README.md

NDVI Time Series Analysis and Forecasting

This project implements an NDVI (Normalized Difference Vegetation Index) time series analysis and forecasting tool using LSTM (Long Short-Term Memory) and Multi-head Attention. It combines satellite imagery data with weather information to predict future NDVI values for specified geographical areas.

Features

• Fetches and processes NDVI data from Sentinel-2 satellite imagery using Google Earth Engine
• Retrieves historical weather data for the specified location
• Applies data cleaning, filtering, and smoothing techniques to NDVI time series
• Implements LSTM models for both original and smoothed NDVI data
• Provides forecasting capabilities for future NDVI values
• Visualizes historical data, predictions, and forecasts using interactive plots

Requirements

• Python 3.7+
• Google Earth Engine account
• CUDA-capable GPU (optional, for faster training)

Installation

1. Clone this repository:

   git clone https://github.com/senthilkumar-dimitra/NDVI-time-series-analysis.git
   cd ndvi-time-series-analysis

2. Install the required packages:

   pip install -r requirements.txt

3. Set up your Google Earth Engine authentication

Usage

Run the main script:

python ndvi_ts_lstm.py

Configuration

You can modify the following parameters in the script:

• start_date: Start date for data retrieval
• end_date: End date for data retrieval
• n_steps_in: Number of time steps used for input sequences
• n_steps_out: Number of time steps to forecast
• lstm_units: Number of units in the LSTM layers
• percentile: Percentile for NDVI filtering
• bimonthly_period: Time interval for filtering
• spline_smoothing: Smoothing parameter for the spline interpolation

Output

The script generates:

• Interactive plots showing historical NDVI data, predictions, and forecasts
• Performance metrics for the LSTM models [WIP]