CE 410 - 5thWk - Lecture Note

2 Precipitation 17 Hydrology

2 PRECIPITATION
2.1 The atmosphere
The lower layer of the atmosphere, up to a height of approximately10 km, known as the
troposphere, is the most interesting part from a hydrologic point of view as it contains almost
all of the atmospheric moisture. The percentage of water in moist air is usually less than 4 %.
The composition of dry air is given in table 2.1.

Mass % Volume %
Nitrogen N2 75.5 78.1
Oxygen O2 23.1 20.9
Argon A 1.3 0.9
Others 0.1 0.1
Table 2.1. Composition of dry air.

The atmosphere depends for its heat content on the radiant energy from the sun. About one-
sixth of the available solar energy is directly absorbed by the atmosphere, more than one third
is reflected into space and less than one half is absorbed by the earth's surface. Heat from the
earth's surface is released to the atmosphere by conduction (action of molecules of greater
energy on those of less), convection (vertical interchange of masses of air) and radiation (long
wave radiation). The process of conduction does not play a significant role. Radiation is an
important link in the energy balance of the atmosphere. The "greenhouse effect", the increase
of the atmospheric temperature as a result of a decrease in the long wave radiation, has grown
into a world-wide concern. It is believed that man-produced gasses such as CO2, CH4 and N2O
obstruct the outgoing radiation, gradually leading to an accumulation of energy and hence a
higher atmospheric temperature. The effect of long wave radiation is further discussed in
Chapter 3.

Convection involves the vertical interchange of masses of air. The process is usually described
assuming a parcel of air with approximate uniform properties, moving vertically without
mixing with the surroundings. When a parcel of air moves upward it expands due to a decrease
of the external pressure. The energy required for the expansion causes the temperature to fall.
If the heat content of the air parcel remains constant, which is not uncommon (the parcel is
transparent and absorbs little radiant heat), the conditions are called adiabatic. The rate of
change of air temperature with height is known as the lapse rate. The average lapse rate is
0.65 oC per 100 m rise and varies from 1.0 under dry-adiabatic conditions to 0.56 under
saturated-adiabatic conditions. The lower value for the lapse rate of saturated air results from
the release of latent heat due to condensation of water vapour. Its value depends on the air
temperature.

The pressure of the air is often specified in mbar and generally taken equal to 1013 mbar or
1.013 bar at mean sea level. In this note the SI unit for pressure Pa (Pascal) will be used, where
1000 mbar = 1 bar = 100,000 Pa = 100 kPa. Hence the pressure at mean sea level is taken as
101.3 kPa. The atmospheric pressure decreases with the height above the surface. For the
lower atmosphere this rate is approximately 10 kPa per kilometre.
2 Precipitation 18 Hydrology

The water vapour

content of the air, or
humidity, is usually
measured as a water
vapour pressure
(kPa). The water
vapour pressure at
which the air is
saturated with water
vapour, the
saturation vapour
pressure of the air,
es is related to the
temperature of the
air as shown in
figure 2.1. If a
certain mass of air is
cooled while the
vapour pressure
remains constant, the
Fig. 2.1 Relation between saturation vapour pressure of the air es and air mass becomes
air temperature Ta saturated at a
temperature known
as dewpoint temperature, Td. The corresponding actual or dewpoint vapour pressure of the air
mass is indicated by ed. The vapour pressure deficit is defined as the difference between the
actual vapour pressure and the saturation vapour pressure that applies for the prevailing air
temperature Ta, thus (es - ed ). The relative humidity of the air is the ratio of actual and
saturation vapour pressure, or when expressed as a percentage

ed
RH = 100 (2.1)
es

2.2 Formation of precipitation

The conditions for precipitation to take place may be summarized stepwise as follows:

1. supply of moisture
2. cooling to below point of condensation
3. condensation
4. growth of particles

The supply of moisture is obtained through evaporation from wet surfaces, transpiration from
vegetation or transport from elsewhere. The cooling of moist air may be through contact with a
cold earth surface causing dew, white frost, mist or fog, and loss of heat through long wave
radiation (fog patches). However, much more important is the lifting of air masses under
adiabatic conditions (dynamic cooling) causing a fall of temperature to near its dew point.
2 Precipitation 19 Hydrology

Five lifting mechanisms can be distinguished:

1. Convection, due to vertical instability of the air. The air is said to be unstable if the
temperature gradient is larger than the adiabatic lapse rate. Consequently a parcel moving
up obtains a temperature higher than its immediate surroundings. Since the pressure on
both is the same the density of the parcel becomes less than the environment and buoyancy
causes the parcel to ascend rapidly. Instability of the atmosphere usually results from the
heating of the lower air layers by a hot earth surface and the cooling of the upper layers by
outgoing radiation. Convective rainfall is common in tropical regions and it usually
appears as a thunderstorm in temperate climates during the summer period. Rainfall
intensities of convective storms can be very high locally; the duration, however, is
generally short.

2. Orographic lifting. When air passes over a mountain it is forced to rise which may cause
rainfall on the windward slope. As a result of orographic lifting rainfall amounts are
usually highest in the mountainous part of the river basin.

3. Frontal lifting. The existence of an area with low pressure causes surrounding air to move
into the depression, displacing low pressure air upwards, which may then be cooled to dew
point. If cold air is replaced by warm air (warm front) the frontal zone is usually large and
the rainfall of low intensity and long duration. A cold front shows a much steeper slope of
the interface of warm and cold air usually resulting in rainfall of shorter duration and
higher intensity (see figure 2.2). Some depressions are died-out cyclones.

4. Cyclones, tropical depressions or hurricanes. These are active depressions, which gain
energy while moving over warm ocean water, and which dissipate energy while moving
over land or cold water. They may cause torrential rains and heavy storms. Typical
characteristics of these tropical depressions are high intensity rainfall of long duration
(several days). Notorious tropical depressions occur in the Caribbean (hurricanes), the Bay
of Bengal (monsoon depressions), the Far East (typhoons), Southern Africa (cyclones), and
on the islands of the Pacific (cyclones, willy-willies). This group of depressions is quite
different in character from other lifting mechanisms; data on extreme rainfall originating
from cyclones should be treated separately from other rainfall data, as they belong to a
different statistical population. One way of dealing with cyclones is through mixed
distributions (see Section 2.8).

5. Convergence. The Inter-Tropical Convergence Zone is the tropical region where the air
masses originating from the Tropics of Cancer and Capricorn converge and lift. In the
tropics, the position of the ITCZ governs the occurrence of wet and dry seasons. This
convergence zone moves with the seasons. In July, the ITCZ lies to the North of the
equator and in January it lies to the South (see figure 2.3). In the tropics the position of the
ITCZ determines the main rain-bringing mechanism which is also called monsoon. Hence,
the ITCZ is also called the Monsoon Trough, particularly in Asia. In certain places near to
the equator, such as on the coast of Nigeria, the ITCZ passes two times per year, causing
two wet seasons; near the Tropics of Capricorn and Cancer (e.g. in the Sahel), however,
there is generally only one dry and one wet season.
2 Precipitation 20 Hydrology

Condensation of water
vapour into small
droplets does not occur
immediately when the
air becomes saturated. It
requires small airborne
particles called
aerosols, which act as
nuclei for water vapour
to condense. Nuclei are
salt crystals from the
oceans, combustion
Fig. 2.2 Frontal lifting products, dust, ash, etc.
In the absence of
sufficient nuclei the air becomes supersaturated.

When the temperature drops below zero, freezing nuclei with a structure similar to ice are
needed for condensation of water vapour into ice crystals. These nuclei are often not available
and the droplets become super-cooled to a temperature of -40 oC.

Cloud droplets with a diameter of 0.01

mm need an updraught of only 1 cm.s-1
to prevent the droplet from falling. A
growth of particles is necessary to
produce precipitation. This growth
may be achieved through coalescence,
resulting from collisions of cloud
droplets. Small droplets moving
upwards through the clouds collide
and unite with larger droplets which
have a different velocity or move
downwards. Another method of
growth known as the Bergeron process
is common in mixed clouds, which
consist of super-cooled water droplets
and ice crystals. Water vapour
condenses on the ice crystals, the
deficit being replenished by the
evaporation of numerous droplets. In
the temperature range from -12 oC to -
30 oC the ice crystals grow rapidly and
fall through the clouds. Aggregates of
many ice crystals may form
snowflakes. Depending on the
temperature near the surface they may
reach the ground as rain, snow, hail or
glazed frost. Fig. 2.3 Position of the Inter-Tropical Convergence
Zone in January (top) and July (below)
2 Precipitation 21 Hydrology

2.3. Rainfall measurement

The standard raingauge as depicted in figure 2.4 is used for daily readings. The size of the
aperture and the height varies between countries but is usually standardized within each
country (The Netherlands: aperture 200 cm2, height 40 cm).

The requirements for gauge construction are:

1. The rim of the collector should have a sharp edge.

2. The area of the aperture should be known with an accuracy of 0.5 %.
3. Design is such that rain is prevented from splashing in or out.
4. The reservoir should be constructed so as to avoid evaporation.
5. In some climates the collector should be deep enough to store one day's snowfall.

Wind turbulence affects the

catch of rainfall. Experiments
in the Netherlands using a
400 cm2 raingauge have shown
that at a height of 40 cm the
catch is 3-7 % less than at
ground level and as much as 4-
16 % at a height of 150 cm.
Tests have shown that
raingauges installed on the roof
of a building may catch Fig. 2.4 Raingauges
substantially less rainfall as a
result of turbulence (10-20%).

Wind is probably the most important factor in rain-gauge accuracy. Updrafts resulting from air
moving up and round the instrument reduce the rainfall catch. Figure 2.5 shows the effect of
wind speed on the catch according to Larson & Peck (1974). To reduce the effects of wind,
raingauges can be provided with windshields. Moreover, obstacles should be kept far from the
rainguage (distance at least twice the height of such an object) and the height of the gauge
should be minimised (e.g. ground-level raingauge with screen to prevent splashing).

Recording gauges are of three different types:

the weighing type, the tilting bucket type and
the float type. The weighing type observes
precipitation directly when it falls (including
snow) by recording the weight of the reservoir
e.g. with a pen on a chart. With the float type
the rain is collected in a float chamber. The
vertical movement of the float is recorded by
pen on a chart. Both types have to be emptied
manually or by automatic means. The tilting or
tipping bucket type (figure 2.4) is a very simple
recording raingauge, but less accurate (only
registration when bucket is full, losses during
tipping and losses due to evaporation).
Fig. 2.5 Effect of wind speed on rain catch
2 Precipitation 22 Hydrology

HYETOGRAPH MASS CURVE

18
6
16

Accumulated depth in mm
Precipitation in mm/hour

5 14
12
4
10
3 8

2 6
4
1
2
0 0
1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
April 4, 2006 April 4, 2006

Fig. 2.6a Hyetograph Fig. 2.6b Corresponding mass curve

Storage gauges for daily rainfall measurement are observed at a fixed time each morning.
Recording gauges may be equipped with charts that have to be replaced every day, week or
month, depending on the clockwork. The rainfall is usually recorded cumulatively (mass
curve) from which the hyetograph (a plot of the rainfall with time) is easily derived (see figure
2.6). The tipping bucket uses an electronic counter or magnetic tape to register the counts
(each count corresponds to 0.2 mm), for instance, per 15 minutes time interval.

Rainfall measurement by radar and satellite

Particularly in remote areas or in areas where increased spatial or time resolution is required
radar and satellite measurement of rainfall can be used to measure rainfall. Radar works on the
basis of the reflection of an energy pulse transmitted by the radar which can be elaborated into
maps that give the location (plan position indicator, PPI) and the height (range height
indicator, RHI) of the storms (see figure 2.7, after Bras, 1990).

A technology, which has a large future, is rainfall monitoring through remote sensing by
satellite. Geostationary satellites that orbit at the same velocity as the earth’s rotation are able
to produce a film of weather development with a time interval between observations of several
minutes. The resolution of the images is in the order of 1 km. This allows to monitor closely
the development of convective storms, depressions, fronts, orographic effects and tropical
cyclones.

Fig. 2.7 Forms of radar display: PPI (left) and RHI (right)
2 Precipitation 23 Hydrology

Successful efforts have been made to correlate rainfall with Cold Cloud Coverage (CCC) and
Cold Cloud Duration (CCD) through simple mathematical regression. Particularly in remote
areas the benefits of such methods are obvious.

Parameters defining rainfall

When discussing rainfall data the following elements are of importance:
1. Intensity or rate of precipitation: the depth of water per unit of time in m/s, mm/min, or
inches/hour (see Section 2.4).
2. Duration of precipitation in seconds, minutes or hours (see Section 2.4).
3. Depth of precipitation expressed as the thickness of a water layer on the surface in mm or
inches (see Section 2.4).
4. Area, that is the geographic extent of the rainfall in km2 (see Section 2.5).
5. Frequency of occurrence, usually expressed by the 'return period' e.g. once in 10 years (see
Section 2.7).

2.4. Intensity, duration and rainfall depth

We have seen that rainfall storms can be distinguished by their meteorological characteristics
(convective, orographic, frontal, cyclonic). Another way to characterize storms is from a
statistical point of view. Distinction is made between "interior" and "exterior" statistics:

 exterior statistics refer to total depth of the storm, duration of the storm, average intensity
of the storm and time between storms;
 interior statistics refer to the time and spatial distribution of rainfall rate within storms.

The exterior statistics can generally be described by probabilistic distributions, which can be
seasonally and spatially dependent. They are generally not statistically independent. Depth is
related to duration: a long duration is
associated with a large depth;
average high intensity is related to
short duration. Moreover there is
spatial correlation between points.

With regard to storm interior,

Eagleson (1970) observed that for
given locations and climatic
conditions some type of storms gave
similar histories of rainfall
accumulation. The percentage-mass
curve of figure 2.8 shows how typical
curves are obtained for
thunderstorms (convective storms)
and cyclonic storms. The derivative
(slope) of these curves is a graph for
the rainfall intensity over time. Such
a graph is referred to as the Fig. 2.8 Typical percentage mass curves of rainfall for
hyetograph and is usually presented thunderstorms and cyclones
in histogram form (see figure 2.6). It
2 Precipitation 24 Hydrology

can be seen from figure 2.8 that convective storms have more or less triangular (moundlike)
hyetographs with the highest intensity at the beginning of the storm (a steep start), whereas
tropical cyclones have a bell-shaped hyetograph with the largest intensity in the middle of the
storm. It should be observed that tropical cyclones have a much longer storm duration and a
much larger rainfall depth than thunderstorms.

Another interior characteristic of a storm is the spatial distribution. The spatial distribution of a
storm generally concentrates around one or two centres of maximum depth. The total depth of
point rainfall distributed over a given area is a decreasing function of the distance from the
storm centre. One can draw lines of equal rainfall depth around the centres of maximum depth
(isohyets).

When two rainfall stations are closely together, data from these stations may show a good
correlation. The further the stations lay apart, the smaller the chances of coincidence become.
In general, correlation is better when the period of observation is larger. For a given period of
observation, the correlation between two stations is defined by the correlation coefficient ( -
1 < < 1). If there is no correlation, is near to zero, if the correlation is perfect, = 1. It is
defined by:

cov (x, y)
 = (2.2)
(x) (y)

For a number of stations in a certain area the correlation coefficient can be determined
pairwise. In general, the larger the distance, the smaller the correlation will be. If r is defined
as the distance between stations, then the correlation between stations at a distance r apart is
often described by Kagan's formula:

r

 (r) =  0 e r0
(2.3)

This is an exponential function which

1
equals o at r = 0 and which decreases
0.8 Very local convective
gradually as r goes to infinity. The
Correlation coefficient

Mixed convective orographic distance ro is defined as the distance

0.6 where the tangent at x = 0 intersects the x-
axis (see figure 2.9). For convective
0.4 storms, the steepness is very large and the
value of ro is small; for frontal or
0.2
orographic rains, the value is larger. Also
the period considered is reflected in the
0
0 20 40 60 80 100 steepness; for a short period of observation
Distance between 2 stations (km)
ro is small. Table 2.2 gives indicative
values of ro and o .
Fig. 2.9 Spatial correlation of daily rainfalls
2 Precipitation 25 Hydrology

Period 1 hour Period 1 day Period 1 month

Rain type ro o ro o ro o
(km) (km) (km)
Very local convective 5 0.80 10 0.88 50 0.95
Mixed convective orographic 20 0.85 50 0.92 1500 0.98
Frontal rains from depressions 100 0.95 1000 0.98 5000 0.99
Table 2.2 Typical spatial correlation structure for different storms

2.5 Areal rainfall

In general, for engineering purposes,
knowledge is required of the average
rainfall depth over a certain area: the
areal rainfall. Some cases where the
areal rainfall is required are: design of a
culvert or bridge draining a certain
catchment area, design of a pumping
station to drain anFigure 1 urbanized
area; design of a structure to drain a
polder, etc.

There are various methods to estimate

the average rainfall over an area (areal Fig. 2.10 Thiessen polygons
rainfall) from point-measurements.

1. Average depth method. The arithmetic mean of the rainfall amounts measured in the area
provides a satisfactory estimate for a relatively uniform rain. However, one of the
following methods is more appropriate for mountainous areas or if the raingauges are not
evenly distributed.

2. Thiessen method. Lines are drawn to connect reliable rainfall stations, including those just
outside the area. The connecting lines are bisected perpendicularly to form a polygon
around each station (see figure 2.10). To determine the mean, the rainfall amount of each
station is multiplied by the area of its polygon and the sum of the products is divided by
the total area.

3. Kriging. D.G. Krige, a mining expert, developed a method for interpolation and averaging
of spatially varying information, which takes account of the spatial variability and which -
unlike other methods - can also indicate the level of accuracy of the estimates made. The
Kriging weights obtained are tailored to the variability of the phenomenon studied. Figure
2.11 shows the comparison between weights obtained through Thiessen's method (a) and
Kriging (b).

4. Isohyetal method. Rainfall observations for the considered period are plotted on the map
and contours of equal precipitation depth (isohyets) are drawn (figure 2.12). The areal
rainfall is determined by measuring the area between isohyets, multiplying this by the
average precipitation between isohyets, and then by dividing the sum of these products by
the total area.
2 Precipitation 26 Hydrology

Fig. 2.11 Comparison of Thiessen (a) and Kriging (b) Fig. 2.12 Isohyetal method
methods

As a result of the averaging process, and depending on the size of the catchment area, the areal
rainfall is less than the point rainfall. The physical reason for this lies in the fact that a
rainstorm has a limited extent. The areal rainfall is usually expressed as a percentage of the
storm-centre value: the areal reduction factor (ARF). The ARF is used to transfer point
rainfall Pp extremes to areal rainfall Pa :

ARF = Pa / Pp (2.4)

Basically the ARF is a function of:

 rainfall depth
 storm duration
 storm type
 catchment size
 return period (see Section 2.7).

The ARF increases (comes nearer to unity) with increasing total rainfall depth, which implies
higher uniformity of heavy storms. It also increases with increasing duration, again implying
that long storms are more uniform. It decreases with the area under consideration, as a result of
the storm-centred approach.

Storm type varies with location, season and climatic region. Published ARF's are, therefore,
certainly not generally applicable. From the characteristics of storm types, however, certain
conclusions can be drawn.

A convective storm has a short duration and a small areal extent; hence, the ARF decreases
steeply with distance. A frontal storm has a long duration and a much larger area of influence;
the ARF, hence, is expected to decrease more slowly with distance. The same applies to
orographic lifting. Cyclones also have long durations and a large areal extent, which also leads
to a more gradual reduction of the ARF than in the case of thunderstorms. In general, one can
2 Precipitation 27 Hydrology

say that the ARF-curve is

steepest for a convective
storm, that a cyclonic storm
has a more moderate slope
and that orographic storms
have an even more
moderate slope.

The functional relationship

between the ARF and
return period is less clear.
Bell (1976) showed for the
United Kingdom that ARF
decreased more steeply for
rainstorms with a high
return period. Similar
findings are reported by
Begemann (1931) for
Indonesia. This is,
however, not necessarily so
in all cases. If widespread
cyclonic disturbances,
instead of more local
convective storms,
constitute the high return
Fig. 2.13 The areal reduction factor as a function of drainage area period rainfall, the opposite
and duration (after NEDECO et al., 1983) may be true.

Again, it should be observed that cyclones belong to a different statistical population from
other storm types, and that they should be treated separately. If cyclones influence the design
criteria of an engineering work, then one should consider a high value of the ARF.

Figure 2.13, as an example, shows ARF's as a function of catchment size and rainfall duration
for Bangkok by Nedeco (1983), and Camp, Dresser and McKee (1968), and for Jakarta by
Nedeco (1973).

2.6 Rainfall data screening

Of all hydrological data, rainfall is most readily available. Though quantitatively abundant its
quality should not, a priori, be taken for granted, despite the fact that precipitation data are
easily obtained. Several ways of specific rainfall data screening are available besides the ones
described in Section 1.5.

daily rainfall

 tabular comparison, maximum values check

 time series plotting and comparison
 spatial homogeneity test
2 Precipitation 28 Hydrology

monthly rainfall

 tabular comparison, maximum, minimum, P80, P20 check

 time series plotting
 spatial homogeneity test
 double mass analysis

tabular comparison
In a table a number of checks can be carried out to screen rainfall data. One way is through
internal operations and another through comparison with other tables. Internal operations are
the computation of the maximum, minimum, mean and standard deviation for columns or
ranges in the table. Spreadsheet programs are excellent tools for this purpose. A table of daily
rainfall organised in monthly columns can be used to compute the monthly sums, the annual
total, and the annual maximum. Comparison of these values with those of a nearby station
could lead to flagging certain information as doubtful. A table showing monthly totals for
different years can be used for statistical analysis. In many cases the normal or lognormal
distribution fits well to monthly rainfall. In such a case the computation of the mean and
standard deviation per month (per column) can be used to compute the rainfall with a
probability of non-exceedence of 20% (a dry year value) or 80% (a wet year value) through the
following simple formula:

P20 = Mean - 0.84 * Std

P80 = Mean + 0.84 * Std

If the lognormal distribution performs better (as one would expect since rainfall has a lower
boundary of 0), than the values of mean and standard deviations should be based on the
logarithms of the recorded rainfall: Lmean, and Lstd. The values of P20 and P80 then read:

P20 = Exp (Lmean - 0.84 * Lstd)

P80 = Exp (Lmean + 0.84 * Lstd)

If individual monthly values exceed these limits too much, they should be flagged as suspect,
which does not, a priori, mean that the data are wrong; it is just a sign that one should
investigate the data in more detail.

double mass
The catch of rainfall for stations in the same climatic region is, for long-time periods, closely
related. An example showing the mass curves of cumulative annual precipitation of two nearby
stations A and B is given in figure 2.14a. The double mass curve of both stations plots
approximately as a straight line (see figure 2.14b). A deviation from the original line indicates
a change in observations in either station A or B (e.g. new observer, different type of
raingauge, new site, etc.). This is called a spurious (= false) trend, not to be mistaken for a real
trend, which is a gradual change of climate. If the cumulative annual rainfall data of station X
are plotted against the mean of neighbouring stations the existence of a spurious trend
indicates that the data of station X are inconsistent (figure 2.15). When the cause of the
discrepancy is clear, the monthly rainfall of station X can be corrected by a factor equal to the
proportion of the angular coefficients.
2 Precipitation 29 Hydrology

time series plotting

Although these
tabular operations
can help to
identify possible
sources of errors,
there is nothing as
powerful to
pinpoint suspect
data as a graphical
plot. In a
spreadsheet a
graphical plot is Fig. 2.14 Mass curves of individual stations A and B (left) and double mass
easy to perform. curve of stations A and B (right)
One can use a bar-
graph for a simple
time plot, a stacked bar-graph to view
two combined sets, or an X -Y relation
to compare one station with another
nearby station. The latter is a powerful
tool to locate strange values.

spatial homogeneity
In the spatial homogeneity test a base
station is related to a number of
surrounded stations. A maximum
distance rmax is defined on the basis
Fig. 2.15 Double mass curve showing apparent trend
of (2.3) beyond which no significant for station X
correlation is found (e.g. < 0.75 or
0.5). To investigate the reliability of
point rainfall, the observed rainfall is compared with an estimated rainfall depth on the basis of
spatial correlation with others (computed from a weighted average through multiple
regression).
A formula often used for the weighted average is by attributing weights that are inversely
proportional to the square of the distance from the station in question:

P
 b 
 r i
Pest  (2.5)
 1
 b 
 r i

where b is the power for the distance, which is often taken as 2, but which should be
determined from experience. If the observed and the estimated rainfall differ more than an
acceptable error criterion, both in absolute and relative terms, then the value should be further
scrutinized, or may have to be corrected (see Workshop on Hydrology).
2 Precipitation 30 Hydrology

Fequency distribution Fequency distribution rainfall

Daily rainfall data Wageningen (1972-1990) Wageningen, The Netherlands (1972-1990)

60
60

Fequency of occurrence (%)

Fequency of occurrence (%)

40
40 K=1
K = 10
K = 30
20
20

0 0
0 1 2 3 4 5 6 7 8 9 10 0 2 4 6 8 10
Rainfall in mm/d Rainfall in mm/d

Fig. 2.16a Frequency distribution daily Fig. 2.16b Frequency distribution of average rainfall data
rainfall data for Wageningen with a K-day duration, Wageninegn

2.7 Frequency analysis

Consider daily rainfall data over a period of many years and compute the percentage of days
with rainfall between 0 - 0.5 mm, 0.5 - 1.0 mm, 1.0 - 1.5 mm, etc. The frequency of occurrence
of daily rainfall data may then be plotted. An example for a station in the Netherlands is given
in figure 2.16A. It shows that the frequency distribution is extremely skew, as days without
rain or very little rain occur most frequently and high rainfall amounts are scarce. The
distribution becomes less skew if the considered duration (k) is taken longer. Table 2.3 gives
an example of the derivation of rainfall data for longer durations (k = 2, 5 and 10 days) from
daily values for a record length of 15 days. An example of frequency distributions of average
rainfall data for k = 1, 10 and 30 days for a rainfall station in the Netherlands is given in figure
2.16B.

no. 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
k
1 0 0 2 8 0 0 0 0 0 12 3 8 24 2 0
2 0 2 10 8 0 0 0 0 12 15 11 32 26 2
5 10 10 10 8 0 12 15 23 47 49 37
10 22 25 33 55 49 49

Table 2.3 Derivation of rainfall data for duration (k) of 2, 5 and 10 days from 15
daily values

For the design of drainage systems, reservoirs, hydraulic works in river valleys, irrigation
schemes, etc. knowledge of the frequency of occurrence of rainfall data is often essential.

The type of data required depend on the purpose; for the design of an urban drainage system
rainfall intensities in the order of magnitude of mm/min are used, while for
agricultural areas the frequency of occurrence of rainfall depths over a period of several days is
more appropriate.
2 Precipitation 31 Hydrology

When dealing with extremes it is usually convenient to refer to probability as the return period
that is, the average interval in years between events which equal or exceed the considered
magnitude of event. If p is the probability that the event will be equalled or exceeded in a
particular year, the return period T may be expressed as

1
T = (2.6)
p

One should keep in mind that a return period of a certain event e.g. 10 years does not imply
that the event occurs at 10 year intervals. It means that the probability that a certain value
(e.g. a rainfall depth) is exceeded in a certain year is 10%. Consequently the probability that
the event does not occur (the value is not exceeded) has a probability of 90%. The probability
that the event does not occur in ten consecutive years is (0.9)10 and thus the probability that it
occurs once or more often in the ten years period is 1 - (0.9)10 = 0.65; i.e. more than 50%.
Similarly, the probability that an event with a return period of 100 years is exceeded in 10
years is 1 - (0.99)10 = 10%. This probability of 10% is the actual risk that the engineer takes if
he designs a structure with a lifetime of 10 years using a design criterion of T = 100. In
general, the probability P that an event with a return period T actually occurs (once or more
often) during an N years period is:

N
 1
P = 1 - 1 -  (2.7)
 T

A frequency analysis to derive intensity-duration-frequency curves requires a length of record

of at least 20 - 30 years to yield reliable results. The analysis will be explained with a
numerical example, using hypothetical data. Consider a record of 50 years of daily precipitat-
ion data, thus 365.25 * 50 = 18262 values (provided there are no missing or unreliable data).
Compute the number of days with rainfall events greater than or equal to 0, 10, 20, 30,
etc. mm. For this example large class intervals of 10 mm are used to restrict the number of
values. In practice a class interval of 0.5 mm (as in fig. 2.16a) or even smaller is more
appropriate.

Table 2.4 shows that all 18262 data are greater than or equal to zero (of course), and that there
is only one day on which the amount of rainfall equalled or exceeded 40 mm. Similar to the
procedure explained in table 2.3, rainfall data are derived for k = 2, 5 and 10 days. For k = 2
this results in 18261 values which are processed as for the daily rainfall data, resulting in two
values with 60 mm or more. The results for k = 2, 5 and 10 are also presented in table 2.4 The
data are used to construct duration curves as shown in figure 2.18.

The procedure is the following. Consider the data for k = 1 which show that on one day in 50
years only the amount of rainfall equals or exceeds 40 mm, hence the probability of occurrence
in any year is P = 1/50 = 0.02 or 2 % and the return period T = 1/P = 50 years. This value is
plotted in figure 2.17. Similarly for rainfall events greater than or equal to 30 mm, P = 5/50 =-
0.1 or T = 10 years and for 20 mm, P = 48/50 = 0.96 or T  1 year. Figure 2.17 shows the
resulting curves for a rainfall duration of one day as well as for k = 2, 5 and 10 days.
2 Precipitation 32 Hydrology

Cumulative frequency or duration

Cumulative frequency curves for
curves often approximate a straight
line when plotted as in figure 2.17 on different durations
250
semi-logarithmic graph paper, which k=1
facilitates extrapolation. Care should 200 k=2

Rainfall depth in mm
be taken with regard to extrapolation k=5
150 k = 10
of frequency estimates, in particular if
the return period is larger than twice 100
the record length.
50
Figure 2.17 shows that for durations
of 1, 2, 5 and 10 days the rainfall 0
1 10 100
events equal or exceed respectively
Logarithm Return Period
20, 30, 60 and 100 mm with a return
period of one year. These values are
plotted in figure 2.18 to yield the Fig. 2.17 Cumulative frequency curves for different
durations (k = 1, 2, 5 and 10 days).
depth-duration curve with a frequency

Depth-Duration-Frequency curves Depth-Duration-Frequency curves

Linear scale Logarithmic scale
250 2.4
Log Rainfall depth (mm)

T=1 T=1
Rainfall depth (mm)

200 2.2
T = 10 T = 10
T = 100 2 T = 100
150
1.8
100
1.6
50 1.4
0 1.2
0 2 4 6 8 10 0 0.2 0.4 0.6 0.8 1 1.2
Duration (days) Log Duration (days)

Fig. 2.18a Depth-Duration-Frequency curves Fig. 2.18b Double logarithmic plot of DDF curves

Intensity-Duration-Frequency curves Intensity-Duration-Frequency curves

Linear scale Logarithmic scale
Log Rainfall intensity (mm/d)

50 1.7
Rainfall intensity (mm/d)

T=1 1.6 T=1

40 T = 10 T = 10
1.5
T = 100 1.4 T = 100
30
1.3
20 1.2
1.1
10
1
0 0.9
0 2 4 6 8 10 0 0.2 0.4 0.6 0.8 1 1.2
Duration (days) Log Duration (days)

Fig. 2.19a Intensity-Duration-Frequency Fig. 2.19b Double logarithmic plot of IDF curves
curves
2 Precipitation 33 Hydrology

of T = 1 year. Similarly the curves for T = 10 and Class

the extrapolated curve for T = 100 years are interval 1 2 5 10
(mm)
constructed. On double logarithmic graph paper 0 18262 18261 18258 18253
these curves often approximate a straight line. 10 384 432 730 2001
Dividing the rainfall depth by the duration yields 20 48 127 421 1539
30 5 52 243 713
the average intensity. This procedure is used to 40 1 12 158 493
convert the depth-duration-frequency curves into 50 5 83 286
intensity-duration-frequency curves (figure 2.19). 60 2 49 221
70 25 170
80 16 96
For the above analysis all available data have 90 9 76
been used, i.e. the full series. As the interest is 100 5 49
110 3 31
limited to relatively rare events, the analysis 120 1 22
could have been carried out for a partial duration 130 1 16
series, i.e. those values that exceed some arbitrary 140 9
150 7
level. For the analysis of 160 5
extreme precipitation amounts a series, which is 170 4
made up of annual extremes, the annual 180 2
190 2
series, is favoured as it provides a good theoreti- 200 1
cal basis for extrapolating the series beyond the
period of observation. The partial duration series
overcomes an objection to the annual series since Table 2.4 Totals of k-day periods
it is made up of all the large events above a given with rainfall greater than the bottom of the
class interval for k = 1, 2, 5 and 10 days
base, thus not just the annual extreme.

Langbein (see Chow, 1964) developed a theory for partial duration series which considers all
rainfall exceeding a certain threshold. The threshold is selected in such a way that the number
of events exceeding the threshold equals the number of years under consideration. Then,
according to Langbein, the relationship between the return period of the annual extremes T and
return period of the partial series Tp is approximately:

1

1
p   1 e p
T
(2.8)
T

Figure 2.20 shows the comparison between the frequency distribution of the extreme hourly
rainfall in Bangkok computed with partial duration series (= annual exceedences) and with
annual extremes. A comparison of the two series shows that they lead to the same results for
larger return periods, say T > 10 years. Hershfield (1961) proposes to multiply the rainfall
depth obtained by the annual extremes method by 1.13, 1.04, and 1.01 for return periods of 2,
5 and 10 years respectively.

The analysis of annual extreme precipitation is illustrated with the following numerical
example. For a period of 10 years the maximum daily precipitation in each year is listed in
table 2.5. For convenience a (too) short period of 10 years is considered. Rank the data in
descending order (see table 2.6). Compute for each year the probability of exceedence using
the formula

m
p = (2.9)
N + 1  2
2 Precipitation 34 Hydrology

Fig. 2.20 Comparison between the method of annual exceedences and annual extremes for
extreme hourly rainfall in Bangkok

where N is the number of years of record and m is the rank number of the event and  a
parameter less than one. Equation (2.9) is also known as the plotting position. In the example
above  is taken equal to zero (Weibull formula). For the Gumbel distribution the Gringerton
formula with  = 0.44 is often used (see Cunnane, 1977). The return period T is computed as
T = 1/p and is also presented in table 2.6. A plot of the annual extreme precipitation versus the
return period on linear paper does not yield a straight line as shown in figure 2.21. Using semi-
logarithmic graph paper may improve this significantly as can be seen in figure 2.22.

Annual rainfall extremes tend to plot as a straight line on extreme-value-probability (Gumbel)

paper (see figure 2.23). The extreme value theory of Gumbel is only applicable to annual
extremes. The method uses, in contrast to the previous example, the probability of non-
exceedence q = 1 - p (the probability that the annual maximum daily rainfall is less than a
certain magnitude). The values are listed in table 2.6.

Gumbel makes use of a reduced variate y as a function of q, which allows the plotting of the
distribution as a linear function between y and X (the rainfall depth in this case).

y = a (X-b) (2.10)
 2 Precipitation 35 Hydrology

Year 1971 72 73 74 75 76 77 78 79 80
56 52 60 70 34 30 44 48 40 38

Table 2.5. Annual maximum daily rainfall amounts.

Rainfall p T q y
Rank amount Probability Return Probability Reduced
(mm) of exceedence period Log T of non-exc. variate
1 70 0.09 11.0 1.041 0.91 2.351
2 60 0.18 5.5 0.740 0.82 1.606
3 56 0.27 3.7 0.564 0.73 1.144
4 52 0.36 2.8 0.439 0.64 0.794
5 48 0.45 2.2 0.342 0.55 0.501
6 44 0.54 1.8 0.263 0.46 0.238
7 40 0.64 1.6 0.196 0.36 -0.012
8 38 0.73 1.4 0.138 0.27 -0.262
9 34 0.82 1.2 0.087 0.18 -0.533
10 30 0.91 1.1 0.041 0.09 -0.875

Table 2.6 Rank, probability of exceedence, return period and reduced variate for the data in table 2.5

The equation for the reduced variate reads:

y = - ln ( - ln ( q )) = - ln ( - ln ( 1 - p )) (2.11)

meaning that the probability of non-exceedence equals:

P ( X  X0 ) = q = exp( - exp(- y )) (2.12)

Fig. 2.21 Annual maximum daily rainfall Fig. 2.22 Annual maximum daily rainfall
(linear plot). (semi-log plot).
2 Precipitation 36 Hydrology

Fig. 2.23 Gumbel paper

The computed values of y for the data in table 2.5 are presented in table 2.6. A linear plot of
these data on extreme-value-probability paper (see figure 2.24) is an indication that the
frequency distribution fits the extreme value theory of Gumbel. This procedure may also be
applied to river flow data.

In addition to the analysis of maximum

extreme events, there also is a need to analyze
minimum extreme events; e.g. the occurrence
of droughts. The probability distribution of
Gumbel, similarly to the Gaussian probability
distribution, does not have a lower limit;
meaning that negative values of events may
occur. As rainfall or river flow do have a lower
limit of zero, neither the Gumbel or Gaussian
distribution is an appropriate tools to analyze
minimum values. Because the logarithmic
function has a lower limit of zero, it is often
useful to first transform the series to its
logarithmic value before applying the theory. Fig. 2.24 Annual daily rainfall (Gumbel
Appropriate tools for analyzing minimum distribution).
flows or rainfall amounts are the Log-Normal,
Log-Gumbel, or Log-Pearson distributions.
2 Precipitation 37 Hydrology

A final remark of caution should be Year Max. Daily Thunderstorm (T)

Rainfall or Cyclone (C)
made with regard to frequency analysis.
None of the above mentioned frequency 1952/53 155.3 T
1953/54 87.1 T
distributions has a real physical 1954/55 103.8 T
1955/56 150.6 T
background. The only information 1956/57 55.3 T
having physical meaning are the 1957/58 46.9 T
1958/59 70.3 T
measurements themselves. 1959/60 54.2 T
1960/61 71.2 T
Extrapolation beyond the period of 1961/62 75.0 T
observation is dangerous. It requires a 1962/63 108.2 T
1963/64 61.2 T
good engineer to judge the value of 1964/65 50.5 T
1965/66 228.6 C (Claud)
extrapolated events of high return 1966/67 90.0 T
periods. A good impression of the 1967/68 64.6 T
1968/69 83.3 T
relativity of frequency analysis can be 1969/70 71.5 T
1970/71 51.3 T
acquired through the comparison of 1971/72 104.4 T
results obtained from different statistical 1972/73 105.4 T
1973/74 103.5 T
methods. Generally they differ 1974/75 92.0 T
1975/76 189.3 C (Danae)
considerably. And finally, in those 1976/77 130.0 C (Emilie)
tropical areas where cyclonic disturban- 1977/78 71.8 T
1978/79 84.6 T
ces occur, one should not be misled by a 1979/80 77.7 T
1980/81 97.1 T
set of data in which the extreme event 1981/82 53.5 T
has not yet occurred at full force. The 1982/83 56.9 T
1983/84 107.0 C (Demoina)
possibility always exists that the cyclone 1984/85 89.2 T
passes right over the centre of the study
area. If that should occur, things may
Table 2.7 Maximum annual precipitation as a
happen that go far beyond the hitherto result of cyclonic storms and
registered events. thunderstorms

2.8 Mixed distributions

We have seen in Section 2.2 that there are four types of lifting mechanisms that cause quite
distinct rainfall types with regard to depth, duration and intensity. Of these, the tropical
cyclones, or a combination of tropical depressions with orographic effects are the ones which,
by far, exceed other lifting mechanisms with regard to rainfall depth and intensity. As a result,
such occurrences, which are generally rare, should not be analyzed as if they were part of the
total population of rainfall events; they belong to a different statistical population. However,
our rainfall records contain events of both populations. Hence, we need a type of statistics that
describes the occurrence of extreme rainfall on the basis of a mixed distribution of two
populations: say the population of cyclones and the population of non-cyclones, or in the
absence of cyclones, the population of orographic lifting and the population of other storms.
Such a type of statistics is presented in this section. Assume that the rainfall occurrences can
be grouped in two sub-sets: the set of tropical cyclone events which led to an annual maximum
daily rainfall C and the set of non-cyclone related storms that led to an annual extreme daily
rainfall T. Given a list of annual maximum rainfall events, much in the same way as one would
prepare for a Gumbel analysis, one indicates whether the event belonged to subset C or T (see
table 2.7). The probability P(C) is defined as the number of times in n years of records that a
cyclone occurred. In this case P(C) = 4/33. The probability P(T), similarly, equals 29/33. The
combined occurrence P(X > X0), that the stochast representing annual rainfall X is larger than
a certain value X0 is computed from:
2 Precipitation 38 Hydrology

PX  X 0   PX  X 0 | C  PC  PX  X 0 | T  PT (2.13)

where:
P ( X > X0 | C ) is the conditional probability that the rainfall is more than X0 given
that the rainfall event was a cyclone;
P ( X > X0 | T ) is the conditional probability that the rainfall is more than X0 given that
the rainfall event was a thunder storm.

In figure 2.25, representing

the log-normal distribution,
three lines are
distinguished. One straight
line representing the log-
normal distribution of
thunderstorms, one straight
line the log-normal
distribution of the cyclones,
and one curve which goes
asymptotically from the
thunderstorm distribution to
the cyclone distribution
representing the mixed
distribution. The data plots
can be seen to fit well to the
mixed distribution curve.
Fig. 2.25 Mixed distribution of combined cyclonic storms and
thunderstorms

2.9 Probable maximum precipitation

In view of the uncertainty involved in frequency analysis, and its requirement for long series of
observations which are often not available, hydrologists have looked for other methods to
arrive at extreme values for precipitation. One of the most commonly used methods is the
method of the Probable Maximum Precipitation (PMP).

The idea behind the PMP method is that there must be a physical upper limit to the amount of
precipitation that can fall on a given area in a given time. An accurate estimate is both
desirable from an academic point of view and virtually essential for a range of engineering
design purposes, yet it has proven very difficult to estimate such a value accurately. Hence the
word probable in PMP; the word probable is intended to emphasize that, due to inadequate
understanding of the physics of atmospheric processes, it is impossible to define with certainty
an absolute maximum precipitation. It is not intended to indicate a particular level of statistical
probability or return period.

The PMP technique involves the estimation of the maximum limit on the humidity
concentration in the air that flows into the space above a basin, the maximum limit to the rate
at which wind may carry the humid air into the basin and the maximum limit on the fraction of
2 Precipitation 39 Hydrology

the inflowing water vapour that can be precipitated. PMP estimates in areas of limited
orographic control are normally prepared by the maximization and transposition of real,
observed storms while in areas in which there are strong orographic controls on the amount
and distribution of precipitation, storm models have been used for the maximization procedure
for long-duration storms over large basins.

The maximization-transposition technique requires a large amount of data, particularly

volumetric rainfall data. In the absence of suitable data it may be necessary to transpose storms
over very large distances despite the considerable uncertainties involved. In this case reference
to published worldwide maximum observed point rainfalls will normally be helpful. The world
envelope curves for data recorded prior to 1948 and 1967 are shown in
figure 2.26, together with maximum recorded falls in the United Kingdom (from: Ward &
Robinson, 1990). By comparison to the world maxima, the British falls are rather small, as it is
to be expected that a temperate climate area would experience less intense falls than tropical
zones subject to hurricanes or the monsoons of southern Asia. The values of Cherrapunji in
India are from the foothills of the Himalayas mountains, where warm moist air under the
influence of monsoon depressions are forced upward resulting in extremely heavy rainfall.
Much the same happens in La Reunion where moist air is forced over a 3000 m high mountain
range. Obviously, for areas of less rugged topography and cooler climate, lower values of PMP
are to be expected.

Fig. 2.26 Magnitude-duration relationship for the world and the UK extreme rainfalls (source: Ward
& Robinson, 1990).
2 Precipitation 40 Hydrology

2.10 Analysis of dry spells

The previous sections dealt with extremely high, more or less instantaneous, rainfall. The
opposite, extremely low rainfall, is not so interesting, since the minimum rainfall is no rainfall.
Although some statistics can be applied to minimum annual or monthly rainfall, in which case
often adequate use can be made of the log-normal distribution, for short periods of observation
statistical analysis is nonsense. What one can analyse, however, and what has particular
relevance for rainfed agriculture, is the occurrence of dry spells. In the following analysis, use
is made of a case in the north of Bangladesh where wet season agriculture takes place on sandy

spell number accum- days number

dura- of ulated per of
tion spells spells season days

t i I n N p q P
days
3 46 165 151 2718 0.0607 0.9393 0.9999
4 20 119 150 2700 0.0441 0.9559 0.9988
5 29 99 149 2682 0.0369 0.9631 0.9963
6 15 70 148 2664 0.0263 0.9737 0.9806
7 11 55 147 2646 0.0208 0.9792 0.9544
8 8 44 146 2628 0.0167 0.9833 0.9150
9 2 36 145 2610 0.0138 0.9862 0.8665
10 5 34 144 2592 0.0131 0.9869 0.8506
11 3 29 143 2574 0.0113 0.9887 0.8022
12 1 26 142 2556 0.0102 0.9898 0.7659
13 2 25 141 2538 0.0099 0.9901 0.7524
14 1 23 140 2520 0.0091 0.9909 0.7230
15 0 22 139 2502 0.0088 0.9912 0.7070
16 1 22 138 2484 0.0089 0.9911 0.7070
17 1 21 137 2466 0.0085 0.9915 0.6901
18 2 20 136 2448 0.0082 0.9918 0.6723
19 2 18 135 2430 0.0074 0.9926 0.6335
20 1 16 134 2412 0.0066 0.9934 0.5901
21 1 15 133 2394 0.0063 0.9937 0.5665
22 0 14 132 2376 0.0059 0.9941 0.5416
23 0 14 131 2358 0.0059 0.9941 0.5416
24 0 14 130 2340 0.0060 0.9940 0.5416
25 0 14 129 2322 0.0060 0.9940 0.5417
26 0 14 128 2304 0.0061 0.9939 0.5417
27 2 14 127 2286 0.0061 0.9939 0.5417
28 0 12 126 2268 0.0053 0.9947 0.4875
29 2 12 125 2250 0.0053 0.9947 0.4875
30 0 10 124 2232 0.0045 0.9955 0.4270
31 1 10 123 2214 0.0045 0.9955 0.4270
33 1 9 121 2178 0.0041 0.9959 0.3941
34 1 8 120 2160 0.0037 0.9963 0.3593
35 1 7 119 2142 0.0033 0.9967 0.3226
42 1 6 112 2016 0.0030 0.9970 0.2838
43 1 5 111 1998 0.0025 0.9975 0.2428
49 1 4 105 1890 0.0021 0.9979 0.1995
50 1 3 104 1872 0.0016 0.9984 0.1536
52 1 2 102 1836 0.0011 0.9989 0.1052
61 1 1 93 1674 0.0006 0.9994 0.0541

Table 2.8 Probability of occurrence of dry spells in the Buri Teesta catchment
2 Precipitation 41 Hydrology

soils. It is known that rainfed agriculture seldom succeeds without supplementary irrigation. In
view of the small water retaining capacity of the soil, a dry spell of more than five days already
causes serious damage to the crop.

The occurrence of dry spells in the wet season is analyzed through frequency analysis of daily
rainfall records in Dimla in northern Bangladesh during a period of 18 years. The wet season
consists of 153 days. In table 2.8 the procedure followed is presented, which is discussed
briefly below.

In the 18 years of records, the number of times i that a dry spell of a duration t occurs has been
counted. Then the number of times I that a dry spell occurs of a duration longer than or equal
to t is computed through accumulation. The number of days within a season on which a dry
spell of duration t can start is represented by n = 153+1 - t. The total possible number of
starting days is N = n * 18. Subsequently the probability p that a dry spell starts on a certain
day within the season is defined by:

I
p = (2.14)
N

The probability q that a dry spell of a duration longer than t does not occur at a certain day in
the season, hence, is defined by:

I
q = 1- (2.15)
N

The probability Q that a dry spell of a duration longer than t does not occur during an entire
season, hence, is defined by:

n
 I
Q = 1   (2.16)
 N

Finally the probability that a dry spell of a duration longer than t does occur at least once in a
growing season is defined by:

n
 I 
P = 1- 1-  (2.17)
 N

In figure 2.27, the duration of the dry spell t is plotted against this probability P. It can be seen
that the probability of a dry spell longer than five days, which already causes problems in the
sandy soils, has a probability of occurrence of 99.6%, meaning that these dry spells occur
virtually every year. Hence, figure 2.27 illustrates the fact that rainfed agriculture is impossible
in the area during the wet season.
2 Precipitation 42 Hydrology

Fig. 2.27. Probability of occurrence of dry spells in the Buri

Teesta catchment