# Storm Nowcasting Hackathon - IndabaX Madagascar 24 — Data

来源：https://www.kaggle.com/competitions/hackathon-DL-IndabaX-MDG-2024/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

## Description

Data source: 
- [EUMETSAT](https://www.eumetsat.int/), Meteosat Second Generation Data on Cloud-Top Temperature
- 2D-wavelet transform by [Klein et al. (2018)](https://agupubs.onlinelibrary.wiley.com/doi/full/10.1002/2017JD027432) to detect storms in real-time and database creation by [Rakotomanga et al. (2024)](https://eps.leeds.ac.uk/maths/pgr/13569/mendrika-rakotomanga).


Each row in the dataset includes the following information as input features:

- The **time of observation** (t<sub>0</sub>), which includes the **year, month, day, hour, and minute**,
- **Latitude (lat) and longitude (lon) (in °)**,
- **Intensity (no unit), size (in km<sup>2</sup>), and distance (km)** 

of the nearest storm at t<sub>0</sub>.

The **training data** contains the true binary labels for storm occurrence at **1-hour (Storm_NosyBe_1h)** and **3-hour (Storm_NosyBe_3h)** lead times, based on the input features. The dataset spans the period from November to April, covering the years 2004 to 2019.

The **test data** contains storm IDs and input features but does not include labels. It covers the period from November to April for the years 2020 to 2024.

## Files

* **train.csv**: The training set.
* **test.csv**: The test set.
* **sample_submission.csv**: An example of submission file in the correct format.