Example 3 - How large a drought should we plan for?
Water managers should plan for very large, rare droughts, beyond the
historic or 50 year droughts estimated in the National Drought
Atlas. The logic for this is simple - ever rare droughts will
happen, so the agencies responsible for water planning should spend at
least a few hours considering how their stakeholders would be impacted.
This generally does not mean that managers must assure that every user
will receive a full supply of water during such extreme droughts; it
often makes more sense to accommodate rare droughts through temporary
curtailments of the least valuable uses of water. The public generally
understands this and responds well to drought curtailments so long as they
are not too frequent.
For that reason, managers may want to determine how frequent drought
curtailments would have to be imposed. Generally speaking, the Atlas
can help answer this question, but other modeling work is
required. If the manager can determine, from modeling, the
characteristic precipitation (amounts, durations) that would trigger a
declaration of drought measures, then the Atlas can be used to determine
the likelihood of that precipitation happening. The same is true if
the manager can relate the declaration to characteristic streamflow at a
gage included in the Atlas analysis.
Some of the steps involved in addressing this question are the same as in assessing the rarity of a
current drought. The principal difference is that the work progresses in the other direction. First, an
acceptable frequency of occurrence is set. Taking the same region of
Alabama (Cluster 105), if it were determined that the criterion should be failure
to meet a certain water use's requirements on the average of once in 20 years,
and that failure characteristically occurs when precipitation for June and
July is less than 75% normal, the analyst would download the Excel file
for Cluster 105, then select turn to the precipitation tables, read a value of 0.49, and multiply the
median precipitation in that area by 0.49. Similarly, if the important parameter were streamflow, one would
turn to the streamflow tables, and, for an appropriate gaging station, read the value for the two-month
period, and multiply this figure by the median streamflow for the station for that period.
If there were no appropriate gaged watershed, the analyst could convert the precipitation to runoff
or streamflow by some model. For information on watershed runoff modelling, see Computer Models for
Water Resources Planning and Management (U.S. Army, 1994). It summarizes brand name models in
eight categories: general purpose software (such as spreadsheets), municipal and industrial water use
forecasting, water distribution systems (pipe networks), groundwater, watershed runoff, stream hydraulics,
river and reservoir water quality, and river and reservoir system operations. The use of runoff
models, of course, adds a significant potential to distort the frequency estimates.
Models for relating precipitation to runoff for durations of one month or more could be based on
statistical or graphical correlation between precipitation and streamflow, or on a deterministic or quasi-deterministic model of some kind, of which Thornthwaite-Mather water balance is the most widely known
(although it is often of low accuracy). Deterministic models, at this time scale, cannot be expected to give
high precision, as they cannot take into account the daily and weekly variations that occur within the one-month period. Deterministic models designed for using time intervals as short as several hours up to one
day must necessarily make assumptions about how precipitation would be distributed during a month in
order to make a calculation. These assumptions might be supported by supplementary calculations
showing the distribution of the number of days of rain occurring in particular months, perhaps conditioned
by the total amount of precipitation occurring in that month.
In order to determine how large a drought to plan for, this basic process should be repeated for
other durations. It is not always obvious which duration will prove to be the most critical.
Additionally, the sensitivity of the planning or design frequency should be tested by using higher and lower frequency
values. It can be seen in some of the clusters and for some stream gaging stations that there is little
difference between the ratios for a 0.05 quantile and a 0.02 quantile, or even a 0.10 quantile and a 0.02
quantile. Where this is the case, designing or planning for the less frequent event may be desirable.
Some clusters display the opposite characteristic, of quite sharp variations between quantiles. In these
cases, if the proposed frequency is near the point where there may be sharp variations, the sensitivity
analysis may show not only the construction or operating costs of using a higher or lower frequency, but
also the societal costs of choosing a frequency that may be too high or low.
Return to Main Page, National Drought Atlas
revised 1 Aug 2006
Return to National Drought Atlas home