Post on 28-Sep-2015
LONG TERM HYDROLOGIC TRENDS IN THE NILE BASIN
A Thesis
Presented to the Faculty of the Graduate School
of Cornell University
In Partial Fulfillment of the Requirements for the Degree of
Master of Professional Studies
by
Zelalem Kassahun Tesemma
May 2009
2009 Zelalem Kassahun Tesemma
ABSTRACT
A study has been conducted to examine if and how streamflow in the Nile Basin has
varied over the period of available records. Streamflow records from 13 flow gauging
stations in four major river basins of the Nile and 38 precipitation stations all over the
Nile basin were studied. Monthly measured discharge (1912-1982) and rainfall data
for those selected stations were collected from four different data sources and Global
Hydro Climate Data Network available at http://dss.ucar.edu/datasets/ds553.2/data/
and Global Historical Climatology Network available at http://gpcc.dwd.de were
selected as the main data sources except those Ethiopian stations. The remaining
recent 20 years data were collected from countries. Monthly and annual streamflows
(up to the year 2000, some up to 2007) were extracted and analyzed for each of the 13
station. The raw data were validated thoroughly by comparing different sources,
corrected and augmented if needed.
The Mann-Kendal and Sens T non-parametric test was used to detect significant
trends in the selected years in combination with the Trend Free Pre-Whitening
(TFPW) method for correcting time series data from serial correlation. The slope of
the data set was computed using the Thiel-Sen Approach (TSA). For this study a 5-
percent level of significance was selected to indicate the presence of statistical
significant trends. Rainfall-Runoff Modeling was done on the upper Blue Nile using
the Thronthwait-Mather model to understand the land cover changes on runoff over
the past 30 years.
The mean annual natural streamflow on the Blue Nile Stations (Bahir Dar, Kessie and
El Diem) show no trend. The rainfall over the basin also shows no significant trend.
The Monthly runoff showed moderate variability at El Diem with 19% and 34% at
Bahir Dar and Kessie. This might be a result that more land was cultivated growing of
different crops as shown by rainfall-runoff modeling over the last 30 years. White Nile
Stations (Jinja, Mongalla and Malakal) show a significant increasing trend on both
rainfall and streamflow. The runoff increased 72%, 67% and 20% of the mean annual
flow at Jinja Mongalla and Malakal respectively. Stations of the Main Nile
(Tamaniate, Hassanab and Dongolla) show significant decreasing trend in streamflow
due to abstraction of flow before reach gauging stations. For water resources
management the key conclusion, that Nile natural streamflows have not changed
significantly during the last 100 years.
iii
BIOGRAPHICAL SKETCH
NAME BIRTH DATE
Mr. Zelalem Kassahun Tesemma April, 23, 1984 EDUCATION
INSTITUTION DEGREE YEAR CONFERRED FIELD OF STUDY
Bahir Dar University B.Sc. 2007
MAJOR RESEARCH AREA OF INTEREST
Water- and Environmental-related research
CURRENT RESEARCH
Long term hydrologic trends in the Nile basin
iv
This research is dedicated to my best friend Mebrahtom Gebre Hiwot.
v
ACKNOWLEDGMENTS
I wish to express appreciation to Dr. Yasir Mohamed and Dr. Tammo Steenhuis, for
their support and assistance, constructive comments and corrections throughout the
period of the research. Special thanks to Dr. Yasir Mohamed for providing me helpful
materials and for his patience and persistence in obtaining data from different
countries.
I also appreciate Dr. Amy Collick for her fruitful comment and continuous support in
materials throughout the research period.
The author was granted access to internet, office and financial support for this research
from International Water Management Institute (IWMI) expresses his gratitude for
this privilege.
I would like to acknowledge the Ethiopian Ministry of Water Resources, NBI-ENTRO
and Ethiopian Metrological Agency and Sudan Ministry of Water Resources and
Irrigation.
vi
TABLE OF CONTENTS
BIOGRAPHICAL SKETCH ......................................................................................... iii
ACKNOWLEDGMENTS .............................................................................................. v
TABLE OF CONTENTS .............................................................................................. vi
LIST OF FIGURES ..................................................................................................... viii
LIST OF TABLES ......................................................................................................... x
LIST OF ABBREVIATIONS ....................................................................................... xi
1. INTRODUCTION ...................................................................................................... 1
1.1 Problem Statement .......................................................................................... 2
1.2 Objective of the research ................................................................................ 3
2. LITERATURE REVIEW ........................................................................................... 4
2.1 Step Trend versus Monotonic Trend .................................................................... 4
2.2 Non parametric versus parametric methods ......................................................... 4
2.3 Trend analysis in the Nile Basin. .......................................................................... 6
3. THE NILE BASIN ..................................................................................................... 7
3.1 Climate, Hydrology and Water Resources of the Nile basin ................................ 8
3.2 Description of the main sub basins ..................................................................... 11
3.2.1 White Nile River sub basin .......................................................................... 11
3.2.2 Blue Nile River sub basin ............................................................................ 12
3.2.3 Atbara River sub basin ................................................................................ 12
4. RESEARCH METHODOLOGY ............................................................................. 13
4.1 Selection of variables ......................................................................................... 13
4.2 Data validation and completion .......................................................................... 14
4.3 Mann-Kendall and Seasonal Kendall test for trend detection ............................ 15
vii
4.4 Sens T test ......................................................................................................... 20
4.5 Rainfall-Runoff modeling of Upper Blue Nile ................................................... 21
5. DATA COLLECTION AND VALIDATION ......................................................... 25
5.1 Data collection and Pre-processing .................................................................... 25
5.2 Selection of stations ............................................................................................ 27
5.3 The Nile stream gauging stations and validation of data ................................... 29
6. RESULTS AND DISCUSSION ............................................................................... 40
6.1 Results of statistical analysis .............................................................................. 40
6.1.1 Pettitt test results .......................................................................................... 40
6.1.2 Runoff trend results ..................................................................................... 41
6.1.3 Precipitation trend results ............................................................................ 47
6.2 Model Results ..................................................................................................... 49
7. CONCLUSIONS ...................................................................................................... 52
8. REFERENCES ........................................................................................................ 54
APPENDIX .................................................................................................................. 61
Appendix I: Location of stations and data availability graphs. ............................... 61
Appendix II: Relation Curves and regression equation comparison plots ............... 65
Appendix III Tables and figures of statistical analysis ............................................ 68
viii
LIST OF FIGURES
Figure 3.2 Map (not to scale) of the Nile Basin which shows the major supply
structures (Adopted Alan, 2005, SOAS). ..................................................................... 10
Figure 5.1 Comparison plot between data sources at Jinja, show a 3rd data source ..... 26
Figure 5.2 Comparison plot between data sources at Tamaniate. ................................ 27
Figure 5.3 Comparison plot between data sources at Aswan ....................................... 27
Figure 5.4 Map of the Nile basin showing the locations of the precipitation station and
stream gauges used in this analysis. ............................................................................. 30
Figure 5.5 Comparison of Mean flow and maximum and minimum daily flow at Bahir
Dar. ............................................................................................................................... 34
Figure 5.6 Comparisons between Outflow at Bahir Dar and Lake Level of Tana. ...... 35
Figure 5.7 Relation curve between flow at Tamaniate and Hassanab .......................... 39
Figure 6.1 Natural and Observed mean annual runoff trend at Sennar station of the
Upper Blue Nile River. ................................................................................................. 46
Figure 6.2 Balance between natural and observed mean annual runoff at Sennar
station. .......................................................................................................................... 46
Figure 6.3 Mean monthly areal and average rainfall distribution over Upper Blue Nile
and White Nile. ............................................................................................................. 47
Figure 6.4 Coefficient of variation in rainfall for Upper Blue Nile and White Nile. ... 47
Appendix I.2 Data availability graphs for Streamflow and rainfall stations
(downloaded from internet). ......................................................................................... 62
Appendix I.3 Data availability graphs for precipitation stations. ................................. 62
Appendix II.1 Regression equations for data validation and completion. ................... 65
ix
Appendix III.1 Pettit change point test result for the average precipitation over
Victoria Nile. ................................................................................................................ 68
Appendix III.2 Pettit change point test result for the areal precipitation over upper
Blue Nile. ...................................................................................................................... 69
Appendix III.3 Pettit change point test result for the runoff over Nile basin. .............. 70
AppendixIII.8 Linear trend in monthly streamflow for the Bule Nile River Basin at
Roseires/El Diem from 1912 to 2000.Statistically significant decreases in flow are
highlighted with heavy line. Total and percent changes are expressed relative to the
beginning of the trend. Annual discharge is expressed m3/s ........................................ 78
AppendixIII.7 Linear trend in monthly streamflow for the Nile River Basin at Jinja
from 1912 to 2000.Statistically significant increases in flow are highlighted with
heavy line. Total and percent changes are expressed relative to the beginning of the
trend. Annual discharge is expressed m3/s. .................................................................. 79
AppendixIII.9 Linear trends in average monthly precipitation for the Victoria Nile
Basin from 1912 to 2000. Statistically significant trend in precipitation are highlighted
with heavy line. Total and percent changes are expressed relative to the beginning of
the trend. Annual Precipitation is expressed in mm ..................................................... 80
AppendixIII.10 Linear trends in areal monthly precipitation for the Upper Blue Nile
Basin from 1965 to 2000. Annual Precipitation is expressed in mm ........................... 81
x
LIST OF TABLES
Table 3.1 Major Nile Basin supply-side structures (Adopted from Alan, 2005). .......... 9
Table 5.1 Total number of data corrected during validation and then completed. ....... 26
Table 5.2 List of data availability. ................................................................................ 28
Table 5.3 List of data sources used in the analysis. ...................................................... 29
Table 6.1 Results of change points with pettitt test for runoff and precipitation. ........ 41
Table 6.2 Results of the runoff trend test for Upper White Nile .................................. 42
Table 6.3 Results of the runoff trend test for Upper Blue Nile .................................... 44
Table 6.4 Results of the precipitation trend test for White and Upper Blue Nile. ....... 48
Table 6.5 Calibrated and validated parameters showing land use/cover change during
the past 30 years. .......................................................................................................... 50
Appendix I.1 Names and Location of rainfall stations. ................................................ 61
Appendix III.4 Trend analysis results for flow gauges in the Nile Basin by months for
different time periods. .................................................................................................. 71
Appendix III.5 Trend analysis results for flow gauges in the Nile Basin from 1912-
2000 by months ............................................................................................................ 75
Appendix III.6 Trend analysis results for average and areal precipitation in the
Victoria Nile and Upper Blue Nile Basin by months. .................................................. 77
Appendix III.11 Trend analysis results for Precipitation Station in the Nile Basin by
Months for different time period. ................................................................................. 82
xi
LIST OF ABBREVIATIONS
TSA-Thiel-Sen Approach
GHCDN- Global Hydro-Climate Data Network
GHCN Global Historical Climatology Network
MN- Mann-Kendall
SM-Soil Moisture
WOM- World Meteorological Organization
1
CHAPTER ONE
1. INTRODUCTION
The hydrology of Nile basin has been studied from many perspectives. Several studies
concerned with the long-term climatologic trends and especially precipitation
(Conway, 2000; Yilma and Demarce, 1995 and Sutcliffe and Parks, 1999). Other
studies relate the effect of climate change and spatial variability of precipitation to
streamflow; (Conway and Hulme, 1993) and developing water balance model for
water resource management (Conway, 1997; Kebede and Travi, 2006) and for
sensitivity analysis of lake level and outflows such as Lake Victoria (Lyons, 1906).
Egypt and Sudan almost completely depend on the Nile water as water source, with
water demand in Egypt alone set to increase (Conway, 1993). It is critical that the role
of future climate change on Nile water management is well understood. On the other
hand, other Nile basin countries want to increase their share of the Nile Water for
economic development which may be a cause for a potential conflict between the
riparian Countries (Shiva, 2002). Similarly, the Nile riparian acknowledges the need
for basin scale management of the Nile water resources (e.g., the Nile Basin
Initiative), if they want to achieve maximum benefit of the resource. Because of these
challenges towards utilization of the Nile waters resources there is great anxiety about
reduction in available water due to future climate changes. Since experts do not agree
on how the streamflow will change in the future, one of the few remaining ways is
studying past trends in streamflow and rainfall. This will help a better understanding
of the change in discharge caused by implementing past practices and use this
information in developing strategies for better utilization of water resources in the
future.
2
The particularities of the trend analysis given in this study it cover the entire Nile
basin as a study domain, consider both the effect of serial correlation and cross
correlation effect on the analysis of the statistical Mann-Kendall test, not only
interblock procedure such as the Mann-Kendall test but also will include an aligned
procedure such as Sens T test to obtain extra confidence in the results and this
research will examined trend characteristics for different time interval at the same
time, such as monthly, seasonal and annual bases, to see whether or not a dramatic
change occurs.
1.1 Problem Statement
The hypothesis is that the assessment of long term trend of time series of discharge
and precipitation at selected locations within the Nile basin will lead to insights into
future Nile water availability. The key questions to be asked in this research are then:
What is the long term trend (100 years) of monthly and annual stream
discharge at key locations in the Nile basin?
What is the long term trend (100 years) of monthly and annual precipitation at key location in the Nile basin?
How is the relation between discharge and rainfall changes with time and also land use/cover change during the last 30 years (1961-1970 and 1991-
2000)?
Using discrete statistical indicators can we determine the trends and long term variability of Nile flows and precipitation?
3
1.2 Objective of the research
The main objective of this research is to determine whether there is evidence of long-
term trends in streamflow as well as precipitation over the entire Nile basin, and if so,
to determine the long term mean annual streamflow from the Nile basin. Another
objective is to investigate relationships between streamflow and precipitation.
This research will provide updated information on the effect of climate change and
climate variability on streamflow in the Nile basin. Such knowledge is vital for the
riparian countries, as well as for their joint effort for cooperative management (e.g.,
Nile Basin Initiative).
4
CHAPTER TWO
2. LITERATURE REVIEW
Information has been gathered from both published, grey literature and basin water
resource management plans such as Abbay river basin master plan (Ethiopia).
Hydrologic studies of the whole basin, workshop proceedings and annual reports of
the countries in the Nile Basin were reviewed. Based on this literature, the
methodology has been developed.
2.1 Step Trend versus Monotonic Trend
Two primary types of long-term trends can be considered in hypothesis testing and
trend estimation. One is the Monotonic trend; the other is the step trend. (Hirsch et al.,
1991). Monotonic trend tests are applied in this study. Step trend test is applied for
those stations with naturally broken in to two distinct periods with relatively long time
gap between them (Helsel and Hirsch, 1992). The other is when human influence or
diversion structure which likely result in a change in streamflow. Monotonic trend test
is applied for those stations with no human influence and diversion.
2.2 Non parametric versus parametric methods
The assumption of the parametric approach (i.e. normality, linearity and
independence) is mostly not satisfied by hydro-climatologic data (Huth and Pokorna,
2004; Van Belle and Hughes, 1984; Helsel and Hirsch, 1988). A statistical trend
analysis will be performed to determine if significant time trends existence for mean
5
monthly and annual streamflow and precipitation at each of the representative
locations. For a detection of the statistical signification of a trend, non-parametric
methods have been used in several studies (Zhang et al, 2001; Huth and Pokorna,
2004; Harry et al, 1999; Kahya, and Kalayci, 2004; Xu et al, 2003; Partal and Kalya,
2006 and Yue and Hashimoto, 2003).
Non-parametric methods were found to be suitable for data commonly skewed, and
the sample size is large. (Hirsch et al, 1982) Non-parametric methods not only tend to
be more resistant to a misbehavior of the data (e.g. outliers) but also are give results
close to their parametric counterparts and lay well within the confidence limits even
the distributions are normal (Huth and Pokorna, 2004). Regarding all the point
discusses above it is suitable to use non parametric methods for trend analysis. Some
of the research used non parametric method and the results were satisfactory (Zhang et
al, 2001; Huth and Pokorna, 2004; Harry et al, 1999; Kahya and Kalayci, 2004; Xu,
2003; Partal and Kalya, 2006 and Yue and Hashimoto, 2003).
Non-parametric tests are more robust compared to their parametric counterpart.
Speaking on the power of the test, i.e. ability to distinguish between the null
hypothesis and alternative hypothesis, the Mann-Kendall tests (Mann, 1945 and
Kendall, 1975) for monotonic trends perform well in comparison to the parametric t-
test (Van Belle and Hughes, 1984). Mann-Kendall test will be used to test for trends
over time. This test is non-parametric test, has been widely used to test for randomness
against trend in hydrology and climatology (e.g. Burn and Elnur, 2002; Zhang et al
2000; and Pokorna, 2004; Harry et al, 1999; Kahya and Kalayci, 2004; Xu, 2003;
Partal and Kalya, 2006 and Yue and Hashimoto 2003).
6
The problem in using Mann-Kendall test is the result is affected by serial correlation
of the time series. If there is a positive serial correlation (persistence) in the time
series, the test will suggest a significant trend in a time series which is actually random
more often than specified by the significance level Kulkarni and Von Storch (1995).
To remove the effect of serial correlation Von Storch (1995) suggest that the series be
pre-whitened before applying the Mann-Kendall test.
2.3 Trend analysis in the Nile Basin.
Some literature has considered climate change and variability of Nile flow. Conway
and Hulme (1996) studied the variability in precipitation and streamflow on the whole
Nile and found causes for the historical fluctuation in main Nile runoff was the
fluctuation in precipitation. In addition they found no correlation in precipitation and
runoff between Blue Nile and White Nile and the precipitation and runoff over the
upper Blue Nile basin displayed no significant temporal trend. Sutcliffe and Parks
(1999) work showed Blue Nile and Atbara flow are variable and declined in the 1970s
and 1980s and Main Nile stations showed high flow up to 1990 and the variable flows
until 1970 and low flow since 1970.
7
CHAPTER THREE
3. THE NILE BASIN
The Nile basin lies to the north east of Africa joining the Lake Victoria to the
Mediterranean Sea. The Nile River, with an estimated length of over 6800km is
longest river in the world flowing from south to north over 35 of latitude. The total
area of the Nile basin (3112400 km2), covers 10% of the area of the African continent
and is shared by 10 riparian countries: Ethiopia, Sudan, Egypt, Tanzania, Burundi,
Democratic Republic of Congo, Eritrea, Rwanda and Uganda (Figure 3.1).
Figure 3.1 Location Map of the study area (Map of Nile basin).
8
3.1 Climate, Hydrology and Water Resources of the Nile basin
The climate and vegetation cover in the Nile basin are highly related with the amount
of precipitation. Precipitation increases southward and with altitude. The common area
with high precipitation about 1200-1600mm/years on the highlands of Ethiopia and
the Equatorial lakes plateaus. The potential Evaporation over the basin increases as
one move downstream which show opposite trend to the precipitation (Mohamed et al,
2005).
In addition to the main route of Nile there are so many tributaries and lakes which feed
into the Nile. After the Nile leaves Lake Victoria it receives water from lakes
(Kayoga, Albert and Edward) and streams. North of Mongalla, the White Nile is
known as the Bahr el Jebel and flows into a vast complex of channels, Lakes, and
Swamps in an enclosed basin. Bahr elghazal, coming from west, has very little
contribution to the Nile flow. A comparison of the historical inflow data at Mongalla
(33.332 km3) and outflow data at Malakal (29.714 km3) shows a Negative balance of
3.619 km3. Taking in to account that the Sobat river contributes on average 13.53 km3
of water per year to the flow at Malakal one can easily conclude that more than half of
the river inflow is lost by evaporation, evapotranspiration and ground water losses
(Sutcliffe and Parks, 1999). White Nile joins Blue Nile at Khartoum and name as main
Nile. The source of Blue Nile is the little Abbay River originated in the Ethiopian
highlands. The little Abbay flows in to Lake Tana, which discharges into the Blue Nile
and runs down through the highlands into Sudan. The long term mean annual flow of
Blue Nile measured at Roseires/El Diem is 48.65 billion m3 and contribute about 60%
of the flow of main Nile.
9
The last tributary of main Nile is Atbara which also originated from Ethiopia and
Eritrea highlands is highly seasonal and contributes on average about 11.05 billion m3
per years. Use of water resource of Nile basin needs agreement and cooperation
between the riparian countries for sustainable utilization. The major reservoirs
constructed on the Nile river basin are Angeber and Koka in Ethiopia for irrigation
and hydropower, Roseires, Sennar and Khashm El Girba in Sudan for irrigation and
Aswan High Dam in Egypt for irrigation. Refer to Figure 3.2 and Table 3.1 for further
information on the reservoirs.
Table 3.1 Major Nile Basin supply-side structures (Adopted from Alan, 2005).
Structure and Location Main function
Date completed
Old Aswan Dam (Egypt)
For irrigation in Egypt, saving some 1 bcm of water; heightened in 1912 and later in 1934, increasing storage capacity to 5.1 bcm.
1902
Sennar Dam (Sudan)
On the Blue Nile in Sudan, 350 km from Khartoum. Completed in 1925 to supply the Gezira Scheme. Storage of 0.8 bcm.
1925
Jebel Aulia (Sudan)
On the White Nile 44 km south of Khartoum to store water for summer irrigation in Egypt. 1937
Owen Falls Dam
(Uganda)
Built at the outlet of the White Nile from Lake Victoria to generate hydroelectricity for Uganda. 1954
Aswan High Dam (Egypt)
To capture an entire years Nile flood, thereby allowing Egypt complete control of Nile flows downstream.
1968
Kashem el-Girba (Sudan)
Built to serve the New Halfa irrigation scheme built to its storage was 1.3 bcm, but fell dramatically because of siltation such that by 1971 this was just 0.97 bcm.
1964
Rosaries (Sudan)
Supplies water to the Gezira Managil extensions and the Rahad scheme. Also produces hydropower for the Sudanese network.
1966
Jonglei Canal Construction halted. Anticipated increase in discharge was expected of l 7.6 bcm per year. Early 1980s
10
Figure 3.2 Map (not to scale) of the Nile Basin which shows the major supply structures (Adopted Alan, 2005, SOAS).
11
3.2 Description of the main sub basins
The Nile basin comprises four main river reaches, White Nile, Blue Nile, Atbara and
Main Nile. The Basin shows a great deal of geographic diversity with rugged
mountains, plateaus, deeply incised gorges, meandering river sections and deserts.
Elevation range from over to 4000m above sea level in the highland areas to several
hundred meters below sea level in depressions. The topography is a major controlling
factor in both climate and water distribution.
3.2.1 White Nile River sub basin
The watershed of the White Nile at Khartoum is 1.7 million km2. It contains Lake
Victoria and comprises a complex of channel, lakes, swamps and wetlands. The
streams which feed the White Nile River are seasonal. The average annual
precipitation in Lake Victoria 1221mm with a bimodal seasonal distribution with
peaks in March-May and November-December. After leaving Lake Victoria the White
Nile flows into Equatorial Lakes (Lake Kyoga and Lake Albert) and then northward
into Sudd sub basin and named Bahir el Jebel. The precipitation falls mostly in one
season from April to October. The vegetation that covers in most of the swamps are
Cyperus papyrus, Vossia cuspidate and Typha Australia (Sutcliffe and Parks,
1999).This part of the sub basin there is more evaporation than rainfall and
consequently the total flow in the river is decreased after it leaves the basin exposed to
loss rather than gain due to the topographic nature of the area.
12
3.2.2 Blue Nile River sub basin
The Blue Nile River starts at the outlet of Lake Tana and flows to Khartoum where it
meets the White Nile with basin area of 324,530km2. Blue Nile contributes about 60%
of the flow of Main Nile (Sutcliffe and Parks, 1999). The topography of the Blue Nile
composed of highlands, hills, valleys and occasional rock peaks. Most of the streams
feeding the Blue Nile are perennial and includes the Dinder and Rahad. The average
precipitation over the Blue Nile subbasin is 1394mm and is higher than the other
subbasin of the Nile basin. The precipitation over the Blue Nile basin varies from
1000mm in the north-eastern part to 1450-2100mm over the south-western part of the
sub basin.
3.2.3 Atbara River sub basin
The Atbara River originates in the Northern Ethiopia and Eritrea and joins the Nile
after the lowland in the eastern Sudan with total basin area of 112,400 km2. The
discharge of the river is extremely torrential. The rainfall is unimodall concentrated in
August and September with mean annual rainfall 900mm relatively high value over
the Ethiopian highlands to less downstream at the confluence with the main Nile.
Generally the average annual precipitation is lowest among the other Nile sub-basins
(Sutcliffe and Parks, 1999).
13
CHAPTER FOUR
4. RESEARCH METHODOLOGY
To be able to access the long term trends in streamflow and precipitation in the Nile
basin. As discussed above the whole Nile basin will be divided into three sub basins:
White Nile, Blue Nile and Atbara sub basins. Each of these sub basins will be divided
further in sub basins according to the long term data availability and completeness.
For each sub basin the steps followed are:
1. Selection of variables to be studied. This is precipitation and streamflow variable
are used.
2. Selection and of stations that have sufficient long record of stream discharge and
obtain the required data.
3. Data analysis and interpretation which include checking for the presence of trend
in the data and to determine the significance of the detected trends.
4.1 Selection of variables
A total of 13 streamflow and precipitation variables have been selected for this
research. Theses variables include the annual mean, flow and precipitation monthly
mean flow and monthly precipitation variables were analyzed in order to gain a bread
understanding of the trends.
14
4.2 Data validation and completion
Data validation started with a good representation of the collected data, tabular or
graphical form using the various techniques provided by HYMOS software packages.
The software was provided from IHE UNISCO and used for data management
validation and completion together with EXCEL. The software provides both tabular
and graphical analysis sheet. Missing data was filled in with regression equation signal
between the neighboring stations is weak (Hastie et al., 2001). In addition for those
stations with having preceding or successor neighboring station, sum or difference of
the time series are used to fill in the missing data. In particular the following steps
were taken.
Screening of data series
This step provides checking the data series against the data limits for total at long
duration. To flag a few very unlikely values from the data set using two standard
deviations from the mean as boundary values. Using the suspect value for the station
considered, the suspicion may be dropped or accepted by confirmed using the
comparison plot of the neighboring stations.
Multi-station validation
Comparison of series may also permit the acceptance of the value flagged as suspect
in screening of data. When two or more station show the same behavior there is strong
evidence to suggest that the value are correct.
15
Relation curve (Regression analysis)
The relation curve between neighboring stations which developed also strengthens the
flagged value by looking the relation curve of two sequential station data values. Then
the suspect values previously identified should be removed before deriving the
relationship, which may then be applied to compute corrected values to replace the
suspect ones. The validation was displayed river by river for the four sub basin. Each
sub basins station was compared with other neighboring stations in the sub basin.
4.3 Mann-Kendall and Seasonal Kendall test for trend detection
First of all, test for the trend in annual series was made so as to get an overall view of
the possible changes in streamflow processes. To determine if the trends found are
significant, the Mann-Kendall trend test was used (Mann, 1945 and Kendall, 1975).
This test was chosen over other trend detection tests based on the following factors:
(1) the Mann-Kendall test is a rank based non parametric test. When compare to
parametric test like t-test the Mann-Kendall test has a higher power for non-normally
distributed data which are frequently encountered in hydrological records (Onoz and
Bayazit, 2003; Yue and Pilon, 2004). (2) In comparison to other non-parametric tests,
like Spearmans rho test, the power of the Mann-Kendall test is similar to the point
where both give indistinguishable results in practice (Yue et al, 2002a, b). (3)The
Mann Kendall test has been extensively used to determine trends in similar hydrologic
studies done in the past (Hirsch et al, 1982; Lins and Slack, 1999; Burn et al, 2004;
Abdul Aziz and Burn, 2006).
The Mann-Kendall test is based on the null hypothesis that a sample of data is
independent and identically distributed, which means that there is no trend or serial
16
correlation among the data points. The alternative hypothesis is that a trend exists in
the data. First the statistic defined by variable S was computed which is the sum of the
difference between the data points for a series {x1...xn} come from a population
where the random variables are independent and identically distributed shown in
Esq.(4.1)
( ) == +=
=1
1 1sgn
n
i
n
ijij xxS , Where
+=
10
1)sgn( If
000
(4.1)
Mann (1945) and Kendall (1975) determined that the statistics S is normally
distributed when n 8 allowing for the computation of the standardized test statistics
Z which represent an increasing or decreasing trends respectively. For the statistical
trend test used in this study a 5-percent level of significance was selected. The 5-
percent level of significance indicates that a 5-percent chance for error exists in
concluding that a trend is statistically significant when in fact no trend exists.
+
=
)(1
0)(
1
SVS
SVS
Z If
0
0
0
S
S
S
(4.2)
Where Var(S), the variance of the data point is given by,
( )( ) ( )( )
++= =
m
tiii tttnnnSVar
1521521
181)( (4.3)
Where m is the number of tied (i.e., equal values) groups in the data set and ti is the
number of data points in the ith tied group. Under the null hypothesis, the quantity z
defined in the following equation is approximately standard normally distributed even
17
for the sample size n = 10. The positive values of S indicate upward trends whereas
negative S value indicate downward trend.
The slope of the data set can be estimated using the Thiel-Sen Approach (4.4). This
equation is used instead of a linear regression because it limits the influence that the
outliers have on the slope (Hirsch et al, 1982). To normalize the slopes calculated for
streams of different size, the mean flow value of each parameter and station was used to
find a percent change in flow rate.
=
ijXX
Median ij For all i
18
lag-1 serial coefficient is calculated after the trend was removed in order to preserve the
initial trend (Yue et al, 2002c).
( )( )( )
=
=+
= nt
tt
kn
ttkttt
k
YYn
YYYYknr
1
2
1
1
1
(4.6)
If lag-1 serial correlation coefficient (rk) is not significant at 5% significance level, then
the Mann-Kendall test is applied to the original time series. Otherwise it is removed from Yt
as:
11 = ttt YrYY (4.7) This procedure is known as the Trend-Free Pre-Whitening (TFPW) procedure. The third
step is to add the trend back to Yt by using Equation (4.8) and then the Mann-
Kendall test is conducted on this new series.
tYY t += (4.8) Seasonal Kendall test
The trend test for annual series gives us an overall view of the change in streamflow
volumes. To examine the possible changes occur in smaller time scale, we need to
investigate the monthly or seasonal flow series. Monthly streamflow usually exhibit
strong seasonality. Trend test techniques for dealing with seasonality of univariate
time series fall into three major categories (Helsel and Hirsh, 1992): (1) fully
nonparametric method, i.e., seasonal Kendall test; (2) mixed procedure, i.e., regression
of deseasonalized series on time; (3) parametric method, i.e., regression of original
series on time and seasonal terms. The first approach, namely, seasonal Kendall test
will be used in this research considering the benefit that the seasonal Kendall test
considers the effect of fluctuation in season in both runoff and precipitation.
19
Hirsch et al, (1982) introduced a modification of the MK test, referred to as the
seasonal Kendall test that allows for seasonality in observations collected over time by
computing the Mann-Kendall test on each of m seasons separately, and then
combining the results. Compute the following overall statistic S.
=
=m
jjSS
1, (4.10)
Where Sj is simply the S-statistic in the MK test for season j (j = 1, 2... m) where Sj
(Equation 4.10). When no serial dependence exhibit in the time series, the variance of
S is defined as
)()(1
=
=m
jjSVarSVar (4.11)
Then the quantity z defined in the following equation is approximately standard
normally distributed:
+
=
)(1
0)(
1
SVS
SVS
Z If
0
0
0
S
S
S
(4.12)
The positive values of S indicate upward trends whereas negative S value indicate
downward trend.
20
4.4 Sens T test
This test is an aligned rank method having procedures that first remove the block
(effect of season) from each time series and sum the data over seasons to produce a
statistic from these sums (Sen, 1968a, b). This procedure is distribution free and not
affected by seasonal fluctuations (Van Belle and Hughes, 1984). The computational
steps are as follows:
1. Compute the average for the month j, nxXn
iijj
==
1 and the average for the year i,
1212
1=
=j
iji xX . Subtract the monthly average from each of the corresponding
months in the n years of data to remove seasonal effects, i.e. calculate Xij - X.j for i = 1, 2, ..., n and j = 1, 2, ...,12.
2. Rank all the differences from 1 to nm (number of Months times the number of
years) to obtain the matrix (Rij), where Rij = rank of (Xij - X.j) among the 12n values
of differences. If t ties occur, the average of the next t ranks is assigned to each of the t
tied values.
3. The ranks for each year are averaged, i.e. 12.12
1=
=j
iji RR . Also the rank for each
month nRRn
iijj
==
1. .
4. Calculate the following test statistic.
+
+
+= =
n
ii
jijij
nmRniRRnn
mT1
2
,
2.
2
21.
21
)()1(12 (4.13)
Where m=number of months or seasons for this case m=12.
For the sample size (n) become large the distribution of T tends toward normality with
mean 0 under the null hypothesis of no trend and unit variance. The statistic test is to
21
reject the hypothesis of no trend if T > z exceeds a pre-specified percentile of the
normal distribution. Where z is the standard normal variate is the level of
significance which is 5% in this work. Van Belle and Hughes (1984) showed by
Monte Carlo simulation that the normal approximation for the statistic T was
reasonable even for small samples. Positive value of T indicates an increasing trend
and negative value indicate decreasing trend.
4.5 Rainfall-Runoff modeling of Upper Blue Nile
Change in land use/cover during the past 30 years was detected using modeling over
the Upper Blue Nile basin. From 1961-1970 and after 30 years 1991-2000 was
divided in to two 5 years for calibration and 5 for validation, then compare and discuss
changes between calibration parameters. Comparing the parameter between the two
time period give some insight on both land use and land cover change.
Based on nature of landscape and other land cover condition, topographic index and
saturation index the watershed will be broken up into a series of contributing areas
with different (SM), or maximum soil moisture storage;
a) Wettest area,
b) Intermediate area,
a) Hillslope area,
22
(Steenhuis et al, 2008)
It is reasonable to assume that water stored in the topmost layer of the soil system for
hillslope and runoff contributories area to be estimated as follows (Collick et al, 2008
and Steenhuis et al, 2008)
SM (t)=S (t-t) + (P AET R Perc) t....(4.14)
Where P is rainfall (mm/month), AET the actual evapotranspiration (mm/month), SM
(t-t) is previous time step storage water in the soil system (mm), R is saturation excess
runoff (mm/month), Perc is percolation to the subsoil (mm/month) and t is monthly
time step. When P (t) is less than PET, water is withdrawn from the soil system. This
results into the exponential soil moisture depletion at time step t is defined by the
following formula (Steenhuis, 2008).
=
(max))()(
)(expSM
tPETPSMSM ttt , When P
23
If P (i) is higher than PET, the actual evapotranspiration (AET) equals potential
evapotranspiration (PET) Steenhuis and Van der Molen, 1986. If not, it is computed
as:
=
(max)
)(
SMSM
PETAET t ....(4.16)
The soil moisture deficit is the difference between PET and AET within the same
month, and is given by:
Deficit (t) = PET (t) AET (t) ............................(4.17)
The soil moisture surplus is the difference between the effective rainfall and the sum
of SM and AET. It is the excess rainfall when the soil layer under consideration is
saturated with water. The surface runoff produced from wettest area and intermediate
area can be computed from the following equation.
Saturation excess runoff (SER (t)) = P (t) [SM + AET].........(4.18)
Where, SM is the difference between the soil moisture content in the current and
previous month. It is given by:
SM = SM (t) - SM (t-t) ............(4.19)
For hillside area, area with high infiltration capacity, the water flow as either interflow
or percolates (Perc) and added to ground water reservoir to form a ground water flow
(baseflow). Steenhuis et al, 2008 assumed that the after the baseflow reservoir is filled
first, the interflow reservoir starts filling. It is possible to assume the ground water
reservoir acts as a linear reservoir and its outflow, GWF and ground water storage
(GWS) is less than the maximum ground water storage, GWR (max) (Steenhuis, 2008).
24
tGWFPercGWSGWS ttttt += )( ()()( ..... (4.20)
[ ]t
tGWSGWF tt
= )exp(1)()(
..........(4.21)
When the storage reaches its maximum storage, GWS (max)
GWS (t) = GWS (max)...(4.22)
[ ]t
tGWSGWFt
= )exp(1(max) ..... (4.23)
Storage of a previous month is available for a surplus of a current month and so on.
Finally, the summation of a direct storm runoff (DR), ground water flow (GWF) in
month gives the total watershed runoff of in month.
Q = DR + GWF (t)......(4.24)
The model was tested and validated for the Blue Nile at the Sudanese border by three
students of the Masters Program in Integrated Watershed Management and
Hydrology and by Steenhuis et al (2009) for three SCCP watersheds and found that
this model gave good predictions.
25
CHAPTER FIVE
5. DATA COLLECTION AND VALIDATION
5.1 Data collection and Pre-processing
Monthly river flow data were collected from different sources: Global Hydro Climate
Data Network (GHCDN) operated by UNESCO/IHP which was used as a main
sources for this research, available at http://dss.ucar.edu/datasets/ds553.2/data/,
Shahin (1985) from Cairo University - Massachusetts Institute of Technology, 1977,
Global Runoff Data Center operated by world metrological organization (WOM)
funded by the Federal Government of Germany available at http://grdc.bafg.de/ as per
requested and from Countries like Sudan, Ethiopia and Uganda Ministry of Water
Resources. The monthly precipitation data for this study were downloaded from
Global Historical Climatology Network (GHCN) available at http://gpcc.dwd.de and
recent data for Ethiopian station from Ethiopia Meteorological Agency. Appendix I.3
for data availability graphs for both river flow and precipitation stations.
After all data from different sources were collected comparison between sources were
made to select the best data sources among them. All data from GRDC, GHCDN and
Shahin (1985) agree perfectly for the period 1912-1982 except the three stations
namely Jinja, Tamaniate and Aswan for some period. For example at Jinja the data
from the Global River Data Center shift by two years as compared with that of data
obtained from Uganda Ministry of Water Resources (Figure 5.1). The data from
GHCDN for Tamaniate in month June 1947 and 1948 were not in agreement with the
two other sources of data (Figure 5.2). Also at Aswan for the period 1912-1945 both
GRDC and GHCDN data are from the dam release but that of Shahin book 1985 is the
26
inflow to the dam and agrees then after 1945 (Figure 5.3). After this initial check of
the data, further data analysis and validation took place before conducting any analysis
as described in the next sections (Table 5.1).
Table 5.1 Total number of data corrected during validation and then completed.
Sub basins
Period of validation Stations name
Numbers of data corrected
( % ) of data corrected
Blue Nile
1959-2007 Lake Tana 13 2 1953-2004 Kessie 15 2
1912-2000
Roseires/El diem 12 1.4
Sennar 6 0.7 Khartoum 2 0.23
Main Nile
1912-1982
Tamaniat 11 1.29 Hassanab 7 0.82 Dongola 2 0.9 Aswan 0 0.5
White Nile
1912-2000
Jinja 0 0 Mongolla 5 0.5 Malakal 4 0.45
Jebel Aulia 4 0.9
Figure 5.1 Comparison plot between data sources at Jinja, show a 3rd data source
27
Figure 5.2 Comparison plot between data sources at Tamaniate.
Figure 5.3 Comparison plot between data sources at Aswan
5.2 Selection of stations
The selection of stations is one of the more important steps in hydro-climate time
series analysis. Stations were selected with record length of above 30 years both for
precipitation and streamflow variables. Burn and Elnur (2002) stated that the selection
28
of stations in a hydro-climate change research is substantial at the initial step and that
a minimum record length of 25 years ensures validity of the trend results statistically.
Hence 15 streamflow stations and 38 precipitation stations were selected. The station
names missing data and years uses for analysis are shown in Tables 5.2 and 5.3 The
location of the precipitation station are shown in Figure 5.4 given appendix I.1
entitled: Names and locations of selected station. Collected data include monthly
discharge and precipitation for the available extended period of time. After the data
had been collected, the entire raw data was imported to Excel sheet accordingly. This
step is a base for conducting any further analysis since data in Excel format can easily
be transferred to any of the software package to be used for the analysis. Table 5.2 List of data availability. No. Flow
station GHCDN GRDC Shahin,
1985 Sudan/ Uganda
Ethiopia
1 Bahir Dar 1959-20072 Kessie 1953-20043 Jinja 1912-2000 4 Mongalla 1912-1982 1912-1982 1912-1973 5 Malakal 1912-1995 1912-1982 1912-1973 1965-2000 6 Jebel Aulia 1973-1982 1912-1973 1966-2000 7 Roseires 1912-1995 1912-1982 1912-1973 1980-2000 8 El Diem 1964-1996 9 Sennar 1912-1995 1912-1982 1912-1973 1980-2000 10 Khartoum 1900-1982 1912-1982 1912-1973 1980-2000 11 Tamaniate 1912-1982 1912-1982 1912-1973 1980-2000 12 Hassanab 1912-1982 1912-1982 1912-1973 1980-2000 13 Kilo-3 1912-1982 1912-1982 1912-1973 14 Dongolla 1912- 1995 1912- 1982 1912-1973 1980-2000 15 Aswan 1912 -1984 1912-1982 1912-1973
29
Table 5.3 List of data sources used in the analysis. No. Flow station GHCDN Sudan Uganda Ethiopia From
relation curve
1 Bahir Dar 1959-2007 2 Kessie 1953-2004 3 Jinja 1912-2000 4 Mongalla 1912-1982 1983-20005 Malakal 1912-1995 1996-2000 6 Jebel Aulia 1966-2000 7 Roseires 1912-1995 1996-2000 8 El Diem 1964-1996 9 Sennar 1912-1995 1996-2000 10 Khartoum 1900-1982 1983-2000 11 Tamaniate 1912-1982 1983-2000 12 Hassanab 1912-1982 1983-2000 13 Kilo-3 1912-1982 14 Dongolla 1912- 1995 1983-2000 15 Aswan 1912 -1984
In the next sections the discharge data is validated and prepared for trend analysis by
dividing the river basin in to four sub-river basins as Blue Nile, White Nile and
Atbara.
5.3 The Nile stream gauging stations and validation of data
White Nile stations
Jinja
The discharge at Jinja, hydrological station located at the outlet of Lake Victoria has
been recorded since 1900 (Kite, 1982 and Conway, 1993). The discharge was
originally regulated by Ripon Falls until the construction of Owen Falls, some 3 km
downstream dam in 1954 but which was began in 1951 with construction of coffer
dam (Sutcliffe and Parks, 1999). The lake level/ discharge relationship remained the
same before and after dam construction.
30
Figure 5.4 Map of the Nile basin showing the locations of the precipitation station and stream gauges used in this analysis.
31
Mongalla
Station indicate the outflow White Nile from the Equatorial Lakes where White Nile
before enters to marshes and swamps. This station is confined to a single channel and
next to Jinja has reliable and long term gauging site data. Data from 1983 onwards is
not available because of the civil war in Sudan and measurements have resumed since
2007.
Helit Dolieb
The discharge gauging station which measures the contribution of Sobat River located
upstream of Malakal. Since the station is located 8 km above the White Nile
confluence, the back water effect of the White Nile due to the rise in Lake Victoria has
affected the levels of measurement. During some years 1965-1967 those flows may
have been overestimated because of lack of the dry season flow otherwise the flow
record is reliable for analysis (Sutcliffe and Parks, 1999). Hence adjustments have
been made by comparing with neighborhood stations.
Malakal
Discharge measuring station on the White Nile basin indicating the contribution of
White Nile, Sobat River, and Bahr al-Ghazel basin. The flows measurements at
Malakal are accurate because the numbers of gauging have been sufficient and the
rating curve is good (Sutcliffe and Parks, 1999).
Jebel Aulia
Streamflow measuring station found 44km south of Khartoum. The Jebel Aulia dam
built forty kilometers upstream of Khartoum in 1937 to store water for later use in
Egypt, has added further evaporation losses along this stretch. The rapid silt up of this
32
reservoir and the construction high Aswan dam in Egypt in 1965 stopped its function
(Shahin, 2002).
Validation of White Nile stations
White Nile at Jinja, Mongalla and Malakal
The discharge measurement at Jinja is relatively in good quality because it is
computed from the developed rating curve between outflow and lake level so no
adjustment was made. The monthly flow at Mongalla was compared with flow at Jinja
by using relation curve to see pronounced outliers. The flow at Mongalla after 1982
was inferred from Jinja in monthly bases using the regression equation between them
r2=0.76. Refer appendix II.1,e for the relation curve and equation. The flow records at
Malakal on March 1944 is very high as compared with the other months and also
advance flow of Mongalla on the same month hence it is clearly an outlier. The value
was corrected by the long term monthly mean value for the same months.
Blue Nile gauging stations
This sub basin has five stations namely, Blue Nile at Lake Tana, Blue Nile at Kessie,
Blue Nile at Roseires /Eddiem dam, Blue Nile at Sennar and Blue Nile at Khartoum.
Khartoum
Discharges measuring station for Blue Nile immediately upstream its junction with the
White Nile. The flow at this station contains seasonal flow Dinder and Rahad which
originated in the highlands of Ethiopia in addition to flow from Sennar.
33
Roseires dam/Eddiem
The hydrological station on the Ethiopia-Sudan border indicates the inflow of Blue
Nile from Ethiopia highlands to the Sudanese areas. This station named Roseires
before the construction of the dam until 1965 and Eddiem afterward which was shifted
upstream of the dam. This station is one of the long term records for Blue Nile.
Sennar
Discharge gauging station located downstream of Sennar dam on the reach of Blue
Nile. This is located 350 km from Khartoum which was completed in 1925 to supply
the Gezira irrigation scheme.
Kessie
The gauging station is located at the bridge where the main road to Addis Ababa from
Bahir Dar crosses Abbay river with Bridge. Discharge was measured at Kessie started
from 1953 to present. The records are complete from 1954 to present. The
construction of the new bridge may affect the gauging section which is directly
affected the stage measurements and then the discharge measurements due to the
constructed temporary coffer dam. The data before the construction is valid and good
in quality for analysis. Conway, 2000 also comment on the quality of the records as it
is fear good.
Lake Tana
The outflow of Blue Nile at Lake Tana is recorded in three separate time period, 1921-
1926, 1928-1933, 1959-present. Considering the data quality, before 1990 records
show errors and needed correction. The construction of the Chara-Chara weir, which
was completed in 1996, has affected the natural flow from the lake.
34
Validation of Blue Nile stations
Lake Tana and Kessie
First visual screening showed a few mistyped decimal points. These were corrected
and flagged in the data files. Then data were subjected to a few validations the data for
upper and lower boundary limit to identify outliers. Sometimes, when data for
complete years were missing, the whole year was omitted which needed further
attention when converting the data format. For flow stations at Bahir Dar (Lake Tana)
and Kessie for which the minimum and maximum of flow data available, three time
series were created, mean monthly discharge (m3/s), maximum daily discharge (m3/s)
and minimum daily discharge (m3/s). The mean monthly series was validated against
the maximum and minimum daily discharges (Figure 5.5). Whenever mean monthly
discharge > maximum daily discharge or mean monthly discharge was smaller than
the minimum daily discharge the record for that month was flagged and further
investigated. Sometimes, errors resulted from mistyping one number or misplacing
the decimal point and were fixed accordingly.
Figure 5.5 Comparison of Mean flow and maximum and minimum daily flow at Bahir Dar.
35
Those records violated the data limit was flagged and wait to be confirmed by the next
step which was comparison of plot between adjoining stations. Comparison was made
between each others and with Roseires/El Diem. For the case of Bahir Dar flow
measuring station has two periods of records, before (1959-1995) and after (1996-
2007) the construction of chara-chara weir. Those streamflow records after the weir
become functional have taken as it is. Those suspect value identified in screening was
further validated using comparison plot between the lake levels records of Lake Tana
at Bahir Dar which is in relatively good quality (Figure 5.6). The streamflow record in
some months of 1984 and 1985 are problematic and hence corrected using average
flow and those missed value was completed using the developed power equation
between lake level and outflow from the lake. Since the lake level records are
completed, the missed monthly discharge for 1984 estimated by regression with lake
level r2=0.95. Refer appendix II.1, a for relation curve and regression equation.
Figure 5.6 Comparisons between Outflow at Bahir Dar and Lake Level of Tana.
36
Blue Nile at Roseires / El Diem, Sennar and Khartoum
These three stations had relatively longer records of streamflow data. The main data
source of this research (GHCDN) compared with two other data sources (GRDC and
Shahin, 1985) has encountered error of using month February as 28 days in leap years
in both Roseires and Sennar stations and corrected by multiplying with 28 first and
divided it by 29. To have a natural flow for after 1963 El Diem station which is
located upstream of Roseires station data was used from Ministry of Irrigation Sudan
and ENTRO since GRCDN data after 1964 shows the release from the Roseires dam.
When the two data plotted together the low flow data from the GRCDN source was
higher the data from Ministry of Irrigation Sudan and ENTRO. Screening was done
independently for the three stations. Value out of the boundary limits flagged and
further validated by comparison of plots between the remaining two stations. The
comparison plot was made in 5 years period to easily identify problematic records.
The suspect values flagged was removed and relation curve developed between the
adjoining stations for correcting the suspect values. The missed data from 1997 to
2000 were filled using the relation curve with Sennar with correlation coefficient
r2=0.99. Refer to appendix II.1 b, c for relation Roseires/Ediem and Sennar.
Main Nile Station
Tamaniat
It is located downstream of confluence of the White Nile and the Blue Nile River. This
station records illustrate the history of flows of both White and Blue Nile. The flow
records are based on gauge-discharge curve from gauging and some few years are base
on a general curve or on interpolation between measurements. From October 1928 to
September 1929 gauging was unreliable hence needed adjustment.
37
Hassanab
Discharge measuring station just above the confluence of Atbara river with the Main
Nile. This station is located downstream of Tamaniat. In between these stations there
is abstraction by Khartoum Hassanab scheme.
Kilo 3
The flow gauging station located at the outlet of the Atbara River immediately
upstream of its Junction with the Main Nile. It joins the main Nile about 320 km
downstream from Khartoum. There is a constructed dam on the river to serve for
irrigation.
Dongola
Discharge measuring station of the Main Nile above Aswan dam. To avoid the
backwater effect of the Aswan dam the station first moved from Wadi-Halfa to
Kajnarty and then to Dongola. The flow records contribution of these stations
expressed as Wadi-Halfa 1912-1939, Kajnarty 1931-1964 and Dongola 1962 to
present. These stations combined to form a single long term records as Dongola 1912
to present by taking into consideration the evaporation loss from Wadi-Halfa/ Kajnarty
to Dongola with its width and length of 450km.
Aswan
The final outlet of the Nile is measured at Aswan. This station provides the longest
historical records. The records are moreover completed and with good quality but it
should be naturalized because there has been considerable usage of water for irrigation
in Sudan.
38
Validation of Main Nile station
Tamaniat and Hassanab
Error of using February as 28 days in leaps years in both Tamaniat and Hassanab
stations and corrected in the same manner like those stations in Blue Nile. Data entry
error such as instead of putting 1169 and 1223 put 116.9 and 122.3 for June 1947 and
1984. These were clearly confirmed by comparing with other data sources Shahin
1985 and GRDC. After correcting the above errors screening was continued to identify
those suspect value violating the data limits. Time series graph of the two stations as
well as sum of time series of Khartoum and Jebel Aulia in five years interval was
plotted in the same graph for comparison. To develop more confidence on the flagged
value relation curve between the two stations was plotted as shown below. The
outliers shown were some of them were problems of Tamaniate and some of them
were problems of Hassanab based on the results from the above comparison analysis
from the three stations (Figure 5.7). All the suspect values were then removed and
completed using regression without including those flagged value with strong
regression coefficients 0.98 for the period 1912-1982. The data after 1982 was
validated and completed using relation curve between Tamaniate and Hassanab from
1980 to 2000 with correlation coefficient 0.967. Refer appendix II.2, d for the relation
curve and equation.
39
Figure 5.7 Relation curve between flow at Tamaniate and Hassanab
Kilo-3, Dongola and Aswan
After screening of each station independently was processed, time series graph of all
the three stations were plotted to see the problematic months together with the
screening results. During plotting the graph, Dongola and Aswan as it is but Kilo-3
plus Hassanab as one series was plotted. Any station value in different behavior with
the remaining two was identified as suspect value. Adjustment of that suspect values
were adjusted using regression analysis between neighboring stations. With respect to
discharge the number of values corrected as a total or a percentage noted for
individual stations.
40
CHAPTER SIX
6. RESULTS AND DISCUSSION
6.1 Results of statistical analysis
The first step in time series analysis is visually inspecting the data. Significant changes
in level or slope usually are obvious. From the visual inspection, it seems that the
annual flow series of the upper Blue Nile at Kessie which exhibits obvious upward
trend but the other Blue Nile station showed no significant change over the period
under consideration. This is also supported from the 5-year moving average curve of
the annual mean discharge at station Kessie exhibits fluctuation in recent 50 years, and
the discharge reached higher record values started from 1995. This procedure also
conducted on the Upper White Nile basin, the visual detection also supported by the
statistical analysis result.
6.1.1 Pettitt test results
Using pettitt the possible change points were examined for the monthly and mean
annual runoff and precipitation and the possible change-points were indicated as upper
values of the probability curve. The change point year is consistent with the monthly
and mean annual runoff and precipitation. Change point years were identified for all
flow gauging stations; 1965 for Upper Blue Nile and 1962 for White Nile basin (Table
6.1). As Sutcliff and Park (1999) stated clearly the phenomena of heavy rainfall in
October-December season in 1961and 1962 showed over the Lake Victoria basin in
the rainfall records. Changes in rainfall over Lake Victoria basin cause for change in
streamflow of White Nile basin stations (Jinja, Mongolla and Malakal). Considering
the validity of the discharge data series, it is divided into two and three stages to have
Mann-Kendall trend test in this research, the results of the computation of pettitt test
41
are given in the next section. Refer also the change point test with probability curves
appendix III.1, 2, and 3. Table 6.1 Results of change points with pettitt test for runoff and precipitation.
Station of River flow series Years Probability Lake Tana (Bahir Dar) 1994 0.98 Kessie 1992 0.99 Roseires/Eddiem 1965 0.9 Jinja 1962 1 Mongolla 1962 1 Malakal 1962 1 Upper Blue Nile Arial precipitation - - Victoria Nile Average precipitation 1961 0.9
6.1.2 Runoff trend results
Significance of serial correlation
Prior to conducting the Mann-Kendall test, the Trend Free Pre-Whitening procedure
was applied to the series in which there was a significant serial correlation. The lag-1
serial correlation test was applied to all the time series data. The majority of the
monthly time series in the data set appear to have a significant lag-1 serial correlation
coefficient at 5% significant level. These indicate the data series violating the
assumption of independence. The White Nile station (Jinja, Mongalla and Malakal)
showed a significant lag-1 serial correlation in all months and annual series. In
addition the station on the Blue Nile (Lake Tana, Kessie and El Diem) showed a
significant lag-1 serial correlation in low flow season and no significant serial
correlation in Flood season.
Seasonal Mann-Kendall and Sens T tests on Upper White Nile
Generally, the Mann-Kendall and Sens T tests confirmed the findings of each other.
According to Sens T test in all White Nile stations (Jinja, Mongalla and Malakal)
there is a upward trend at 5% significant level (T= 22.10, 18.83 and 13.98) indicating
42
a tendency towards much more water on annual basis (Table 6.2). According to Mann-
Kendall test in all White Nile station (Jinja, Mongalla and Malakal) there is a upward
trend at 5% significant level (z = 8.6, 8.11 and 6.02) indicating a tendency towards
much more water on annual basis (Table 6.2). Among the findings of seasonal trend
analysis, the result of Flood and Low flow series showed upward trend indicating an
increasing in amount of flow at 5 % significant levels according to both the Sens and
Seasonal Mann- Kendall test.
Table 6.2 Results of the runoff trend test for Upper White Nile
BN station Season T values of Sens T test
z values of Mann-Kendall test
Jinja Flood 15.57 19.64 Low flow 15.73 20.23 Annual 22.10 8.6 Mongalla Flood 13.92 18.0 Low flow 12.67 16.29 Annual 18.83 8.11 Malakal Flood 9.22 12.86 Low flow 11.13 10.87 Annual 13.98 6.02
Note: Bold figures are significant at 5% significance level.
White Nile Stations
Summary of the Mann-Kendall test for monotonic trend and non-Parametric Sens
slope estimates for the streamflow series are given in Table 6.2. An increase of 7.12
m3/s/year (P
43
Jinja and Malakal the higher increasing slope were estimated and runoff changed at
the rate of 7.67 m3/s/year and 3.71 m3/s/year respectively since 1912. Also in the
month of April at Mongalla the highest runoff changed was exhibited and runoff
changed at the rate of 10.37 m3/s/year since 1912. However, monotonically decreasing
trend in all months was estimated for the analysis period 1962-2000 after the change
point at Jinja, Mongalla and Malakal. For example Annual mean runoff decreased at a
rate 13.13 m3/s/year, 25.72 m3/s/year and 8.96 m3/s/year at Jinja and Mongalla in the
month of September and at Malakal in the month of November respectively. Refer
appendix III.4 for result of the trend analysis.
Seasonal Mann-Kendall and Sens T tests on Upper Blue Nile
Upper Blue Nile stations (Lake Tana and El Diem) there are no trend at 5% significant
level in flood season (T= 0.34 and -1.77), low flow season (T= 0.01 and 0.4) and
annually (T= 1.04 and -1.24) and indicating flow stay the same but at station Kessie
there is an upward trend at 5% significant level in flood season (T= 2.67), low flow
season (T= 0.72) and annually (T= 3.37) indicating the amount of flow increasing on
annual basis . According to Mann-Kendall test low flow season in all Upper Blue Nile
stations (Lake Tana, Kessie and El Diem) displayed no trend at 5% significant level (z
= 0.31, 0.79 and 0.72) indicating no additional flow added to the basin, but for flood
season the lake Tana and Kessie displayed an upward trend at the same significance
level (z = 2.34 and 2.87). Among the findings of seasonal trend analysis from the
Sens T test and seasonal Mann-Kendall test at Lake Tana and Kessie. This is due to
the trend heterogeneity between months imply that there is an upward and downward
trend with in the season. For such case the result from Sens T test seems better to
explain the trend since it is not affected by seasonal blocks.
44
Table 6.3 Results of the runoff trend test for Upper Blue Nile BN station Season T values of
Sens T test z values of
Mann-Kendall test Lake Tana Flood -0.75 0.47 Low flow 1.74 3.45 Annual 1.04 0.4 Kessie Flood 2.02 1.5 Low flow 2.56 3.23 Annual 3.37 1.45 El Diem Flood -1.77 -1.78 Low flow 0.40 0.72 Annual -1.24 -1.10
Note: Bold figures are significant at 5% significance level.
Upper Blue Nile Stations
Summary of the Mann-Kendall test for monotonic trend and non-Parametric Sens T
test for the streamflow series for both Low and Flood flow season are given in Table
6.3. In general no mean annual runoff trend was recorded at those stations in the
Upper Blue Nile. Lake Tana flow station showed increasing trend for months April,
May, June and July at a rate of 0.37, 0.31, 0.4 and 1.14 m3/s/yr respectively during
1959-2007. In Upper Blue Nile river, at Kessie significantly increasing trend in
monthly flow from May to June at a rate of 0.99 and 1.18 m3/s/yr respectively.
But the down stream station from Kessie which is Roseires/El Diem no trend in all
months except September with significantly decreasing trend at 9.78 m3/s/yr (appendix
III.5) which also true for the natural monthly flow at Sennar down stream of
Roseires/El Diem with 13.52 m3/s/yr. The observed monthly flow at Sennar decreased
significantly between September to February and June but increased significantly
during April and May. The trend physically explained by comparing with the flow to
Roseires and its release, the dam hold water during high flow period and release
during low flow period. The same pattern of trend with observed flow at Sennar was
estimated at Khartoum station. Both stations exhibited significantly decreasing trend
45
in mean annual runoff at a rate of 5.77 m3/s/yr and 5.87 m3/s/yr at Sennar and
Khartoum (Figure 6.1). When comparing the Natural mean annual runoff (appendix
III.5) with the observed annual runoff, the slopes of the trend line for the Sennar
station are noticeably reduced. In other word, the runoff reduction directly due to
human activities has distinctly increased during the last four decades, especially after
1964. The results of Mann-Kendalls trend test for natural monthly runoff (appendix
III.5) show some differences with that of observed monthly runoff. The increasing
trend in observed flow for April and June at Sennar disappeared, which means this
increasing trend resulted from regulated stream flow due to dam control if the
estimation of natural runoff is accurate. The decreasing trend is still exists, but the
magnitude is remarkably reduced, which is reflected by the smaller z values and fewer
months with a significant decreasing trend observable when comparing natural and
observed flow in (appendix III.5) This indicates that the Upper Blue Nile runoff has a
decreasing trend even after deducting the water withdrawn for human uses (Figure
6.2).
Main Nile Stations
The trend analysis was made on observed runoff at three sequential stations on the
main rout of the Main Nile Basin, at Tamaniate, Hassanab and Dongolla. The mean
annual runoff display a strong significant decreasing trend at a rate of 5, 5.6 and 7.6
m3/s / yr respectively. The result also showed that the rate increased as downstream
from Tamaniate to Dongolla. Monthly runoff from July to January significantly
decreased, while it significantly increased from March to May. The increasing trend
exhibited from March to May do not imply the actual trend patter instead due to the
regulation effect of dams upstream of the stations. But the peak flow period display
significantly decreasing trend also identified in those unregulated upstream station.
46
Figure 6.1 Natural and Observed mean annual runoff trend at Sennar station of the Upper Blue Nile River.
Figure 6.2 Balance between natural and observed mean annual runoff at Sennar station.
47
6.1.3 Precipitation trend results
The pattern of the mean monthly rainfall on Upper Blue Nile and Upper White Nile
differ due to seasonal variation (Figure 6.3). Upper Blue Nile has shorter wet season
and highly variable during dry season than wet season as compared with Upper White
Nile. The rainfall in upper White Nile have two rainfall season one having high
amount of rainfall than the other. The seasonal coefficients of variation are displayed
in Figure 6.4 for both the Upper Blue Nile and White Nile.
Figure 6.3 Mean monthly areal and average rainfall distribution over Upper Blue Nile and White Nile.
Figure 6.4 Coefficient of variation in rainfall for Upper Blue Nile and White Nile.
48
To give insight on the rainfall trend over the Upper White Nile and Upper Blue Nile
basin the average rainfall over the lake Victoria Nile and the areal Rainfall were taken
and the seasonal Mann-Kendall and Sens T statistical test were conducted. It is self
evident that the results of the Mann-Kendall and Sens T test are in good agreement
about the existence of significant trends over the Upper White Nile basin and also no
significant trend over the Upper Blue Nile basin (Table 6.4). For the Upper White Nile
basin, there is a positive trend indicating a tendency towards wet condition on seasonal
basis at 5% significance level according to the Sens T test (T = 2.58) and Mann-
Kendall test (z = 2.61), respectively. Significantly increasing trend in monthly rainfall
wet season was display. Out of ten stations selected for rainfall trend analysis over the
Upper White Nile basin seven of them were displayed significantly increasing trend in
November, three in October, three in January and two in mean annual. These indicate
much of the wet season month showed a significant increasing trend. Refer appendix
III.11 for the full results of the trend analysis. According to Sens T test on Upper
Blue Nile basin rainfall there is a no trend at 5% significant level on both wet and dry
season as well in annual basis (T= -0.42, 0.27 and 0.40) indicating a no tendency
towards wet condition. According to Mann-Kendall test on Upper Blue Nile basin
there is no trend at 5% significant level on both wet and dry season as well in annual
basis (z = -1.14, 0.14 and -0.57) indicating no change in rainfall on both seasonal and
annual basis.
Table 6.4 Results of the precipitation trend test for White and Upper Blue Nile.
Basin Season T values of Sens T test
z values of Mann-Kendall test
White Nile Wet 2.58 2.61Dry -1.2 -0.9Annual 1.35 1.37
Upper Blue Nile Wet -0.42 -1.14Dry 0.27 0.14Annual 0.4 -0.57
Note: Bold figures are significant at 5% significance level.
49
6.2 Model Results
Model running started before the beginning of the rainfall period; January was
selected. The calibration parameters were selected according to suggestions from
Steenhuis, 2008 and Collick, 2008. The data from 1961 to 1970 used for calibration
and validation then after 30 years from 1991 to 2000 used for calibration and
validation (Figure 6.4 A and B). The parameters calibrated are the soil moisture and
the percentage of contributing areas. The root mean square error and coefficient of
regression were used as a relation criterion between the observed and simulated runoff
and the efficiency criterion by Nash and Sutcliffe (1970) represented as E is used for
the model efficiency. The evapotranspiration was set according to Steenhuis, 2008
recommendation 3.3mm/ day for wet season (June to September) and 5mm/day for dry
season (October to May) the = 0.21month-1. These all parameters have been used to
analyze the relationship between different land uses/covers with the soil physical
properties and crop properties.
A
50
Figure 6.5 Calibration (A) from 1991 1995 and validation (B) from 1996 - 2000 of the model. Table 6.5 Calibrated and validated parameters showing land use/cover change during the past 30 years. Parameters 1961-1965 1966-1970 1991-1995 1996-2000
Calibration Validation Calibration Validation Nash-Sutcliffe model eff. (e) 0.8 0.7 0.78 0.75 Root mean square error (RMSE) 13.52 15.5 12.67 15.06 Correlation coefficient (R2) 0.79 0.8 0.87 0.83 AR (Saturated zone) 0.10 0.10 0.15 0.15 AR (Intermediate zone) 0.20 0.20 0.20 0.20 AR ( Hillslope zone) 0.70 0.70 0.65 0.65 SM (Saturated zone) 400.00 400.00 300.00 300.00 SM (Intermediate zone) 40.00 40.00 10.00 10.00 SM (Hillslope zone) 600.00 600.00 450.00 450.00 PET wet season 3.30 3.30 3.50 3.50 PET dry season 4.50 4.50 5.00 5.00 Note: AR-area ratio, SM-soil moisture.
The result from the model is good indicator for land use/land cover change over the
past 30 years. The soil moisture and percent of contributing area is a good indicator of
land use change and the potential evapotranspiration is good indicator of the land
cover changes. The soil moisture in forest is very high as compared with the other land
cultivated or plantation. Hence the model result indicates clearly that more land
B
51
changed from forest and forest related land to plantations. More land from forest and
related land use type changed to cultivation land. In addition to the land use change
the land cover change from low to high water consuming crops. This may also be due
to temperature change over the past 30 years.
52
CHAPTER SEVEN
7. CONCLUSIONS
Based on the above analysis, some conclusions are drawn as follows:
1) Most data sources for the Nile basin shows a good agreement between them
especially the Global Hydro Climate Data Network have good data for most of
the streamflow considering some errors.
2) Years through 1960 are the period experiencing abrupt changes in runoff in the
White Nile Basin stations. Pettit test proves that years are periods most
changes in runoff were observed. The reason for the change is due to abrupt
changes in Rainfall over the region.
3) All the stations located in White Nile presents an obviously increasing trend in
the annual mean runoff, and also an increasing trend to different degrees in
monthly mean runoff from 1912-2000. But in the last four decades all of that
station showed a significant decreasing trend in both annual and monthly flow
series.
4) The Upper Blue Nile Basin presents no trend in the mean annual and the flood
season runoff but increasing trend in the low flow season at two of the station
located on the route of the river. In recent years saw a significant increasing
trend in low flow period April to May due to regulation in the upstream.
5) The trends of the runoff variation are basically consistent with the changing
trend of precipitation in the Victoria Nile even though no correlation was
found between the average precipitation and streamflow due regulation and
large area lake effects.
6) No temporal trend in precipitation over the Upper Blue Nile was determined;
hence the display in no trend in the streamflow was expected.
53
7) Trends of low flow significantly increase after chara-chara weir started
operation. The reason for the increment is not due to an actual increase in
runoff but regulation effect.
8) There was land use/cover change during the past 30 years in the upper Blue
Nile basin. More land went for cultivation and also change in crop type which
has high water consumption. The model is useful for evaluating the effect of
land cover/land use scenarios on discharge and soil moisture regimes.
54
CHAPTER EIGHT
8. REFERENCES
Abdul Aziz, O., Burn, D., 2006. Trends and variability in the hydrological regime of
the Mackenzie river basin. Journal of Hydrology 319, 282-294.
Alan Nicol, 2005. The Nile: Moving beyond cooperation. Water policy program, ODI