4th Year Project_john Yumbya

A DECISION SUPPORT SYSTEM FOR FLOOD HAZARD PREPAREDNESS FOR NZOIA RIVER BASIN, KENYA

BY

MUTUA JOHN YUMBYA

GEO/048/05

A DISSERTATION SUBMITTED TO THE SCHOOL OF ARTS AND SOCIAL SCIENCES IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR A BACHELOR OF ARTS DEGREE IN GEOGRAPHY

MOI UNIVERSITY SCHOOL OF ARTS AND SOCIAL SCIENCES DEPARTMENT OF GEOGRAPHY P.O. BOX 3900 ELDORET - KENYA

Date Submitted: 30th April, 2009

DECLARATION

This is my original work and has never been presented for any award in any institution. No part of this dissertation may be produced without permission from the author or the Department of Geography, Moi University.

NAME

REG. NO.

MUTUA JOHN YUMBYA

GEO/048/05

Signature:………………………….

Date……………………

SUPERVISOR

This dissertation has been submitted as an examination requirement with my approval as the university supervisor.

MR. TOM ESIPILA

DEPARTMENT OF GEOGRAPHY

MOI UNIVERSITY Signature:………………………….

Date:…………………….

i

DEDICATION

To my Mum, Dad and siblings: Brian, Immaculate, Joseph, Patrick and Peter, for their support and prayers during the compilation of this dissertation.

ii

ACKNOWLEDGEMENT

The entire research process and the compilation of this report would not have been possible without the efforts and contribution by some key individuals. I am therefore indebted to express my gratitude to the following people.

First and foremost is Mr. Tom Esipila whose supervisory work lent credence to this dissertation. Secondly I am grateful to Johnson Maina, the Assistant Director Hydrometeorology Division, Kenya Meteorological Department, my dear friend for his constant assistance in the analysis of meteorological data and Hydrological modeling. Lastly my gratitude goes to Mr. Luke Kanda, the Moi University Geographical Information Systems laboratory technician for his unfailing support throughout the writing of this dissertation.

Without forgetting, more thanks to my family for their financial support and my dear colleagues and students in the Department of Geography, Moi University.

May God reward them abundantly!

iii

ABSTRACT

Weather, water and climate-related hazards are being experienced more frequently and extensively the world over and the spatial and temporal scales of these hazards vary widely. Extreme hydrometeorological conditions are not inherently catastrophic, since the natural environment shows remarkable resilience through adoption and rejuvenation. However, hazards posed by these events cause potential disasters by seriously affecting the physical infrastructure and economic activities of human beings. In one stroke, disasters can wipe out a lifetime of development and deprive the countries of resources, which could otherwise be used for economic and social development. Nzoia River basin and particularly Budalang‟i area has been experience periodical flood disasters which cause death and loss of property. The study seeks to research on the predictability and forecasting of floods in Nzoia river basin as a means of decision making for flood risk preparedness. The study hypothesized that flood hazards can be forecasted to reduce risks of death and property damage. To investigate into the problem the study employed various research tools to gather data. The primary data sources were observation and measurements. Secondary data was provided by library research and historical data from the Ministry of Environment.

Study findings however, show that the employment of modeling techniques by modeling rainfall-runoff processes using historical rainfall, evaporation and discharge data for the basin and real-time river flow forecasting is the only way to manage floods in Nzoia river basin since it has been documented that Budalang‟i floods are yearly and the population is very high in the area, they can only live with the floods. The study showed a very small error in the observed and forecasted river depths. iv

However the challenges for flood forecasting in Kenya is that there is no real-time data since the equipments for capturing such data are expensive, data used for modeling is usually outdated and usually the results in the forecast are not close to accurate.

In the light of this the study concludes that there is need for the government and various agencies to employ forecasting techniques in not only floods but most occurring disasters in order to reduce death and damage. In general, the recommendations focus towards the integration of the entire community for a sustainable solution to the environmental problem of floods in Budalang‟i.

v

TABLE OF CONTENTS DECLARATION .......................................................................................................................................... i DEDICATION ............................................................................................................................................ ii ACKNOWLEDGEMENT ............................................................................................................................ iii ABSTRACT .............................................................................................................................................. iv TABLE OF CONTENTS .............................................................................................................................. vi LIST OF TABLES ......................................................................................................................................viii LIST OF FIGURES ..................................................................................................................................... ix ABBREVIATIONS AND ACROYNMS ........................................................................................................... x CHAPTER ONE ......................................................................................................................................... 1 1.0 Introduction ...................................................................................................................................... 1 1.1 Statement of Problem........................................................................................................................ 3 1.2 Objectives of the study ...................................................................................................................... 4 1.2.1 Main objective ............................................................................................................................ 4 1.2.2 Specific objectives ....................................................................................................................... 4 1.3 Hypothesis......................................................................................................................................... 5 1.4 The Study area ................................................................................................................................... 6 1.4.1 The Nzoia River Basin .................................................................................................................. 7 1.4.2 Physiography and Geology .......................................................................................................... 8 1.4.3 Climate ....................................................................................................................................... 9 1.4.4 Economy ..................................................................................................................................... 9 1.4.5 Hydrology ................................................................................................................................. 10 1.5 The Case of Floods in Budalang’i ...................................................................................................... 12 1.6 Justification and Significance of the Study ........................................................................................ 13 CHAPTER TWO ...................................................................................................................................... 15 2.0 Conceptual Frame Work and Literature Review ............................................................................... 15 2.1 Introduction .................................................................................................................................... 15 2.2 Conceptual frame work ................................................................................................................... 16 2.2.1 Review of Rainfall Runoff Process and Models .......................................................................... 16 2.2.1.1 Introduction ....................................................................................................................... 16 2.2.1.2 Model Classification ........................................................................................................... 16 2.2.2 Watershed Characteristics and Hydrologic Models .................................................................... 22 2.2.3 Watershed Soil.......................................................................................................................... 23 2.2.4 Watershed Vegetation .............................................................................................................. 25 2.3 Rainfall Runoff Processes and Component Models ........................................................................... 26 2.4 Previous hydrological modelling studies in Lake Victoria basin ......................................................... 27 2.5 Current Modelling status in the study area ...................................................................................... 30 2.6 GIS-Based Floodplain Management ................................................................................................. 31 2.7 A review of the Galway Flow Forecasting System. ............................................................................ 34 2.7.1 Description of Rainfall-Runoff Models used in the GFFS ............................................................ 34 2.7.1.1 The SMAR Conceptual Model ............................................................................................. 34 2.7.1.2 The Simple Linear Model (SLM) .......................................................................................... 36 2.7.1.3 The Linear Perturbation Model (LPM) ................................................................................ 37 2.7.1. 4 Linearly-Varying Gain Factor Model (LVGFM) .................................................................... 38 2.7.1. 5 Artificial Neural Network (ANN) Model .............................................................................. 40 vi

2.7.2 Conclusion on Galway Flow Forecasting System Models............................................................ 45 CHAPTER THREE .................................................................................................................................... 46 3.0 Methodology ................................................................................................................................... 46 3.1 Data Sources.................................................................................................................................... 46 3.1.1 Primary Data Sources ................................................................................................................ 46 3.1.1.1 Daily Rainfall Data .............................................................................................................. 46 3.1.1.2 Stream flow data ................................................................................................................ 48 3.1.1.3 Potential Evapotranspiration Rates .................................................................................... 48 3.1.2 Secondary Data sources ............................................................................................................ 50 3.1.2.1 Historical Rainfall Data ....................................................................................................... 50 3.1.2.2 Historical Evapotranspiration Data ..................................................................................... 50 3.1.2.3 Historical River Discharge Data ........................................................................................... 50 3.2 ArcView 3.2 ..................................................................................................................................... 51 3.3 ArcView Spatial Analyst ................................................................................................................... 51 3.3.1 Interpolation Using Inverse Distance Weighted (IDW) ............................................................... 51 3.4 Modelling and Analysis .................................................................................................................... 52 3.5 Calculation of Discharge values from Observed water levels ............................................................ 53 3.6 Modelling in Galway Flow Forecasting System ................................................................................. 54 3.6.1 Selection of Model .................................................................................................................... 54 3.6.1.1 The Simple Linear Model (SLM) .......................................................................................... 55 3.6.1.2 Justification of the Model Used .......................................................................................... 57 3.6.1.3 The Soil Moisture Accounting and Routing (SMAR) Model .................................................. 57 3.6.1.4 Displaying of the Model Outputs ........................................................................................ 59 3.6.1.5 Optimization of the Model Output ..................................................................................... 60 3.6.1.6 Running the Model in Updating Mode ................................................................................ 60 3.6.1.7 Real-time Flow Forecasting ................................................................................................ 61 3.7 Calculation of Forecasted Water levels ............................................................................................ 62 3.8 Conclusion ....................................................................................................................................... 62 CHAPTER FOUR ..................................................................................................................................... 63 4.0 Results and Findings ........................................................................................................................ 63 4.1 Introduction .................................................................................................................................... 63 4.2 Daily Reported Rainfall Amounts ..................................................................................................... 63 4.3 Rainfall Gridding using GIS ............................................................................................................... 67 4.3.1 Discussion ................................................................................................................................. 73 4.4 Calculation of Daily Discharge from reported Water levels............................................................... 73 4.4.1 Discussion ................................................................................................................................. 75 4.5 Calculation of Daily Water Levels from Forecasted Discharge values ................................................ 76 4.5.1 Discussion ................................................................................................................................. 78 CHAPTER FIVE........................................................................................................................................ 80 5.0 Summary, Conclusion and Recommendations .................................................................................. 80 5.1 Summary ......................................................................................................................................... 80 5.2 Conclusion ....................................................................................................................................... 81 5.3 Recommendations ........................................................................................................................... 81 5.3.1 Recommendations at National Level ......................................................................................... 81 5.3.2 Recommendations at Local Level .............................................................................................. 82 REFERENCES .......................................................................................................................................... 84 vii

LIST OF TABLES

Table 1: Rainfall stations in Nzoia River Basin

Table 2: Selected River gauging stations

Table 3: Daily Potential Evapotranspiration rates

Table 4: Daily reported Rainfall amounts at Nzoia River basin

Table 5: Rainfall grid values showing Mean Areal Rainfall

Table 6: Daily discharge values for Nzoia River as at Rwambwa gauging station

Table 7: Daily forecasted water levels for Nzoia River

viii

LIST OF FIGURES

Figure 1: Map showing Nzoia River basin

Figure 2: Digital Elevation Model of Nzoia River basin

Figure 3: Classification of Models based on process description

Figure 4: Classification of Models based on space time scale

Figure 5: Classification of Models based on solution technique

Figure 6: Nzoia River basin Rainfall Grids

Figure 7: Graph showing Daily River discharge against Rainfall amounts

Figure 8: Graph showing daily forecasted water levels against Rainfall amounts

ix

ABBREVIATIONS AND ACROYNMS

GFFS

Galway Flow Forecasting System

APFM

Associated Programme on Flood Management

WMO

World Meteorological Organization

WKCDD&FMP Project

Western Kenya Community Driven Development and Flood Mitigation

IFM

Integrated Flood Management

GIS

Geographical Information System

LVN

Lake Victoria North

ITCZ

Inter-Tropical Convergence Zone

QPF

Quantitative Precipitation Forecasts

SWMM

Storm Water Management Model

SHE

Systeme Hydrologique Eurepeen model

PWP

Permanent Wilting Point

LVDSS

Lake Victoria Decision Support System

FAO

Food and Agriculture Organization

AML

Arc Macro Language

WRMA

Water Resources Management Authority

PSLM

Parametric Simple Linear Model

PLPM

Linear Perturbation Model

SMAR

Soil Moisture Accounting and Routing Model

ANNM

Artificial Neural Network Model

LVGFM

Linearly Varying Gain Factor Model x

OLS

Ordinary Least Squares

ANN

Artificial Neural Network Model

RTMOCM

Real-Time Model Output Combination Method

API

Antecedent Precipitation Index

MAP

Mean Areal Precipitation

RE

Relative Error

IVF

Index of volumetric fit

UH

Unit Hydrograph

LTF

Linear Transfer Function

xi

CHAPTER ONE 1.0 Introduction

Water is the key element in economic, social and cultural development of any society. Throughout history, people have settled next to waterways and in flood plains because of the advantages they offer. In spite of these benefits, water can also cause destruction and damage. Flood devastation results in loss of lives, widespread crop destruction and associated economic disasters. During the last couple of decades, Kenya has experienced serious incidents of flood disasters, in different parts of the country and caused major disturbances, destroying property and resulting in loss of life. Recurring floods in the rivers Nzoia, Yala, Nyando, Sondu and Kuja, sub-basins of Lake Victoria, cause large scale devastation of crops, property and physical infrastructure and hamper developmental activities. However, loss of human lives is the most catastrophic impact of floods. Often people are taken unawares when the flood causing heavy rainfall occur in the upper catchments while the plains lower down receive relatively low or even scanty rainfall.

Preparedness and response actions of the various disaster management authorities to prevent or mitigate flood-related disasters are highly dependent on the overall flood management strategy adopted by the country. Absence of a clear strategy and policy for flood management contributes to an increase in the adverse impacts of flood disasters socially, economically as well as environmentally. The availability and proper use of accurate and timely meteorological and hydrological monitoring and forecast products and dissemination of adequate and relevant information to authorities responsible for civil protection and the general public for effective 1

disaster response, play an important part in the overall strategy. The difficulties are compounded when the infrastructure on which to build early warning and response systems is rudimentary, as is the case in most developing countries including Kenya.

Kenya has been experiencing some of its worst flood events during recent years. The water Act of 2002 of the Government of Kenya took certain concrete steps towards the development of Flood Management Strategy for Lake Victoria under the Associated Programme on Flood Management (APFM). APFM was a joint initiative of World Meteorological Organization (WMO) and Global Water Partnership funded by the Government of Japan and the Government of the Netherlands. It was to promote the concept of Integrated Flood Management (IFM), which aimed at maximizing net benefits from flood plains and minimizing loss of life by reducing the vulnerability of the society to flood risks through an optimal mix of structural and non-structural measures. In 2007 through the Western Kenya Community Driven Development and Flood Mitigation Project (WKCDD&FMP) sponsored by the World Bank started more emphasis to shift from structural to non-structural measures or at least to realize the relevance of those strategies as essential components of a more comprehensive approach for floodplain management. Non-structural measures like zoning, flood insurance, relocation, and flood forecasting, warning, and response systems are intended for land use management in the floodplain to prevent and/or properly face potential damage. In general, after a structural flood control plan is in place, complete control of flood water or prevention of all damages is normally not feasible. There is always a residual flood damage risk that remains and which might be addressed by nonstructural measures. The development of Flood Hazard Maps for zoning and insurance programs is part of these non-structural components. The first product that needs to be 2

obtained for these measures is a flood inundation map depicting the extent of the inundated areas for a given real or synthetic storm event. The evaluation of flood damage reduction plans and the definition of nonstructural measures require the consideration of hydrologic, hydraulic, and economic components, Government of Kenya and World Bank Report (2007).

1.1 Statement of Problem

Nzoia River experiences frequent flooding almost every rainy season. The frequency has of late increased and the magnitudes of the floods discharges are on an increasing trend. This has been compounded further by the environmental degradation in the upper reaches of the river leading to increased erosion and siltation. Forest clearing for increased agriculture in the upper parts of the basin is on the increase due to the ever increasing population in the basin. This pressure on land has led to settlements being established in the flood plains thereby exposing settlers to inundation by floods over the years. Floods have over the years caused loss of both human life and property in the Budalang‟i and its vicinity. Previous attempts to control floods through construction of dikes have not successfully solved the flooding problem. Flood waters have overtopped the dykes and in other cases breaches in dikes have led to continued loss of lives and property besides destruction of social life, infrastructure disruption and destruction.

The study seeks to manage the floods for the benefit of the affected communities other than just protecting them from flood related hazards. In this vision floods are to be harnessed and used for other purposes e.g. irrigation, hydropower generation and recreation.

3

It is in this regard that it is necessary to develop a flood forecasting and monitoring system to assist in reducing loss of human life and property destruction in the lower reaches of the Nzoia River. The floods make it difficult for the people and the government to plan their development activities due to frequent disruptions.

1.2 Objectives of the study 1.2.1 Main objective The main objective of this research was to develop a Decision Support System for Flood Hazard Preparedness and response using Geographical Information Systems.

1.2.2 Specific objectives This study specifically aims at:



Developing a Geographical Information Systems Database for Nzoia river basin.



Calculating the daily mean areal rainfall over Nzoia river basin and development of rain grids showing rainfall intensity over the basin.



Calculating the daily river discharge of Nzoia River to be incorporated in the model from observed daily water levels.



Preparing graphs showing observed and simulated water levels in Nzoia River as at Rwambwa and Webuye gauging stations.



To forecast the outflow at Budalangi using the Black box catchment model and generation of surface runoff.

4

1.3 Hypothesis

Floods frequently occurring in a river basin can be modeled and forecasted to reduce damaging to life and property.

5

1.4 The Study area

Figure 1: Map Showing Nzoia River Basin

6

1.4.1 The Nzoia River Basin The domain of the study will constitute River Nzoia sub-basin which is one of the sub-basin in Lake Victoria. River Nzoia is one of the largest and longest rivers in western Kenya and lies between latitude 000 02‟N; 01014‟N and 330 54‟E; 350 35‟E in the northern parts of the Lake Victoria Basin. The catchment can be divided into three major areas namely: Upper Catchment covering two major water towers called Cherengani Hills and Mt. Elgon, Middle Catchment is mainly covered by undulating hills which are dissected by the river and its tributaries, and Lower Catchment. This comprises the flood plains extending up to Lake Victoria (Okoola, 1999; Mutemi, 2003.

According to Kabanda and Jury, (1999) Nzoia River is 334 km long and has a catchment area of 12,903 km2 with an annual discharge of 1,777,106 m3/Sec/year. The river originates from Cherengani Hills which form the northern part of the watershed dividing the Keiyo valley from the Lake Basin and transverses Trans Nzoia, Bungoma, Butere-Mumias, Siaya and Busia Districts. The main upper course tributary is the Moiben River. Many other rivers feed the Nzoia before it discharges into Lake Victoria. The major ones are the Kwoitobos, the Little Nzoia, the Ewaso Rongai, the Kibisi and the Kipkaren. The other tributaries into the Nzoia are the Kuywa, the Chwele and the Khalaba discharging into the Nzoia from the north and the Lusumu and Viratsi flowing into the Nzoia from the southern part of the basin. The Nzoia River empties into the Lake Victoria in the South western corner of Lake Victoria North (LVN) catchment area. Therefore, Nzoia River is of international importance as it contributes enormously to the shared waters of Lake Victoria.

7

1.4.2 Physiography and Geology The average Basin elevation is however estimated at 1,917 Meters above mean sea level, while the length of the mainstream is 275km from source to mouth. The river follows the 275km course with a mean slope of 0.010% from its source to discharge into Lake Victoria at about 1,000m above sea level Gadain, H. M et al (2000). Figure 2 below shows a Digital Elevation Model of Nzoia River Basin.

Figure 2: Digital Elevation Model of Nzoia River Basin 8

1.4.3 Climate The climate of the Basin is mainly tropical humid characterized by day temperatures varying between 16ºC in the highland areas of Cherangani and Mt. Elgon to 28º C in the lower semi-arid areas on annual basis. The mean annual night temperatures vary between 4º C in the highland areas to 16º C in the semi-arid areas. Mean annual rainfall varies from a maximum of 1100 to 2700 mm and a minimum of 600 to 1100 mm (Ogallo, 1980).

The area experiences four seasons in a year as a result of the inter-tropical convergence zone (ITCZ). There are two rainy seasons and two dry seasons, namely, short rains (October to December) and the long rains (March to May). The dry seasons occur in the months of January to February and June to September. However the local relief and influences of the Lake Victoria modify the regular weather pattern (Ininda, 1995)

1.4.4 Economy According to the Western Kenya Community Driven Development Project Plan, (2007), the economy of the region is still largely rural and more than 90% of the population earns its living from agriculture and livestock. The farms are privately owned and on average 1 – 3 hectares. However, large commercial farms with an average of 50 – 100 hectares or more characterize such districts as Trans Nzoia and Uasin Gishu. The main food crops include maize, sorghum, millet, bananas, groundnuts, beans, potatoes, and cassava while the cash crops consist of coffee, sugar cane, tea, wheat, rice, sunflower and horticultural crops. Dairy farming is also practiced together with traditional livestock keeping. The River Basin is of great economic importance at local as well as national levels especially in such sectors as agriculture, tourism, fishing, forestry, mining and transport. It is also the main source of water for domestic, (rural and urban water 9

supply), agriculture and commercial sectors, as well as for very important industrial establishments in Western Kenya, namely Pan Paper Mills, Nzoia Sugar Company, Mumias Sugar Company, and West Kenya Sugar. In addition there are numerous minor sugar factories (jageries), coffee roasters, wood processors and tea factories. Other factories are found in Eldoret, Kitale and Kapsabet. The local communities provide labor to these industries from which they obtain income to supplement that from their subsistence activities. Budalang‟i has a total population of about 56,000 people (1999 Census), who inhabit the region mostly affected by floods. The population at the time of study was concentrated in ten camps located in 7-8 sites. The entire Budalang‟i has six administrative divisions with six locations and eighteen sub-locations The areas affected by rivers Nzoia and Yala, which divide Budalang‟i into two sections, suffer a lot during the over flooding of these rivers since the dykes no longer contain the floods as required. The main challenges in the basin include soil erosion and sedimentation, deforestation, flooding, wetland degradation, pollution and solid waste, river bank cultivation, sand harvesting, brick making, human-wildlife conflict and poorly developed infrastructure, Western Kenya Community Driven Development Project Plan, (2007).

1.4.5 Hydrology The accumulation of many years of sediment in the river bed has made the channel of the river course to be above the general level of the flood plain as a result over bank flow across the 400 – 600m wide dykes causes massive flooding. The width of the channel decreases gradually from about 50m at 140Km inland to around 40m in the upper reaches of Kakamega, Bungoma, Trans Nzoia and Uasin Gishu Districts where the altitude is high and the slopes steep. The sediment

10

accumulates and reduces the capacity of river channel so that the river overflows the banks forming the delta. The flow regime of the Nzoia is varied and is occasionally as low as 20m3/s and with extreme floods that may surpass 1,100m3/s, which is the proposed protection level for the dykes for a 25 year return flood. Siltation is heavy. Earlier estimates of siltation rates by ItalConsult are in the order of 158 tonnes per day ItalConsult (1980), however, recent assessments by Lake Basin Development Authority (LBDA) for the period 2000 to 2003 puts it at 574 tonnes per day which is thrice the ItalConsult value. The discharge varies from a low flow of 2.8m3??/s to a 100-year flood flow of 930m3/ s (ItalConsult, 1980, 1982). The soils of the flood-plain in the lower reaches of the Nzoia River are all alluvial. The river meanders in the flood-plain depositing silt during seasonal floods. The major parts contain black cotton soils while other areas have coarse textured sand silt mixture. In some places, there exist saline soils (ItalConsult, 1980, 1982).

According to ItalConsult, (1980, 1982), the upper Nzoia Basin contains extensive seasonal swamp areas in the high and medium rainfall zones that are mainly utilized for grazing due to poor drainage. Once the river bursts its banks, the resultant flooding displaces close to 30,000 persons every rainfall season. People are usually evacuated to safer higher grounds during the floods. Entire homesteads are swept away, property and crops worth hundreds of thousands of shillings are lost and many people perish as rivers break their banks, rendering large areas of land inaccessible. The floods enhance poverty, because crops and businesses are destroyed.

11

1.5 The Case of Floods in Budalang’i Budalang‟i division lies on the shores of Lake Victoria partly on the mouth of Yala River but mainly on the mouth of River Nzoia. River Nzoia is one of the largest rivers in Western Kenya. The main stream of the river flows from the western side of the Elgeyo Escarpment (Sergoi, Sosiani and Kipkelion tributaries) and the Cherangani Hills (Chepkotet and Kaisungur tributaries) from an elevation of approximately 2,286 Metres above sea level. Its tributaries, which flow from the high slopes of Mount Elgon attain maximum elevation in the river‟s basin and is estimated at about 4,300m above mean sea level. The tributaries in Mt. Elgon include Kuywa, Sosio, Ewaso, Rogai and Koitobos).

The average Basin elevation is however estimated at 1,917km above mean sea level, while the length of the mainstream is 275km from source to mouth. The river follows the 275km course with a mean slope of 0.010% from its source to discharge into Lake Victoria at about 1,000m above sea level. It enters the Lake a short distance to the north of the Yala Swamp. The upper catchment is a high rainfall zone having a mean annual rainfall of between 1500 to 1700mm. per annum, while the flood plains is a low rainfall zone with rainfall below 800mm per year. The total catchment area drained by the basin‟s river network is about 12,950 km2 when measured from Rwambwa Ferry area. Of this 12,950 km2, about 90.8% (11,667 km2??) is composed of land area, which is relatively flat, rolling, or sloping with fairly deep soil, continuous vegetative cover and shows a relatively stable landscape. The mean slope of the basin is 0.010%. The remaining land area is either covered by swamp (5.4%), areas of localized instability, sheet or gully-eroded areas, or steep sloping areas with rock outcrops. 12

1.6 Justification and Significance of the Study

There has been lack of application of meteorological radars to quantitative rainfall measurement and hydrological forecasting in Kenya. Radar has obvious advantages for rainfall estimation in hydrology: detailed spatial and temporal resolution over an extensive spatial domain, collected by a single remote device, with the ability to make informed casts of future rainfall. However, limitations on the accuracy of conventional single polarization radar in rainfall measurement have been acknowledged for some time. A number of studies have addressed rainfall-runoff modelling in the Lake Victoria basin, but little work has been done using distributed models. As a result of civilization and industrialization, we have upset the equilibrium of many aspects of our environment, including the water cycle. Environmental protection, sustainable development and climate change are becoming issues of major concern to nations across the world. Besides the emission of greenhouse gases to the atmosphere, politicians and scientists are also interested in the implications of land use changes, agricultural practices, afforestation or deforestation etc.

Global and regional climate models have revealed that there are increases (positive trends) in seasonal precipitation amounts in most of these locations during some seasons. Such increased rainfall coupled with enhanced runoffs is a good combination for generating increased flooding downstream. Therefore in order to effectively study, the impacts of land use changes, surface and groundwater exploitation, climate changes, and subsurface migration of industrial and agricultural chemicals, on our river basins, there has been a trend towards developing fully distributed, physically based hydrologic models and GIS based.

13

Better understanding and accurate prediction of river flood forecasting is of paramount importance in the policy planning and implementation of early warning systems, development and management of agricultural, water resources and other rainfall-dependent sectors of the economy. With the growth and advancement of analysis methods and models, coupled with the availability of quantitative precipitation forecasts (QPF), the relationship between daily rainfall events and atmospheric convective patterns seems due for a modern scientific visitation.

14

CHAPTER TWO 2.0 Conceptual Frame Work and Literature Review 2.1 Introduction

The necessity for estimating river flows from measurable causative factors, principally rainfall, has perhaps provided the most important driving force in developing hydrology as a discipline of science. As early as the seventeen century a little known French scientist, Pierre Perraualt (Dooge, 1959), quantitatively showed that rainfall and snowmelt were sufficient to maintain flow in the River Seine. This contracted the classic belief then that water originating in the earth‟s interior provided a substantial component of stream flow. Two centuries later, Mulvaney (Dooge, 1973) attempted to relate the storm peak of river flow with rainfall records by what is known as the “rational method” that still finds the application in the design of urban storm drainage network in parts of the world. Since then, a plethora models have been developed for different purposes, mainly to simulate and forecast the runoff from watersheds. A model is a mathematical or physical description, which represents a physical, biological or social system. All models simplify the complexity of the real world by selectively exaggerating the fundamental aspects of a system at the expense of incidental detail. A model never completely represents the real world. A mathematical model of a natural system like a complex, large and imperfectly understood catchment is unlikely to represent fully every process occurring at every point in the system. The simplest model that reflects the system‟s behaviour in an adequate way and satisfies the question raised, in the „best‟ model.

15

2.2 Conceptual frame work 2.2.1 Review of Rainfall Runoff Process and Models 2.2.1.1 Introduction Hydrology is concerned with study of the motion of the earth‟s waters through the hydrologic cycle, and the transport of constituents such as sediment and pollutants in the water as it flows. A hydrological model is a simplified simulation of the complex hydrological system. Modelling is one field of scientific activity which has developed the capability of delivering customized solutions through identifying a variety of arrangements or changes within a system to comply with both external and internal highly developed mathematical capabilities and versatile software tools. The rainfall runoff process is a component of the hydrologic system; therefore, it is a valid approach to develop either „detailed models” serving a wide range of modelling requirement, or “parsimonious models” meeting a specific requirement. Since rainfall data is generally in abundance in comparison to runoff data, the attempt has always been to convert rainfall to runoff.

2.2.1.2 Model Classification Clarke (1973) described the rainfall-runoff models by one or more of the characteristic. Firstly are the deterministic or probabilistic models. It is based on the characteristics of model parameters and variables. If the model parameters or variables are considered random variables with probability distributions, the model is called probabilistic. If the model parameters or variables are free from any random variation, the model is deterministic. Secondly are the lumped or distributed models. It is based on the geometric or probabilistic. If the model parameters or variables vary spatially within the watershed, the model is a distributed model; 16

otherwise, it is a lumped model. Thirdly are linear or nonlinear models. It is system-theory sense or statistical regression sense. If the principle of superposition is not violated, the model is linear in system-theory sense. If the model parameters are linear, the model is linear in statistical regression sense. Otherwise, the model is nonlinear. Fourthly are continuous or discrete models. If the model uses continuous functions in the formulation of the physical phenomena, it is continuous. Otherwise, it is discrete. Fifthly are the black-box or process models. It is based on analyses of rainfall input, runoff output, and a transfer function, which simulates the relationship between rainfall and runoff in a watershed, such as unit hydrograph and time-area methods (McCuen, 1997).

Conceptual models are combinations of process and black-box models. In these models, the physical processes are defined using process models and model parameters are optimized employing a black-box approach (Singh, 1988). Examples where the conceptual models are used include the Clark, Nash and Stanford Watershed models. Lastly are the event-driven or continuous process models. It is based on the simulation period. If the model is designed to simulate a single event, it is called an event-driven model. The focus of event-driven models is on the evaluation of surface runoff and direct infiltration. The most commonly used event-driven models are HEC-HMS, developed by the US Army Corps Engineers; Storm Water Management Model (SWMM), developed by US Environmental Agency, among others.

Therefore in order to effectively study, the impact of land-use changes, surface water and groundwater exploitation, climate changes, and subsurface migration of industrial and agricultural chemicals, on our river basins, there has been a trend towards developing fully distributed, physically based hydrologic models. For the last two decades, different causal 17

models have been built with an attempt to fill in the gap of lumped models. A good example is the European Hydrological System -systeme Hydrologique Eurepeen model (SHE), developed by Abbott et al., in 1986, which unfortunately has little real world applications since its data requirements often far exceed what is available. As an improvement to lumped conceptual models, Amerman (1965) developed extension of lumped, non-linear synthesis model based on 'unit source' areas in which the catchment is broken down into a system of sub-areas of relatively homogeneous soils, topography and land use. A similar approach, known as 'hydrological response zones', was adopted by England and Stephenson (1970) to account for spatial variability across the watershed. However these models do not allow for the interaction between sub-areas and the resulting runoff was estimated by summing up the contributions from the individual elements. Beven and Kirkby (1979) also developed a physically based model which takes into account the distributed effects of the channel network topology and dynamic contributing areas. Based on the concept of unit sources proposed by Amerman (1965), semidistributed hydrologic models have evolved recently as spatially distributed hydrologic data become more readily through remote sensing (e.g., HYDROTEL of Fortin et al., 1986; TOPMODEL of Beven et al, 1987; and SLURP model of Kite, 1995).

Without spatially distributed hydrologic information retrievable from many space platforms launched in recent years, distributed or semi-distributed hydrologic models would have little or no practical application. Remotely sensed data, initially collected from truck mounted and airborne sensors and later from space platforms, have been used for wide ranges of applications in water resources problems. (Kite and Pietroniro 1996) provide an excellent review of the current uses of remotely sensed data in various processes of a hydrologic model and indicates the 18

likely future development in this aspect. Studies also indicated potential benefits of using satellite data on the migration of flood, damage improved planning of hydropower production, and irrigation (e.g., Castruccio et al., 1980).

Even though there is potential to review spatially distributed hydrologic information from satellite data, other than mapping of land cover and snow extent, the current use of remotely sensed data in hydrologic modelling is very limited. According to Kite and Pietroniro (1996), reasons for limited use of remotely sensed information in hydrologic models are such as a lack of universally applicable operational methods of deriving hydrological variables from remotely sensed information at different resolutions from different platforms, and insufficiency of appropriate education and training.

Watershed models are generally classified according to the method they use to describe the hydrologic processes, the spatial and temporal scales for which they are designed, and any specific conditions or intended use for which they are designed. The model processes include all the hydrologic processes that contribute to the system output. Based on the description of those processes, in conjunction with the system characteristics, the models can be described as lumped or distributed, deterministic or stochastic or mixed. Lumped or lumped-parameter models treat an entire watershed as one unit and take no account of the spatial variability in processes, input, boundary conditions, or the hydrologic properties of the watershed. In contrast, distributed models ideally account for all spatial variability in the watershed explicitly by solving the governing equations, for instance, for each pixel in a grid (Castruccio et al., 1980).

19

The description of the hydrologic processes within a watershed model can be deterministic, stochastic, or some combination of the two. Deterministic models are models in which no random variables are used, i.e. for each unique set of input data the model will compute fixed, repeatable results. The governing equations describing the hydrologic and soil erosion processes in a deterministic model should be a major factor in selecting a model. Models with equations based on fundamental principles of physics or robust empirical methods are the most widely used in computing surface runoff and sediment yield. Stochastic models, in contrast, use distributions for each variable to generate random values for model input.

As a result, the output from a stochastic model is itself random, with its own distribution, and can thus be presented as a range of values with confidence limits. Fully stochastic models, in which all components of the model are stochastic, are virtually non-existent.

Figure 3: Classification of models based on process description

20

Figure 4: Classification of models based on space and time scale

Figure 5: Classification of models based on solution technique

The watershed models can also be classified based on the time scale of models. The time scale can be defined as a combination of two time-intervals. One of the time intervals is used for input and internal computations. The second is the time interval used for the output and calibration of the model. Hydrologic processes may have to be computed at different time scales; therefore it is important to consider models that operate from event to daily to yearly time scales. At the event time scale, models typically do not compute inter-storm soil moisture conditions and therefore this information must be provided as an initial condition to initiate the model run. Event based models may be employed for storm events of relatively short duration or to finalize the design of 21

technically complex structural and nonstructural management practices. On the other hand, continuous-time hydrologic models can simulate precipitation, available surface storage, snowmelt, evapotranspiration, soil moisture, and infiltration in a seasonal framework. These models typically operate on a time interval ranging from a fraction of an hour to a day. The principal advantage of continuous modelling is that it can provide long-term series of water and pollutants loadings (Kite and Pietroniro 1996).

The spatial scale can be used as a criterion to classify models into small-watershed, medium-size watershed, and large watershed models. Spatial scale is an important criterion in the selection of a model because the storage characteristics may vary at different watershed scales, that is, large watersheds have well developed channel networks and channel phase, and thus, channel storage is dominant. Such watersheds are less sensitive to short duration, high intensity rainfalls. On the other hand, small watersheds are dominated by the land phase and overland flow, have relatively less conspicuous channel phase, and are highly sensitive to high intensity, short duration rainfalls.

2.2.2 Watershed Characteristics and Hydrologic Models The watershed can be viewed as an aggregate of media between which flux of water, nutrients, energy, etc., occurs. Examples of such media are: the soil formation, the groundwater aquifers, the various water bodies, the plants, etc. It is exactly the description of movement of water and nutrients in and between these media that is sought in developing models (Amerman, 1965).

22

2.2.3 Watershed Soil The soil formation plays a fundamental role in the rainfall-runoff process. The precipitation that is received at the ground surface either infiltrates into the soil or gets stored temporarily on the ground surface, until it flows down into water courses or evaporates back into the atmosphere. Thus, it is mainly the characteristics of the soil formation that governs, for instance, how much of the rainfall becomes effective in producing runoff. As rainfall is intermittent, the soil formation experiences random periods of recharge by rainfall, and depletion by evapotranspiration as well as movement of water to deeper and down-the-slope layers.

The soil properties that have a direct influence on the movement of water are the hydraulic conductivity and the soil water-retention characteristics. Hydraulic conductivity under saturated condition is termed as saturated hydraulic conductivity while that under unsaturated conditions is called unsaturated hydraulic conductivity. The unsaturated hydraulic conductivity under field conditions, thus, is not a constant value but rather changes depending upon the water content of the soil. Moreover, the hydraulic conductivity is spatially variable in the watershed.

The soil water-retention property describes the ability of the soil to withhold or release water and is defined as the relationship between the soil moisture content and the soil water suction or matric potential. The matric potential, thus, measures the relative ease with which water within the soil pores can readily flow at different moisture content and as such a very important parameter of the soil formation in modelling soil water movement ( Gadain et al. 2000).

The watershed exhibits considerable spatial variability in soil properties. Often there are different soil types, and within the same soil type, variations have been observed in hydraulic as well as 23

physical properties. Many research works have showed that there is a great deal of spatial variability in soil hydraulic properties. However, very limited information is available on the spatial variability of soil hydraulic properties in most watersheds. Thus, when it comes to the incorporation of the spatial heterogeneity of such soil hydraulic properties, then evaluation and modelling of the variability and efficient incorporation of the same into hydrologic models become important issues related to the overall model development and application exercises. With the development and proliferation of GIS, the capacity to handle spatial heterogeneity of model parameters has been significantly enhanced (Anderson et al, 2002)

The two soil hydraulic properties discussed above are employed by physically-based models for soil-water movement, such as the Richards‟ equation (Anderson et al, 2002). Many other models that fall in the conceptual models category rather use other derived soil properties among which Field Capacity (FC), and the Permanent Wilting Point (PWP), are very common. Although not necessarily in an efficient way, the FC has become one of the most important parameter used in conceptual models for describing the soil water movement.

For instance, the ratio of actual evapotranspiration from a drying soil to the potential evapotranspiration is assumed to be a function of the soil moisture status. There have been many empirical functions describing the ratio; many of them use the FC as the upper limit of the soil moisture above which evapotranspiration proceeds at the potential rate, while the PWP is taken as the limiting soil moisture below which evapotranspiration is assumed to cease.

24

2.2.4 Watershed Vegetation The land use/cover of a watershed greatly affects the hydrologic processes within the watershed. While land cover describes the type of vegetation in an area, land use refers to the type of use to which the land is made. Particularly, processes like interception, transpiration, infiltration, and overland flow are the processes that are directly affected by the type of land use and cover in the watershed.

Exchanges of energy, mass, and momentum between the atmosphere and vegetation are controlled by plant canopies. The influence of the vegetation type and density on hydrologic processes has been studied by conceptualizing the same in some parameters. Primary vegetation characteristics which affect water and energy balance are the leaf area index (LAI), the stomatal conductance, the rooting depth, albedo and surface roughness (Surkan, 1974).

The interaction between vegetation in watershed and hydrologic processes may be summarized as:

1. The climate influences the soil moisture via evaporation and transpiration (which depend on the vegetation) 2. Soil moisture and climate determine the type of vegetation that may grow in a region. 3. The vegetation determines how it may control soil moisture (depending on the climate). The effects of vegetation on hydrologic processes change with time, while at the same time, the vegetation characteristics may be viewed as parts of the responses of the watershed to moisture and energy input.

25

This dynamic interaction is often not reproduced in most of the hydrologic models available today. Often seasonal indices are assigned to vegetation-dependent parameters affecting hydrologic processes (Anderson et al, 2002).

2.3 Rainfall Runoff Processes and Component Models

The water that constitutes a stream flow may reach the stream channel by any of the several paths from the point where it first reaches the earth as precipitation. Some water flows over the soil surface as surface runoff and reaches the stream after the occurrences as runoff. Other water infiltrates through the soil surface and flows beneath the surface to the stream. This water moves more slowly than surface runoff and contributes to sustained flow of the stream during periods of dry weather.

When rain starts falling on more or less pervious areas, there is an initial period during which:

The rainfall is intercepted by buildings, trees, shrubs or other objects, and is thus prevented from reaching the ground: known as interception It infiltrates into the ground: known as infiltration It finds its way to innumerable small and large depressions, filling them to their overflow level: called depression storage.

The maximum rate at which a soil, when in a given condition, can absorb water is its infiltration capacity. The storage may either evaporate or used by vegetation, or it infiltrates into the soil. The difference between the total rainfall, P, and that which is intercepted is called effective rainfall. However, after the depression storage is filled, the rain intensity exceeds the infiltration 26

capacity of the soil and the difference is then called rainfall excess, Pe. Runoff is obviously the residual of precipitation after the demands of interception; infiltration; depression storage and evapotranspiration are met (Anderson et al, 2002).

During the period of rainfall excess, the actual route followed by a specific water particle from the time it reaches the ground until it enters a stream channel is devious. It is convenient to visualize three main routes of travel: overland flow or surface runoff is that water which travels over the ground surface to the channel. Once rainfall excess occurs, it first accumulates on the ground as surface detention, and then surface water originates as overland flow and begins to move down the slopes into small channels, then into larger channels, and finally as channel flow to the water shed outlet. This movement is called overland flow, and the water that thus reaches the stream channels is the surface runoff.

Surface runoff can occur only as a result of storms having rainfall excess. All water contained within the permanent stream channel is called channel storage, Sc. It has been mentioned earlier that the rainfall runoff process can be viewed as set of successive component processes. Most hydrologic models attempt to describe hydrologic processes in a detailed manner through combining these components. In each of the subsequent sections, process descriptions are given first followed by models developed in the past to simulate the process.

2.4 Previous hydrological modelling studies in Lake Victoria basin

Lake Victoria is the largest lake in Africa and the second largest lake in the world. The total catchment area of the lake tributaries is 194,000 km2. The basin is shared by Kenya, Tanzania, Uganda, Burundi and Rwanda, while the lake itself (70,000 Km2) falls within Kenya, Tanzania 27

and Uganda. The tributaries drain a variety of areas: the forested slopes of the escarpment to the northeast; the drier plains of the Serengeti to the Southeast; the Kagera draining the mountains of Rwanda and Burundi to the West; and the swamps of Uganda to the Northeast. The lake and inflowing rivers from the land catchment area contribute significantly to the development of the Lake Victoria basin and beyond. The hydrology of Lake Victoria has been a matter of great importance for several African countries since ancient times. Earlier studies include those of Hurst and Phillips (1933), Hurst (1946), de Baulny and Baker (1970) and Kite (1981, 1982; during the WMO Hydrometeorological Survey), while the Egyptian government carried out early measurements. A comprehensive programme of measurement and analysis was started by the WMO Survey in 1967, and this included gauging stations on all the major tributaries to supplement the stations on the Kagera from 1940 and four Kenya tributaries from 1956. Rainfall stations were established around the lake and on islands; index basin were selected to study catchment hydrology and mathematical models were developed to study tributary inflows and the lake water balance. The water balance studies described by Kite (1981) were unable to reproduce the sharp jump in lake levels observed in early 1960's. It was also deduced that an increase in rainfall of 25-30% over that recorded was necessary to explain the later rise of the lake between 1977 and 1980. The hydrological model developed and applied during the WMO Hydrometeorological Survey in 1982 was the Sacramento model, which was originally developed by Bumash in 1973 with primary purpose of determining catchment discharge. This model uses the Sacramento Soil Moisture Accounting Model to generate surface and sub-surface runoff components with mean areal rainfall and potential evaporation as inputs (Bumash 1995). The data used was established by the WMO project and the model was tested on most of the big catchments in the lake basin for the modelling period 1970-1974. The time frame used for 28

calibrating the model was concurrent with the rainfall and evaporation data availability. Since the WMO survey, the rainfall, evaporation and discharge stations remained operational except for the few cases where there is war or bad economic situations. Georgakakos modified the same model in 1993 and 1995 to operate as a distributed model driven by remote sensed and on site data (Georgakakos et al, 1993). The most upper part of the model is fully distributed, GIS based model of surface runoff and upper soil moisture.

The model uses the Sacramento soil moisture accounting scheme to produce runoff within the sub-basins of interest, the geomorphologic unit hydro graph, and the kinematic channel routing (for steeper slopes) or Muskingum-Cunge routing (for milder slopes) scheme to route the flows through main channel segments. In addition to stream flow, the model produces runoff and soil moisture estimates within each watershed (Moges et al, 1999).

The Lake Victoria Decision Support System (LVDSS) was developed under the auspices of the Food and Agriculture Organization (FAO) of the United Nations for the Governments of Kenya, Tanzania and Uganda to explore various planning and management scenarios in the Lake Victoria basin. The decision support system integrates conventional and remotely sensed data, Geographic Information Systems, and various models for rainfall-runoff, agricultural planning, hydropower scheduling, and lake regulation (LVDSS, 1999).

The hydrological model developed and used in the LVDSS is the modified version of the Sacramento model. The model was applied to the big ten catchments in the lake basin: using the same WMO data. The time frame modeled was also 1970-1974. The model performance was

29

presented in terms of average monthly cycles and daily observed and estimated discharges. The model was only calibrated and not verified on any of the modeled catchments (LVDSS,1999).

Moges et al, (1999) developed a GIS-based distributed model that operates on monthly time step at a spatial resolution of 10 minutes by 10 minutes grid cells. The model was applied to the key catchments identified by the WMO survey and using the same data for calibrating the model. The model consists of two components, grid based water balance model and a flow accumulation and routing model. The performance of the model has been found to be reasonable. Gadain et al, (2000) investigated the hydrologic model part of the LVDSS and compared the results with a conceptual rainfall runoff model. The results obtained were far better than using the Sacramento model in the LVDSS. The problem seems to be in the data availability for calibrating the distributed Sacramento model. It seems that the time frame is not sufficient for calibrating such a model.

2.5 Current Modelling status in the study area

Early this year (2008) WKCDD & FMP established Flood Diagnostics and Forecasting Centre (FDFC) at Kenya Meteorological Department (KMD). The main responsibility of FDFC is monitoring and forecasting on daily basis and so far it has calibrated Geospatial Stream Flow Model (GeoSFM) using concurrent daily historical rainfall, evapotranspiration, and discharge data sets. GeoSFM is a „Galway Real-Time River Flow Forecasting System‟ software package, developed at the Department of Engineering Hydrology, of the National University of Ireland, Galway, (O‟ Conner et al, 2001) and it comprises a suite of black box and lumped conceptual models. The GeoSFM model has given encouraging simulations and the forecasting is done by 30

updating daily rainfall and river levels and the use Quantitative Precipitation Forecasts (QPF) from KMD.

2.6 GIS-Based Floodplain Management

Given that river and floodplain aspects of floodplain management have a spatial component, a GIS-based approach is suitable to manipulate and visualize the spatial distribution of flood project components. Traditionally, GIS overlay functionalities and computational engines have been used in automated floodplain delineation systems. Several automated GIS-based floodplain delineation systems have been developed to support flood damage assessment components (Noman, et al, 2001). Some of the most well-known systems are: Arc/Info MIKE11-GIS, Arc/Info Floodplain delineation, ArcView MIKE11-GIS, Watershed ModellingSystem, flood mapping functionalities in FLOODWAVE, and the HEC-GeoRAS post-processing delineation. In contrast with the abundant systems for automated floodplain delineation, GIS-based flood damage assessment systems have not proliferated. By the end of the 1980s, many countries started to experience a worsening trend in recurrent flood problems mostly attributed to urban and land use developments that substantially change runoff characteristics and drainage configurations. Some of the attempts to address the increasing pattern of severe flood occurrences by means of GIS-integrated systems are summarized below to provide a conceptual framework for the present research.

To formulate appropriate floodplain management strategies in the form of basin management plans, the government of Hong Kong started to develop in 1990 a system for flood risk assessment (Brimicombe and Bartlett, 1996). The proposed flood risk assessment system was 31

based on the transfer of GIS-based parameterization to stand alone hydrologic/hydraulic modelling systems whose output is passed back to GIS for output visualization and reporting. The spatial extent of flood was superimposed to land used configurations to define flood hazard maps as the main foundation for a spatial decision support system. Even though GIS-based, the system does not represent a true integration of GIS and modelling with central execution of all the involved processes. This system achieves integration by means of data exchange only and not by means of a central and unified execution of chained systems representing the modelling workflow of processes and data. An early attempt to integrate “industry standard” hydraulic numerical modelling and geographic information systems was done through the ArcView GIS software (Muller and Rungoe, 1995). An interface between the 1-D numerical hydraulic model of the Danish Hydraulic Institute, MIKE 11 and Arc View 3.x was developed, the MIKE11-GIS ArcView interface. MIKE 11 was coupled with ArcView to generate 2D and 3D water level and flood inundation maps. The system allows for rapid generation of inundation boundaries for different flood scenarios, including scenarios with or without flood protection measures. It provided a systematic protocol for locating the inundated land under alternative mitigation strategies. The system allows for making multiple runs and testing a number of scenarios efficiently. Almost simultaneously with the previous approach in 1996, the Delft Hydraulics research institute developed a flood hazard assessment model for the river Meuse case study in south Netherlands as a direct response to the flooding events of December 1993 (Jonge et al, 1996). Recognizing the fact that the important river related aspects of flooding and managing floods (safety, agriculture, industry, etc.) all have

32

a spatial component, the GIS package ARC/INFO was selected at the time as the central framework to develop the model (Tineke De et al, 1996).

The Galway Flow Forecasting System (GFFS) provides for a somewhat less aggregated approach to evaluate the depth-damage relationship. In it, each structure is given a particular depth-damage assessment based on a more detail definition of the properties‟ first floor and ground levels. Specification of first floor stages and beginning damage depth stages for each property in the inventory allow for a more realistic approach. However, the depth-damage relationship is aggregated at each damage index location station and the effect of a given depth at the index location relies on very good quality field surveys and evaluations. This approach even though more realistic relies on very hard to get depth-damage relationships that can quickly become obsolete given the dynamic nature of floodplains regularly affected by new regulations, alleviation plans, and changing land use configurations. So, even though the GFFS provides a distributed approach for definition of depth-damage curves (based on a distributed structure inventory)

By defining the three basic functions at damage reach index locations a supposedly uniform floodplain section gets aggregated to each index location station. Through this approach any changes on the floodplain configuration are difficult to introduce and will imply a new definition of the basic functions. As part of this research, an alternative approach for flood damage assessment is sketched that takes full advantage of the distributed strength of GIS to obtain the depth-damage relationship at each structure based on the current depth at that spatial location as given by the integrated floodplain delineation process (The Map2Map application). By doing this, a more realistic assessment is expected that may keep up with the dynamic development of 33

the floodplains without having to redefine the aggregated depth-damage curves at the land use, modelling cell, or index location level.

2.7 A review of the Galway Flow Forecasting System.

The Galway river Flow Forecasting System (GFFS) software package was used. The Hydrological models used in the study to simulate the process of runoff generation were two system theoretic black box models namely, the Parametric Simple Linear Model (PSLM) and the Parametric Seasonally based Linear Perturbation Model (PLPM), and a physically inspired conceptual model namely, the Soil Moisture Accounting and Routing (SMAR) Model.

2.7.1 Description of Rainfall-Runoff Models used in the GFFS 2.7.1.1 The SMAR Conceptual Model The Soil Moisture Accounting and Routing (SMAR) Model is a development of the „Layers‟ conceptual rainfall-runoff model introduced by O‟Connell et al, (1970), its water-balance component being based on the „Layers Water Balance Model‟ proposed in 1969 by Nash and Sutcliffe (Clarke, p.307, 1994). Typical of its class, the SMAR model is a lumped quasi-physical conceptual rainfall-evaporation-runoff model, with quite distinct water-balance and routing components (hence it‟s generic name). Using a number of empirical and assumed relations which are considered to be at least physically plausible, the non-linear water balance (i.e. soil moisture accounting) component ensures satisfaction of the continuity equation, over each time-step, i.e. it preserves the balance between the rainfall, the evaporation, the generated runoff and the changes in the various elements (layers) of soil moisture storage. The routing component, on the other hand, simulates the attenuation and the diffusive effects of the catchment by routing the various 34

generated runoff components, (which are the outputs from the water balance component), through linear time-invariant storage elements. For each time-step, the combined output of the two routing elements adopted (i.e. one for generated „surface runoff‟ and the other for generated „groundwater runoff‟) becomes the simulated (un-updated) discharge forecast produced by the SMAR model. The version of SMAR used in the present study, Figure 5, is that which incorporates the suggested modifications of both Khan (1986) and Liang (1992).

In the GFFS version presented here, three two-parameter distribution options are available for routing the generated „surface runoff‟ component of the SMAR model, namely, the classic gamma distribution (Nash-cascade) model (Nash, 1957), its discrete counterpart, the Negative Binomial distribution (O‟Connor, 1976), and the sharp-peaked Inverse Gaussian distribution (Bardsley, 1983) for flashy catchments. The version of SMAR used in the present study has nine parameters, five of which control the overall operation of the water-budget component, while the remaining four parameters (including a weighting parameter which determines the amount of generated „groundwater runoff‟) control the operation of the routing component. The SMAR is calibrated to the observed data using the user‟s choice optimisation procedure to minimise the selected measure of error between the observed and the model estimated discharges. In the context of the SMAR model, the selected measure of model error used for this study is a weighted combination of the sum of squares of the discharge forecast errors and the corresponding index of volumetric fit (i.e. the ratio of the total volume of the estimated discharge hydrograph to that of the corresponding observed hydrograph). As the Nash-Sutcliffe (1970) model efficiency criterion.

35

2.7.1.2 The Simple Linear Model (SLM) Nash and Foley (1982) introduced the Simple Linear Model (SLM) not as a substantive rainfallrunoff model in its own right but rather as a naïve black-box model to be used mainly for the purpose of model efficiency comparisons. The intrinsic hypothesis of the SLM is the assumption of a linear time-invariant relationship between the total rainfall Ri and the total discharge Qi . In its discrete non-parametric form, the SLM, including the forecast error term e i , is expressed by the convolution summation relation (Kachroo and Liang, 1992),

m

Qi =

m

R i - j+1h j + ei = G j=1

m

R i - j+1B j where j=1

Bj = 1 j 1

(F1)

…

Where Qi and Ri are the discharge and rainfall respectively at the i-th time-step,

hj

is the j-th

discrete pulse response ordinate or weight, m is the memory length of the system and G is the gain factor. This can be viewed as a multiple linear regression model of the observed discharge on the previous observed rainfall values and hence estimates of the unit pulse response ordinates can therefore be obtained directly by the method of ordinary least squares (OLS) (Nash and Foley (1982); Kachroo and Liang (1992)). In the SLM, when the rainfall and discharge are expressed in the same units of measurements (e.g. mm/day over the catchment area, for daily data), then the arithmetic sum of the discrete pulse response ordinates B j defines the „gain factor‟ G of the model, which may also be considered as the long term „coefficient of runoff‟ approximately reflecting the ratio of the total volume of the observed discharge hydrograph to that of the observed rainfall input (Kachroo and Liang, 1992). The SLM is included in the GFFS for use as a naïve model in model performance intercomparison studies with the option; in the 36

context of discharge forecast combination, of its inclusion as a very primitive rainfall-runoff model. Any model that does not perform better than the SLM can hardly be considered as a serious rainfall-runoff model.

2.7.1.3 The Linear Perturbation Model (LPM) This model exploits the seasonal information inherent in the observed rainfall and discharge series. It was originally introduced in the context of rainfall-runoff modelling by Nash and Barsi (1983). Initially referred to as the hybrid model, in a series of subsequent publications it is referred to as the Linear Perturbation Model (LPM) (Kachroo et al., 1988; Liang and Nash, 1988; Kachroo, 1992; Kachroo et al., 1992; Liang et al.,1992; Liang and Guo, 1994; Abdo et al., 1996; Elmahi and O‟Connor, 1996; Shamseldin et al., 1997).

In the LPM, it is assumed that, during a year in which the rainfall is identical to its seasonal expectation, the corresponding discharge hydrograph is also identical to its seasonal expectation. However, in all other years, when the rainfall and the discharge values depart from their respective seasonal expectations, these departures series are assumed to be related by a linear time invariant system. Hence, the LPM structure reduces reliance on the linearity assumption of the SLM and gives substantial weight to the observed seasonal behaviour of the catchment. A schematic diagram of the LPM is presented in Formulae 2. The relation between the departures/perturbation series of the LPM, incorporating an output error term e i , may be represented algebraically by the convolution summation equation

m

Qi =

…

Ri - j +1h j + ei j =1

37

(F2)

Where R i and Q i are the rainfall departures and the corresponding discharge departures from their seasonal expectations, respectively. As in the case of the SLM, the ordinary least squares (OLS) method is used to estimate the pulse response ordinates of the LPM, provided that the values of these departures are known. The seasonal expectations (365 values for daily data) are determined directly from the total rainfall and discharge series, the expectations being smoothed by harmonic analysis by omitting the high-frequency harmonics before subtracting them from their respective series to obtain the corresponding departures series.

2.7.1. 4 Linearly-Varying Gain Factor Model (LVGFM) The Constrained Linear System with Thresholds (CLS-T) of Todini and Wallis (1977) and also the Multi-Linear Systems of Becker and Kundzewicz (1987), Kachroo and Natale (1992) and Ahsan (1993) addressed the constant gain factor deficiency of the SLM by using a threshold concept to decompose the input rainfall vector into non-concurrent component input vectors, each of which are fitted to the rainfall-runoff data using separate linear time-invariant systems. However, an abrupt switch from one system response function to another whenever the threshold of the selected control variable (e.g. the antecedent precipitation or catchment wetness) is crossed, involving a substantial change both in the shape of the response function and in the magnitude of its gain factor G, is physically unrealistic. Although a response function which varied gradually with the catchment wetness, both in scale G and in shape, would appear to be much more sensible, our experience of testing many models in the Galway workshops had indicated that getting the water balance (i.e. the volume of runoff) right is far more important in producing high model efficiency than the actual distribution of that volume over time. So, the Linearly-Varying Gain Factor Model (LVGFM), proposed by Ahsan and O'Connor (1994) for 38

the single-input to single-output case, involves only the variation of the gain factor with the selected index of the prevailing catchment wetness, without varying the shape (i.e. the weights) of the response function. The model output has the familiar convolution summation structure (based on the concept of a time-varying gain factor G i );

m

Qi = G j

m

R i - j+1B j ,

where

j=1

Bj

1

…

j 1

(F3)

The multiple-input to single-output form of this model was investigated by Liang et al. (1994). In its simplest form, G i is linearly related to an index of the soil moisture state z i of the catchment by the equation

Gi = a + bz i

(a and b being constants)

…

(F4)

…

(F5)

The overall operation of the LVGFM has the mathematical form

m

m

Qi = a

R i - j+1B j + b z i j=1

j=1

m

m

R i - j+1 aB j + j=1

(zi R i - j+1 ) bB j + ei j=1

m

=

R i - j+1B j + j=1

where

m

R i - j+1B j + ei =

Bj

R i - j+1B j + ei j=1

a B j R i - j+1 = z i R i - j+1

,

Bj

b Bj

,

and

Bj

1.0

.

Although the antecedent precipitation index (API) provides a crude index of the current soil moisture state zi, Ahsan and O‟Connor (1994) suggested that zi be conveniently obtained from the outputs of the naïve SLM, operating as an auxiliary model (See Formulae F3), according to the relation 39

zi =

ˆ G Q

m

R i - j+1hˆ j

(F6)

…

j=1

h Where G and j are estimates of the gain factor and the pulse response ordinates respectively of the SLM and Q is the mean discharge in the calibration period. Although, in the systems sense, the overall LVGFM is non-linear, in terms of the model weighting sequences

Bj

and

Bj

,

Eq. (F5) can be viewed as a multiple linear regression model (Shamseldin et al., 1997). Hence, the weighting sequences

Bj

and

Bj

can be estimated directly by the method of ordinary least

squares (OLS).

2.7.1. 5 Artificial Neural Network (ANN) Model The widely-used artificial neural networks (ANN) provide a flexible non-linear mapping of the network input (or set of inputs) into the network output (or set of outputs) without specifying a priori the mathematical nature of the relation between inputs and outputs. In the GFFS, the ANN can be used in four quite distinct contexts. Firstly, as an option, it may be used as a basic blackbox rainfall-runoff model (Shamseldin, 1997) to produce un-updated discharge forecasts (i.e. as an alternative in the GFFS to the SLM, LPM, LVGFM or SMAR models). Secondly, it may be used in the GFFS in the model-output-combination context, wherein the un-updated discharge forecasts of selected basic models are combined by the network to produce better un-updated discharge forecasts than those provided by the individual basic models included in the combination (i.e. as an alternative to either the WAM or SAM forecast combination methods) (Shamseldin et al., 1997). Thirdly, the ANN may be used as a real-time discharge forecast updating technique (Shamseldin and O‟Connor, 2001), wherein the ANN operates on both the 40

discharge forecasts and on the recent observed discharge values in order to produce updated forecasts, these input discharge forecasts being either those of an individual basic rainfall-runoff model, or those produced by a forecast combination method (i.e. as an alternative to using an AR model for forecast error updating). Fourthly, the ANN can be used as a real-time rainfall-runoff model having an inbuilt updating structure, using the observed discharges and the traditional input information (i.e. rainfall and /or upstream flow hydrograph data) to directly produce the updated discharge forecasts. It is also planned, as a fourth updating option, to incorporate the Real-Time Model Output Combination Method (RTMOCM) based on a multiple-input Linear Transfer Function Model (Shamseldin and O‟Connor, 1999), into the GFFS in the near future (to function as an alternative to its existing AR and ANN updating techniques). The type of neural network used in the GFFS is the “multi-layer feed-forward network” which is considered to be very powerful in function modelling. It consists of an input layer, an output layer and only one “hidden” layer located between the input and the output layers. A layer consists of a set of neurons each having the same pattern of connection pathways to the other neurons of the adjacent layers. Each neuron of a particular layer has connection pathways to all the neurons in the following adjacent layer, but none to those of its own layer or to those of the previous layer (if any), i.e. nodes within a layer are not inter-connected. Likewise, nodes in nonadjacent layers are unconnected. Only one hidden layer is used in the version of the ANN included in the GFFS. The number of neurons in the input layer equals the number of the elements in the external input array to the network. There is only one neuron, for the single output, in the output layer.

41

In the context of the ANN as a basic rainfall-runoff model, instead of using the rainfall series R i as the input, Shamseldin (1997) used a form of antecedent rainfall index, comprising a weighted sum of the current and immediately previous rainfall values, as the single external input to the network. As the neural network itself does not incorporate storage effects, storage is implicitly accounted for by the use of such an index. Instead of using the classic Antecedent Precipitation Index (API), involving a geometric weighting series, as the rainfall index, Shamseldin used the output series of the naïve SLM, i.e. the convolution summation of the rainfall with a more realistic weighting series, for this purpose. So, the ANN effectively enhances the output forecast of the SLM by means of a suitable non-linear transformation.

In the context of the un-updated forecast combination method (ANNM), the estimated discharge output of each the chosen rainfall runoff models is assigned, at each time-step, to one (and only one) neuron in the input layer, resulting in one neuron in the input layer for each model included (Shamseldin form Yi

et

al.,

1997).

ˆ ,Q ˆ ,...,Q ˆ ,T Q i,1 i,2 i, n

In this

case,

the

ANN

input-array vector

has

the

, where n is the number of individual model forecasts included in the

combination. Note that, for network parsimony, the number of neurons in the hidden layer is usually two or three. Note also that the combination output forecast is still an un-updated forecast.

In the context of using the ANN as a forecast updating procedure, as new observations of discharge become available, the input array vector has the form

Y

i

Qi-1,Qi- 2 , ,Qi-p ,Qˆi ,Qˆi-1, ,Qˆ i-q

T

42

ˆ where Qi is observed discharge series and Qi is the simulation mode discharge forecast produced by the substantive rainfall-runoff model or by model output combination. Shamseldin and O‟Connor (2001) consider such ANN updating as a form of non-linear Autoregressive Exogenous-Input model, i.e. as a non-linear generalisation of the Real-Time Model Output Combination Method (RTMOCM) proposed by Shamseldin and O‟Connor, (1999).

In the context of using the ANN as a real-time forecasting model having a built-in updating mechanism, involving the production of updated forecasts as new observed discharge values become available; the input array vector has the form Yi = (Ri, Ri-1,…,Ri-p, Qi-1, Qi-2,…,Qi-q,)T,

where Ri and Qi are the rainfall and observed discharge series respectively (Xiong et al., 2001). For a neuron either in the hidden or in the output layer, the received inputs y i are transformed to its output y out by a mathematical transfer function of the form …

(F7)

M

yout = f (

wi yi

wo )

i =1

Where f ( ) denotes the transfer function, w i is the input connection pathway weight, M is the total number of inputs (which usually equals the number of neurons in the preceding layer), and w o is the neuron threshold (or bias), i.e. a base-line value independent of the input.

43

In the GFFS, the non-linear transfer function adopted for the neurons of the hidden layer and also that of the output layer is the widely-used logistic function, i.e. a form of sigmoid function, given by

M

f(

wi yi

1

wo ) =

i =1

…

M

-

1+ e

i =1

(F8)

wi y i wo

Which is bounded in the range [0,1], implying that the network output is likewise bounded in that range, σ being a scaling parameter of the transfer function. The weights w i , the threshold

w o and the σ of the different neurons can all be interpreted as the parameters of the selected network configuration. These parameters may be determined using the method of backpropagation or the conjugate gradient method (Shamseldin, 1997) but, as the Simplex method is used in the GFFS for calibration of the rainfall-runoff models, it is also used for calibrating the ANN in that package.

The effective range of the logistic function is generally less than that indicated by Eq. (F8) and in order to facilitate the comparison of the actual observed discharges Qi and the network-estimated outputs, the following equation is adopted for training (i.e. calibrating) the ANN;

Qsi = 0.1 + 0.75(

Qi ) Qmax

…

(F9)

i.e. a rescaling of the observed discharges Q i so that the resulting rescaled discharges Qs i are bounded between 0.1 and 0.85, Q max being the maximum observed discharge of the calibration period. The discharge forecast of the ANN is given by the inverse of Eq. (F9). 44

2.7.2 Conclusion on Galway Flow Forecasting System Models The performance of the NP-SLM, which is a very crude and simplified model of input-output transformation process, is clearly inferior to that of all other models. The LVGFM, which is an improvement over the NP-SLM, performs better in very large catchments (O‟Connor et al 1992). For the catchments characterised by strong seasonality, the NP-LPM outperforms the LVGFM. For large catchments with such seasonality, the NP-LPM seems to perform even better than the SMAR model.

For smaller catchments, however, the SMAR conceptual model performs

consistently better than the NP-LPM. The ANNM, although characterised by a large number of weights (parameters), does not generally perform better than the simpler models. According to Goswami, M et al, (1992), the SMAR model, also having nine parameters, fails to simulate the hydrological behaviour of the large catchments adequately. Nzoia River basin having a catchment of 12,903 Km2 is a small catchment hence the study used the SMAR Model.

45

CHAPTER THREE 3.0 Methodology 3.1 Data Sources 3.1.1 Primary Data Sources 3.1.1.1 Daily Rainfall Data Daily Rainfall data came from The Kenya Meteorological Department. A total of 22 rain gauge stations within and in the periphery of the basin were used as shown in Table 1. Table 1 shows the daily readings as per the various rainfall stations at Nzoia River basin in the National Meteorological Centre.

46

Table 1: Rainfall stations in Nzoia River basin

Latitude in Longitude

Year

Altitude in

Degrees

in Degrees

Started Metres

No

Name

Station ID

1

NABKOI F STATION

8934139

2

CHEBIEMIT AGRIC OFF

8935104

0.867

35.500

1950

2439

3

ADC CHORLIM FARM

8834013

1.033

34.800

1926

1981

4

ELDORET AIRPORT

8935115

0.400

35.230

1997

2084

5

ELDORET MET

8935181

0.533

35.283

1972

2120

6

KADENGE YALA

8934140

0.033

34.183

1968

1167

7

KAIMOSI TEA

8934078

0.217

34.950

1939

1707

8

KAKAMEGA MET

8934096

0.267

34.750

1957

1585

9

KAPENGURIA /WRMA

8835004

1.249

35.113

1930

2134

10

KIMLILI AGRIC

8934098

0.867

34.683

1959

2073

11

KIPKABUS FS

8935117

0.317

35.517

1951

2499

12

KITALE MET

8834098

1.000

34.983

1950

1840

13

LUGARI F STN

8934016

0.667

34.900

1932

1680

14

CHEPTONGEI

15

MUMIAS SUGAR

8934133

0.367

34.500

1967

1301

16

NANDI F STN

8935112

0.200

35.067

1950

1981

17

NZOIA SUGAR

8934183

0.567

34.650

1980

1462

18

UHOLO CHIEF‟S CAMP

8935232

19

PORT VICTORIA

8935172

20

KAPSARA TEA F

8934189

1.088

35.159

47

1893

3.1.1.2 Stream flow data Observed data on stream flow by Ministry of Water Resources at Webuye and Rwambwa Bridge was used for calibrating and verification process in the rainfall-runoff model. Table 2 shows gauging stations and locations where daily water level observations were made.

Table 2: Selected River Gauging Stations

Year

Years

in River

GPS Location

Started Operation Gauge Latitude

Longitude

in Degrees in Degrees 21.02.74 24

Nzoia

at 0.121330N 34.0910E

Altitude

in

Meters 1325

Rwambwa Ferry 02.04.47 40

Nzoia

at 0.585970N 34.806860E 1457

Webuye 3.1.1.3 Potential Evapotranspiration Rates Estimation of Evapotranspiration rates is important in determining expected rates of stream discharge. The concept of potential Evapotranspiration is the possible loss of water without any limits imposed by the supply of water. The data was collected from Kenya Meteorological Department. Table 3 shows potential Evapotranspiration rates for Nzoia River basin for the month of October, 2008.

48

Table 3: Daily Potential Evapotranspiration Rates.

Date

Evaporation rate in Millimeters Per day

10/1/2008

4.3

10/2/2008

4.1

10/3/2008

4

10/4/2008

3.8

10/5/2008

4.2

10/6/2008

4.6

10/7/2008

4.3

10/8/2008

4.1

10/9/2008

3.6

10/10/2008 3.5 10/11/2008 4.6 10/12/2008 4.5 10/13/2008 4.1 10/14/2008 4.1 10/15/2008 3.7

49

Date

Evaporation rate in Millimeters Per day

10/16/2008

4

10/17/2008

4.2

10/18/2008

4

10/19/2008

4.5

10/20/2008

4.5

10/21/2008

4.5

10/22/2008

4.7

10/23/2008

4.8

10/24/2008

4.9

10/25/2008

4.9

10/26/2008

4.7

10/27/2008

5

10/28/2008

5

10/29/2008

4.9

10/30/2008

5.4

10/31/2008

5.4

3.1.2 Secondary Data sources 3.1.2.1 Historical Rainfall Data The observed daily concurrent rainfall for a period of 15 years (1995 – to date) was used for calibration and verification, while Quantitative Precipitation Forecast (QPF) grids will be used for the forecasting purpose. In the Nzoia basin there are four synoptic and sixteen rainfall stations distributed uniformly. Rainfall data was obtained from Kenya Meteorological Department.

3.1.2.2 Historical Evapotranspiration Data The observed daily concurrent Evapotranspiration data for a period of 15 years (1995 – to date) was used for calibration and verification. In the Nzoia basin there synoptic and 16 rainfall stations distributed uniformly. Evapotranspiration data was obtained from Kenya Meteorological Department.

3.1.2.3 Historical River Discharge Data The observed daily concurrent River discharge data for a period of 15 years (1995 – to date) was used for calibration and verification In the Nzoia basin there are two River gauging stations which are Webuye and Rwambwa Bridge. Rating curves and discharge data was obtained from The Ministry of Water, Water Resources Management Authority (WRMA).

50

3.2 ArcView 3.2

ArcView is a Geographical Information System (GIS), which runs on a wide variety of computer systems or platforms. It was used for mapping purposes in the project. The ultimate goal for ArcView 3.2 in the project was for calculating the basin‟s Daily Mean Areal Rainfall to be incorporated in modelling process. The specific extension used is discussed below.

3.3 ArcView Spatial Analyst

ArcView Spatial Analyst is an extension in ArcView 3.2. Some of the options of Spatial Analyst include Interpolating Grid from points, Creating contours from and Digital Elevation Model, Deriving slope from a Terrain Elevation Model, Derive Aspect from a Terrain Elevation Model, Computing Hillshade from a Digital Elevation Model and Calculating Viewshade from a Digital Elevation Model. This project only used the Interpolate Grid option to get the Daily Rain grids of the Nzoia basin in order to compute the Mean Areal Rainfall to be used in modelling flows. The method to be used is the Inverse Distance Weighted technique, the Z value field will be the particular day e.g. Day 5. Interpolation will be calculated by the nearest neighbour statistics of 12 neighbours.

3.3.1 Interpolation Using Inverse Distance Weighted (IDW) Interpolation predicts values for cells in a raster from a limited number of sample data points. It is used to predict unknown values for geographic data point data, where in the case of this project; four synoptic and sixteen rainfall stations were selected. These were the points that were used in production of Daily rainfall grids. Unknown values of other rainfall amounts for other

51

places were predicted with the help of this raster surface, using a mathematical formula that used values of nearby known points.

Interpolation was done so as to create a rainfall amounts surface to calculate the mean Areal rainfall for Nzoia river basin.

Inverse Distance Weighted method of interpolation was the method that proved more viable. This was because IDW estimates cell values by averaging the values of the vicinity of each cell. The closer a point is to the centre of the cell being estimated, the more influence it has in the averaging process. This method assumes that the variable being mapped decreases in influence with distance from its sample location.

3.4 Modelling and Analysis

This project focuses on Real-time river flow forecasting of Nzoia river with two Lead Times, giving two day forecasts. Modelling tools used as mentioned earlier will include Daily Mean areal rainfall, Discharge as at gauging stations, Potential daily Evaporation and Quantitative Precipitation Forecasts

GIS was used in mapping the whole Nzoia basin, drainage network, Synoptic rainfall stations and Rainfall stations creating spatial data sets of the basin using ArcView 3.2. Each Station was given an Identification Number. Daily rainfall amounts for the 21 stations will be inputted in Microsoft Excel. Daily rainfall data was got from the National Meteorological Centre. The rainfall data inputted in Excel was saved in Text format and transferred to ArcView 3.2 where it was added as a table and subsequently added as a new Event theme in a View. It was converted 52

to a Shape file. Once the data is in Shape format, Station Interpolation was done to get a surface of the point theme. The technique here is the ArcView Spatial analyst extension for the GIS program. The method used was the Inverse Distance Weighted technique, the Z value field was the particular day e.g. Day 5. Interpolation was calculated by the nearest neighbour statistics of 12 neighbours. This gave a weight to the point with the highest number in the theme. Once the surface was gotten, the themes were arranged in a manner that all the themes were visible from one view. The basin shape file should be the top most theme. In order to get the Mean Areal rainfall over the basin, all the stations have to be summarised which lie in the basin. In this case the basin shape file is the determinant. From the Analysis menu in ArcView summarise zone is chosen and base line of summarizing is chosen, in this case Nzoia basin shape file. The statistic to get was Mean Areal rainfall.

3.5 Calculation of Discharge values from Observed water levels

To calculate the discharge of Nzoia river measurements of the levels of water were taken at Webuye and Rwambwa gauging stations. The area of the flow and the speed of the flow were calculated. To get the area, the depth and width of the river at a particular cross-section perpendicular to the flow direction were measured. Since most rivers have irregular bottoms, measurements of the depth at a number of points across the bed were taken, a cross section was drawn. Then the area of the section was calculated in square feet. The speed of the flow at the same section was also measured with a flow meter; the speed is calculated in feet per second. The total flow is the area times the speed, which is in cubic feet per second. This was converted into cubic meters per minute. The flow rate is in units volume per unit time, usually cubic meters per second, or cubic feet per second. The equation for flow is 53

Q W D V

Where: Q = Flow rate, W = Width, D = depth, V = Velocity

3.6 Modelling in Galway Flow Forecasting System 3.6.1 Selection of Model For modelling to start a working directory/folder was created and all the required data files (discharge, rainfall, and evaporation) in standard GFFS format were transferred to the working directory, any model(s) within GFFS can be run. A model was to be chosen from the programs menu, the following 3 options appeared:

i.

Calibration and verification in simulation mode

ii.

Calibration and verification in updating mode

iii.

Real-time flow forecasting

Because the models were being run for the first time, all models in GFFS were ran so as to check which one fitted well for Nzoia river basin from the show of the Unit Hydrographs (UH‟s) of the outputs in relation with the hydrographs drawn from observed discharges in the basin. The following models below were run.

i.

Simple Linear Model (SLM)

ii.

Soil Moisture Accounting and Routing Model (SMAR)

54

Any models above had the chance to be selected. However, for efficiency it was advisable to run the SLM first and to save the estimated discharge output in a user-defined file name, which was subsequently used as an auxiliary input to other models.

3.6.1.1 The Simple Linear Model (SLM) To run the SLM Model, two categories of the Model; Non-parametric Simple Linear Model and Parametric Simple Linear Model were run from the same Model. Whereas the entire response function series of the parametric form of the SLM is defined by an equation of just a few parameters, this is not so for the non-parametric form. A brief description on running the Nonparametric SLP is given below.

The Model can be run by inputting modelling parameters which include Discharge, Potential Evapotranspiration and Mean Areal Rainfall of Nzoia River basin. The parameters can be inputted through three methods which include

i.

Entering manually on the screen using keyboard (for calibration)

ii.

Entering from file containing input information (for calibration)

iii.

Entering from file containing pulse response values (no calibration)

Because the SLM was being run for the first time, option (i) was selected. When option (i) above was selected, the systems required to fill the below parameters.

Name of the catchment and number of inputs to the catchment (maximum allowable 5) (e.g. 1 for the single-input series case).

55

File for input information and File for estimated discharge. The former is used for selecting input information option (ii) in subsequent runs of the model for that catchment, as mentioned above, and the latter is useful for using the output of the SLM (as an auxiliary model) as an auxiliary input to other models such as the LVGFM and the ANN.

NB. It was advisable to select the name of these two files (*.inp and *.out) in such a way that they will be immediately recognized. For example, for the Nzoia catchment and NP-SLM model, these file names were "NzoiaNPSLM.inp" and "NzoiaNPSLM.out".

There are two methods of calibration; these are Split record calibration, verification option and Calibration with whole series. For Split record calibration and verification, data number corresponding to the data number for the calibration and verification periods is entered. These data numbers correspond to the data number in the data file (discharge, rainfall), e.g. 1000 correspond to the 1000th element in the series.

After the above parameters were entered, a dialogue box requiring the information about the data file for Rainfall data popped up, the name of the rainfall file from the working directory for the Nzoia project was selected and also the catchment area and memory length (an initial guess of the number of elements in the SLM pulse response series). Parameters about the corresponding observed discharge file has to be inputted in the SLM Model.

The model ran, after performing the run operation successfully. Following a similar procedure to that described above and filling in some extra information of the general input information, the Parametric Simple Linear Model (PSLM) can likewise be run.

56

3.6.1.2 Justification of the Model Used For Nzoia river basin Soil Moisture Accounting and Routing (SMAR) Model is used since the Unit Hydrograph produced from the simulation showed the same trend with Unit Hydrographs drawn from observed statistics.

3.6.1.3 The Soil Moisture Accounting and Routing (SMAR) Model SMAR model gave two options for calibrating the Model, these were;

i.

Entering manually on the screen using keyboard.

ii.

Entering from existing file.

During an earlier run of the SMAR model and the Simple Linear Model, a specific file name for the general input information on the selected catchment, with the default .inp extension was created in the system hence the system contains all the parameters needed in the modelling process. The options below were selected to run the model. The parameters were either to be entered from;

i.

File containing input information

ii.

File containing parameter values

In calibration of the model using an optimization technique, option (ii.) was selected and the name of the general input file is entered, which had been saved during an earlier run of the SMAR model. However because SMAR model was being run for the first time, option (i) was selected where by there were the below options which were used.

i.

Calibration and verification (parameter optimization) 57

ii.

Parameter values (not to be optimized)

For calibrated SMAR model using an optimization technique, and optimized model parameter values in a separate file, the second option was used. The only difference in the two options is that, in the former case you have to enter the lower limit, upper limit and starting values of the model parameters, whereas in the later case, you have to enter only the optimised values of the model parameters, which have been obtained from the earlier model calibration/run. In either case, calibration and verification or parameters value, the system required;

i.

Catchment name

ii.

Catchment area in sq. Km2

For the Catchment name there were 9 parameters for SMARG model and 10 parameters for the SMARK model, a variant of SMAR model for the Karstic case or for the case when the water balance of the data set (Rainfall, Evaporation & Discharge) is not satisfied for the catchment. The default value of the lower limit, upper limit and starting values of the parameters were used as they are, for the first run of the SMAR model or they can be changed. For Nzoia river basin, the name of Rainfall, Evaporation and Discharge data files were inputted. For the calibration type, either Split record calibration and verification or Calibrate with whole series. For Split record calibration and verification named data number for the calibration and verification periods was entered (discharge, rainfall), e.g. 1000 correspond to the 1000th element in the series. The options for optimization for Nzoia river modelling included Genetic Algorithm, Simplex search, Rosenbrock and the fourth for selecting all three methods. By default, the Simplex search is selected. The Warm-up period (No. of Time steps) and Memory Length (No. of Time steps). For 58

the warm-up period, the no. of steps taken is roughly about 3% of the total no. of elements for calibration period, e.g., if there are 2000 elements for calibration, then the warm-up period will be 60. Similarly, the memory length is also entered by observing the shape of the response function/unit hydrograph. By default, the warm-up period is 60 and the memory length is 15.

The SMAR model runs.

3.6.1.4 Displaying of the Model Outputs The system provided three options for displaying Model outputs they include: Tabular display, Graphical display, and Values of efficiency criteria.

In Tabular Display, results were given in 5 categories

i.

Results summary

ii.

Estimated discharge values

iii.

Pulse response

iv.

Discharge component corresponding to each input

v.

Contents of file containing input information

In Graphical display results were given in 4 categories

i.

Observed and estimated discharge

ii.

Scatter diagram of residual and observed discharge

iii.

Scatter diagram of residual and estimated discharge

iv.

Pulse response

59

3.6.1.5 Optimization of the Model Output In order to optimize the model efficiency, Coefficient of Correlation (R2) for Nzoia river basin, the model was run several times, by changing the memory length (i.e., the number of ordinates in the pulse response series) each time and visually analysing the graph of the corresponding pulse response to refine the estimate of the memory length for the next calibration run. For Project purpose SMAR was used. The appropriate value of the memory length was determined by observing the shape of the unit hydrograph for Nzoia river basin.

3.6.1.6 Running the Model in Updating Mode The model was run in updating mode by selecting Calibration and Verification in Updating mode with the following 3 methods for updating:

i.

Autoregressive (AR) Method

ii.

Linear Transfer Function (LTF) method

iii.

Neural Network Method

A description of the Linear Transfer Function (LTF) method application used in the project is given below.

Running the Model in Updating Mode was done in two ways;

i.

Entering modelling parameters manually on the screen using keyboard.

ii.

Entering modelling parameters from file containing input information in the system.

Because LTF model was being run for the first time, option (i) was selected. However, if the model had already being run and had given some specific file name for the general input 60

information on the selected catchment, with the default .inp extension enter the name of the file containing the general input information.

3.6.1.7 Real-time Flow Forecasting After all the information had been filled and the model had been run, Real Time Flow Forecasting for Nzoia Bain was done so as to forecast for flows and discharge which were used in calculating the Water levels for Nzoia river. The process below was used. From the system the option for Real – Time River flow forecasting was used and a list of 57 different combinations for real time flow forecasting was displayed. The two options for all of the 57 combination were that they either lie in the substantive models of the updating model. Before running this model and because some or all of the substantive models (SMAR) and updating models (LTF) had been run and the model estimated discharge output files had been saved in unique file names for each of the model it was easy to select the option.

For Nzoia river basin River Flow modelling option 29 which included SMAR for substantive models and LTF model for updating models was selected. Once the model was ran it gave estimated discharges as three Lead times for the three consecutive days from the day the model was run. For Nzoia river basin due to the unreliability of the data one day forecast was efficient since the error was big for three days.

61

3.7 Calculation of Forecasted Water levels

Estimated water levels were calculated using the daily estimated discharges from the model. The formula for calculation was the inverse of the formulae used for calculation of the discharges to be inputted in the model from the observed daily morning water levels. The equation for flow is

Q W D V

Where: Q = flow rate, W = width, D = depth, V = velocity

There for the formulae is

D Q WV

3.8 Conclusion

The results for forecasts of discharge and water levels for Nzoia River basin in Chapter 4 are the results for SMAR Model for the substantive model and LTF for the updating model both in the Galway Flow Forecasting System.

62

CHAPTER FOUR 4.0 Results and Findings 4.1 Introduction

This chapter presents results of the study and the discussion of findings. Some of the issues discussed here include daily rainfall observations in Nzoia river basin for the 22 rainfall stations, Nzoia basin rainfall gridding, daily discharge calculation for reported levels, graphs showing observed water level at Nzoia River, River depth calculation from forecasted levels and graph showing forecasted river depths. The results given in this chapter are results for SMAR for the substantive model and LTF for the updating model used in the project.

4.2 Daily Reported Rainfall Amounts

Daily reported rainfall amounts from the 22 rainfall stations were recorded in the Nzoia River basin. The reported rainfall amounts were used for rainfall gridding using ArcView 3.2 so as to get the Daily Mean Areal rainfall. Table 4 shows the daily reported rainfall amounts for the 22 rainfall stations.

63

Table 4: Daily reported Rainfall amounts at Nzoia river basin STATION PORT VICTORIA CHEBIEMIT F STATION NABKOI F STATION LUGARI F STATION UHOLO CHIEF'S KADENGE YALA KAIMOSI TEA F MUMIAS SUGAR KAPSARA TEA F KIMILILI AGRIC ADC CHORLIM NZOIA SUGAR KIPKABUS F S KAPENGURIA WRMA NORTH NANDI F S CHEPTONGEI KITALE MET KAKAMEGA AGROMET ELDORET MET ELDORET AIRPORT KISUMU

Lon Lat 34.011 0.146

Day1 13.2

Day2 1.4

Day3 4.2

Day4 4.7

Day5 0

Day6 0

Day7

35.501 35.469 34.900 34.365 34.173 34.930 34.500 35.112 34.710 34.800 34.650 35.546 35.112 35.016 35.515 34.983

0.855 0.136 0.667 0.202 0.024 0.155 0.317 1.250 0.789 1.033 0.567 0.284 1.250 0.383 0.944 1.000

29 9.1 19 12.6 0 10.5 3 19.5

25.5 0 0 2.5 0 7.7 2 2

5.2 8.3 7.3 15 68.3 2.7 25 10

3.2 0 10 25.5 56.7 7.3 31.5 19

17.8 0 9.3 10 0 14.3 7.1 10.5

0 0 0 2.7 0.1 4.6 3 10.8

0 9.9 0 1.5 6.2 2.4 7.7 9

9.5 10.3 13.9 8.5

3.5 10.1 26.5

3 10.4 3.3 10.5

20 10.5 17.5 14.1

3 1 4.3 22.6

3.5 1.4 1.2 0

0 5.1 24.5 0

3.4

3.1

0.6

27.8

2.9

0.2

0

34.750 35.283 35.230 34.580

0.280 0.530 0.400 -0.100

10.6 12.4 6.1 22.8

2.8 4.6 0.7 10.1

5.7 1.8 0.7 0.01

23.2 16.2 56.3 65.1

0.01 6 0.1

0 0

19.6 0 0 0

STATION Lon Lat PORT VICTORIA 34.011 0.146 CHEBIEMIT F STATION 35.501 0.855 NABKOI F STATION 35.469 0.136 LUGARI F STATION 34.900 0.667 UHOLO CHIEF'S 34.365 0.202 KADENGE YALA 34.173 0.024 KAIMOSI TEA F 34.930 0.155 MUMIAS SUGAR 34.500 0.317 KAPSARA TEA F 35.112 1.250

Day8 1.5

Day9 5

Day10 10

Day11 0

Day12 0

Day13 8.3

Day14 0

0 0 0 5.2 12.2 2.4 2.5

3.3 0 0 26 0 3.3 10.3

0 0 0 0 0 6.7 0

6.8 0 0 0 0 0 0

10.4 1.9 14.7 8.8 0 2.2 1.9

3.6 1.8 0 2.9 7.2 7.4 1.2

0 1 0 2.3 14.2 6.8 6.5

Continuation….

64

KIMILILI AGRIC ADC CHORLIM NZOIA SUGAR KIPKABUS F S KAPENGURIA WRMA NORTH NANDI F S CHEPTONGEI KITALE MET KAKAMEGA AGROMET ELDORET MET ELDORET AIRPORT KISUMU

34.710 34.800 34.650 35.546 35.112 35.016 35.515 34.983

0.789 1.033 0.567 0.284 1.250 0.383 0.944 1.000

34.750 35.283 35.230 34.580

0.280 0.530 0.400 -0.100

Lon 34.011

Lat 0.146

Day15 11.5

Day16 0.5

Day17 0

Day18 0

Day19 0.7

Day20 1

35.501 35.469 34.900 34.365 34.173 34.930 34.500 35.112 34.710 34.800 34.650 35.546 35.112 35.016 35.515 34.983

0.855 0.136 0.667 0.202 0.024 0.155 0.317 1.250 0.789 1.033 0.567 0.284 1.250 0.383 0.944 1.000

2.4 0 0 0.5 0 0 0

0 5.2

4.9 0 2 2.8 0 0 1.5 0

8 3.2 13 5 0 0 40.6 12

15.7 0 5.5 11 2.2 3.2 0.4 10

33.5 4.3 0 9 0 40 18.3 0

0 0 0 0

0.4 4 0 9.5

2 10.1 9.9 7.5

16 9.3 0 0

12 12.3 13.4 0

9 12.1 0 10.1

0.01

0.01

0.6

0.7

4.9

12.6

34.750 35.283 35.230

0.280 0.530 0.400

0 0 0

1.7 0.5 15.5

0.5 0 1.9

2.6 9.3 1.5

17.1 1 10.7

5.8 7.2 0.8

0

0

0

11.5

25

2.5

0

0 0

0.5 2.5

11.5 0

14.1 0

1 5.2

4.8

3.2

0.6

0

17.6

0.01 0 0 0

21

3.8 1.4 3.5 0

0 0 0

7.5 6 6.1 0.01

0.4 4.5

0 1.1 0 0 0

0.6 0.2 1.2

2.8 0 0

Continuation….. STATION PORT VICTORIA CHEBIEMIT F STATION NABKOI F STATION LUGARI F STATION UHOLO CHIEF'S KADENGE YALA KAIMOSI TEA F MUMIAS SUGAR KAPSARA TEA F KIMILILI AGRIC ADC CHORLIM NZOIA SUGAR KIPKABUS F S KAPENGURIA WRMA NORTH NANDI F S CHEPTONGEI KITALE MET KAKAMEGA AGROMET ELDORET MET ELDORET AIRPORT

0.5 0 6.3 31.8

65

KISUMU

34.580

-0.100

0

1.7

0

33.5

Lon 34.011

Lat 0.146

Day21 0.1

Day22 43.1

Day23 0

Day24 0

Day25 0

Day26 3.2

35.501 35.469 34.900 34.365 34.173 34.930 34.500 35.112 34.710 34.800 34.650 35.546 35.112 35.016 35.515 34.983

0.855 0.136 0.667 0.202 0.024 0.155 0.317 1.250 0.789 1.033 0.567 0.284 1.250 0.383 0.944 1.000

5.5 0 1.1 1.4 0 1.2 38.4 5 11

10.4 11.2 0 1.5 4.6 48.5 1.8

3 6.9 14.5 0 10.1 0 55.4

23.5 26.6 20.8 0 0 0 2.8

11.1 43.7 34 0 0 3.1 0

0 0 0 0 0 8.5 6.7

2.9

32 32.4 24 18.2 43

0.5 0 11.2 3 3.5

21 0 10 0.2 0

34.750 35.283 35.230 34.580

Continuation…….. STATION PORT VICTORIA CHEBIEMIT F STATION NABKOI F STATION LUGARI F STATION UHOLO CHIEF'S KADENGE YALA KAIMOSI TEA F MUMIAS SUGAR KAPSARA TEA F KIMILILI AGRIC ADC CHORLIM NZOIA SUGAR KIPKABUS F S KAPENGURIA WRMA NORTH NANDI F S CHEPTONGEI KITALE MET KAKAMEGA AGROMET ELDORET MET ELDORET AIRPORT KISUMU

18 2.2 4.3 1.2

0.5 34.5 6.5 2.2

6.5 5.8 2.5 15 0 10.9

0.5

12.2

0.8

35.8

4.2

0.7

0.280 0.530 0.400 -0.100

38.5 10 20.7 0

0.2 12.8 6.1 47.9

0 7.1 37.4 0

17.8 24.2 4.4 4.9

0 14.2 8.8 0

0 0 0 12.6

Lat 0.146

Day27 25.3

Day28 7.3

Day29 0.5

Day30 0

Day31 0

0.855 0.136 0.667 0.202 0.024 0.155

0 0 0 0 2.2 2.5

0 0 0 0 0 2.4

9.5 8.9 0 0 0 12.2

7.7 0 5.1

0 0 0 0 0 0

Continuation……. STATION Lon PORT VICTORIA 34.011 CHEBIEMIT F STATION 35.501 NABKOI F STATION 35.469 LUGARI F STATION 34.900 UHOLO CHIEF'S 34.365 KADENGE YALA 34.173 KAIMOSI TEA F 34.930

66

0 7

MUMIAS SUGAR KAPSARA TEA F KIMILILI AGRIC ADC CHORLIM NZOIA SUGAR KIPKABUS F S KAPENGURIA WRMA NORTH NANDI F S CHEPTONGEI KITALE MET KAKAMEGA AGROMET ELDORET MET ELDORET AIRPORT KISUMU

34.500 35.112 34.710 34.800 34.650 35.546 35.112 35.016 35.515 34.983

0.317 1.250 0.789 1.033 0.567 0.284 1.250 0.383 0.944 1.000

0

21

0.5

0

0.6

0 0 0 0

0 0

0

0 0 0

7.9 9 21.5 4.5 16.5 5

0 3.1 6 13.8 5.6 9.5

0 0 0 0 0 0

0

0.01

15.6

7.4

0

34.750 35.283 35.230 34.580

0.280 0.530 0.400 -0.100

0.01 0 0

0 0 0

8.1 0.01 0 1.7

3.6 1.9 5.8 0

0 0 0 0

4.3 Rainfall Gridding using GIS

The main goal for rainfall gridding was to get the mean areal rainfall amount over the basin to be used in SMAR Model of the GFFS. Daily rainfall gridding was done using ArcView 3.2. First the rainfall data was input in Microsoft Excel. Some of the data was missing but it did not create a big error in the final values. The Excel sheet was saved in Text format to be inputted in ArcView 3.2 as a table. Mean Areal Rainfall amounts for the entire month are shown in the table below.

67

Table 5: Rainfall Grid Values showing Daily Mean Areal rainfall

Date 1-Oct-08 2-Oct-08 3-Oct-08 4-Oct-08 5-Oct-08 6-Oct-08 7-Oct-08 8-Oct-08 9-Oct-08 10-Oct-08 11-Oct-08 12-Oct-08 13-Oct-08 14-Oct-08 15-Oct-08 16-Oct-08 17-Oct-08 18-Oct-08 19-Oct-08 20-Oct-08 21-Oct-08 22-Oct-08 23-Oct-08 24-Oct-08 25-Oct-08 26-Oct-08 27-Oct-08 28-Oct-08 29-Oct-08 30-Oct-08 31-Oct-08

Mean Areal Rainfall (mm) 12.2 8.1 5.88 22.25 6.35 2.2 4.51 1.48 3.23 1.3 2.74 10.73 2 1.08 0.42 5.62 2.61 8.44 7.87 9.57 11.84 8.65 6.96 22.22 8.51 3.09 0.47 1.49 7.22 4.76 0

Highest Rainfall Amount (mm) 29 31.7 68.3 65.1 44.4 18.5 93 12.2 26 19.5 22.7 50.9 19.1 14.2 11.5 70.1 10.1 40.6 4404 40 66.3 48.5 55.4 43 43.7 21 25.3 21 21.5 13.8 46.3

68

Lowest Rainfall Amount (mm) 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

69

70

71

Figure 6: Nzoia River Basin rainfall Grids

72

4.3.1 Discussion Table 5 and Figure 6 above shows the rainfall grid values for October 2008. Much of the basin received light rainfall from day 1 to day 6 except Chebiemit and North Nandi which recorded heavy rainfall. The rest of the basin recorded moderate to average rainfall amounts during the period. The highest areal rainfall for the period is 22.25 mm in day 4. Below 5mm rainfall amount were recorded in day 6.

From day 8 to 19, most of the basin received moderate to no rainfall. The lower parts of the basin received moderate rainfall while the middle parts of the basin received light rainfall amounts. The highest mean areal rainfall recorded was 10.73mm. The station which recorded the highest amounts includes; Mumias Sugar 31.8mm in day 16 and 40mm in day 18. Nzoia Sugar recorded 25mm in day 12. The lowest Mean areal rainfall was 0.42mm in day 15.

From day 20 to 31, Vey high rainfall amounts were recorded in day 22 and 23. Mumias Sugar recorded 55.4mm in day 23 and Chebiemit recorded 33.5mm in day 20. High rainfall amounts were recorded at the start of the period but reduced as the month ended. The highest Mean rainfall amounts were in day 24 and 21, 22.22mm and 11.84mm respectively. Kakamega, Butere, Siaya and Keiyo districts received very heavy rainfall amount between day 20 and 22.

4.4 Calculation of Daily Discharge from reported Water levels

The table below shows daily discharge calculation from reported water levels. The formula used is discussed in the previous chapter. Findings were presented as graphs and tables.

73

Table 6: Daily Discharge values for Nzoia River as at Rwambwa gauging station. RWAMBWA BRIDGE RGS

Day 1-Oct-08 2-Oct-08 3-Oct-08 4-Oct-08 5-Oct-08 6-Oct-08 7-Oct-08 8-Oct-08 9-Oct-08 10-Oct-08 11-Oct-08 12-Oct-08 13-Oct-08 14-Oct-08 15-Oct-08 16-Oct-08 17-Oct-08 18-Oct-08 19-Oct-08 20-Oct-08 21-Oct-08 22-Oct-08 23-Oct-08 24-Oct-08 25-Oct-08 26-Oct-08 27-Oct-08 28-Oct-08 29-Oct-08 30-Oct-08 31-Oct-08

AREAL RAIN RWAMBWA (mm) 12.2 8.1 5.88 22.25 6.35 2.2 4.51 1.48 3.23 1.3 2.74 10.73 2 1.08 0.42 5.62 2.61 8.44 7.87 9.57 11.84 8.65 6.96 22.22 8.51 3.09 0.47 1.49 7.22 4.76 0

Water LevelS (m) MEAN 9AM 4PM (M) 1.86 1.80 1.83 2.1 2.25 2.18 2.2 2.2 2.20 2.34 2.3 2.32 2.48 2.6 2.54 3.46 3.62 3.54 3.30 3.1 3.20 2.82 2.80 2.81 2.75 2.70 2.73 2.65 2.60 2.63 2.55 2.50 2.53 2.45 2.45 2.45 2.3 2.45 2.38 2.3 2.25 2.28 2.1 2 2.05 2.2 2.1 2.15 2.1 2.15 2.13 2.3 2.35 2.33 2.3 2.25 2.28 2.7 2.75 2.73 2.86 2.95 2.91 3.4 3.25 3.33 2.9 2.95 2.93 3 3.1 3.05 3.15 3.25 3.20 4 4.35 4.175 3.74 3.7 3.72 3.5 3.4 3.45 3.2 3.1 3.15 2.95 2.9 2.925 2.86 2.8 2.83

ALERT (M) 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8 2.8

74

WARNING (M) 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5

BANKFULL (M) 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5

DYKE LEVEL (M) 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8 5.8

DISCHARGE (M3/sec) 131.87 163.59 165.95 177.38 198.78 302.52 266.14 225.79 217.20 207.20 197.31 189.96 182.68 173.08 151.92 161.24 158.90 177.86 173.08 217.20 235.47 279.39 237.53 250.44 266.14 373.20 322.20 292.78 260.88 237.53 227.82

Figure 7: Graph showing daily Discharge values against Rainfall amounts for Nzoia river basin.

4.4.1 Discussion At the start of the month the basin didn‟t record very high amounts of rainfall. From day 1 to day 6 there was increasing discharge in the river due to moderately high amounts of rainfall recorded in the following stations; Uholo Chiefs Camp, Chebiemit, Kipkabus, Kapenguria WRMA, Kadenge yala, Kakamega Agromet, Eldoret Met and Kisumu, rainfall amounts went as high as 34.6 mm.The highest water level was 3.2m which had surpassed the alert level and flood level by 0.2m, no flooding was recorded in the basin. From day 8 to 19 there was average rainfall recorded over the basin, the lowest discharge at Nzoia river was recorded this period, water level went as below as 2.05m the highest in this period being 2.81 a distance from the flood level. 75

Stations which recorded no rainfall during this period were Port Victoria, Chebiemit, Nabkoi, and Lugari. The rest of the basin recorded moderate to no rainfall.

From day 20 to day 31 very high rainfall amounts were recorded in Chebiemit Filling station, as high as 33.5mm in day 20 and 15.7mm in day 19. Kaimosi recorded the highest amounts of rainfall in the period having 40mm and 48.5mm in day 22. Mumias sugar recorded very high amounts, 55.4mm in day 23 and 38.4mm in day 21. Kipkabus Filling station recorded 34.5mm in day 22 and 24mm in day in day 24. Kitale on the upper parts of the basin recorded 35.8mm and 15.6mm in the period. This conditions led to the highest water levels recorded during the period, 4.18m in day 25, this surpassed the flood level by 0.68m other high levels were recorded in day 26 and 27, 3.72 and 3.33m respectively. The water levels from day 27 went down due to the moderate to no rainfall over the basin.

4.5 Calculation of Daily Water Levels from Forecasted Discharge values

The table below shows Water level calculation from forecasted Discharge values. The formula used is discussed in the previous chapter. Findings were presented as graphs and tables.

76

Table 7: Daily Forecasted Water level values for Nzoia River.

Date 1-Oct-08 2-Oct-08 3-Oct-08 4-Oct-08 5-Oct-08 6-Oct-08 7-Oct-08 8-Oct-08 9-Oct-08 10-Oct-08 11-Oct-08 12-Oct-08 13-Oct-08 14-Oct-08 15-Oct-08 16-Oct-08 17-Oct-08 18-Oct-08 19-Oct-08 20-Oct-08 21-Oct-08 22-Oct-08 23-Oct-08 24-Oct-08 25-Oct-08 26-Oct-08 27-Oct-08 28-Oct-08 29-Oct-08 30-Oct-08 31-Oct-08

F. Discharge 3 (m /sec) 129.19 170.69 165.95 175.47 204.71 311.23 255.65 224.77 214.69 204.71 194.85 189.96 189.96 170.69 147.3 156.56 161.24 180.27 170.69 219.72 240.1 271.42 240.1 255.65 271.42 393.26 320 287.41 255.65 234.96 224.77

lnq 4.861284 5.139849 5.111687 5.167468 5.321594 5.740532 5.543809 5.415078 5.369195 5.321594 5.27223 5.246814 5.246814 5.139849 4.992471 5.053439 5.082894 5.194456 5.139849 5.392354 5.481056 5.603667 5.481056 5.543809 5.603667 5.974471 5.768321 5.66091 5.543809 5.459415 5.415078

ln (56.756878) 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847 4.038776847

77

A=x/1.3 0.632698 0.846979 0.825315 0.868224 0.986783 1.309043 1.157717 1.058693 1.023399 0.986783 0.94881 0.929259 0.929259 0.846979 0.733611 0.78051 0.803167 0.888984 0.846979 1.041213 1.109445 1.203762 1.109445 1.157717 1.203762 1.488995 1.330419 1.247795 1.157717 1.092799 1.058693

B=exp(A) 1.882683 2.332589 2.2826 2.382676 2.68259 3.702627 3.18266 2.882601 2.782636 2.68259 2.582635 2.532632 2.532632 2.332589 2.082588 2.182584 2.2326 2.432656 2.332589 2.832652 3.032675 3.332631 3.032675 3.18266 3.332631 4.432641 3.782626 3.482654 3.18266 2.98261 2.882601

Forecasted Levels (m) 1.88 2.33 2.28 2.38 2.68 3.70 3.18 2.88 2.78 2.68 2.58 2.53 2.53 2.33 2.08 2.18 2.23 2.43 2.33 2.83 3.03 3.33 3.03 3.18 3.33 4.43 3.78 3.48 3.18 2.98 2.88

Observed Levels (m) 1.83 2.18 2.2 2.32 2.54 3.54 3.2 2.81 2.73 2.63 2.53 2.45 2.38 2.28 2.05 2.15 2.13 2.33 2.28 2.73 2.91 3.33 2.93 3.05 3.2 4.18 3.72 3.45 3.15 2.93 2.83

Error -0.05 -0.15 -0.08 -0.06 -0.14 -0.16 0.02 -0.07 -0.05 -0.05 -0.05 -0.08 -0.15 -0.05 -0.03 -0.03 -0.10 -0.10 -0.05 -0.10 -0.12 0.00 -0.10 -0.13 -0.13 -0.25 -0.06 -0.03 -0.03 -0.05 -0.05

Figure 8: Graph showing daily Forecasted water level values against Rainfall amounts for Nzoia river basin 4.5.1 Discussion The Nzoia River Basin received above average rainfall during October 2008. Moderate rainfall occurred throughout the month with a few days recording heavy rainfall. In response the Nzoia river level at Rwambwa Bridge reached 4.2m which was above the bankfull level of 3.5m resulting to flooding. Forecasts from the model were close to the observed values with at certain intervals giving an exact forecast like in the 22nd October, 2008. The smallest error from the model was -0.02m on the 7th October, while the biggest being -0.25m in 26th October. The forecast for 26th October where there was flooding was close to the observed levels reported of 78

4.18m, the forecast being 4.43m. A comparison of the Observed and Forecasted level graphs show a similarity in the trend of the lines The peaks of the periods between day 1 and day 6 registering an observed level of 3.54m compared to the 3.7m forecasted and period between day 7 and day 20 having a level of 2.05m compared to 2.06m forecasted and finally day 21 and day 31 registering a level of 4.18m surpassing the flood level of 3.5m was also shown in the forecasted graph recording a level of 4.43m.

79

CHAPTER FIVE 5.0 Summary, Conclusion and Recommendations 5.1 Summary Understanding the dynamics of rainfall-runoff process constitutes one of the most important problems in hydrology, in order to predict or forecast stream flow for purposes such as water supply, power generation, flood control, water quality, irrigation, drainage, recreation, and fish and wildlife propagation. During the past decades, a wide variety of approaches, such as conceptual models, has been developed to model rainfall-runoff process. However, an important limitation of such approaches is that treatment of the rainfall-runoff process as a realization of stochastic and statistical process means that only some statistical features of the parameters are involved. Therefore, what is required is an approach that seeks to understand the complete dynamics of the hydrologic process, capturing not only the overall appearance but also the intricate details. The study area was Nzoia river basin. The main objective of this research was to develop a Decision Support System for Flood Hazard Preparedness and response using the new innovations in technology. Historical rainfall, river discharge and evaporation were modeled in Galway Flow Forecasting System to simulate rainfall run-off processes in Nzoia river basin. Daily rainfall, discharge and evaporation were incorporated in the system for real-time river flow forecasting. The results showed a very small error in the observed and forecasted discharge and water level values. The results were given in graphs and tables showing the observed and forecasted discharge and water level values.

80

5.2 Conclusion The conclusions for the study were enumerated in point form and are as follows: 1. Floods cause death and damage of property and thus should be managed by both physical and non-physical methods. 2. To deal with floods, Flood control and Early warning systems have to be set up both and national and local levels. 3. There is lack of hydrometeorological data in the country since values used for modelling were read a day before hence are outdated. 5.3 Recommendations The main recommendation for the study carried out at Nzoia river basin is that there is need to set up a Flood Early warning system for floods in the country and particularly Nzoia river basin since floods occurring at Budalang‟i are perennial and cause a lot of damage to the community. Recommendations were divided into national and local level. 5.3.1 Recommendations at National Level 1. There is need for the government to set up a Ministry of Disaster Preparedness and Management which will implement policies on disasters. 2. Need for the government to formulate policies on protecting watersheds and catchment areas to minimize causes of flooding. 3. Need for the government to formulate policies on sustainable land use management to reduce causes of flooding. 81

4. There is need for the government to formulate policies on the use of GIS in all government ministries on issues of the environment for a sustainable environmental management. 5. Upgrading of rainfall/meteorological and river gauging stations to provide real time data transmission mode. 6. Setting up an integrated Hydrometeorological data collection system at the Flood Forecasting Centers, uplinking with other sources of data such as satellite and radar. 7. Calibration and validation of hydrological models for Nzoia basin. 8. Preparation of flood hazard maps for the Nzoia river basin. 5.3.2 Recommendations at Local Level 1. Setting up of District based Flood Management Committees [FMCs] 2. New approach to flood management and control is required. The starting point would be an open dialogue between all stakeholders, focusing on the concept that flooding can be seen as an important resource that needs improved management. Under this new approach, the floods will be utilized for fish farming, watering of livestock, spreading and settling the silt load, and farming for wealth creation. 3. Any development has to be in close dialogue with the communities, in the light of traditional coping strategies, and taking account of population pressures. It is conceivable that communities could revert to their traditional practices of moving between high grounds and flood plains – depending on the extent and magnitude of the floods.

82

4. Develop and institutionalize a proactive mechanism for a community based flood early warning system 5. Flood rescue and relief: Establish well-laid out procedures at the local. These involve evacuation, rescue, providing food, drinking water, health care, temporary shelter, clothing, financial assistance for repair/rebuilding of damaged houses, relief works

83

REFERENCES Government of Kenya, 2002: “The Water Act”, Ministry of Water Resources Management and Development

WMO, 2004: Strategy for Flood Management for Lake Victoria Basin, Kenya Government of Kenya and World Bank, 2007: “Western Kenya Community Driven Development and Flood Mitigation Project”, Flood control for Nzoia and Yala Rivers Report. Government of Kenya, 2002: “National Water Master Plan-1992” prepared by Japan International Co-operation Agency

Clarke R. T. (1973): Mathematical models in hydrology. Flood & Agricultural Organization (FAO), Irrigation & Drainage paper no. 19, Rome 282pp. College Galway, Ireland, from April 24 to June 30, 1995.

Nash, J. E. and Barsi, B. I. (1983): A hybrid model for flow forecasting on large catchments. Journal of hydrology, 65, pp. 125 - 137. Ininda, J.M., 1995: Numerical simulation of the influence of sea surface temperature anomalies on the East African seasonal rainfall. PhD. thesis, Department of Meteorology, University of Nairobi, Kenya

Mukabana, J. R., 1992: Numerical simulation of the influence of the large-scale monsoon flow on the weather patterns over Kenya using a three-dimensional limited area model. PhD. thesis, Department of Meteorology, University of Nairobi, Kenya

84

Mutemi, J. N., 2003: Climate anomalies over Eastern Africa associated with various ENSO evolution phases. PhD. thesis, Department of Meteorology, University of Nairobi, Kenya

Ogallo, L. J., 1980: Regional classification of the East African Rainfall stations into homogeneous groups using the method of Principal Component Analysis. Stat. Climatol. Devel. Atmos. Sci., 13, 255 – 266

Okoola, R. E., 1999: Synoptic systems affecting Eastern Africa. Lecture notes of the first Drought Monitoring Centre (DMC), Nairobi climate prediction capacity building training workshop for the Greater Horn of Africa, Nairobi, Kenya. Drought Monitoring Centre – Nairobi. 51 – 62

Owiti, Z. O. 2005: Use of the Indian Ocean dipole indices as a predictor of East African rainfall anomalies. MSc. thesis, Department of Meteorology, University of Nairobi, Kenya

Abbort, M. B., Bathurst, J. c., Cunge, J.A., O'connell, P. E., and Rasmussen, J. (1986): An introduction to the European Hydrological System - Systeme Hydrogique Europeen, 'SHE' 1: History and philosophy of a physically - based, distributed modelling system. Journal of hydrology, 87 (1/2), pp. 45 - 59.

Amerman, C. M., (1965): The use of unit-source watershed data for runoff prediction. Water Resources Research, vol., NO.4, pp. 499 - 507.

Amorocho and Orlob, G. T. (1986): Nonlinear analysis of hydrologic systems, Water Resources Center, University of California, contribution No. 40, Berkeley.

85

Beven, K. J. and 0' Connell P. E. (1982): On the role of physically - based distributed modelling in hydrology. Report No. 81, Institute of Hydrology (IHE).

Beven, K. J. (1989): Changing ideas of hydrology. The case of physically - based models, Journal of hydrology, 105, pp. 157 -172.

Burnash R. J. C and Singh V. P. (1995): The NWS River forecast systemmcatchment modelling, pp 311 - 366, water resources publications, Colarado.

Chow, V.T., Maidment, D.R., and Mays, L.W. (1988): Applied Hydrology, McGraw Hill, NY, pp 465-470

Castruccio, P. A. Loats, Journal of hydrology L. Lioyd, D., and Newsman, P. B. (1980): Cost/benefit analysis for the operational applications of satellite snow observations (OASSO). Proc. Final workshop on operational applications of satellite snow cover observations, NASA CP - 2116, pp: 239 - 254.

Clark, C. O. (1945): Storage and the unit hydro graph transformation. American Society of Civil Engineering, 110, pp. 1419 -1488.

David R. Maidment (1999): Flooding forecasting for the Buffalo Bayou Using CRWR - PrePro and HEC-HMS (http://www.hec.usace.mil/software/.)

De Coursey, D. G. (1991): Integrated quality and quantity modelling - Receiving waters. In: Bowels, D. S. and O'Connell, P. E. Recent Advances in the modelling of hydrologic systems. Kluwer academic publishers, Dordrecht, The Netherlands, pp.323 - 353. 86

Dolezal, F. (1997): Workshop Lecture Notes on Evaporation and Infiltration, University College Galway.

Dooge, J. C. (1959). A general theory of the unit hydrograph. J. Geophys. Res., 64 (2),241 - 256.

Dooge, J. C. (1973). Linear theory of hydrologic systems, Tech. Bull, No. 1468, Agri.,Res. Serv., pp. 117 - 124, October, U.S. Department of Agriculture, Washington, D. C.

Dooge, J. C. I. (1986). Looking for hydrologic laws. Water Resources Research, 22, pp. 46 - 58.

Eagleson, P. E., Mejiar, R. and March, F. (1965): The computation of optimum realizable unit hydro graph from rainfall and runoff data. Hydrodynamic. Lab. Rep. No. 84, MIT.

87