
• datafiles/alldatafiles_2023.mat: 2023 data file containing sensor and spatial data.

Results

• Predicted spatial ET data saved as CSV files in the SpatialPredET directory.
• Irrigation prescription saved as a CSV file in the Final output directory.

Contact

For any questions or issues, please contact Farshina at [email protected]