KO | EN

GitLite

Minimalist Open Source Repository Search Engine

gitlite — search

#dsh #dsh-plugin #ai-agents #developer-tools #ai-agent #claude-code #mcp #typescript #codex #cli #rust #deepseek-harness

Weather-Forecasting-AI-Model-with-METAR-Data

★ 11 Open GitHub ↗

Weather Forcasting AI Model with METAR Data

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# 🌤️ Weather Prediction with METAR + LSTM
This project focuses on enhancing **short-term weather forecasting** at airfields using **METAR (Meteorological Aerodrome Reports)** and **Long Short-Term Memory (LSTM)** neural networks.

Accurate forecasting is crucial for flight safety, especially for student pilots, whose training heavily depends on **calm and clear weather conditions.** Using structured hourly METAR reports, this project builds a data-driven machine learning pipeline to predict:

- Surface wind direction and speed
- Visibility
- Temperature and dew point
- Atmospheric pressure

&nbsp;

## 📂 Project Folder Structure

```
demo/
├── metar_raw/
│   └── raw_metar.csv            # Your input METAR data
├── metar_processed/             # Stores parsed & scaled data
├── input_splits/                # Train/val/test data splits
├── model_weights/               # Saved model and scaler
└── predict/                     # Temporary data for predicting
```

&nbsp;

## 🚀 Quick Start

**1. Set up Environment**
```bash 
pip install -r requirements.txt
```

&nbsp;

**2. Prepare Input Data**

Place your METAR data as "raw_metar.csv" inside:
```
demo/metar_raw/raw_metar.csv
```

The file should contain at least these columns:
- valid: Timestamp of report
- metar: Raw METAR string

&nbsp;

**3. Run the Full Pipeline**
```bash 
bash Train.sh
```

This script will:

1. Parse raw METAR reports -> structured CSV

2. Scale (normalize) and encode data

3. Create sequences for LSTM input/output

4. Train the LSTM model

5. Evaluate the model on the test set

&nbsp;

**🔮 Predicting with New Data**

You can use the trained model to generate a 12-hour weather forecast using the most recent 7 days (168 hours) of METAR data.
```bash 
bash Predict.sh
```

This script will:

1. Parse and scale your data

2. Prepare the last valid 168-hour input sequence

3. Run the trained LSTM model

4. Output the predicted weather for the next 12 hours

| Hour | WindDir (°) | Var | WindSpd | Vis (m) | Temp (°C) | DewPt | Pressure (hPa) |
|------|-------------|-----|---------|---------|------------|--------|----------------|
| 1    | 22.9        | 0   | 4.86    | 9687    | 8.14       | 1.23   | 1030.20        |
| 2    | 26.9        | 0   | 4.76    | 9805    | 7.81       | 1.26   | 1030.12        |
| 3    | 27.7        | 0   | 4.44    | 9768    | 7.08       | 1.22   | 1029.99        |
| ...  |             |     |         |         |            |        |                |
| 12   | 22.6        | 0   | 2.93    | 8667    | 2.42       | 1.28   | 1029.47        |

&nbsp;

## 📊 Model Details
- Architecture: 2-layer LSTM (128 + 64 units) + Dense output

- Input: Past 168 hours of METAR-derived weather data

- Output: 12-hour prediction for:
    - Wind direction (as sin/cos)
    - Wind speed
    - Visibility
    - Temperature
    - Dew point
    - Pressure
    - Wind direction variability

&nbsp;

## 🖼️ Results
The following chart shows the normalized **mean absolute error (MAE)** of the model's predictions, **scaled to the [0, 1] range**. Wind direction is shown as a fraction of 360°.

![Normalized MAE Chart](https://raw.githubusercontent.com/OmerZeyveli/Weather-Forecasting-AI-Model-with-METAR-Data/main/docs/normalized_mae_chart.png)

| Feature        | Normalized MAE |
|----------------|----------------|
| Wind Direction | 0.0383         |
| Wind Var       | 0.0949         |
| Wind Speed     | 0.0597         |
| Visibility     | 0.0829         |
| Temperature    | 0.0338         |
| Dew Point      | 0.0392         |
| Pressure       | 0.0265         |

&nbsp;

## 📬 Contact
Questions or suggestions? Feel free to open an issue or contribute via pull request.

&nbsp;

## ⭐ Star if you find it useful!

Weather-Forecasting-AI-Model-with-METAR-Data

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README