Return periods and extreme wind

The return period of an extreme event, such as a '50-year gust', is the average interval between occurrences of an event of that magnitude or greater. It does not imply that an event will occur exactly every 50 years, but rather that, on average, over a very long period, such an event would be expected to happen once every 50 years. More formally, a 50-year gust has a 1 in 50 (or 2%) chance of being exceeded in any given year.

This concept is crucial for infrastructure design, where structures must withstand extreme loads over their operational lifetime. For instance, a building designed to a 50-year return period wind speed is expected to have a low probability of failure due to wind within its design life, typically 50 years or more. However, it is essential to understand that an event with a 50-year return period could occur in consecutive years, or not at all for a century; the definition is statistical, not a fixed schedule.

The calculation of return periods relies on statistical analysis of long-term wind records, often annual maximum gust speeds. These records are then fitted to extreme value distributions to estimate the likelihood of events beyond the observed period. The longer the historical record, the more robust the return period estimates become, particularly for very rare events like 100-year or 250-year gusts.

To determine return periods, meteorologists and engineers typically use extreme value analysis. One common approach is the annual maxima (AM) method. This involves extracting the highest wind speed (usually a 3-second gust at 10 m) recorded in each year from a long-term dataset. This series of annual maximum values forms the basis for statistical modelling.

For example, if you have 40 years of daily maximum gust data, you would select the single highest gust from each of those 40 years. This results in a dataset of 40 annual maximum gusts. This method simplifies the analysis by focusing only on the most extreme event within each block (year), discarding other high, but not annual-maximum, events.

An alternative is the peaks-over-threshold (POT) method, also known as the block-maxima method. This method considers all events that exceed a certain high threshold, rather than just the annual maximum. While more complex to implement, POT can sometimes make more efficient use of data, especially when the number of extreme events is low, or when the annual maxima method might miss significant events that occur in the same year as an even higher one.

Both methods aim to provide a robust dataset for fitting extreme value distributions, which then allow for the extrapolation to return periods beyond the length of the observed record.

Once a series of annual maximum wind speeds is established, these data points are fitted to a statistical distribution designed for extreme values. The most commonly used distributions for this purpose are the Gumbel distribution (also known as Extreme Value Type I) and the Generalised Extreme Value (GEV) distribution.

The Gumbel distribution is a special case of the GEV distribution. The GEV distribution combines three types of extreme value distributions (Gumbel, Fréchet, and Weibull) into a single form, making it more flexible. It is defined by three parameters: location, scale, and shape. The shape parameter determines which of the three types the distribution represents.

Fitting these distributions involves estimating the parameters that best describe the observed annual maxima. This is typically done using methods like maximum likelihood estimation. Once the parameters are determined, the distribution can be used to calculate the probability of exceeding a certain wind speed, or conversely, the wind speed corresponding to a given return period.

For example, if a GEV distribution is fitted to 40 years of annual maximum gust data for a site in Ireland, it might yield a 50-year return period gust of 45 m/s (approximately 87 knots). This value is then used in design codes and standards to ensure structures can withstand such extreme conditions.

The accuracy of return period estimates is highly dependent on the length and quality of the historical wind record. Short records introduce significant uncertainty, particularly for longer return periods. For instance, estimating a 100-year return period gust from only 10 or 20 years of data involves substantial extrapolation beyond the observed range, leading to wider confidence intervals.

Consider a site with only 20 years of data. The highest observed gust in that period is the 20-year return period gust by definition. Estimating a 50-year or 100-year gust from this limited dataset requires assuming that the underlying statistical distribution remains valid for events much rarer than those observed. This assumption may not hold true, especially in regions with complex topography or changing climate patterns.

Commonly cited guidance suggests that for reliable estimates of a T-year return period event, the record length should ideally be at least 2T years, or at minimum T years. For a 50-year return period, a record of 100 years would be ideal, and 50 years a minimum. Many sites, especially newer ones, do not possess such long records of direct observation. This is why reanalysis datasets like ERA5, which provide consistent, long-term modelled data, are often preferred for return period studies, despite being models rather than direct measurements.

Worked Example: If a site has 30 years of annual maximum gust data, and the fitted GEV distribution estimates a 50-year return period gust of 40 m/s, the uncertainty around this estimate will be higher than if 60 years of data were available. The confidence interval for the 50-year gust might range from 37 m/s to 43 m/s, indicating the spread of possible values due to data limitations.

A frequent misinterpretation of return periods is to view them as a schedule. For example, a '100-year flood' does not mean that if one occurred last year, another will not happen for 99 more years. Instead, it means that in any given year, there is a 1% chance (1/100) of an event of that magnitude or greater occurring. The events are considered independent from year to year.

This is a critical distinction for risk assessment. The probability of an event with a return period T occurring at least once in n years can be calculated using the formula:

P(at least one in n years) = 1 - (1 - 1/T)^n

Let's apply this to a 50-year return period gust:

• Probability of occurrence in any single year: 1/50 = 0.02 or 2%.
• Probability of occurrence at least once in 10 years: 1 - (1 - 1/50)^10 = 1 - (0.98)^10 ≈ 1 - 0.817 = 0.183 or 18.3%.
• Probability of occurrence at least once in 50 years: 1 - (1 - 1/50)^50 = 1 - (0.98)^50 ≈ 1 - 0.364 = 0.636 or 63.6%.
• Probability of occurrence at least once in 100 years: 1 - (1 - 1/50)^100 = 1 - (0.98)^100 ≈ 1 - 0.133 = 0.867 or 86.7%.

This demonstrates that for a structure with a 50-year design life, there is a significant chance (over 60%) that it will experience at least one gust of 50-year return period magnitude or greater during its lifetime. This understanding guides engineers in setting appropriate design standards.

In Ireland, as in many countries, building codes and standards specify design wind speeds based on return periods. These values are critical for ensuring the structural integrity of buildings, bridges, cranes, and other infrastructure. The Eurocodes, specifically EN 1991-1-4: Eurocode 1: Actions on structures – Part 1-4: General actions – Wind actions, are the primary standards used.

For general building design, a 50-year return period gust is commonly specified for ultimate limit state (ULS) design, which concerns the safety of the structure. For more critical structures, or those with very long design lives or high consequences of failure (e.g., nuclear power plants, major bridges), longer return periods such as 100, 150, or even 250 years may be used. For serviceability limit state (SLS) design, which concerns comfort and functionality (e.g., limiting excessive vibrations), shorter return periods like 10 years might be applied.

These design wind speeds are typically expressed as the 3-second gust speed at 10 m above ground in open terrain. They then need to be adjusted for factors such as terrain roughness, topography (hills, cliffs), building height, and proximity to other structures, using coefficients provided in the Eurocode. The Wind Agent's return period chart provides the baseline 10 m gust values, which are the starting point for these detailed engineering calculations.

The Wind Agent's return_period chart (chart ID: return_period) presents the estimated extreme gust speeds for various return periods at your selected location. This chart is a direct visualisation of the GEV distribution fitted to long-term reanalysis data for that specific point on the map.

How to interpret the chart:

• X-axis: Represents the return period in years (e.g., 10, 20, 50, 100, 250 years).
• Y-axis: Shows the corresponding 3-second gust speed at 10 m above ground level, typically in metres per second (m/s) or kilometres per hour (km/h).
• The curve: This line indicates the gust speed that, on average, will be equalled or exceeded once every X years. For example, if the curve shows 45 m/s at the 50-year mark, it means a gust of 45 m/s or greater has a 2% chance of occurring in any given year.
• Confidence intervals: The shaded area around the main curve represents the confidence interval (e.g., 90% confidence). This band illustrates the uncertainty in the estimate, which widens for longer return periods due to the extrapolation involved. A wider band indicates greater uncertainty.

This chart is a valuable tool for initial design considerations, risk assessment, and understanding the extreme wind climate of a location. It provides the baseline data required for more detailed engineering analysis, always remembering that these are statistical estimates derived from modelled data.