We develop a forecast to predict when severe storms are likely to occur using data science techniques. We take a representative sample of all storms in a database, where each data point represents the time when a storm had wind speeds exceeding a specific threshold according to the Saffir-Simpson scale. In this process, we discover explainable artificial intelligence utilizing statistical analysis.
The above button navigates to a Python Notebook where we can fully replicate the forecast.
- NHC HURDAT2 data
- The official data from the National Hurricane Center (NHC) and their 2nd version of the Hurricane Database, or Wallet.
- Severe Storms (Category 3 and above)
- Anything above 96 knots
- Calendar Day Number Analysis
- The calendar day number is the total count of days since the beginning of the year. For example, the calendar day number for July 31st is 212.
- Based on this, we can assign a value for every time there’s a severe storm record in the database.
- We can test to see if this sample is uniform, or whether the calendar day number for all severe storm records in the database have a uniform probability. With a 95% confidence value, we can assume that the sample is not uniform.
- Descriptive statistics
