Frequency of exceedance

The frequency of exceedance is the number of times a stochastic process exceeds some critical value, usually a critical value far from the process' mean, per unit time.[1] Counting exceedance of the critical value can be accomplished either by counting peaks of the process that exceed the critical value[1] or by counting upcrossings of the critical value, where an upcrossing is an event where the instantaneous value of the process crosses the critical value with positive slope.[1][2] This article assumes the two methods of counting exceedance are equivalent and that the process has one upcrossing and one peak per exceedance. However, processes, especially continuous processes with high frequency components to their power spectral densities, may have multiple upcrossings or multiple peaks in rapid succession before the process reverts to its mean.[3]

Consider a scalar, zero-mean Gaussian process y(t) with variance σ_y² and power spectral density Φ_y(f), where f is a frequency. Over time, this Gaussian process has peaks that exceed some critical value y_max > 0. Counting the number of upcrossings of y_max, the frequency of exceedance of y_max is given by[1][2]

${\displaystyle N(y_{\max })=N_{0}e^{-{\tfrac {1}{2}}\left({\tfrac {y_{\max }}{\sigma _{y}}}\right)^{2}}.}$

N₀ is the frequency of upcrossings of 0 and is related to the power spectral density as

${\displaystyle N_{0}={\sqrt {\frac {\int _{0}^{\infty }{f^{2}\Phi _{y}(f)\,df}}{\int _{0}^{\infty }{\Phi _{y}(f)\,df}}}}.}$

For a Gaussian process, the approximation that the number of peaks above the critical value and the number of upcrossings of the critical value are the same is good for y_max/σ_y > 2 and for narrow band noise.[1]

For power spectral densities that decay less steeply than f⁻³ as f→∞, the integral in the numerator of N₀ does not converge. Hoblit gives methods for approximating N₀ in such cases with applications aimed at continuous gusts.[4]

As the random process evolves over time, the number of peaks that exceeded the critical value y_max grows and is itself a counting process. For many types of distributions of the underlying random process, including Gaussian processes, the number of peaks above the critical value y_max converges to a Poisson process as the critical value becomes arbitrarily large. The interarrival times of this Poisson process are exponentially distributed with rate of decay equal to the frequency of exceedance N(y_max).[5] Thus, the mean time between peaks, including the residence time or mean time before the very first peak, is the inverse of the frequency of exceedance N⁻¹(y_max).

If the number of peaks exceeding y_max grows as a Poisson process, then the probability that at time t there has not yet been any peak exceeding y_max is e^−N(y_max)t.[6] Its complement,

${\displaystyle p_{ex}(t)=1-e^{-N(y_{\max })t},}$

is the probability of exceedance, the probability that y_max has been exceeded at least once by time t.[7][8] This probability can be useful to estimate whether an extreme event will occur during a specified time period, such as the lifespan of a structure or the duration of an operation.

If N(y_max)t is small, for example for the frequency of a rare event occurring in a short time period, then

${\displaystyle p_{ex}(t)\approx N(y_{\max })t.}$

Under this assumption, the frequency of exceedance is equal to the probability of exceedance per unit time, p_ex/t, and the probability of exceedance can be computed by simply multiplying the frequency of exceedance by the specified length of time.

• Probability of major earthquakes[9]
• Weather forecasting[10]
• Hydrology and loads on hydraulic structures[11]
• Gust loads on aircraft[12]

• 100-year flood
• Cumulative frequency analysis
• Extreme value theory
• Rice's formula

• Hoblit, Frederic M. (1988). Gust Loads on Aircraft: Concepts and Applications. Washington, DC: American institute of Aeronautics and Astronautics, Inc. ISBN 0930403452.CS1 maint: ref=harv (link)
• Feller, William (1968). An Introduction to Probability Theory and Its Applications. Vol. 1 (3rd ed.). New York: John Wiley and Sons. ISBN 9780471257080.CS1 maint: ref=harv (link)
• Leadbetter, M. R.; Lindgren, Georg; Rootzén, Holger (1983). Extremes and Related Properties of Random Sequences and Processes. New York: Springer–Verlag. ISBN 9781461254515.CS1 maint: ref=harv (link)
• Rice, S. O. (1945). "Mathematical Analysis of Random Noise: Part III Statistical Properties of Random Noise Currents". Bell System Technical Journal. 24 (1): 46–156. doi:10.1002/(ISSN)1538-7305c.CS1 maint: ref=harv (link)
• Richardson, Johnhenri R.; Atkins, Ella M.; Kabamba, Pierre T.; Girard, Anouck R. (2014). "Safety Margins for Flight Through Stochastic Gusts". Journal of Guidance, Control, and Dynamics. AIAA. 37 (6): 2026–2030. doi:10.2514/1.G000299. hdl:2027.42/140648.CS1 maint: ref=harv (link)