< Back to Print Version Table of Contents <

The statistical representation of flood events is affected by both length of the period of record and consistency of the hydrologic conditions within the drainage basin. Unrepresentative data for either of these reasons can lead to inaccurate guidance from flood frequency analysis.

This section will introduce basic concepts used in flood frequency analysis and demonstrate calculation of flood statistics.

A house near a stream

In this section you will learn to:

Topics in this section include:

Period of Record Issues
The Colorado River Compact
Confidence in Return Period Estimates
Exceedance Probability
Example of Exceedance Probability
Independent and Homogeneous Data
Impact of Basin Alterations
Review Questions

Choose a section above by clicking it, or scroll down to continue the module.

< Back to Print Version Table of Contents <

Period of Record Issues

Satellite image showing typical flow vs. great flood of 1993 for Missippi and Missouri River confluence.

Sometimes the record length of river flows may be insufficient for representing the river's entire flow history. Floods or droughts that occur infrequently may be under-represented in a limited streamflow record.

For example, the Mississippi flood of 1993 was a rare event that may only occur once every 100 years. Thus, if the period of record is only 40 years, there is a good chance that a flood like 1993 is not represented in the statistics. On the other hand, if the period of record is 300 years, it is likely that more than one flood like 1993 is in the flood frequency statistics. Thus, a longer period of record results in more representative flood statistics.

Photo of drought stricken Lake Roosevelt, Stevens County, WA

For accurate and reliable statistical hydrologic guidance about possible events, it is necessary to use a dataset that includes a representative sample of as many different events as possible.

The longer the period of record, the better the likelihood of capturing the range of possible events.

There are several different return periods traditionally used by hydrologists.

Common return periods include the 2-, 10-, 25-, 50-, 100-, and even 500-year flood. Values for each of these return periods can be calculated based on the statistics of the flow record. However, there is a question of just how representative extreme values, such as the 500-year flood value, might be.

If possible, it is best to avoid estimating return period flood values that are greater than twice the record length. So you might not want to put much faith in your 500-year flood estimate unless you have at least 250 years of data.

Return to Top of Page

The Colorado River Compact

Photo of Colorado River Commission 1922

One dramatic example on the impact of record length is illustrated by The Colorado River Compact.

In 1922 an agreement was signed between seven states in the western United States governing the use of the water of the Colorado River.

Colorado River at Lees Ferry, AZ

This agreement calls for the river water as measured at Lees Ferry, Arizona to be distributed as follows:

Key points of the Colorado River Compact of 1922

The problem is that studies have shown that the Colorado River long-term average is much less than the water allocation total. In other words, the river usually does not have enough water for all the allocations.

Original flow estimates were based on a very short period of record, during a relatively high period of flow. Actually the river's long term flow average is about 13 million acre-feet per year. The result is that the river is over-allocated by about 4.5 million acre-feet per year for an average year!

In truth, the annual river flow volumes are highly erratic, ranging from 4.4 million acre-feet to over 22 million acre-feet per year.

Return to Top of Page

Confidence in Return Period Estimates

Guidelines for length of data record vs. expected error rate

Estimates of flood return periods can be made with relatively short periods of record. But the associated confidence level in the flood frequency statistics is much higher with a longer period of data. For example, to estimate a 10-year flood with no more than a ±10 percent error, one would need 90 years of record. If you are willing to accept a ±25 percent error, then only 18 years of record is needed. Here we can see the length of data record needed to be within either ±10 percent or ±25 percent errors for the 10-, 25-, 50-, and 100-year floods.

Return to Top of Page

Exceedance Probability

High water mark for July 28, 1997 flood in Ft. Collins, CO

Sometimes a hydrologist may need to know what the chances are over a given time period that a flood will reach or exceed a specific magnitude. This is called the probability of occurrence or the exceedance probability.

Let's say the value "p" is the exceedance probability, in any given year. The exceedance probability may be formulated simply as the inverse of the return period. For example, for a two-year return period the exceedance probability in any given year is one over two = 0.5, or 50 percent.

Exceedance probability = 1 - (1 - p)n

But we want to know how to calculate the exceedance probability for a period of years, not just one given year. To do this, we use the formula
1- (1-p)n .

In this formula we consider all possible flows over the period of interest "n" and we can represent the whole set of flows with "1." Then (1-p) is the chance of the flow not occurring, or the non-exceedance probability, for any given year.

(1-p)n is all the flows that are less than our flood of interest for the whole time period.

Finally, "1," all possible flows, minus (1-p)n, all flows during the time period than are lower than our flood of interest, leaves us with 1 - (1-p)n, the probability of those flows of interest occurring within the stated time period.

Relationship between return period adn annual exceedance and non-exceedance probability

This table shows the relationship between the return period, the annual exceedance probability and the annual non-exceedance probability for any single given year.

So, if we want to calculate the chances for a 100-year flood (a table value of p = 0.01) over a 30-year time period (in other words, n = 30), we can then use these values in the formula for the exceedance probability.

We can also use these same values of p and n to calculate the probability of the event not occurring in a 30-year period, or the non-exceedance probability.

Return to Top of Page

Example of Exceedance Probability

Relationship between return period and annual exceedance and non-exceedance probability

Let's say you want to know what the probability is for a 50-year flood over a 50-year period. It's not 100 percent!

Calculation for Probability of 50-Year Flood Over 50-Year Period

1 - (1 - p)n

n = 50

p = 0.02

We know that n = 50 since we are looking at a 50-year period of time and using the probability of occurrence table we see that p=0.02 for a 50-year return period.

1 - (1 - 0.02)50

= 1 - (0.98)50

So, applying these values in the equation, the (1-p) value is (1-0.02), or 0.98.

= 1 - 0.36

= 0.64 or 64%

(1-p) to the n is 0.98 raised to the 50th power. That comes out to 0.36.

Now we have (1-0.36), which is 0.64.

There is a 64 percent chance of a 50-year flood in a 50-year period. That means there is a 36 percent chance we won't see a 50-year flood in the 50-year period.

Now let's determine the probability of a 100-year flood occurring over a 30-year period of a home mortgage where the home is within the 100-year floodplain of a river.

Calculation for Probability of 100-Year Flood Over 30-Year Period

1 - (1 - p)n

n = 30

p = 0.01

n=30 and we see from the table, p=0.01 .

1 - (1 - 0.01) 30

= 1 - (0.99) 30

= 1 - 0.74

(probability of non-occurrence = 0.74)

= 0.26 or 26% probability of occurrence

The 1-p is 0.99, and .9930 is 0.74.

There is a 0.74 or 74 percent chance of the 100-year flood not occurring in the next 30 years.

But 1-0.74 is 0.26, which shows there is a 26 percent chance of the 100-year flood in that time.

Return to Top of Page

Independent and Homogeneous Data

Illustration of flood independence

Flood frequency analysis requires that data be independent and homogeneous. The independence requirement means that floods occur individually and do not influence each other. For example, two peak flows that are above flood stage are independent floods if the flow completely returned to baseflow in between those two events.

On the other hand, if there is not a return to baseflow level between the peaks, the floods are not independent because the first flood has influenced the second flood.

Hurricane Floyd evacuation 1999

The homogeneous requirement means that each flood needs to occur under the same type of conditions. Two flood events are homogeneous if both are caused by rainfall only.

Damage from overtopping of Ft. Meade Dam near Sturgis, South Dakota 1972.

An example of non-homogeneity is when one flood is caused by rainfall while another flood is caused by a dam failure.

Return to Top of Page

Impact of Basin Alterations

Basin changes such as land development may change the hydrologic properties of a drainage basin over time, thus changing the way a river responds to storms. Aerial photo of Bear Creek in Story County Iowa
Aerial photo of Las Vegas, Clark County, Nv Such trends may result in future flow behavior that is different from that observed in the past.

 

Rough channel example

Significant basin alterations violate the requirement of homogeneity for flood frequency analysis.

Smooth urbanized channel example

Basin alterations include urbanization or other land use changes, water diversions, or the construction of reservoirs.

Illustration of runoff processes, rural vs. urban settings

Such changes could affect the behavior and frequency of floods. Consequently, flood frequency statistics generated prior to the basin changes no longer apply.

Flooding at Sonoma State University, Rohnert Park, CA 2005

This can have the effect of dramatically reducing the length of the period of record where homogeneous conditions exist. In other words, if you have a 100-year record, but major urbanization took place 20 years ago, then you really only have a 20-year homogeneous record for your now urbanized basin.

Return to Top of Page

Review Questions:

1. It is best to avoid estimating return period flood values that are greater than twice the length of the data record.
(Choose the best answer.)

a) True
b) False

2. To be within a 25 percent error for a 50-year flood estimate, the length of your data record should be at least _____. (Please refer to the table below.)
(Choose the best answer.)

a) 18 years
b) 39 years
c) 50 years
d) 110 years

Guidelines for length of data record vs. expected error rate

3. Land use changes within a watershed can violate the basic assumption of homogeneity and can change flow behavior within the stream.
(Choose the best answer.)

a) True
b) False

4. What is the probability of occurrence of a 25-year flood in the current year if that same magnitude flood just happened last year?
(Choose the best answer.)

a) Less than 0.01
b) 0.01
c) 0.04
d) 0.25

5. The chance of a flood of a specific return period not occurring is called the _____.
(Choose the best answer.)

a) non-exceedance probability
b) exceedance probability
c) null frequency probability
d) exceedance non-probability

6. Two large, but separate, peak flows occur at the same gauging station one day apart during the passing of a tropical storm remnant. The flow remained above baseflow during the minimum flow level between the peaks. Using both values in flood frequency analysis would violate the principle of _____.
(Choose the best answer.)

a) return period
b) homogeneity
c) confidence limits
d) independence

Return to Top of Page

Review Question Feedback

1. It is best to avoid estimating return period flood values that are greater than twice the length of the data record.
(Choose the best answer.)

a) True
b) False

The correct answer is a).

2. To be within a 25 percent error for a 50-year flood estimate, the length of your data record should be at least _____. (Please refer to the table below.)
(Choose the best answer.)

a) 18 years
b) 39 years
c) 50 years
d) 110 years

The correct answer is b).

Guidelines for length of data record vs. expected error rate

3. Land use changes within a watershed can violate the basic assumption of homogeneity and can change flow behavior within the stream.
(Choose the best answer.)

a) True
b) False

The correct answer is a).

4. What is the probability of occurrence of a 25-year flood in the current year if that same magnitude flood just happened last year?
(Choose the best answer.)

a) Less than 0.01
b) 0.01
c) 0.04
d) 0.25

The correct answer is c). The 25-year flood has a 1/25 = 0.04, or 4% chance of occurring in any given year.

5. The chance of a flood of a specific return period not occurring is called the _____.
(Choose the best answer.)

a) non-exceedance probability
b) exceedance probability
c) null frequency probability
d) exceedance non-probability

The correct answer is a).

6. Two large, but separate, peak flows occur at the same gauging station one day apart during the passing of a tropical storm remnant. The flow remained above baseflow during the minimum flow level between the peaks. Using both values in flood frequency analysis would violate the principle of _____.
(Choose the best answer.)

a) return period
b) homogeneity
c) confidence limits
d) independence

The correct answer is d).

Return to Top of Page

End of Section Two: Statistical Representation of Floods