The mean **Relative Operating Characteristic Skill Score (ROCSS)** of the two lead times (1 hour and 3 hours) will be used to evaluate your models. The **ROCSS** for a lead time is calculated as:

$$
\text{ROCSS} = 1 - \frac{AUC_{\text{NFLICS}}}{AUC_{\text{model}}}
$$

where **AUC** is the [Area Under the ROC Curve](http://en.wikipedia.org/wiki/Receiver_operating_characteristic). 

**NFLICS** is an operational storm nowcasting system developed by [Anderson et al. (2024)](https://rmets.onlinelibrary.wiley.com/doi/full/10.1002/qj.4607) and is used as the benchmark for comparison.

**A positive ROCSS value indicates that your model has better skill than the reference model, NFLICS**. 

**Since the NFLICS predictions are only available during inference, you can focus on optimising the [AUC-ROC](https://scikit-learn.org/1.5/modules/generated/sklearn.metrics.roc_auc_score.html) for each lead time during the model development phase.**

# Competition Submission Format

For each ID in the test set, you are required to predict the probability of the storm occurring at both 1-hour and 3-hour lead times separately. 

The submission file must be a CSV file and:
- Include a header with three columns: `storm_id`, `Storm_NosyBe_1h`, and `Storm_NosyBe_3h`.
- Follow this example:

storm_id,Storm_NosyBe_1h,Storm_NosyBe_3h
storm_202001011115,0.1,0.09
storm_202001011130,0,1
storm_202001011145,0.78,0

**Hint: You can train two different models, one for each lead time.**

**While the primary requirement is to submit the CSV file with your predictions, we would really appreciate it if you could also briefly describe the model you used for your nowcasting approach (e.g., logistic regression, deep learning, etc.). This is optional but helps us understand your methodology and approach.**