How does the fractions skill score credit a near-miss storm location?
The fractions skill score compares the fraction of cells with an event around each grid point. A displaced storm can receive partial credit when the chosen neighborhood contains both forecast and observed echoes.
Start with How do you calculate the critical success index for rare storms?.
| Neighborhood | Forecast cells | Observed cells | Fractions (forecast, observed) |
|---|---|---|---|
| A | 1, 0 | 1, 1 | (0.5, 1) |
| B | 0, 1 | 0, 0 | (0.5, 0) |
Set up the experiment
Use two sample centers, with forecast neighborhood fractions f = (0.5, 0.5) and observed fractions o = (1, 0). For example, these can represent two-cell windows whose forecast event patterns are [1,0] and [0,1] and whose observed patterns are [1,1] and [0,0]. Keep the event threshold and eligible centers fixed. Predict the score before calculating. Compute FSS = 1 − Σ(f−o)²/Σ(f²+o²). The numerator sum is 0.50 and the denominator sum is 1.50, so the score is 2/3. Smoothing can improve the score for a location error. Record the neighborhood size and treat edge handling consistently.
Practice questions
- What FSS results if the forecast neighborhoods match observations exactly?
- Use forecast fractions (0,1) and observed fractions (1,0). Calculate FSS.
- If forecast and observed fractions are both zero everywhere, what happens?
Worked answers
Answer 1
The numerator is zero and the denominator is positive. FSS = 1.
Answer 2
The numerator and denominator both equal 2. FSS = 0.
Answer 3
The reference denominator is zero. Report the all-zero case using the verification system’s stated convention and avoid silently treating it as a perfect event match.
- The numerator is zero and the denominator is positive. FSS = 1.
- The numerator and denominator both equal 2. FSS = 0.
- The reference denominator is zero. Report the all-zero case using the verification system’s stated convention and avoid silently treating it as a perfect event match.