Solar Wind Time Series Gap Filling Interface

A web application that demonstrates different methods for filling gaps in solar wind time series data.

Features

• Synthetic Data Generation: Creates realistic solar wind data using multiple cosine waves + noise
• Gap Simulation: Introduces artificial gaps (up to 1 hour) to simulate missing data
• Multiple Gap-Filling Methods:
  • Linear Interpolation
  • FFT Reconstruction
  • LSTM Neural Networks
  • Transformer Models
  • Bayesian Modeling with Gumbel Softmax
• Interactive Visualization: 2D time series and 3D vector plots
• Real-time Comparison: Compare original, gapped, and filled data

Installation

1. Install dependencies:

pip install -r requirements.txt

2. Run the application:

streamlit run solar_wind_app.py

Usage

1. Select a time range using the slider (0-168 hours)
2. Choose a gap-filling method from the dropdown
3. View the results in the plots:
   • Left panel: B vector (magnetic field) components
   • Right panel: V vector (velocity) components
   • Bottom: 3D vector visualizations

Methods

• Linear Interpolation: Simple linear interpolation between gap boundaries
• FFT Reconstruction: Uses Fourier analysis of surrounding data to reconstruct gaps
• LSTM Prediction: Recurrent neural network trained on non-gap data
• Transformer: Attention-based sequence model for gap prediction
• Bayesian + Gumbel Softmax: Probabilistic modeling with smooth sampling

Data

The synthetic solar wind data includes:

• B vector (magnetic field): Bx, By, Bz components
• V vector (velocity): Vx, Vy, Vz components
• Generated using 5 different cosine frequencies
• Gaussian noise added for realism
• 1-minute sampling resolution starting from 1981