Content uploaded by Hesbon Otieno
Author content
All content in this area was uploaded by Hesbon Otieno
Content may be subject to copyright.
International Journal of Water Resources and Environmental Engineering Vol. 4(12), pp. 397-403, December 2012
Available online at http://www.academicjournals.org/IJWREE
DOI: 10.5897/IJWREE12.086
ISSN 1991-637X ©2012 Academic Journals
Full Length Research Paper
Coupling agricultural non-point source (AgNPS) model
and geographic information system (GIS) tools to
predict peak runoff and sediment generation in the
upper River Njoro catchment in Kenya
Otieno, Hesbon1* and Onyando Japheth Ogalo2
1Department of Hydrology and Water Resource Management, South Eastern University College, Kitui, Kenya.
2Department of Agricultural Engineering, Egerton University, P. O. Box 536-20115, Egerton, Kenya.
Accepted 14 February, 2012
Human interference in the upper River Njoro catchment has led to the increased exposure of the land to
accelerated erosion. An application that combined the capabilities of remote sensing, geographic
information system (GIS) and agricultural non-point source (AgNPS) model was used to estimate peak
runoff rate and sediment yield from the upper River Njoro catchment. Remotely sensed Landsat
Thematic Mapper (TM) images were used to obtain land cover and associated AgNPS model input
parameters. Other input parameters for the model were extracted from GIS layers using the agricultural
non-point source-integrated land and water information system (AgNPS-ILWIS) interface. Surface water
quantity and quality data including peak runoff and sediment yield of selected storm events were
obtained from two gauging stations, within the catchment. Base flow separation was done so that
measured direct peak runoff rate and sediment yield generated by direct runoff could be determined
and compared directly with the model simulated results. Simulated peak runoff rates in Upstream
(Treetop) station were satisfactory with an EFF of 0.78 and a percent error of 4.1%. The sediment yield
was also reasonably estimated with an EFF of 0.88 and a 2% error. The downstream (Egerton) station
results were also satisfactorily predicted with peak runoff rate having an EFF of 0.69 and a 5.5% error of
estimates, while the estimated sediment yield had an EFF of 0.86 and a 2.5% error.
Key words: Peak runoff, sediment, agricultural non-point source (AgNPS), geographic information
system (GIS), integrated land and water information system (ILWIS), soil loss.
INTRODUCTION
Soil erosion and sedimentation are major environmental
problems which cause degradation of natural resources
in river basins. Such degradation may be in the form of
reduction of land productivity due to loss of fertile soil
from agriculturally productive land and undesirable
deposition of eroded material in the lower reaches of the
river channels increasing frequency of floods and
depletion of ground water resources. Sediment deposits
also cause accumulation of silts in lakes or reservoirs
*Corresponding author. E-mail: otienohz@gmail.com. Tel:
+254712792101.
reducing their useful life. Fertilizer chemicals are transported
into water bodies together with sediments causing
excessive growth of water plants, which result in clogging
of water courses, loss of aquatic life and other related
problems. Scouring of river channels has also been noted
to destroy hydraulic structures along the river courses.
The destruction of soil through erosion is becoming of
particular concern because soil formation is an extremely
slow process. Serious soil erosion is occurring in Kenya’s
major agricultural regions and the problem is growing as
more land is brought under agricultural production.
Surface runoff is a critical variable in determining the rate
of soil erosion and sediment transport. Its turbulence is
known to be an influential factor in detachment of soil by
398 Int. J. Water Res. Environ. Eng.
overland flow. The rapid growth in the world population,
which leads to the need for more crops, will only intensify
the water problem, particularly if soil erosion is not
contained.
The River Njoro catchment is part of the larger Lake
Nakuru catchment, and one of the rivers originating from
the Eastern Mau forest of the Mau Complex and draining
into the saline Lake Nakuru. The River Njoro catchment
is a high potential area and is under intensive cultivation.
The forested hill slopes of the catchment have undergone
extensive deforestation, which has led to increased soil
erosion, low recharge and remarkable fluctuation in
stream flows. Through erosion, the fertile topsoil and the
sediment generated are transported by the stream and
get deposited in the lower reaches in the river and the
Lake. Lake Nakuru is a protected area for biodiversity
conservation. As a habitat for various flora and fauna, its
degeneration in quantity and quality has adverse effects
on biodiversity which it supports.
Recent advancements in computer technology have
provided new techniques to study and tackle
environmental problems for effective and efficient
environmental systems management. It is now easier to
analyze and process enormous amount of data within a
very short time. The computer has become a versatile
tool for studying and modeling our environment. In
addition, since the early 1970’s, satellites have scanned
the earth and furnished digital images of several
wavelengths ranging from the visible part of the spectrum
to the middle infra red (De Jong and Riezebos, 1992).
Developments in the field of erosion studies have led to
the design of new models that can handle large number
of parameters and perform large number of calculations.
These models are often linked to geographical
information system (GIS) thus simplifying the modeling
task.
In the present study, the integrated land and water
information system (ILWIS) developed by International
Institute for Aerospace Survey and Earth Sciences ITC
(Meijerink et al., 1988) was used in a GIS-Model link to
determine and handle the distributed input and output of
the agricultural non-point source (AgNPS) pollution
model. The distributed model was used to predict
sediment yield and runoff for single storm events. A large
depth of literature on AgNPS model exists as expounded
on by Finn et al. (2006). GIS capabilities in incorporating
catchment data through remote sensing and spatial
analysis with hydrological data have been shown to be
effective in determining peak flows at various points
along a stream channel (Onyando et al., 2005).
Study area
The River Njoro catchment is located approximately 150
km North West of Nairobi, Kenya. The catchment is part
of the larger Lake Nakuru catchment. The area of the
catchment is approximately 250 km2 and varies in altitude
from 2700 m above mean sea level (a. m. s. l.) on the
eastern side of the Mau complex, one of Kenya’s major
water towers to 1700 m a. m. s. l at the outlet in Lake
Nakuru. The mean annual precipitation is 1200 mm
distributed bimodally with peaks in May and October. The
study focused on the upper catchment which is
approximately 127 Km2; it is shown in Figure 1, which
also includes drainage network and the sampling sites at
Treetop and Egerton River gauging stations.
The Egerton gauging station is located at (00.37347°S,
35.94077°E) with an altitude of 2203 m a. m. s. l while
treetop gauging station located at 3.7 km upstream of
Egerton station has coordinates of (00.37528°S,
35.92029°E) and altitude of 2285 m a.m.s.l. The geology
and soils of the area is influenced by the volcanic nature
of the Rift valley. The upper part of the catchment is
predominantly loamy soils that developed from ashes and
other pyroclastic rocks of recent volcanoes (Ralph and
Helmidt, 1984), whereas the lower catchment is covered
by erosive lucustrine soils.
MATERIALS AND METHODS
Rainfall, stream flow and water quality
The rainfall data used in the study were taken fr om the readings of
non-recording rain gauges of the Egerton University weather station
(00°23’S, 35°55’E) at 2238 m a. m. s. l within the sub-catchment.
Eleven storm events were considered in the study. The
corresponding stream flow was also taken at the two monitoring
sites selected for the study. Stream flow measurements were made
by recording the height of the surface of water read from a staff
gauge after the storms. The elevation readings were then converted
to discharge using a rating equation.
Runoff sampling was also done at the two monitoring sites after
storm events for sediment concentration analysis in the laboratory.
A depth integrating hand sampler was lowered into the stream at a
constant vertical speed, and also raised to the surface at a constant
speed. Three traverses were made across the stream section to
come up with the suspended sediment load for the section. The
samples were then filtered, oven dried and weighed. Using the
direct runoff volume, the weight of the s ediment was converted into
a sediment yield in tonnes for a particular storm.
Agricultural non-point source (AgNPS) model simulations
The process of predicting the required outputs via the model-GIS
link was divided into three phases.
Spatial database preparation
The AgNPS model requires a large volume of data from various
sources. The basic input data into the GIS were contours, drainage
network, boundary and land cover maps. The contour, drainage
and boundary maps were created by vectorization fr om topographic
maps of the study area, whereas the land cover map was
processed from a Landsat TM image of 2001 by multi spectral
image processing capabilities of the ILWIS software. It was
assumed that there were no major changes in land use that
occurred between the year 2001 and 2005 when the data collection
was done. Three spectral bands, bands 3, 5 and 7 were used in the
Otieno and Ogalo 399
Figure 1. Map showing Kenya, River Nj oro catchment in Nakuru District and Upper catchment with drainage network and gauging stations
at Treetop and Egerton,
supervised classification using Gaussian maximum likelihood
classifier method to extract thematic information from the s atellite
imagery. The other parameters r equired by the model were either
derived from the basic input maps or input as constants.
Derivation of spatial layers
This phase utilized the capabilities of GIS to come up with other
spatial layers r elated to the basic input layers. The contour,
drainage and boundary maps were rasterized (converted to raster
maps) and used to generate the digital elevation model (DEM)
which was later used to generate dependent raster layers of cell
number, slope length, slope shape, flow direction, channel indicator
and channel gradient. The land cover map was used to derive
USLEs’ K (soil erodibility), C (Cropping), and P (Conservation
Practice) factors, SCS-CN (Soil Conservation Service-Curve
Number), SCC (Surface Condition Constants), COD (Chemical
Oxygen Demand) and Manning’s coefficient maps as shown in
Figure 2. The values of these parameters for the study area were
determined and are presented in the Table 1.
The C (crop management) factor values, which compares the
ratio of soil loss under a given crop to that of bare soil for the study
area were found to range from 0.038 to 0.35 from literature (Crops
C-factor manual). The forest cover had a lower value of C indicating
good protection to the soil as c ompared to the shrub land,
settlement and agriculture. A high C-factor has the effect of
reducing infiltration during a storm, resulting to an increase in
surface runoff and sediment concentration in the runoff.
Another parameter for the model is the chemical oxygen demand
(COD) which is a measure of oxygen required to oxidize organic
and oxidizable inorganic compounds in water, is an indicator of the
degree of pollution and varies with land cover. AgNPS assumes
soluble COD. COD estimates in runoff is based on average
concentration of COD in runoff, and allowed to decay with time
once they enter a channel according to an exponential decay. The
values of this parameter for the various land cover ranged from 20
to 170. The COD factor was found to increase in the order of
forests, settlement, shrubs and agriculture, implying that agricultural
land required more oxygen to oxidize organic and oxidizable
inorganic compounds in runoff.
Soil erodibility factor (K) is a soil dependent parameter. It is a
function of the percentage of silt and c oarse sand, soil structure,
permeability of soil and the percentage organic matter. Based on
soil Nomographs by Morgan (1986), the K-factors for the various
land covers were found to be in the r ange 0.035 to 0.29. Another
parameter, the soils manning coefficients refers to the soils
roughness and thus implies the resistance of flow of water in and
on the s oils. Therefore it is one of the important parameters for
describing water flow over the ground. Lower values of this
coefficient denote less resistance to flow and vice versa. Fost er et
al. (1981) estimated the average Manning’s coefficient values for
different land uses, and based on their tabulated values, the spatial
distribution of the coefficient for the various land covers in the st udy
area was derived.
The c urve number (CN) is a dimensionless index that describes
runoff as a range between 1 and 100, with 100 indicating maximum
runoff. It is dependent on the hydrological soil cover complex of the
catchment. This cover complex comprises of a combination of the
hydrologic soil group and land use and treatment class. Curve
Number values were assigned to each complex to indicate their
specific runoff potential. The curve number values for the
catchment ranged from 25 to 90 based on literature by Chow et al.
(1988). The greater the curve number, the greater the surface
400 Int. J. Water Res. Environ. Eng.
Figure 2. Spatial distribution of land use related parameters.
runoff volume.
Geographic information system (GIS) and model interface
This phase involved the coupling of the interface GIS with the
model. An extraction program was used to extract data from the
GIS environment and tr ansform and take it to the model. An
interface program (GRIPs) developed by ITC-W ater Resources and
Environmental Studies (WRES) was used to convert ILWIS map
files, containing the parameter data, to an input data file acceptable
by the model. This program reads the map values for the respective
cells and converts them to AgNPS data format.
The input data was then entered in the interface and check ed for
flow routing, by eliminating sink holes and ensuring that the
catchment drains through the outlet. The GRIPs then run the model
and creates ILWIS files which contain the model outputs to be
compared with the observed outputs.
RESULTS
Peak runoff rates
The simulated peak runoff rates generated by the model
are presented in Table 2 together with the observed ones.
Otieno and Ogalo 401
Table 1. Land use and its related variables.
Parameter SCC SCS COD-factor K-factor C-factor Manning coefficient
Forest 0.29 25 20 0.035 0.038 0.04
Agriculture 0.29 78 170 0.290 0.350 0.04
Shrubs 0.22 58 80 0.090 0.087 0.04
Settlement 0.14 90 37 0.150 0.320 0.15
Table 2. Peak runoff rate (m3/s) results.
Event Rainfall
(mm)
Treetop Egerton
Observed Simulated Error (%) Observed Simulated Error (%)
15/01/04 40.6 2.966 3.065 -3.3 2.547 2.622 -2.9
11/04/04 42.5 2.981 3.088 -3.6 2.534 2.603 -2.7
23/05/04 45.0 3.055 3.179 -4.1 2.714 2.876 -6.0
22/07/04 37.0 2.267 2.406 -6.1 2.018 2.160 -7.0
11/08/04 28.0 2.843 2.692 5.3 2.543 2.321 8.7
14/11/04 8.4 2.522 2.406 4.6 2.192 2.019 7.9
23/11/04 9.5 2.747 2.617 4.7 2.349 2.191 6.7
25/11/04 17.5 2.521 2.670 -5.9 2.253 2.369 -5.2
16/12/04 25.6 2.910 2.882 1.0 2.887 2.790 3.4
26/01/05 16.5 2.468 2.521 -2.2 2.239 2.352 -5.1
22/03/05 26.5 2.863 2.747 4.1 2.469 2.358 4.5
Sediment yield
The simulations of sediment yield by the same model are
presented in Table 4. In the same Table, observed
sediment yield is presented for comparison of the level of
the accuracy of the simulations.
DISCUSSION
Peak runoff rates
The results presented in Table 2 showed the model to
have satisfactorily predicted the peak runoff rates in both
the stations. On an individual basis all the storm events
were predicted with a percent error ranging between 1
and 8.7. Events 22/07/04 and 11/08/04 were the most
poorly predicted in the two stations with percent error
values of -7.04 and 8.7 (Treetop) and -6.1 and 5.3
(Egerton) respectively. However, the gross errors in peak
flow rates estimation represented an overestimate of only
0.139 (Treetop) and 0.142 m3/s (Egerton) for event
22/07/04 and an underestimate of only 0.15 (Treetop)
and 0.222 m3/s (Egerton) for event 11/08/04, this could
be due to fact that these two were complex storms with
rainfall following an irregular pattern. The event of
16/12/04 was the best predicted. An error of 3.4%
(Treetop) and 1% (Egerton) was produced for this
simulation. The results were subjected to statistical
analysis by comparing the observed and predicted output
data for individual storm events at the two monitoring
stations.
The goodness of fit between the predicted and
observed peak runoff rates was assessed for the
stations. For the upstream station the average
percentage error between observed and predicted values
was 4.1%. This agreement is confirmed by satisfactory
EFF values of 0.78. For Egerton station, there is
generally a good correlation between the observed and
predicted peak runoff rates. The average percent error
between observed and predicted values was 5.5%, and
EFF value of 0.69 as shown in Table 3.
The proximity of the EFF to 1 and scatter smaller
percentage errors is an indication of good predictive
ability of the model as far as peak runoff rate is
concerned. The statistics presented above were found to
be consistent with those for similar work involving AgNPS
done in other parts of the world. Khoelliker and Humbert
(1989) in their work in northeast Kansas found a percent
error of 3%, Young et al. (1987) in an agricultural
catchment in Minnesota had a 1.6% error in their
estimations and a correlation coefficient of 0.81. Lee and
White (1992) also found a good agreement between
observed and simulated data for runoff. Suttles et al.
(1999) simulation of runoff rate in Georgia coastal plain
had an EFF of 0.85, whereas Mostaghini et al. (1997)
402 Int. J. Water Res. Environ. Eng.
Table 3. Statistical parameters of observed and predicted Peak runoff rate.
Catchment Area (km2) No. of events Peak runoff rate
Efficiency Error (%)
Egerton 127 11 0.69 5.5
Treetop 110 11 0.78 4.1
Table 4. Sediment Yield (Tonnes) results.
Event Rainfall
(mm)
Treetop Egerton
Observed Simulated Error (%) Observed Simulated Error (%)
15/01/04 40.6 326.12 331.51 -1.7 276.21 282.21 -2.2
11/04/04 42.5 335.45 327.62 2.3 288.43 283.33 1.8
23/05/04 45.0 340.66 336.62 1.2 279.57 286.77 -2.6
22/07/04 37.0 290.22 295.14 -1.7 239.38 248.73 -3.9
11/08/04 28.0 331.98 325.44 2.0 264.14 270.06 -2.2
14/11/04 8.4 288.14 295.55 -2.6 239.37 246.72 -3.1
23/11/04 9.5 290.64 296.89 -2.2 241.36 247.14 -2.4
25/11/04 17.5 295.84 306.26 -3.5 266.35 258.34 3.0
16/12/04 25.6 318.88 324.83 -1.9 281.30 275.71 2.0
26/01/05 16.5 307.23 303.34 1.3 248.42 243.51 2.0
22/03/05 26.5 312.00 306.73 1.7 266.11 259.63 2.4
simulation of runoff was nearly 100% of the observed. In
Hesse, central Germany 3.8% was the error found for
runoff simulations.
Sediment yield
Only a fraction of the sediment eroded within a catchment
finds its way to the outlet as sediment yield. Large storms
were generally seen to result in high sediment yield as
shown in Table 4. As is the case of runoff, the sediment
yield was also satisfactorily predicted. The individual
percent error values for the various rainfall-runoff events
ranged between 1.2 and 3.9. Sediment results presented
in Table 4 indicate that Event 25/11/04 was the worst
predicted for the two stations, with an error of -3.5% for
the Treetop station and 3.0% for Egerton station. The
best predicted event was that of 26/01/05, which had an
error of 1.3 and 2.0% for Treetop and Egerton stations
respectively. The difference in errors between the best
and worst sediment yield predictions was found to be
small, in addition the errors for sediment yield were lower
than those for the runoff results. This could be due to the
fact that, all particles were allowed to participate in the
channel scouring, and not the AgNPS default that allows
only sand particles to erode. The study area also falls in
the tropics where storms are intense. The choice of storm
type influences the energy-intensity value, and hence the
erosive potential of a storm. Four AgNPS storm types
exist according to AgNPS model classification of storms,
these are 1a, 1, 2 and 3 in increasing amount of the
energy intensity-values. The lower percent errors in the
sediment yield results were realized when storm type 2 is
used in the simulations. This type of storm is normally
common in tropical environments and has a high
potential to erode due to its high intensity. The storm type
and channel scouring assumptions were deemed v alid in
this work because of the resulting increase in the
accuracy of predictions upon their use. This argument
has also been confirmed by Perrone and Madramootoo
(1999) when using AgNPS for watershed modeling in
Quebec. They used storm type 1 which was appropriate
for Quebec conditions. The sediment yield results were
then subjected to further statistical analysis to establish
the model efficiency and the average percentage error for
the predictions.
As in the case of peak runoff rates, the goodness of fit
between the predicted and observed sediment yield was
assessed for each of the two stations separately. For the
upstream station the average percentage error between
observed and predicted values was 2.0% with a
satisfactory EFF value of 0.88. For Egerton station, there
was also a good correlation between the observed and
predicted sediment yields. The average percent error
between observed and predicted values was 2.5% while
for the model efficiency, an EFF value of 0.86 was
obtained as summarized in Table 5.
The results obtained in the study compared favourably
Otieno and Ogalo 403
Table 5. Statistical parameters of observed and predicted sediment yield.
Catchment Area (km2) No. of events Sediment yield
Efficiency Error (%)
Egerton 127 11 0.86 2.5
Treetop 110 11 0.88 2.0
to works using AgNPS by Young et al. (1987), which over
predicted sediment yield by 2.5% and had an EFF of 0.95
in the Trevor watershed. In Hesse, central Germany,
there was a 5% error in sediment estimation. Walling et al
(2003) compared the performance of AgNPS and
ANSWERS models coupled to GIS in estimating
sediment concentration and found the two models to be
reasonably consistent with the recorded values, although
the AgNPS model appeared to provide closer agreement
between observed and simulated values.
Conclusion
Deterioration of surface water quality and quantity is a
problem in the River Njoro catchment. In the present
study the peak runoff rate and sediment yield from the
upper River Njoro catchment was predicted using a
combination of Remote sensing and AgNPS model in a
GIS environment. The GIS platform provided a faster and
better method for spatial modeling and availed output
maps that are easy to understand.
The stream flow is an integrating measure of all the
hydrological processes operating within the catchment.
Most catchments in the developing countries are not
gauged with the appropriate instruments for runoff
measurement. However, AgNPS, a storm event based
model can be used to predict the runoff and sediment
yields based on the rainfall information and land use and
its related parameters which can be derived in a GIS
environment using Remote sensing information. The
combination of AgNPS, GIS and Remote Sensing has
shown to be a useful substitute to the in situ
measurement of runoff and sediment yield based on the
results realized in the study.
ACKNOWLEDGMENTS
The authors would like to thank Prof. Robert Becht and
Remco Dost of International Institute for Geo-information
and Earth Sciences (ITC), Netherland for their technical
and financial support that ensured the understanding and
application of the AgNPS model in the current research.
REFERENCES
Chow VT, Maidment DR, Mays LW (1988). Applied Hydrology.
McGraw-Hill, Singapore.
De Jong SM, Riezebos HTH (1992). Assessment of erosion risks using
Multispectral Remote Sensing Data and an empirical erosion model.
Proc. 3rd Euro. Conf. GIS (EGIS’92). 23-27th April, Munich. pp. 895.
Finn MP, Usery EL, Scheidt DJ, Jaromack GM, Krupinski TD (2006). An
interface between the Agricultural Non-Point Source (AGNPS)
pollution model and the ERDAS Imagine Geographic Information
System (GIS). Int. Ass oc. C hin. Pr ofessionals Geogr. Inf. Sci.
(CPGIS) 12(1):9-19.
Foster GR, Lane LJ, Nowlin JD, Laflen JM, Young RA (1981).
Estimating erosion and sediment yield on field sized areas. Trans.
ASAE 24(50).
ILWIS 3.1. (2001). User guide. ITC Enschede, The Netherlands
Khoelliker JK, Humbert CE (1989). Applicability of AGNPS model for
water quality planning. ASAE, St. Joseph, MI.
Lee MT, White DC (1992). Application of GIS databases and water
quality modelling for agricultural non point source pollution control.
Research Report 014. W ater Resource Center, University of Illinois
Urbana-Campaign, Urbana, IL. pp. 24.
Meijerink AMJ, Valenzuela CR, Stewart A (1988). The Integrated Land
and W ater Information System (ILWIS). Scientific Status Report on
the project Geo-information System for land use zoning and
watershed management. ITC Pub. No. 7. International Institute of
Aerospace Survey Earth Science (ITC). The Netherlands pp. 115.
Morgan RPC (1986). Soil Erosion and Conservation. Longman. Hong
Kong.
Mostaghini SS, Park W, Cooke RA, Wang SY (1997). Assessment of
management alternatives on a small agricultural watershed. Water
Res. 31(8).
Onyando JO, Musila F, Awer M (2005). The use of GIS and Remote
sensing techniques to analyse water balance of Lake Bogoria under
limited data conditions. J. Civ. Eng. Res. Pract. 2:1.
Perrone J, Madramootoo CA (1999). Sediment yield prediction using
AGNPS. J. Soil Water Conserv. 54(1).
Ralph J, Helmidt S (1984). Agricultural Management Handbook.
Ministry of Agriculture and Livestock Development, Nairobi, Kenya. p.
397.
Suttles JB, Vellidas G, Bosch DD, Lowrance R, Sheridan JM, Usery EL
(1999). W atershed scale simulation of s ediment and nutrient loads in
Geogia Coastal plain stream using the Annualized AGNPS model.
ASAE Paper 40(5).
Walling DE, He Q, W helan PA (2003). Using 137Cs measurements to
validate the application of the AGNPS and ANSWERS erosion and
sediment yield models in two small Devon c atchments. Soil Till. Res.
69(1/2).
Young RA, Onstad CA, Bosch DD, Anderson W P (1987). AgNPS:
Agricultural Non point source pollution model. A Watershed analysis
tool. USDA, Conserv. Res. Rep. No. 35.
