User Tools

Site Tools


metrics:crpss

This is an old revision of the document!


Mean Continuous Ranked Probability Skill Score (CRPSS)

The Continuous Ranked Probability Score (CRPS) measures the integral square difference between the cumulative distribution function (cdf) of the forecast $\mathrm{F}_S \left( q \right)$, and the corresponding cdf of the observed variable $\mathrm{F}_Q \left( q \right)$,

\begin{equation} \mathrm{CRPS} = \int_{-\infty}^{\infty} \left\lbrace \mathrm{F}_S \left( q \right) - \mathrm{F}_Q \left( q \right) \right\rbrace \mathrm{d} q \textrm{.} \end{equation}

The mean CRPS comprises the CRPS averaged across $J$ pairs of forecasts and observations,

\begin{equation} \overline{\mathrm{CRPS}} = \frac{1}{J} \sum \limits_{j=1}^{J} \mathrm{CRPS}_j \textrm{.} \end{equation}

The Continuous Ranked Probability Skill Score (CRPSS) is a function of the ratio of the mean CRPS of the main prediction system, $\overline{\mathrm{CRPS}}$, and a reference system, $\mathrm{\overline{CRPS}_{\mathrm{ref}}}$,

\begin{eqnarray} \mathrm{CRPSS} &=& \frac{ \mathrm{\overline{CRPS}} - \mathrm{\overline{CRPS}}_\mathrm{ref} }{ \mathrm{\overline{CRPS}}_\mathrm{perfect} - \mathrm{\overline{CRPS}}_\mathrm{ref} } \\ &=& \frac{ \mathrm{\overline{CRPS}} - \mathrm{\overline{CRPS}}_\mathrm{ref} }{ 0 - \mathrm{\overline{CRPS}}_\mathrm{ref} } \nonumber \\ &=& \frac{ \mathrm{\overline{CRPS}}_\mathrm{ref} - \mathrm{\overline{CRPS}} }{ \mathrm{\overline{CRPS}}_\mathrm{ref} } \nonumber \\ &=& 1 - \frac{ \mathrm{\overline{CRPS}} }{ \mathrm{\overline{CRPS}}_\mathrm{ref} } \nonumber \end{eqnarray}

metrics/crpss.1486456477.txt.gz · Last modified: 2017/02/07 09:34 by jan