Download-manuals-hydrometeorology-data processing-09howtocarryoutsecondaryvalidationofrainfall

World Bank & Government of The Netherlands funded
Training module # SWDP - 09
How to carry out secondary
validation of rainfall data
New Delhi, November 1999
CSMRS Building, 4th Floor, Olof Palme Marg, Hauz Khas,
New Delhi – 11 00 16 India
Tel: 68 61 681 / 84 Fax: (+ 91 11) 68 61 685
E-Mail: dhvdelft@del2.vsnl.net.in
DHV Consultants BV & DELFT HYDRAULICS
with
HALCROW, TAHAL, CES, ORG & JPS

HP Training Module File: “ 09 How to carry out secondary validation of rainfall.doc” Version Nov.99 Page 1
Table of contents
Page
1. Module context 2
2. Module profile 3
3. Session plan 4
4. Overhead/flipchart master 6
5. Handout 7
6. Additional handout 9
7. Main text 10

1. Module context
While designing a training course, the relationship between this module and the others,
would be maintained by keeping them close together in the syllabus and place them in a
logical sequence. The actual selection of the topics and the depth of training would, of
course, depend on the training needs of the participants, i.e. their knowledge level and skills
performance upon the start of the course.

2. Module profile
Title : How to carry out secondary validation of rainfall data
Target group : Assistant Hydrologists, Hydrologists, Data Processing Centre
Managers
Duration : Three sessions of 60 minutes each
Objectives : After the training the participants will be able to:
• Perform secondary validation of rainfall data
Key concepts : • Spatial correlation structure of rainfall for various durations
• Spatial homogeneity
• Entries at wrong days
• Accumulated rainfall
• Transposed entries
• Number of rainy days
• Double mass analysis
• Auto correlation and spectral density functions
Training methods : Lecture, exercises, software
Training tools
required
: OHS, Computers
Handouts : As provided in this module
Further reading
and references
:

3. Session plan
No Activities Time Tools
1 General
• Secondary validation – General points
• Spatial correlation – Influencing factors
• KHEDA Catchment
• Spatial correlation – Daily interval
• Spatial correlation – Ten-daily interval
• Spatial correlation – Monthly interval
• Spatial correlation – Region A
• Catchment map (Region A)
• Spatial correlation – Extended region
• Catchment map (Extended region)
• Use of spatial correlation
• Validation procedures
15 min
OHS 1
OHS 2
OHS 3
OHS 4
OHS 5
OHS 6
OHS 7
OHS 8
OHS 9
OHS 10
OHS 11
OHS 12
2 Screening of data
• Screening
• Example 2.1- Tabular
10 min
OHS 13
OHS 14
3 Scrutiny of multiple time series graphs
• Example 3.1 - Multiple plot (stacking side-by-side)
• Example 3.1 - Multiple plot (shifted vertically)
10 min
OHS 15
OHS 16
4 Scrutiny of tabulations of daily rainfall at multiple stations
• Example 4.1 - Scrutiny of tabulation
10 min
OHS 17
5 Checking against data limits for totals at longer durations
• Example 4.1 - Scrutiny by tabulation
5 min
OHS 18
6 Spatial homogeneity testing of rainfall (nearest neighbour
analysis)
• General
• Nearest neighbour test - Definition
• Spatial estimate
• Example 6.1(a) – Selection of test station
• Example 6.1(b) – Tabular results
15 min
OHS 19
OHS 20
OHS 21
OHS 22
OHS 23
7 Identification of common errors
• Example 7.1 - Shift in entries (test results)
• Tabulation of suspect period
• General – Unidentified accumulation
• Example 7.2 – Test results
• Example 7.3 – Missed entries (test results)
10 min
OHS 24
OHS 25
OHS 26
OHS 27
OHS 28
OHS 29
OHS 30

No Activities Time Tools
8 Checking for systematic shifts using double mass curves
• General principle
• Detection of in-homogeneity
• Test procedure
• Double mass curve - Consistent data set
• Example 12.1 – Inconsistent data set
9 Exercise:
• For KHEDA data set detect at least one suspect situation
each by employing four methods: (1) Screening, (2) Multiple
graphs, (3) Tabulation, and (4) Limits for totals on longer
duration
• Employ spatial homogeneity test for a station in KHEDA
catchment and infer the test results
• Apply double mass analysis and detect in-homogeneous
data set at one or more stations in KHEDA catchment
30 min
30 min
30 min
Computer

4. Overhead/flipchart master

5. Handout

Add copy of Main text in chapter 8, for all participants.

6. Additional handout
These handouts are distributed during delivery and contain test questions, answers to
questions, special worksheets, optional information, and other matters you would not like to
be seen in the regular handouts.
It is a good practice to pre-punch these additional handouts, so the participants can easily
insert them in the main handout folder.

7. Main text
Contents
1. General 1
2. Screening of data series 5
3. Scrutiny by multiple time series graphs 7
4. Scrutiny by tabulations of daily rainfall
series of multiple stations 8
5. Checking against data limits for totals
at longer durations 9
6. Spatial homogeneity testing of rainfall
(Nearest neighbour analysis) 12
7. Identification of common errors 16
8. Checking for entries on wrong days -
shifted entries 17
9. Entries made as accumulations 20
10. Missed entries 23
11. Rainfall missed on days with low
rainfall - rainy days check 25
12. Checking for systematic shifts using
double mass analyses 27

How to carry out secondary validation of rainfall data
1. General
• Rainfall data received at Divisional offices have already received primary
validation on the basis of knowledge of instrumentation and conditions at the field
station and information contained in Field Record Books.
• Secondary validation now puts most emphasis on comparisons with neighbouring
stations to identify suspect values. Some of the checks which can be made are
oriented towards specific types of error known to be made by observers, whilst others
are general in nature and lead to identification of spatial inconsistencies in the data.
• Secondary validation is mainly carried out at Division. However since comparison
with neighbouring stations is limited by Divisional boundaries, the validation of
some stations near the Divisional boundaries will have to await assemblage of
data at the State Data Processing Centre.
• Rainfall poses special problems for spatial comparisons because of the limited or
uneven correlation between stations. When rainfall is convectional in type, it may rain
heavily at one location whilst another may remain dry only a few miles away. Over a
month or monsoon season such spatial unevennesss tends to be smoothed out and
aggregated totals are much more closely correlated.
Spatial correlation in rainfall thus depends on:
duration (smaller at shorter durations),
distance (decreasing with distance),
type of precipitation, and
physiographic characteristics of a region.
For any area the correlation structure for different durations can be determined on the basis
of historical rainfall data. A study for determining such correlation structures for yearly
duration for the entire country has been made (Upadhaya, D. S. et al, (1990) Mausam 41, 4,
523-530). In this the correlation field has been determined for 21 meteorological
homogeneous regions which cover almost the entire country using 70 years of data (1900 -
1970) for about 2000 stations. However, for the purpose of data validation and especially for
hourly and daily data such correlation structures are not readily available. It will be possible
to determine such structures on the basis of available rainfall data, though.
Example 1.1:
The effect of aggregation of data to different time interval and that of the inter-station
distances on the correlation structure is illustrated here.
The scatter plot of correlation between various rainfall stations of the KHEDA catchment for
the daily, ten daily and monthly rainfall data is shown in Fig. 1.1, Fig. 1.2 and Fig. 1.3
respectively.
From the corresponding correlation for same distances in these three figures it can be
noticed that aggregation of data from daily to ten daily and further to monthly level increases
the level of correlation significantly. At the same time it can also be seen that the general
slope of the scatter points becomes flatter as the aggregation is done. This demonstrates
that the correlation distance for monthly interval is much more than that for ten daily interval.

And similarly the correlation, which sharply reduces with increase in distance for the case of
daily time interval, does maintain its significance over quite longer distances.
Fig. 1.1: Plot of correlation with distance for daily rainfall data
Fig. 1.2: Plot of correlation with distance for ten-daily rainfall data
Fig. 1.3: Plot of correlation with distance for monthly rainfall data
Spatial Correlation - Daily Rainfall (Kheda Catchment)
Distance [km]
120110100908070605040302010
Correlation
1.0
0.8
0.6
0.4
0.2
0.0
Spatial Correlation - Monthly Rainfall (Kheda Catchment)
Distance (km)
12011511010510095908580757065605550454035302520151050
Correlation
1.0
0.8
0.6
0.4
0.2
0.0
Spatial Correlation - 10 Daily Rainfall (Kheda Catchment)
Distance (km)
12011511010510095908580757065605550454035302520151050
Correlation
1.0
0.8
0.6
0.4
0.2
0.0

Example 1.2:
Effect of physiographic characteristics over the correlation structure is illustrated by
considering monthly rainfall for two groups of stations in the PARGAON catchment.
Fig. 1.4 shows the scatter plot of the correlation among some 20 stations in small hilly region
(elevations ranging from 700 m to 1250 m) in the lower left part of the catchment (see,
Fig.1.5). This small region can be considered as homogeneous in itself and which is also
substantiated by the scatter plot of the correlation. Monthly rainfall data has been considered
for this case and as is clear from the plot there is a very high level of correlation among
stations and the general slope of the scatter diagram indicates a high value of the correlation
distance.
However, Fig. 1.6 shows the scatter plot of the correlation among monthly rainfall at some
34 stations in a region which includes the hilly region together with an extended portion in
the plain region (the plains ranging from 700 m to 600 m with very low and scattered hills in
between) of the catchment (see Fig. 1.7).
It is apparent from Fig. 1.6 that in case such a combination of stations, in which there are a
few stations from the hilly region and another lot from the adjoining plain region, is taken
then the resulting correlation shows a weaker correlation structure. The correlation decays
very fast against distance and even for shorter distances it is very much diffused. In fact, the
level of variability for the group of stations in the hilly region is much lower than that of the
remaining stations in the plain region. This is what is exhibited by Fig. 1.6 in which lot of
scatter is shown even for smaller inter station distances.
Fig. 1.4: Scatter plot of correlation for monthly rainfall in the small hilly region
Spatial Correlation - Monthly Rainfall (Region A)
Distance (km)
2522.52017.51512.5107.552.50
Correlation
1.0
0.8
0.6
0.4
0.2
0.0

Fig. 1.5: Selection of a group of some 20 stations in the hilly region of the catchment.
Fig. 1.6: Scatter plot of correlation for monthly rainfall in the extended region
Spatial Correlation - Extended Region
Distance (km)
80757065605550454035302520151050
Correlation
1.0
0.8
0.6
0.4
0.2
0.0

Fig. 1.7: Selection of a group of some 34 stations in the extended region of the
catchment.
• Spatial correlation can be used as a basis for spatial interpolation and correction.
However, there is a danger of rejecting good data which is anomalous as well as
accepting bad data. A balance must be struck between the two. In considering this
balance, it is well to give weight to the previous performance of the station and the
observer.
One must particularly be wary of rejecting extreme values, as true extreme values are for
design purposes the most interesting and useful ones in the data series. True extreme
values (like false ones) will often be flagged as suspect by validation procedures. Before
rejecting such values it is advisable to refer both to field notes and to confer with Sub-
divisional staff.
• The data processor must continue to be aware of field practice and
instrumentation and the associated errors which can arise in the data, as described
in Module 8.
2. Screening of data series
After the data from various Sub-Divisional offices has been received at the respective
Divisional office, it is organised and imported into the temporary databases of secondary
module of dedicated data processing software. The first step towards data validation is
making the listing of data thus for various stations in the form of a dedicated format. Such
listing of data is taken for two main objectives: (a) to review the primary validation
exercise by getting the data values screened against desired data limits and (b) to get the
hard copy of the data on which any remarks or observation about the data validation
can be maintained and communicated subsequently to the State/Regional data
processing centre.

Moreover, for the case of validation of historical data for period ranging from 10 to 40 years
this listing of the screening process is all the more important. This screening procedure
involves, for example for daily rainfall data, flagging of all those values which are beyond the
maximum data limits or the upper warning level. It also prepares the data in a well-organised
matrix form in which various months of the year are given as separate columns and various
days of the month are given as rows. Below this matrix of data the monthly and yearly basic
statistics like total rainfall, maximum daily rainfall, number of rainy days etc. are listed. Also,
the number of instances where the data is missing or has violated the data limits is also
given.
This listing of screening process and basic statistics is very useful in seeing whether the data
has come in the databases in desired manner or not and whether there is any mark
inconsistency vis-à-vis expected hydrological pattern.
Example: 2.1:
An example of the listing of screening process for MEGHARAJ station of KHEDA catchment
for the year 1991 is given in Table 2.1. The flagging of a few days of high rainfall shows that
these values have crossed the Upper Warning Level. Such flagged values can then be
subsequently attended to when comparing with adjoining stations. This particular year shows
a few days of very heavy rainfall, one in fact making the recorded maximum daily rainfall (i.e.
312 mm on 27 July). Monthly and yearly statistics are also viewed for appropriateness.
Table 2.1: Result of the screening process of daily rainfall data for one year
Daily data and statistics of series MEGHARAJ MPS Year = 1997
Day Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May
1 .0 .0 192.5* .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
2 .0 .0 15.0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
3 .0 .0 1.0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
4 .0 .0 .0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
5 .0 .0 .0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
6 .0 .0 .0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
7 .0 .0 1.0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
8 .0 .0 32.0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
9 .0 .0 1.0 25.0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
10 .0 .0 .0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
11 .0 .0 .0 14.5 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
12 .0 .0 7.0 1.5 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
13 .0 .0 1.0 4.0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
14 .0 .0 .5 .5 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
15 .0 .0 1.0 1.0 5.5 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
16 14.0 .0 .0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
17 .0 .0 .5 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
18 .0 .0 .0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
19 .0 10.0 12.0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
20 .0 .0 1.0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
21 .0 2.0 6.5 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
22 .0 1.0 .0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
23 12.0 .0 9.5 2.0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
24 9.0 .0 125.5 27.5 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
25 138.0* 1.0 11.0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
26 132.0* 4.0 54.5 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
27 38.0 312.0* 1.0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
28 54.0 32.5 .0 .0 .0 -99.0* -99.0* -99.0* -99.0* -99.0* -99.0* -99.0*
29 .0 4.5 .5 .0 .0 -99.0* -99.0* -99.0********** -99.0* -99.0* -99.0*
30 .0 12.0 .5 .0 .0 -99.0* -99.0* -99.0********** -99.0* -99.0* -99.0*
31********* 22.0 .0 ********* .0 ********* -99.0* -99.0********** -99.0********** -99.0*
Data 30 31 31 30 31 30 31 31 28 31 30 31
Eff. 30 31 31 30 31 0 0 0 0 0 0 0
Miss 0 0 0 0 0 30 31 31 28 31 30 31
Sum 397.0 401.0 474.5 76.0 5.5 -99.0 -99.0 -99.0 -99.0 -99.0 -99.0 -99.0
Mean 13.2 12.9 15.3 2.5 .2 -99.0 -99.0 -99.0 -99.0 -99.0 -99.0 -99.0
Min. .0 .0 .0 .0 .0 -99.0 -99.0 -99.0 -99.0 -99.0 -99.0 -99.0
Max. 138.0 312.0 192.5 27.5 5.5 -99.0 -99.0 -99.0 -99.0 -99.0 -99.0 -99.0
High 130.0 130.0 130.0 130.0 130.0 .0 .0 .0 .0 .0 .0 .0
Numb 2 1 1 0 0 0 0 0 0 0 0 0
Low .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0 .0
Numb 0 0 0 0 0 0 0 0 0 0 0 0
Annual values:
Data 365 * Sum 1354.0 * Minimum .0 * Too low 0
Effective 153 * Mean 8.8 * Maximum 312.0 * Too high 4
Missing 212
Exceedance of:
- Lower bound ( .00) marked with *
- Upper bound ( 130.00) marked with *
- Rate of rise ( 320.00) marked with +
- Rate of fall ( 320.00) marked with -

Comparison of Daily Rainfall at Multiple Stations
MPS_ANIOR MPS_BHEMPODA MSP_RELLAWADA MPS_MEGHARAJ
Time
15/08/96
14/08/96
13/08/96
12/08/96
11/08/96
10/08/96
09/08/96
08/08/96
07/08/96
06/08/96
05/08/96
04/08/96
03/08/96
02/08/96
01/08/96
31/07/96
30/07/96
29/07/96
28/07/96
27/07/96
26/07/96
25/07/96
24/07/96
23/07/96
22/07/96
Rainfall(mm)
240
220
200
180
160
140
120
100
80
60
40
20
0
3. Scrutiny by multiple time series graphs
Inspection of multiple time series graphs may be used as an alternative to inspection
of tabular data. Some processors may find this a more accessible and comprehensible
option. This type of validation can be carried out for hourly, daily, monthly and yearly rainfall
data. The validation of compiled monthly and yearly rainfall totals helps in bringing out those
inconsistencies which are either due to a few very large errors or due to small systematic
errors which persist unnoticed for much longer durations. The procedure is as follows:
(a) Choose a set of stations within a small area with an expectation of spatial correlation.
(b) Include, if possible, in the set one or more stations which historically have been more
reliable.
(c) Plot rainfall series as histograms stacked side by side and preferably in different
colours for each station. Efficient comparison on the magnitudes of rainfall at different
stations is possible if the individual histograms are plotted side by side. On the other
hand a time shift in one of the series is easier to detect if plots of individual stations are
plotted one above the other. Stacking side-side is presently possible with the software.
(d) After inspection for anomalies and comparing with climate, all remaining suspect
values are flagged, and comment inserted as to the reason for suspicion.
Example 3.1:
Consider that a few of the higher values at ANIOR station of KHEDA catchment during July
and August 1996 are suspect. Comparison with adjoining available stations BHEMPODA,
RELLAWAD and MEGHARAJ is made for this purpose. Fig. 3.1 gives the plot of daily rainfall
for these multiple stations during the period under consideration.
It may be noticed that rainfall of about 165 mm and 70 mm are observed at ANIOR and
BHEMPODA stations which are virtually not more than 5 kms. apart. Though it is not that
such variation could not be possible but at least such deviations are sufficient for one to
cross check with other information. On checking with the hourly observations available at
ANIOR station it is noticed that the compiled daily rainfall is only 126 mm. This substantiates
the earlier suspicion of it being comparatively larger.
Fig. 3.1: Comparison of multiple time series plot of daily rainfall data
Further it may be noticed from the plot that the daily rainfall for 12th
and 13th
August at
ANIOR seems to shifted ahead by a day. This shifting is also confirmed when the ARG
record is compared with the SRG record. The time shifting error is clearly in the SRG record
of ANIOR station. Thus inspection of the record sheets, visit to site and interaction with the

observation can be helpful in getting more insight into the probable reasons of such
departures.
4. Scrutiny by tabulations of daily rainfall series of multiple
stations
In the case of rainfall (unlike other variables), a tabular display of daily rainfall in a
month, listing several stations side by side can reveal anomalies which are more
difficult to see on multiple time series graphs (see below), plotted as histograms.
Scanning such tabular series will often be the first step in secondary data validation.
Anomalies to look out for are:
• Do the daily blocks of rainydays generally coincide in start day and finish day?
• Are there exceptions that are misplaced, starting one day early or late?
• Is there a consistent pattern of misfit for a station through the month?
• Are there days with no rainfall at a station when (heavy) rainfall has occurred at all
neighbouring stations?
Field entry errors to the wrong day are particularly prevalent for rainfall data and
especially for stations which report rainfall only. This is because rainfall occurs in dry
and wet spells and observers may fail to record the zeros during the dry spells and hence
lose track of the date when the next rain arrives. When ancillary climate data are available,
this may be used to compare with rainfall data. For example a day with unbroken sunshine in
which rain has been reported suggests that rainfall has been reported for the wrong day.
However, most comparisons are not so clear cut and the processor must be aware that there
are a number of possibilities:
• rainfall and climate data both reported on the wrong day - hence no anomaly between
them but discrepancy with neighbouring stations.
• rainfall data only on the wrong day - anomalies between rainfall and climate and
between rainfall and neighbouring rainfall
• rainfall and climate both reported on the correct day - the anomaly was in the occurrence
of rainfall. For example no rainfall at one site but at neighbouring sites. In this case
climatic variables are likely to have been shared between neighbouring stations even if
rainfall did not occur.
Example 4.1:
As a routine process of scrutinising daily data for a common error of time shift in one or more
data series, consider KAPADWANJ, KATHLAL, MAHISA, SAVLITANK and VADOL stations
of KHEDA catchment. These stations are within a circle of 25 kms. diameter and thus are
expected to experience similar rainfall on an average.
For an easy scrutiny of the data series for possible time shift in one or more series the data
series are tabulated side by side as shown in Table 4.1 for a period of 1st
August to 20th
August 1984. A very casual look at this tabulation reveal that there is very high possibility of
a one day time shift in the data of SAVLITANK station. Data series of SAVLITANK station
appears to be having a lag of one day in consequent rainfall events. Exactly same shift is
persisting for all 20 days and is confirmed by closely looking at the start and end times of five
rainfall events (highlighted by underlining) one after another.
Such a finding then must be followed by first a closer look at the manuscript record and to
see if the shift has been during entering or managing the data series. If it is found that the
this shift has been due to data handling during or after data entry then it is corrected
accordingly. If the manuscript record also shows the same series then the observer can be

asked to tally it from the field note book. The feed back from the observer will help in settling
this type of discrepancy and also will encourage observer to be careful subsequently.
Table 4.1: Tabulation for scrutiny of possible error in the timing of daily rainfall data
Tabulation of series, Year 1984
==========Data==========
Year mth day hr si KAPADWANJ KATHLAL MAHISA SAVLITANK VADOL
PH PH PH PH PH
1984 8 1 .0 .0 .0 .0 .0
1984 8 2 .0 .0 .2 .0 .0
1984 8 3 152.4 99.3 157.4 .0 39.3
1984 8 4 104.1 50.2 87.0 150.0 59.2
1984 8 5 7.7 12.0 18.0 76.0 13.1
1984 8 6 1.5 35.0 .0 16.0 .0
1984 8 7 .0 .0 .0 3.0 .0
1984 8 8 1.3 .0 .0 .0 .0
1984 8 9 .0 13.0 .0 .0 .0
1984 8 10 231.2 157.0 179.0 .0 17.3
1984 8 11 43.2 18.3 64.0 201.0 63.2
1984 8 12 .0 .0 .0 26.0 33.3
1984 8 13 .0 .0 .0 .0 13.1
1984 8 14 .0 .0 20.0 .0 .0
1984 8 15 .0 .0 .0 30.0 .0
1984 8 16 2.6 8.3 16.5 .0 16.3
1984 8 17 .0 .0 .0 20.0 20.2
1984 8 18 32.0 50.3 25.6 .0 37.2
1984 8 19 16.5 8.2 15.0 27.0 19.3
1984 8 20 .0 .0 .0 13.0 .0
1984 8 21 .0 .0 .0 .0 .0
5. Checking against data limits for totals at longer durations
5.1 General description:
Many systematic errors are individually so small that they can not easily be noticed.
However, since such errors are present till suitable corrective measures are taken, they tend
to accumulate with time and therefore tend to be visible more easily. Also, some times when
the primary data series (e.g. daily rainfall series) contains many incorrect values frequently
occurring for a considerable period (say a year of so) primarily due to negligence of the
observer or at the stage of handling of data with the computer then also the resulting series
compiled at larger time interval show the possible incorrectness more visibly. Accordingly, if
the observed data are accumulated for longer time intervals, then the resulting time series
can again be checked against corresponding expected limits. This check applies primarily to
daily rainfall at stations at which there is no recording gauge.
5.2 Data validation procedure and follow up actions:
Daily data are aggregated to monthly and yearly time intervals for checking if the resulting
data series is consistent with the prescribed data limits for such time intervals.
Together with the upper warning level or maximum limit, for monsoon months and yearly
values use of lower warning level data limit can also be made to see if certain values are
unexpectedly low and thus warrants a closer look. Aggregated values violating the
prescribed limits for monthly or annual duration are flagged as suspect and appropriate

remarks made in the data validation report stating the reasons for such flagging. These
flagged values must then validated on the basis of data from adjoining stations.
Example 5.1:
The daily data of VADOL station (in KHEDA catchment) is considered and the yearly totals
are derived. The period of 1970 to 1997 is taken for the compilation wherein two years of
data, i.e. 1975 & 1976, is missing.
The plot of these yearly values is shown in Fig. 5.1. In this case of yearly rainfall data the
values can be validated against two data limits as upper and lower warning levels. The
values of such limits can be drawn from the experience of the distribution of the yearly
rainfall in the region. In this case, the mean of the 26 yearly values is about 660 mm with an
standard deviation of 320 mm with a skewness of 0.35. With an objective of only flagging a
few very unlikely values for the purpose of scrutiny, a very preliminary estimate of the upper
and lower warning levels is arbitrarily obtained by taking them as:
Lower warning level = mean – 1.5 x (standard deviation) = 660 – 1.5 x 320 = 180 mm
and
Upper warning level = mean + 2.0 x (standard deviation) = 660 + 2.0 x 320 = 1300 mm
The multipliers to the standard deviation for the lower and upper warning levels have been
taken differently in view of the data being positively skewed with a finite lower bound. Such
limits can be worked out on a regional basis on the basis of the shape of distribution and
basically with the aim to demarcate highly unlikely extremes.
These limits have been shown in the plot of the yearly values and it may be seen that there
are a few instances where the annual rainfall values come very close or go beyond these
limits. For example, in the year 1997 a large value of yearly rainfall more than 1329 mm is
reported and similarly for year 1974 the reported rainfall is as low as 92.6 mm.
Fig. 5.1: Plot of rainfall data compiled at an yearly interval
After screening such instances of extreme values in the data series compiled at longer time
intervals, it is then essential that for such instances the values reported for the station under
consideration is compared with that reported at the neighbouring stations. For this, the yearly
Plot of Yearly Rainfall with Data Limits
Vadol Lower Warning Level Upp. Warning Level
Time
9897969594939291908988878685848382818079787776757473727170
Rainfall(mm)
1,400
1,300
1,200
1,100
1,000
900
800
700
600
500
400
300
200
100
0

data at five neighbouring stations including the station under consideration, i.e. VADOL, is
tabulated together as Table 5.1 for an easier comparison.
Table 5.1: Tabulation of yearly rainfall at five neighbouring stations
Tabulation of series, Year 1970 - 1997
==========Data==========
Year mth day hr si BALASINOR KAPADWANJ SAVLITANK VADOL VAGHAROLI
MPS MPS MPS MPS MPS
1970 802.8 927.2 -99.0 739.8 -99.0
1971 546.7 569.5 -99.0 475.0 -99.0
1972 338.2 291.0 -99.0 198.2 -99.0
1973 1061.2 1305.0 1226.0 1186.4 1297.4
1974 338.1 421.0 268.5 92.6 -99.0
1975 -99.0 -99.0 -99.0 -99.0 -99.0
1976 -99.0 -99.0 -99.0 -99.0 -99.0
1977 1267.2 1217.5 1168.9 1083.5 1575.8
1978 672.8 507.5 517.0 801.4 1347.0
1979 437.5 428.5 525.5 455.6 1197.0
1980 551.3 661.6 378.0 545.7 892.0
1981 917.7 1273.6 1004.0 950.7 722.0
1982 302.1 540.2 376.0 320.1 267.0
1983 1028.0 1088.5 1020.0 1099.1 1110.0
1984 523.1 882.9 888.0 475.1 649.6
1985 438.9 661.5 1101.0 510.8 1173.0
1986 526.9 474.9 256.0 470.7 505.0
1987 257.0 256.0 209.0 227.5 232.0
1988 -99.0 1133.0 826.0 734.5 849.4
1989 1088.0 1064.0 787.0 840.8 -99.0
1990 1028.1 971.0 1042.0 761.0 1174.0
1991 451.0 815.0 523.0 618.1 628.0
1992 421.1 1028.0 469.0 459.6 606.0
1993 531.0 410.5 781.0 512.8 781.0
1994 1085.0 1263.0 1039.0 1083.3 1332.0
1995 590.0 528.0 422.0 399.6 525.0
1996 1397.0 968.0 760.0 762.6 1050.0
1997 1272.0 1876.0 1336.2 1329.0 950.0
It may be seen from this table that for the year 1997 at most of the neighbouring stations the
reported rainfall is very high and is even about 1875 mm for KAPADWANJ station. At two
other stations also it is in the range of 1200 to 1300 mm except that for VAGHAROLI it is
only 950 mm for this year. Thus, as far as the suspect value of 1329 mm at VADOL station is
concerned, the suspicion may be dropped in view of similar higher values reported nearby.
Comparison for the year 1974 shows that though all the stations seems to have experienced
comparatively lower amount of rainfall (about 340, 420 and 270 mm), the rainfall at VADOL
station is extremely low (i.e. 92.6 mm). Such a situation warrants that the basic daily data for
this test station must be looked more closely for its appropriateness.
For looking at the daily data for the year 1974 a tabulation is again obtained as given in
Table 5.2 for the neighbouring stations. Only a portion of the year for a brief period in May is
given the Table.
Though, there are comparatively more zeros reported for the VADOL station then other
stations for many rain events during the season but looking at the variability in the
neighbouring stations it might be accepted. However, there is one significant event in the
month of May which is reported elsewhere and for which zero rainfall is reported at VADOL.
This may seem to have an error due to non-observation or incorrect reporting. It is
necessary to refer the manuscript for this year and to see if data in the database

corresponds with it. It may also be possible that the observations have not really been taken
by the observer on this particular station for this period during which it is normally not
expected to rain. On the basis of the variability experienced between various stations in the
region it may then be decided to consider some of the reported zero values as doubtful at
VADOL station.
Table 5.2: Tabulation of daily rainfall at VADOL station.
1974 5 23 .0 .0 .0 .0 -99.0
1974 5 24 .0 .0 .0 .0 -99.0
1974 5 25 .0 .0 .0 .0 -99.0
1974 5 26 4.2 75.0 73.0 .0 -99.0
1974 5 27 23.0 30.0 19.0 .0 -99.0
1974 5 28 .0 .0 .0 .0 -99.0
1974 5 29 12.0 .0 .0 .0 -99.0
1974 5 30 .0 .0 .0 .0 -99.0
1974 5 31 .0 .0 .0 .0 -99.0
6. Spatial homogeneity testing of rainfall (Nearest neighbour
analysis)
As mentioned above, rainfall exhibits some degree of spatial consistency. The degree of
consistency is primarily based on the actual spatial correlation. The expected spatial
consistency is the basis of investigating the observed rainfall values at the individual
observation stations. An estimate of the interpolated rainfall value at a station is obtained on
the basis of the weighted average of rainfall observed at the surrounding stations. Whenever
the difference between the observed and the estimated values exceed the expected limiting
value then such values are considered as suspect values. Such values are then flagged for
further investigation and ascertaining the possible causes of the departures.
6.2 Data validation procedure and follow up actions
First of all, the estimation of the spatially interpolated rainfall value is made at the station
under consideration. The station being considered is the suspect station and is called the
test station. The interpolated value is estimated by computing the weighted average of the
rainfall observed at neighbouring stations. Ideally, the stations selected as neighbours
should be physically representative of the area in which the station under scrutiny is situated.
The following criteria are used to select the neighbouring stations (see Fig. 6.1):
(a) the distance between the test and the neighbouring station must be less than a
specified maximum correlation distance, say Rmax kms.
(b) a maximum of 8 neighbouring stations can be considered for interpolation.
(c) to reduce the spatial bias in selection, it is appropriate to consider a maximum of only
two stations within each quadrant.
The estimate of the interpolated value at the test station based on the observations at N
neighbouring stations is given as:

P t
P t D
D
est
i
i
N
i
b
i
b
i
N
( )
( ) /
/
= =
=
∑
∑
1
1
1
Where:
Pest(t) = estimated rainfall at the test station at time t
Pi(t) = observed rainfall at the neighbour station i at time t
Di = distance between the test and the neighbouring station i
N = number of neighbouring stations taken into account.
b = power of distance D
Fig. 6.1: Definition sketch of Test and Base (neighbouring) stations
This estimated value is compared with the observed value at the test station and the
difference is considered as insignificant if the following conditions are met:
Pobs(t) - Pest(t)  ≤ Xabs
Pobs(t) - Pest(t)  ≤ Xrel * S Pest (t)
where:
Xabs = admissible absolute difference
SPest(t) = standard deviation of neighbouring values
Xrel = multiplier of standard deviation and
2
1
)( ))()(( tPtPS ii
N
i
tPest
−= ∑=
+
+
+
+
+
++
+
+
+
+
+
+
+
+ III
III IV
+
+
Neighbouring
stations
Test station
Selected
neighbouring
stations
II etc. Quadrant

Where departures are unacceptably high, the recorded value is flagged “+” or “-”, depending
on whether the observed rainfall is greater or less than the estimated one. The limits Xabs
and Xrel are chosen by the data processor and have to be based on the spatial variability of
rainfall. They are normally determined on the basis of experience with the historical data with
the objective of flagging a few values (say 2-3%) as suspect values.
It is customary to select a reasonably high value of Xabs to avoid having to deal with a large
number of difference values in the lower range. In the example, illustrated below, Xabs = 25
mm. This value may be altered seasonally. It should be noted that where Xrel only is applied
(i.e., Xabs is large), the test also picks up an excessive number of anomalies at low rainfalls
where Xrel x S has a small absolute value. Such differences at low rainfall are both, more
likely to occur and, have less effect on the overall rainfall total, so it is important to select a
value of Xrel to flag a realistic number of suspect values. In the example shown Xrel = 2.
This check for spatial consistency can be carried out for various durations of rainfall
accumulations. This is useful in case smaller systematic errors are not detectable at lower
level of aggregation. The relative limit Xrel is less for daily data than for monthly data because
of relatively higher SPest.
Typical rainfall measurement errors show up with specific patterns of “+” and “-“ in the spatial
homogeneity test and will be mentioned in the following sections to aid interpretation of the
flagged values.
Example 6.1:
A test is performed for reviewing the spatial homogeneity of the daily rainfall data at
SAVLITANK station in KHEDA catchment. An area within a radius of 25 kms. around
SAVLITANK station is considered for selecting the base stations (see Fig. 6.2). Absolute and
relative errors admissible for testing are kept as 50 mm and a multiplier of 2 with standard
deviation respectively. Report on the result of the analysis of spatial homogeneity test is
given in Table 6.1.
Fig. 6.2: Selection of test station SAVITANK and neighbouring base stations

Table 6.1: Results of the spatial homogeneity test.
Spatial homogeneity check
====================================
Test station SAVLITANK PH
Start date: 1984 6 1 0 1 End date: 1985 10 31 0 1
Radius of circle of influence : 25.000 (km)
Station weights proportional to : 1/D^2.00
Admissible absolute error : 50.000
Multiplier to stdv of neighbours: 2.000
Selected neighbour stations:
Quadrant Station Distance (km)
1 VADOL PH 9.225
2 KAPADWANJ PH 8.139
3 MAHISA PH 13.480
3 KATHLAL PH 13.895
4 VAGHAROLI PH 17.872
4 THASARA PH 21.168
Year mth day hr si P_obs flag P_est Stdv n
1984 6 14 0 1 9.00 + .00 .00 6
1984 6 15 0 1 14.00 + .00 .00 6
1984 6 16 0 1 23.00 + .00 .00 6
1984 7 2 0 1 52.00 + 14.52 9.71 6
1984 7 6 0 1 47.00 + 2.13 4.51 6
1984 7 25 0 1 25.00 + .32 1.21 6
1984 8 3 0 1 .00 - 96.59 65.70 6
1984 8 4 0 1 150.00 + 78.44 38.47 6
1984 8 5 0 1 76.00 + 20.64 36.20 6
1984 8 10 0 1 .00 - 128.36 93.57 6
1984 8 11 0 1 201.00 + 59.25 42.04 6
1984 8 15 0 1 30.00 + .50 1.89 6
1984 8 19 0 1 27.00 + 16.81 4.91 6
1984 8 28 0 1 8.00 + .00 .00 6
1985 6 13 0 1 9.00 + .00 .00 6
1985 6 14 0 1 14.00 + .00 .00 6
1985 6 16 0 1 8.00 + .00 .00 6
1985 7 2 0 1 21.00 + .07 .37 6
1985 7 6 0 1 47.00 + .73 3.73 6
1985 7 19 0 1 60.00 + 16.05 15.49 6
1985 7 21 0 1 29.00 + 10.41 7.93 6
1985 7 23 0 1 12.00 + .15 .75 6
1985 7 25 0 1 25.00 + 3.15 3.78 6
1985 8 1 0 1 10.00 + .48 1.97 6
1985 8 4 0 1 150.00 + 82.57 76.84 6
1985 8 5 0 1 76.00 + 15.06 37.51 6
1985 8 11 0 1 201.00 + 11.39 53.59 6
1985 8 15 0 1 30.00 + .29 1.49 6
1985 8 17 0 1 20.00 + 1.09 5.59 6
1985 8 19 0 1 27.00 + 1.75 8.94 6
1985 8 28 0 1 8.00 + .00 .00 6
1985 9 14 0 1 17.00 + .00 .00 6
1985 9 15 0 1 3.00 + .00 .00 6
1985 10 8 0 1 145.00 + 70.17 67.38 6
1985 10 9 0 1 .00 - 86.03 116.43 6
Legend
n = number of neighbour stations
+ = P_obs - P_est > 0
- = P_obs - P_est < 0
* = P_est is missing
Six neighbouring stations are considered eligible for making the spatial estimate.
Comparison of observed and estimated daily rainfall value is made and those instances
where the difference between observed and estimated value is more than the test criteria
(i.e. absolute or relative difference) a flag is put. Listing of these instances can be seen in the
analysis report given above.

Following can be easily deduced from the above listing:
(a) There are quite a few very large differences in the observed and the estimated
values e.g. those on 3rd
, 4th
, 10th
, 11th
August 1984 and 4th
, 11th
August 1985 and 8th
,
9th
October 1985 (highlighted in the table). Such large differences warrant a closer
look at the observed values in conjunction of the rainfall at the neighbouring stations.
(b) A few of these instances of large differences are preceded or followed by 0 rainfall
values at the test station which indicates that either the rainfall is accumulated or
there is a possibility of time shift in the data. However, presence of a large amount of
standard deviation points to the fact that the variability of rainfall at these instances is
quite high among the neighbouring stations and it may not be impossible to observe
such large variations at the test station as well. However, another possibility is that
there have been some time shift in the data of one or more of the base stations as
well. When all the stations considered are also likely to have similar errors this aspect
can be ruled out. Tabulation of data at these base stations in fact reveal possibility of
such shiftings.
(c) Some of the instances when the rainfall has been very low and the standard
deviation among the neighbouring stations is also very low are also listed (specially
those with zero rainfall at all the neighbouring stations and thus zero standard
deviation and a very low rainfall at the test station). Such differences would normally
be picked up by the relative error test owing to very small standard deviations and
can be overlooked if the value at test station is also meagre. However, in the present
example, another possibility is indicated at least for those in the month of June. It can
be noticed that on all the instances of June, the estimated rainfall is 0 implying that
there has been zero rainfall reported at all the six neighbouring stations. And since
the resulting standard deviation is also zero all these instances have been short
listed. In fact, it is very likely that at all these neighbouring stations observation of
rainfall is started from 16th
June of every year and thus the first observation is
available only for 17th
of June and inadvertently all these missing data on and before
16th
June has been reported as 0 mm. Further, SAVLITANK station being on a
reservoir site might have an arrangement of having the observation throughout the
year and thus the reported rainfall values may be correct.
(d) As explained above, for the listed inconsistencies possible scenarios are required to
be probed further and only then a judicious corrective measure can be forthcoming.
In case, none of the corroborative facts substantiates the suspicion further then either
the value can be left as suspect or if the variability of the process is considered very
high such suspect values can be cleared of subsequently.
7. Identification of common errors
In the following sections, procedures for identification of common errors in rainfall data are
discussed with reference to either:
• Graphical and tabular (Section 3 and 4)
• Spatial homogeneity tests (Section 6)
Typical errors are:
• Entries on the wrong day - shifted entries
• Entries made as accumulations
• Missed entries
• Rainfall measurement missed on days of low rainfall.

8. Checking for entries on wrong days - shifted entries
Since the record of rainfall data is interspersed with many entries having zero values, values
may be entered against wrong days. This is due to the fact that while entering the data one
or more zero entries may get omitted or repeated by mistake. For daily data, such mistakes
are more likely when there are a few non-zero values in the middle and most of the entries at
the beginning and end of the month as zero values. This results in shifting of one or more
storms by a day or two, which normally tend to get corrected with the start of the new month.
This is because for the next month the column or page starts afresh in the manuscript from
which the data is being entered.
Shift errors in rainfall series can often be spotted in the tabulated or plotted multiple series,
especially if they are repeated over several wet/dry spells. It is assumed that no more than
one of the listed series will be shifted in the same direction in the same set. With respect to
spatial homogeneity testing, application of the test will generate a + at the beginning of a wet
spell and a - at the end (and possibly others in between) if the data are shifted forward, and
the reverse if the data are shifted backward.
A shift to coincide with the timing of adjacent stations and rerun of the spatial homogeneity
test will generally result in the disappearance of the + and - flags, if our interpretation of the
shift was correct.
The re-shifted series is then adopted as the validated series for the station/period in
question.
Example 8.1:
Spatial homogeneity test for daily rainfall series of VADAGAM station in KHEDA catchment
is carried out with neigbouring stations MODASA, RAHIOL, BAYAD and ANIOR as base
stations. The result of this test is reported as given in Table 8.1 below:
It may be noticed from above listing that a –ve flag together with 0 mm observed rainfall
followed by a +ve flag, both with very high value of absolute difference between the
observed and estimated daily rainfall is shown on 5th
and 7th
August 1988. Such flagging
indicates a possible shift in the data at this station VADAGAM. Other instances listed in the
test report are primarily due to very small standard deviation among base stations during low
rainfall days and may be overlooked.
This suspicion is confirmed after looking at the tabulation of this station data alongwith the
other four base stations as given in Table 8.2. It may be seen that except for the event
starting on 5th
August, most of the other rain events at these five stations correspond
qualitatively with respect to timings. Data for this event seems to have shifted forward (i.e.
lagging in time) by one day. This shifting has been the reason for –ve flag and 0 observed
rainfall and followed later with a +ve flag in the recession phase of the event.

Table 8.1: Result of the spatial homogeneity test at VADAGAM station.
====================================
Test station VADAGAM PH
Start date: 1988 7 1 0 1
End date: 1988 9 30 0 1
1 RAHIOL PH 12.606
1 MODASA PH 18.689
4 BAYAD PH 12.882
4 ANIOR PH 21.829
1988 8 1 0 1 .50 - 8.32 3.83 4
1988 8 5 0 1 .00 - 181.97 45.70 4
1988 8 7 0 1 161.00 + 14.23 8.32 4
1988 8 8 0 1 4.00 - 11.98 3.06 4
1988 8 9 0 1 18.00 + 7.12 1.72 4
1988 8 11 0 1 4.20 + .59 1.43 4
1988 8 25 0 1 32.00 + 1.97 4.34 4
1988 9 6 0 1 9.50 + .00 .00 4
1988 9 29 0 1 12.00 + 1.09 1.30 4
Legend
+ = P_obs - P_est > 0
- = P_obs - P_est < 0
This shift was confirmed by looking at the manuscript and thus implies that this has occurred
at the time or after the data has been entered into the computer. The shift was corrected by
removing one day lag in this storm event and stored as a temporarily (Data type TMA).
When the spatial homogeneity test was carried out again with this corrected series following
results were obtained (Table 8.3):
It may now be seen that there is no negative or positive flag with 0 observed rainfall and
large difference in observed and estimated value. The rainfall on 6th
August is still flag
because of larger difference in observed and estimated rainfall as against the permissible
limit. Thus in this way the time shifts may be detected and removed by making use of spatial
homogeneity test.

Table 8.2: Tabulation of daily rainfall at neighbouring stations.
==========Data==========
Year mth day hr si ANIOR BAYAD MODASA RAHIOL VADAGAM
PH PH PH PH PH
1988 7 12 .0 .0 .0 .0 .0
1988 7 13 .0 .0 70.0 .0 .0
1988 7 14 33.0 65.0 75.0 30.0 14.0
1988 7 15 8.0 17.8 12.5 5.0 3.0
1988 7 16 26.8 14.0 31.0 60.6 40.0
1988 7 17 5.4 1.2 10.0 2.0 1.0
1988 7 18 .0 2.0 .0 .0 1.0
1988 7 19 40.0 57.8 2.5 50.8 35.0
1988 7 20 54.2 46.0 60.0 32.8 46.0
1988 7 21 7.0 17.0 4.0 4.0 19.0
1988 7 22 113.0 78.4 124.0 91.8 82.0
1988 7 23 .0 11.2 15.0 6.8 16.3
1988 7 24 13.0 .0 29.0 7.4 .0
1988 7 25 8.0 14.0 43.5 35.8 23.1
1988 7 26 18.0 27.0 1.0 .0 4.2
1988 7 27 31.0 1.0 .0 3.4 1.2
1988 7 28 29.0 42.0 7.0 10.0 23.0
1988 7 29 .0 14.0 15.0 4.0 10.0
1988 7 30 13.4 .0 43.0 2.0 .0
1988 7 31 4.2 17.0 6.0 .0 .0
1988 8 1 8.0 3.0 13.0 11.4 .5
1988 8 2 4.0 .0 2.0 .0 .0
1988 8 3 .0 .0 17.0 22.0 4.0
1988 8 4 .0 1.0 1.0 .0 .0
1988 8 5 253.0 135.0 161.0 212.8 .0
1988 8 6 139.0 94.0 112.0 110.6 140.0
1988 8 7 20.0 24.0 4.0 7.6 161.0
1988 8 8 11.2 8.0 11.0 16.5 4.0
1988 8 9 9.0 8.0 9.0 4.8 18.0
1988 8 10 2.6 3.0 8.0 1.0 1.2
1988 8 11 3.5 .0 1.0 .0 4.2
1988 8 12 .0 .0 3.0 .0 3.0
1988 8 13 .0 .0 .0 .0 .0
Table 8.3: Results of the spatial homogeneity test on the corrected series
====================================
Test station VADAGAM TMA
Start date: 1988 7 1 0 1
End date: 1988 9 30 0 1
1 RAHIOL PH 12.606
1 MODASA PH 18.689
4 BAYAD PH 12.882
4 ANIOR PH 21.829

1988 8 1 0 1 .50 - 8.32 3.83 4
1988 8 6 0 1 161.00 + 108.49 16.13 4
1988 8 9 0 1 1.20 - 7.12 1.72 4
1988 8 25 0 1 32.00 + 1.97 4.34 4
1988 9 6 0 1 9.50 + .00 .00 4
1988 9 29 0 1 12.00 + 1.09 1.30 4
Legend
+ = P_obs - P_est > 0
- = P_obs - P_est < 0
9. Entries made as accumulations
The rainfall observer is expected to take rainfall observations every day at the stipulated
time, without discontinuity for either holidays, weekends or sickness. Nevertheless, it is likely
that on occasions the raingauge reader will miss a reading for one of the above reasons.
The observer may make one of three choices for the missed day or sequence of days.
• Enter the value of the accumulated rainfall on the day on which he/she returned from
absence and indicate that the intervening values were accumulated (the correct
approach).
• Enter the value of the accumulated rainfall on the day on which he/she returned and
enter a zero (or no entry) in the intervening period.
• Attempt to guess the distribution of the accumulated rainfall over the accumulated period
and enter a positive value for each of the days.
The third option is probably the more common as the observer may fear that he will be
penalised for missing a period of record even for a legitimate reason. The second also
occurs. Observers must be encouraged to follow the first option, as a more satisfactory
interpolation can be made from adjacent stations than by the observer’s guess.
If accumulations are clearly marked by the observer then the accumulated value can readily
be distributed over the period of absence, by comparison with the distribution over the same
period at adjacent stations.
For unindicated accumulations with a zero in the missed values, the daily tabulation will
indicate a gap in a rainy spell in comparison to neighbouring stations. Of course, an absence
during a period of no rain will have no impact on the reported series. Spatial homogeneity
testing will show a –ve flag on days on which there was significant rain during the period of
accumulation and a +ve flag on the day of accumulation.
The data processor should inspect the record for patterns of this type and mark such
occurrences as suspect. In the first instance, reference is made to the field record sheet to
confirm that the data were entered as recorded. Then, this being so, a search is made

backward from the date of the accumulated total to the first date on which a measurable
rainfall has been entered and an apportionment made on the basis of neighbouring stations.
The apportioning is done over the period which immediately preceded the positive departure
with negative departures and zero rainfall. The accumulated rainfall is apportioned in the
ratio of the estimated values on the respective days as:
∑
=
=
accN
i
iest
totiest
iappor
P
PP
P
1
,
,
,
*
Where:
Ptot = accumulated rainfall as recorded
Nacc = number of days of accumulation
Pest,i = estimated daily rainfalls during the period of accumulation on the basis of adjoining
stations
Pappor,i = apportioned value of rainfall for each day of accumulation period
Where it is not possible to adequately reason in favour or against such an accumulation then
the suspect value can be left labelled as doubtful. On the other hand if the period of such
accumulation is clearly marked by the observer then apportionment for the said period can
be done directly without checking for the period of accumulation.
The field supervisor should be informed of such positively identified or suspicious
accumulations and requested to instruct the field observer in the correct procedure.
Example 9.1:
As a routine secondary validation, spatial homogeneity test for station DAKOR (KHEDA
catchment) for the year 95 is carried out considering a few neighbouring stations. The test
results are as given below (Table 9.1):
On examining the above results, it can be apparent that there are a few “–ve” flags having nil
observed rainfall which is followed by a “+ve” flag having a very high rainfall value. Such
combination indicate a possible accumulation of rainfall for one or more days prior to 28 July
95 and warrants a closer look at this suspect scenario at DAKOR station.
The listing of the daily rainfall for neighbouring stations considered for the above spatial
homogeneity test is as given in Table 9.2.
Upon careful examination it can be seen that at DAKOR station the rainfall recorded for few
consecutive days during 11 July 1995 to 27 July 1995 is nil while most of other neighbouring
stations have received significant rainfall on these days. On the next day that is 28 July there
has been a very large value recorded for DAKOR station whereas the other nearby stations
are not experiencing that high rainfall. Such situation does not rule out an un-indicated
accumulation of rainfall at DAKOR for one or more days prior to 28 July.
At this stage the manuscripts of the daily rainfall at DAKOR station must be revisited to
confirm if the data in the databases are properly recorded. If the data are as per the records
then based on the feed back from the observer about his absence/holidays etc. and upon
overall reliability of the station in the past, it can be decided to flag such un-indicated
accumulations for subsequent correction using spatial interpolation (see Module 10).

Table 9.1: Result of spatial homogeneity test at DAKOR station
====================================
Test station DAKOR PH
Start date: 1995 6 1 0 1
End date: 1995 9 30 0 1
1 THASARA PH 8.252
1 VAGHAROLI PH 18.976
2 MAHISA PH 13.948
2 KATHLAL PH 22.216
2 MAHUDHA PH 22.694
2 SAVLITANK PH 23.403
1995 7 15 0 1 .00 - 56.64 20.50 6
1995 7 18 0 1 .00 - 8.79 3.34 6
1995 7 19 0 1 .00 - 21.24 8.73 6
1995 7 20 0 1 .00 - 36.82 15.42 6
1995 7 28 0 1 97.50 + 18.12 13.28 6
1995 7 30 0 1 6.80 - 48.59 16.20 6
Legend
+ = P_obs - P_est > 0
- = P_obs - P_est < 0
Table 9.2: Tabulation of daily rainfall for neighbouring stations
Year mth day hr si DAKOR KATHLAL MAHISA MAHUDHA SAVLITANK THASARA
1995 7 11 .0 7.0 10.0 1.5 27.0 9.0
1995 7 12 .0 .0 3.0 2.0 3.0 17.0
1995 7 13 .0 45.0 .0 .0 .0 .0
1995 7 14 .0 10.0 20.0 7.5 .0 7.0
1995 7 15 .0 14.0 50.0 33.5 24.0 77.0
1995 7 16 .0 .0 8.0 9.5 25.0 8.0
1995 7 17 .0 20.0 4.0 1.0 .0 22.0
1995 7 18 .0 10.0 8.0 1.0 6.0 11.0
1995 7 19 .0 23.0 20.0 43.0 27.0 16.0
1995 7 20 .0 .0 35.0 32.5 14.0 48.0
1995 7 21 .0 57.0 27.0 23.0 14.0 56.0
1995 7 22 .0 .0 6.0 7.0 4.0 .0
1995 7 23 .0 .0 4.0 12.0 2.0 27.0
1995 7 24 .0 10.0 .0 .0 .0 .0
1995 7 25 .0 11.0 10.0 3.0 6.0 3.0
1995 7 26 .0 25.0 .0 10.0 5.0 8.0
1995 7 27 .0 18.0 3.0 4.0 25.0 9.0
1995 7 28 97.5 25.0 24.0 46.0 3.0 12.0
1995 7 29 16.7 40.0 4.0 6.0 .0 .0
1995 7 30 6.8 45.0 34.0 22.0 62.0 52.0
1995 7 31 .0 10.0 3.0 13.0 39.0 9.0

9.3 Screening for accumulations on holidays and weekends
To screen for accumulated values on holidays and weekends it may be appropriate to
prepare a list of all holidays and weekends. Then a comparison is made between observed
and estimated values of daily rainfall of the station under consideration for the period of
holidays and weekends and a day following it. While comparing the two sets, the data points
having significant positive difference between observed and estimated values on the day
following the holidays or weekends are picked up.
10. Missed entries
10.1 General description
Values may be missed from a record either by the observer failing to do the observation,
failing to enter a value in the record sheet or as the result of a mis-entry. A zero may have
been inserted for the day (or days). Similarly, some longer periods may have missed
readings without an accumulated value at the end, for example resulting from breakage of
the measuring cylinder.
10.2 Data validation procedure and follow up actions
For rainy periods such missed values will be anomalous in the multiple station tabulation and
plot and will be indicated by a series of “-ve” departures in the spatial homogeneity test.
Where such missed entries are confidently identified, the missed values will be replaced by
the estimates derived from neighbouring stations by the spatial homogeneity test. Where
there is some doubt as to the interpretation, the value will be left unchanged but flagged as
suspect.
Example 10.1:
The spatial homogeneity test for BHEMPODA station (KHEDA catchment) for the year 1997
is carried out. The results of the test are as given below in Table 10.1:
On examining the above tabular result of the test it can be noticed that there are very many
instances in succession which are flagged “-ve” and also have nil (0 mm) observed rainfall.
At the same time, on these days of “-ve” flag and 0 mm observed rainfall a considerable
rainfall at the neighbouring stations has been reported. Such an inference leads to suspicion
that at this test station BHEMPODA the rainfall has either not been observed and wrongly
reported as 0 mm or has been observed but has been wrongly reported/entered.
The above suspicion is very strongly corroborated after looking at the tabulation of these
neighbouring stations as given in Table 10.2.
It is almost certain that the rainfall at BHEMPODA station has been reported/entered
incorrectly from the second week of August 97 onwards for most of the rainy days reported
at the neighbouring stations. These rainfall values must checked with the records of the data
at BHEMPODA and if the values available in the records are different then those available in
the database then the same must be corrected. Instead, if the manuscript also shows same
values then these have to be flagged for necessary correction subsequently using spatial
interpolation (see Module 10).

Table 10.1: Result of spatial homogeneity test at BHEMPODA station
====================================
Test station BHEMPODA PH
Start date: 1997 6 1 0 1
End date: 1997 9 30 0 1
1 MEGHARAJ PH 20.897
2 RAHIOL PH 17.898
3 ANIOR PH 4.535
3 BAYAD PH 23.253
1997 6 9 0 1 9.00 + .00 .00 4
1997 6 14 0 1 3.00 + .00 .00 4
1997 6 22 0 1 20.00 + 4.79 2.38 4
1997 6 23 0 1 17.00 + 2.11 4.20 4
1997 6 25 0 1 165.00 - 205.65 33.94 4
1997 6 27 0 1 173.00 + 71.55 37.77 4
1997 7 10 0 1 .00 - 1.31 .65 4
1997 7 20 0 1 3.00 + 1.34 .65 4
1997 7 21 0 1 29.00 - 80.48 34.46 4
1997 7 26 0 1 1.00 - 12.73 4.42 4
1997 7 27 0 1 125.00 - 225.13 58.75 4
1997 7 28 0 1 280.00 - 376.98 153.43 4
1997 8 2 0 1 94.00 + 36.15 21.21 4
1997 8 8 0 1 .00 - 20.98 5.32 4
1997 8 9 0 1 .00 - 2.37 .56 4
1997 8 11 0 1 .00 - .44 .22 4
1997 8 14 0 1 .00 - 2.66 1.14 4
1997 8 19 0 1 .00 - 48.96 18.63 4
1997 8 24 0 1 .00 - 87.56 42.17 4
1997 9 11 0 1 .00 - 18.50 6.03 4
1997 9 13 0 1 .00 - 15.36 5.79 4

Table 10.2: Tabulation results for daily rainfall at neighbouring stations
Year mth day hr si ANIOR BAYAD BHEMPODA MEGHARAJ RAHIOL
1997 7 25 .0 .0 .0 1.0 .0
1997 7 26 13.0 11.0 1.0 4.0 16.0
1997 7 27 225.0 147.5 125.0 312.0 209.5
1997 7 28 420.5 194.5 280.0 32.5 60.0
1997 7 29 4.0 1.5 3.0 4.5 5.5
1997 7 30 16.5 9.0 13.0 12.0 7.0
1997 7 31 3.0 4.0 3.0 22.0 1.5
1997 8 1 290.0 257.0 275.0 192.5 129.5
1997 8 2 38.5 57.5 94.0 15.0 2.5
1997 8 3 11.5 28.5 24.0 1.0 11.5
1997 8 4 .0 .0 .0 .0 .0
1997 8 5 15.0 .0 23.0 .0 .0
1997 8 6 .0 .0 .0 .0 .0
1997 8 7 1.0 1.5 .0 1.0 .0
1997 8 8 20.5 25.0 .0 32.0 18.0
1997 8 9 2.5 2.0 .0 1.0 1.5
1997 8 10 .0 .0 .0 .0 .0
1997 8 11 .5 .0 .0 .0 .0
1997 8 12 4.0 1.0 .0 7.0 8.0
1997 8 13 2.5 6.0 .0 1.0 .0
1997 8 14 3.0 1.0 .0 .5 .0
1997 8 15 .0 .0 .0 1.0 .0
1997 8 16 .0 2.0 .0 .0 .0
1997 8 17 .0 .0 .0 .5 1.0
1997 8 18 .0 .0 .0 .0 3.0
1997 8 19 54.0 33.0 .0 12.0 7.0
1997 8 20 .0 7.0 .0 1.0 30.0
1997 8 21 1.0 .0 .0 6.5 .0
1997 8 22 .0 .0 .0 .0 .0
1997 8 23 3.0 .0 .0 9.5 19.5
1997 8 24 91.0 13.5 .0 125.5 50.0
1997 8 25 16.5 33.0 .0 11.0 31.0
1997 8 26 29.0 19.0 .0 54.5 .0
1997 8 27 2.5 5.5 .0 1.0 .0
1997 8 28 .0 .0 .0 .0 .0
1997 8 29 .0 .0 .0 .5 .0
1997 8 30 15.5 33.0 .0 .5 31.5
1997 8 31 .0 .0 .0 .0 .0
11. Rainfall missed on days with low rainfall - rainy days check
Whilst it is required that observers inspect the raingauge for rain each day, the practice of
some observers may be to visit the gauge only when they know that rain has occurred. This
will result in zeros on a number of days on which a small amount of rain has occurred. Totals
will be generally correct at the end of the month but the number of rainy days may be
anomalously low. In addition spatial homogeneity testing may not pick up such differences.
-Owing to spatial homogeneity with respect to the occurrence of rainfall within the day, it is
expected that the number of rainy days in a month or year at the neighbouring stations will
not differ much. Presently, there are two definitions for number of rainy days: some agencies
consider a minimum of 0.1 mm (minimum measurable) in a day to be eligible for the rainy
day whereas some use 2.5 mm and above as the deciding criteria. The later is used more

often in the agriculture sector. For the hydrological purpose it is envisaged that the definition
of minimum measurable rainfall (i.e. 0.1 mm) will be used for the data validation.
It is good to utilise this fact to see if the observed data follow such characteristic. A graphical
or tabular comparison of the difference in the number of rainy days for the neighbouring
stations for the monthly or yearly period will be suitable in bringing out any gross
inconsistency. The tolerance in the number of rainy days between the stations has to be
based on the variability experienced in the region and can easily be established using
historical data. If the difference is more than the maximum expected, the data may be
considered suspect. Any gross inconsistency noticed must then be probed further by looking
at the manuscript and seeking a report on, or inspecting the functioning and behaviour of the
observer.
First of all, with the help of historical daily rainfall data, belonging to a homogenous region,
the expected maximum variation in the number of rainy days for each month of the year and
for year as a whole is found out. A group of stations being validated is then chosen and the
number of rainy days at each station within the month(s) or year obtained. The number of
rainy days at each station is then compared with every other station in the group. All those
instances when the expected variation is exceeded by the actual difference in the number of
rainy days is presented in tabular or graphical form. It is appropriate to present the output in
a matrix form in which the stations are listed as rows and columns of the table or the graph.
In case the presentation is on the monthly basis then each tabular or graphical matrix can
accommodate a period of one year.
Any glaring departure in the number of rainy days, at one or more stations, can be apparent
by inspecting the matrix. The station for which the number of rainy days is much different
from others will have the column and row with lower (or occasionally higher) values. The
data pertaining to such months or years of the station(s) for which the difference in the
number of rainy days is beyond the expected range is considered suspect and has to be
further probed. The original observer’s manuscript for the suspect period can be compared
with the values available in the database. Any discrepancy found between the two can be
corrected by substituting the manuscript values. Where the manuscript matches with the
data available in the database then comparison with other related data like temperature and
humidity at the station, if available, can be made. Together with analytical comparison,
feedback from the observer or supervisor will be of a great value in checking this validation
especially where it is done within one or two months of the observations. If the related data
corroborate the occurrence of such rainy days then the same can be accepted.
Where there is strong evidence to support the view that the number of rainy days derived
from the record is incorrect, then the total may be amended by reference to neighbouring
stations. Such action implies that there are unreported errors remaining in the time series,
which it has not been possible to identify and correct. A note to this effect should be included
with the station record and provided with the data to users.
As a follow up measure a report can be sought on the functioning and behaviour of the
observer.

12. Checking for systematic shifts using double mass analyses
Double mass analysis is a technique that is effective in detecting a systematic shift,
like abrupt or gradual changes in the mean of a series, persisting in the record for a
considerable period of time. Rainfall record contains such inconsistencies which may exist
for a considerable period of time. Inconsistencies present in the rainfall data of a station
can occur for various reasons:
•
• The raingauge might have been installed at different sites in the past
• The exposure conditions of the gauge may have undergone a significant change due to
the growth of trees or construction of buildings in its proximity
• There might have been a change in the instrument, say from 125 mm to 200 mm
raingauge
• The raingauge may have been faulty for a considerable period etc.
Such inhomogeneity in the data set must be removed before any statistical inference can be
drawn. The double mass analysis tests the record for its inconsistency and accuracy and
provides a correction factor to ensure that the data series is reasonably homogeneous
throughout its length and is related to a known site. A note may be available in the station
registers of the known changes of site and instruments and can corroborate the detection of
inconsistency using this technique. The application of double mass analysis to rainfall data
will not be possible until a significant amount of historical data have been entered to the
database.
12.2 Description of method
Double mass analysis is a technique to detect possible inhomogeneities in series by
investigating the ratio of accumulated values of two series, viz.:
- the series to be tested, and
- the base series
The base series is generally an artificial series, i.e. the average of reliable series of nearby
stations (usually 3 as minimum) which are assumed to be homogenous.
First of all the accumulated test and base series are obtained as two vectors (say Yi and Xi
respectively, for i = 1, N). The double mass analysis then considers the following ratio:
or expressed as a ratio of the percentages of the totals for N elements:
pc
Y
Y
X
X
i
j
j
i
j
j
N
j
j
N
j
j
i
=
=
=
=
=
∑
∑
∑
∑
1
1
1
1
.
.
∑
∑
=
=
= i
j
j
i
j
j
i
X
Y
rc
1
1

These ratios in absolute and percent form gives the overall slope of the double mass plot
from origin to each consequent duration of analysis.
A graph is plotted between the cumulative rainfall of the base series as abscissa and the
cumulative rainfall of test station as the ordinate. The resulting plot is called the double mass
curve. If the data of test station is homogeneous and consistent with the data of the base
series, the double mass curve will show a straight line. An abrupt change in the test-series
will create a break in the double mass curve, whereas a trend will create a curve. Graphical
inspection of the double mass plot provides the simplest means of identifying such
inconsistencies but significance tests may also be used to identify breaks and jumps. A
change in slope is not usually considered significant unless it persists for at least 5 years
and there is corroborating evidence of a change in location or exposure or some other
change. There is a regional consistency in precipitation pattern for long periods of time but
this consistency becomes less pronounced for shorter periods. Therefore the double mass
technique is not recommended for adjustment of daily or storm rainfalls. It is also important
to mention here that any change in regional meteorological or weather conditions would not
have had any influence on the slope of the double mass curve because the test station as
well as the surrounding base stations would have been equally affected.
It must also be emphasised here that the double mass technique is based on the
presumption that only a part of the data under consideration is subjected to systematic error.
Where the whole length of the data being considered has such an error then the double
mass analysis will fail to detect any error.
For analysing the rainfall data for any persistent systematic shift, the accumulated rainfall for
longer duration at the station under consideration (called the test station) is compared with
another accumulated rainfall series that is expected to be homogeneous. Homogeneous
series for comparison is derived by averaging rainfall data from a number of neighbouring
homogenous stations (called base stations).
Accumulation of rainfall can be made from daily data to monthly or yearly duration. The
double mass plot between the accumulated values in percent form at test and base station is
drawn and observed for any visible change in its slope. The tabular output giving the ratio
between the accumulated values at test and base station in absolute and percent is also
obtained. In case, there are some missing data points within each duration of analysis, a
decision can be made about the number of elements which must essentially be present for
that duration to be considered for analysis. The analysis, if required, can also be carried for
only a part of the years or months.
Where there is a visible change in the slope of the double mass plot after certain period then
such a break must be investigated further. Possible reasons for the inhomogeneity in the
data series are explored and suitable explanation prepared. If the inhomogeneity is caused
by changed exposure conditions or shift in the station location or systematic instrumental
error then the data series must be considered suspect. The data series can then be made
homogeneous by suitably transforming it before or after the period of shift as required.
Transformation for inconsistent data is carried out by multiplying it with a correction factor
which is the ratio of the slope of the adjusted mass curve to the slope of the unadjusted
mass curve (see Module 10 for details).

Example 12.1:
Double mass analysis for VADAGAM station (in KHEDA catchment) is carried out
considering two stations MEGHARAJ and BAYAD as the base stations for the period from
1968 to 1996. A period of only three months from July to September (92 days) has been
taken into consideration while carrying out the analysis. Though the reliability of records and
the homogeneity of these base stations have to be ascertained before considering them for
the analysis but here it has been assumed that they are reliable stations.
It can be seen from double mass plot of this analysis, as shown in Fig. 12.1, that the data of
VADAGAM station is fairly consistent throughout the period of analysis (1968 to 1997) with
respect to the other two base stations. Baring a few short-lived very small deviations from
the ideal curve (of 45°), the plot shows a similar trend throughout the period.
The result of this analysis on yearly basis is given in Table 12.1. The yearly rainfall and the
rainfall accumulated in time for the base and test station is given in columns 2, 3 and 5, 6
respectively. These cumulative rainfall values are then expressed in percent form in columns
4 and 7 respectively. The ratio of these cumulated values in absolute in percent form are
given in the last two columns 8 & 9.
Fig. 12.1: Double mass plot showing near consistent trend at test station
Double Mass Curve - (1968 - 1997)
Base Series (%) - (MEGHARAJ & BAYAD)
10095908580757065605550454035302520151050
TestSeries(%)-(VADAGAM)
100
95
90
85
80
75
70
65
60
55
50
45
40
35
30
25
20
15
10
5
0

Table 12.1: Analysis result of the double mass analysis
Test series: VADAGAM PH
Weight
Base series: MEGHARAJ PH .50
BAYAD PH .50
1 2 3 4 5 6 7 8 9
Period BASE TEST Ratios
Amount Cum Perc Amount Cum Perc (6)/(3) (7)/(4)
1968 451.5 452. 2.5 382.4 382. 2.2 .85 .88
1969 487.5 939. 5.3 437.0 819. 4.8 .87 .90
1970 957.4 1896. 10.7 743.1 1563. 9.1 .82 .85
1971 462.3 2359. 13.3 443.4 2006. 11.7 .85 .88
1972 332.1 2691. 15.2 339.1 2345. 13.7 .87 .90
1973 1124.8 3816. 21.5 1266.3 3611. 21.0 .95 .98
1974 247.8 4063. 22.9 214.9 3826. 22.3 .94 .97
1976 910.2 4974. 28.0 831.6 4658. 27.1 .94 .97
1977 751.0 5725. 32.3 1124.1 5782. 33.7 1.01 1.04
1978 735.0 6460. 36.4 748.2 6530. 38.0 1.01 1.05
1979 576.0 7036. 39.6 389.1 6919. 40.3 .98 1.02
1980 205.3 7241. 40.8 234.3 7154. 41.7 .99 1.02
1982 323.6 7565. 42.6 417.7 7571. 44.1 1.00 1.03
1983 766.3 8331. 46.9 817.4 8389. 48.9 1.01 1.04
1984 737.8 9069. 51.1 737.0 9126. 53.2 1.01 1.04
1985 312.4 9381. 52.8 198.4 9324. 54.3 .99 1.03
1986 313.8 9695. 54.6 229.6 9554. 55.7 .99 1.02
1987 337.3 10032. 56.5 261.9 9816. 57.2 .98 1.01
1988 986.0 11018. 62.1 837.7 10653. 62.1 .97 1.00
1989 605.8 11624. 65.5 493.0 11146. 64.9 .96 .99
1990 1047.8 12672. 71.4 1065.5 12212. 71.1 .96 1.00
1991 481.0 13153. 74.1 508.5 12720. 74.1 .97 1.00
1992 596.8 13749. 77.5 697.0 13417. 78.2 .98 1.01
1993 598.0 14347. 80.8 599.0 14016. 81.7 .98 1.01
1994 1101.0 15448. 87.0 1079.5 15096. 87.9 .98 1.01
1995 592.5 16041. 90.4 478.5 15574. 90.7 .97 1.00
1996 746.8 16788. 94.6 647.6 16222. 94.5 .97 1.00
1997 963.0 17751. 100.0 944.0 17166. 100.0 .97 1.00
Total number of periods analysis: 28
Example 12.2:
The long term data series of rainfall for the period 1970 to 1996 is considered at VADOL
station (in KHEDA catchment) for double mass analysis taking three nearby stations
KAPADWANJ, MAHISA and THASARA. Unlike the previous example, which is a case of the
test station being homogeneous in time, this example illustrates a case where the test
station records shows that there has been a significant change in the amount of rain over a
period of time.
It can be easily seen from the double mass curve shown in Fig. 12.2, that the behaviour of
the test station suddenly changes after about half of the time period under consideration.
This turning point corresponds with the year 1984 and is also apparent from the values of
the ratios of accumulated rainfall at test and base stations as given in Table 12.2 showing
the results of the test.

Fig. 12.2: Double mass curve for VADOL station showing significant change of slope
of the curve after about half the period under consideration.
Table 12.2: Results of the double mass analysis
Double mass analysis
Test series: VADOL PH
Weight
Base series: KAPADWANJ PH .33
MAHISA PH .33
THASARA PH .33
1 2 3 4 5 6 7 8 9
Period BASE TEST Ratios
Amount Cum Perc Amount Cum Perc (6)/(3) (7)/(4)
MM MM MM MM - -
1970 767.4 767. 4.6 624.4 624. 4.5 .81 .98
1971 454.0 1221. 7.3 426.0 1050. 7.6 .86 1.04
1972 372.5 1594. 9.5 197.9 1248. 9.0 .78 .94
1973 935.3 2529. 15.1 1114.2 2363. 17.0 .93 1.13
1974 240.3 2769. 16.6 72.8 2435. 17.6 .88 1.06
1977 843.8 3613. 21.6 882.8 3318. 23.9 .92 1.11
1978 646.4 4260. 25.5 758.8 4077. 29.4 .96 1.15
1979 436.7 4696. 28.1 370.2 4447. 32.1 .95 1.14
1980 450.2 5147. 30.8 388.9 4836. 34.9 .94 1.13
1981 950.0 6097. 36.5 898.1 5734. 41.4 .94 1.13
1982 403.6 6500. 38.9 320.1 6054. 43.7 .93 1.12
1983 801.4 7302. 43.7 882.1 6936. 50.0 .95 1.15
1984 806.0 8108. 48.5 475.1 7411. 53.5 .91 1.10
1985 364.2 8472. 50.7 82.8 7494. 54.1 .88 1.07
1986 281.5 8753. 52.3 234.0 7728. 55.7 .88 1.06
1987 257.7 9011. 53.9 227.5 7956. 57.4 .88 1.06
1988 866.1 9877. 59.1 734.5 8690. 62.7 .88 1.06
1989 877.0 10754. 64.3 693.3 9384. 67.7 .87 1.05
1990 1145.0 11899. 71.2 746.0 10130. 73.1 .85 1.03
1991 682.7 12582. 75.2 618.1 10748. 77.5 .85 1.03
1992 697.7 13279. 79.4 422.2 11170. 80.6 .84 1.01
1993 639.8 13919. 83.2 512.8 11683. 84.3 .84 1.01
1994 1350.0 15269. 91.3 1083.3 12766. 92.1 .84 1.01
1995 525.0 15794. 94.5 371.6 13137. 94.8 .83 1.00
1996 926.7 16721. 100.0 725.0 13862. 100.0 .83 1.00
Total number of periods analysis: 25
Double Mass Curve - (1970 - 1996)
Base Series (%) - (KAPADWANJ, MAHISA & THASARA)
10095908580757065605550454035302520151050
TestSeries(%)-(VADOL)
100
95
90
85
80
75
70
65
60
55
50
45
40
35
30
25
20
15
10
5
0

It is amply clear that from the year 1985 onwards the test station, i.e. VADOL, started
receiving rainfall which is comparatively lesser than what it use to receive before that time.
And this change in behaviour is not short lived but is continuos thereafter. The reasons for
such variations are required to be ascertained. Various factors which could results in such a
change can be: (a) a systematic error in the observation of rainfall after the year 1983 or (b)
a possible change in the meteorologic factors around the test station (which is very unlikely
since any meteorologic changes would generally be spread wide enough to cover all and
more neighbouring stations). For both the possibilities the reasons have to be identified
beyond doubt before any corrective measure can be taken. A visit to the station and looking
at the exposure conditions etc. and taking the history from the observer will be very useful in
trying to establish the reasons of this change in the behaviour.

Download-manuals-hydrometeorology-data processing-09howtocarryoutsecondaryvalidationofrainfall

More Related Content

What's hot (18)

Viewers also liked (20)

Similar to Download-manuals-hydrometeorology-data processing-09howtocarryoutsecondaryvalidationofrainfall (20)

More from hydrologyproject001 (20)

Recently uploaded (20)

Download-manuals-hydrometeorology-data processing-09howtocarryoutsecondaryvalidationofrainfall