Extreme precipitation

Extreme precipitation statistics (also known as design rainfall, precipitation frequency, intensity-duration-frequency, depth-duration-frequency or high intensity precipitation) are estimates of the depth or intensity that will occur at a certain frequency, and are used for modeling the flooding that might occur at a location.

Customizing the information you see

We recognise that users have different needs, and/or are accustomed to seeing the data in a certain way. For this reason extreme precipitation data can be customized as follows:

  • Choose between units of millimeters, millimeters/hour, inches, or inches/hour
  • Choose between average recurrence interval (ARI) and annual exceedance probability (AEP)
  • Select the durations and frequencies you wish to see:
    • Durations range from 1 hour to 18 weeks
    • Any frequency can be entered, subject to:
      • a minimum ARI of 1 year or maximum AEP of 63.2%
      • a maximum ARI of 250 years or minimum AEP of 0.4%
  • Switch the horizontal and vertical axis (transpose the table)

Users can also:

  • Copy or download data (using the same configuration as for on-screen data)
  • Plot depth or intensity against duration or frequency

How were these statistics derived?

The process used to generate XRain extreme precipitation statistics is described in the paper Oliver (2022) “Extreme Precipitation Statistics on a Global Scale”.

Extreme precipitation statistics are derived from a regionalized analysis of 20 years of historical data. A generalised extreme value (GEV) distribution is fitted to annual maxima, allowing us to deduce the depth or average intensity of rare events. However the uncertainty of these estimates increases with increasing ARI or decreasing AEP, and for this reason XRain applies a limit of 250 year ARI or 0.4% AEP.

Because XRain is based on remotely-sensed precipitation data, it is by nature less accurate than data derived from rain gauges. We recommend that you calibrate extreme precipitation statistics against known data where possible. Read about our suggested calibration procedure.

The statistical analysis used implicitly makes the assumption that climatic conditions have not changed over the period in question—a “stationary” climate. This is not true, for at least two reasons:

  • Long-term (multi-decade) climate oscillations such as the Atlantic Multidecadal Oscillation and the Indian Ocean Dipole can lead to increases in precipitation in some places and decreases in others. These effects will invariably be seen in extreme precipitation events, and therefore will influence the extreme precipitation statistics calculated.
  • Climate change: there is strong consensus that global surface temperatures are increasing, leading to more frequent weather extremes, including precipitation extremes (droughts and heavy rainfall).

Users will need to account for these as necessary in their use of the data. Read more on climate change adjustment.


How accurate is the data?

XRain data is based on satellite precipitation measurements, which is by nature less accurate than data derived from rain gauges. But it can work hand-in-hand with local data to allow you to clearly understand how rainfall patterns vary in areas that otherwise have little information.

Section 6 of the XRain technical report includes a comparison against datasets from USA, UK, Australia, and New Zealand. It demonstrates that:

  • although absolute values differ, these differences are relatively consistent spatially; meaning that calibrating XRain statistics against any available local data can provide robust estimates of absolute values at locations without local data.
  • the ratio between XRain depths and depths from local datasets does not vary dramatically across durations or ARIs/AEPs; meaning that XRain can be used to extrapolate “missing” values from a local dataset. For example, you might have you have 24 hour depths for your site that were derived from a rain gauge record, but you need to know sub-daily durations too.

We are happy to discuss this with you on a video call, if that would be helpful. If so, please get in contact.

If you only have 20 years of historical data, how robust are your 250 year ARI estimates?

Let’s be up front: 20 annual maxima is not much to go with. An analysis like this assumes that annual maxima are representative of long term behavior, but what if a location had unusually low or unusually high rainfall over many of those years, for instance?

To address this issue, XRain uses a regionalization approach. For each cell, annual maxima are pooled together with annual maxima together from 20 neighboring cells, giving a total of 400 maxima. After appropriately normalizing these values, the GEV distribution is fitted to these pooled maxima. This results in much more robust and consistent estimates of the GEV parameters. You can read more about the regionalization process in Section 5 of the XRain report.

How good are XRain estimates in mountainous areas?

The resolution of the underlying satellite precipitation dataset is approximately 11 km by 11 km at the equator and 6 km by 11 km at ± 60° latitude. In broad terms, you can consider seasonal and extreme statistics to represent an average of the precipitation over each cell.

Precipitation generally increases as elevation increases. Consequently, if there are large elevation differences within a cell, XRain statistics will be unable to give you information on the highest precipitation (mountain peaks) or the lowest precipitation (valleys). However, with appropriate calibration its estimates may still provide helpful information for such contexts.

Why is a 30 minute duration not included?

The process used to derive extreme precipitation statistics involves identifying the maximum depths in each year of the GPM_3IMERGHH_06 data.

XRain does not include extreme precipitation statistics at the 30 minute (0.5 hour) duration because the underlying time step is 30 minutes, which cannot adequately resolve the largest 30 minute event. This is because the largest 30-minute event observed in a year could start and finish between time steps.

For example, let’s imagine a 100 mm event starting at 1:15am and finishing at 1:45am. This event could be recorded as 50 mm at the 1:00 time step and 50 mm at 1:30. If another event starting at 6:00 and finishing at 6:30 had just 60 mm, it would be identified as being larger, even though it was 40% smaller.

This effect is still present for longer durations, but is less severe.