2.4 Statistical analysis of precipitation data

Statistical analysis of precipitation data

Rainfall patterns can be unpredictable, but statistics help us make sense of them. By analyzing precipitation data, you can figure out how often certain storm events are likely to occur and design infrastructure around those estimates. This section covers four core tools: frequency analysis, IDF curves, DAD analysis, and probable maximum precipitation (PMP).

Frequency analysis

Frequency analysis is how hydrologists figure out the likelihood of a given rainfall event happening. You take historical precipitation records and use statistics to estimate how often storms of a certain magnitude will occur in the future.

Frequency distribution is the starting point. You rank all observed rainfall events (say, annual maximum daily rainfall over 50 years) from largest to smallest and plot how often each magnitude occurs. This gives you a picture of which rainfall amounts are common and which are rare. Common distributions used include the Gumbel, Log-Pearson Type III, and normal distributions.

Return period (also called recurrence interval) is the average number of years between events of a given magnitude or greater. A "100-year storm" has a return period of 100 years, meaning on average it occurs once every 100 years. This does not mean it can't happen two years in a row.

Probability and exceedance probability are directly tied to return period:

• Probability of exceedance in any given year = $P = \frac{1}{T}$, where $T$ is the return period
• A 50-year storm has $P = \frac{1}{50} = 0.02$, or a 2% chance of being equaled or exceeded in any single year
• The probability of a $T$-year event occurring at least once in $n$ years is $P_n = 1 - \left(1 - \frac{1}{T}\right)^n$

So a 100-year storm actually has about a 26% chance of occurring at least once during a 30-year mortgage period. That's why "100-year storm" is a misleading name for the public.

Intensity-duration-frequency (IDF) curves

IDF curves are one of the most widely used tools in hydrology and civil engineering. They relate three variables for a specific location:

• Intensity: the rate of rainfall, typically in mm/hr or in/hr
• Duration: how long the rainfall lasts (e.g., 5 min, 30 min, 1 hr, 24 hr)
• Frequency: the return period of the event (e.g., 2-year, 10-year, 100-year)

On a typical IDF curve, duration is on the x-axis, intensity is on the y-axis, and each curve represents a different return period. Two key patterns to remember:

1. For a given return period, intensity decreases as duration increases. A short, intense burst is more common than sustained heavy rainfall.
2. For a given duration, intensity increases as the return period increases. Rarer storms are more intense.

Uses of IDF curves:

• Designing storm drainage systems, culverts, and detention basins
• Estimating peak runoff using methods like the Rational Method ($Q = CiA$, where $i$ comes from the IDF curve)
• Setting design standards for roads, bridges, and urban infrastructure
• Flood risk assessment for a chosen level of protection (e.g., designing a culvert for the 25-year, 1-hour storm)

IDF curves are developed from local rainfall records, so they're site-specific. You can't use an IDF curve from one city for a project in another region.

Depth-area-duration (DAD) analysis

DAD analysis addresses the fact that rainfall isn't uniform over a large area. A rain gauge at the storm center might record 150 mm, but the average depth over a 1,000 km² watershed will be much less. DAD analysis quantifies this relationship.

The three variables:

• Depth: total precipitation depth (mm or inches) averaged over the area
• Area: the size of the region being considered (km² or mi²)
• Duration: the time period of accumulation (e.g., 6 hr, 12 hr, 24 hr)

DAD curves show that as the area increases, the average precipitation depth decreases for a given duration. And for a given area, longer durations accumulate greater total depth.

Uses of DAD analysis:

• Estimating the average rainfall over an entire watershed rather than relying on a single point measurement
• Designing large dams and spillways where the whole catchment area matters
• Converting point rainfall data to areal estimates for flood studies
• Providing input for PMP estimation (see below)

Probable maximum precipitation (PMP)

PMP is the theoretically greatest depth of precipitation that is physically possible over a given area and duration at a particular location and time of year. It represents the worst-case scenario for rainfall, not a statistical extrapolation but a physical upper limit.

Methods to estimate PMP:

1. Meteorological method (storm maximization): Take the most severe observed storms in a region, then maximize their moisture content by assuming the atmosphere was fully saturated. You transpose major storms from nearby similar regions and adjust for local conditions.
2. Statistical method: Extrapolate from the frequency distribution of observed rainfall using a frequency factor (e.g., the Hershfield method applies the formula $PMP = \bar{X} + K \cdot S$, where $\bar{X}$ is the mean, $S$ is the standard deviation, and $K$ is a frequency factor, typically around 15).
3. Generalized PMP charts: Published by agencies like the WMO or the U.S. National Weather Service (e.g., HMR series), these provide PMP estimates for broad regions based on compiled storm data.

Uses of PMP values:

• Designing spillways for large dams where failure would be catastrophic
• Estimating the Probable Maximum Flood (PMF) by routing PMP through a watershed model
• Setting the upper bound for flood risk analysis on critical infrastructure (nuclear plants, major dams)
• PMP is used where the consequences of failure are so severe that designing to a statistical return period (even 10,000 years) isn't considered safe enough