Content uploaded by Liesbeth I. Wilschut
Author content
All content in this area was uploaded by Liesbeth I. Wilschut on Oct 01, 2015
Content may be subject to copyright.
ISRIC – World Soil Information has a mandate to serve the international community as custodian of
global soil information and to increase awareness and understanding of soils in major global issues.
More information: www.isric.org
ISRIC – World soil Information has a strategic association
with Wageningen UR (University & Research centre)
Green Water Credits Report 9
L.I. Wilschut
Land Use Map of the Upper Tana, Kenya
Based on remote sensing
Green Water Credits
Land Use Map of the Upper Tana, Kenya
Based on remote sensing
Author
L.I. Wilschut
Contributors
R.A. MacMillan
J.H. Kauffman
R. de Jong
Series Editor
s
W.
R.S. Critchley
E.
M. Mollee
Green Water Credits
Report 9
Wageningen, 2011
Ministry of Agriculture
Water Resources Management
Authority
Ministry of Water and Irrigation
©
2011, ISRIC Wageningen, Netherlands
All rights reserved. Reproduction and dissemination for educational or non
-commercial purposes are permitted without any prior
written permission provided the source is fully acknowledged. Reproduction of materials for resale or other commercial purposes
is prohibited without prior written permission from ISRIC. Applications for such permission should be addressed to:
Director, ISRIC
– World Soil Information
PO B0X 353
6700 AJ Wageningen
The Netherlands
E
-mail: soil.isric@wur.nl
The designations employed and the presentation of materials do not imply the expression of any opinion whatsoever on the part
of ISRIC concerning the legal status of any country, territory, city or
area or of is authorities, or concerning the delimitation of its
frontiers or boundaries.
Despite the fact that this publication is created with utmost care, the authors(s) and/or publisher(s) and/or ISRIC cannot be
held
liable for any damage caused by t
he use of this publication or any content therein in whatever form, whether or not caused by
possible errors or faults nor for any consequences thereof.
Additional information on ISRIC
– World Soil Information can be accessed through http://www.isric.org
Citation
Wilschut, LI 2010.
Land Use Map of the Upper Tana, Kenya; Based on remote sensing
. Green Water Credits Report 9 / ISRIC
Report 2010/03. ISRIC
– World Soil Information, Wageningen.
Green Water Credits Report 9
Green Water Credits Report 9 3
Foreword
ISRIC – World Soil Information has the mandate to create and increase the awareness and understanding of
the role of soils in major global issues. As an international institution, ISRIC informs a wide audience about the
multiple roles of soils in our daily lives; this requires scientific analysis of sound soil information.
The source of all fresh water is rainfall received and delivered by the soil. Soil properties and soil
management, in combination with vegetation type, determine how rain will be divided into surface runoff,
infiltration, storage in the soil and deep percolation to the groundwater. Improper soil management can result
in high losses of rainwater by surface runoff or evaporation and may in turn lead to water scarcity, land
degradation, and food insecurity. Nonetheless, markets pay farmers for their crops and livestock but not for
their water management. The latter would entail the development of a reward for providing a good and a
service. The Green Water Credits (GWC) programme, coordinated by ISRIC – World Soil information and
supported by the International Fund for Agricultural Development (IFAD) and the Swiss Agency for Development
and Cooperation (SDC), addresses this opportunity by bridging the incentive gap.
Much work has been carried out in the Upper Tana catchment, Kenya, where target areas for GWC intervention
have been assessed using a range of biophysical databases, analysed using crop growth and hydrological
modelling.
This report addresses the need for a more updated and higher resolution land use map than the Africover
2000 map used to date. Remote sensing analysis was applied using two classification methods. On the basis
of overall accuracy, the Support Vector Machine method was selected for the classification of land use. The
SVM map is based on satellite images from 2000; however land use changes have occurred between 2000
and 2009. The Green Water Credits Pilot Implementation phase will require an updated detailed land use map.
Dr ir Prem Bindraban
Director, ISRIC – World Soil Information
4 Green Water Credits Report 9
Green Water Credits Report 9 5
Key Points
– The Green Water Credits Pilot Operation Phase for the Upper Tana catchment requires a more updated and
higher resolution land use map than the Africover 2000 map used to date.
– Remote sensing analysis was applied using two classification methods. On the basis of overall accuracy,
the Support Vector Machine method was selected for the classification of land use.
– The single land use classes given by the Africover map are, in reality, a mix of land cover types. The new
higher resolution map provides an improved description of the mixture of land use types for each zone.
– Rangeland is dominant in the lower elevation dry area. The Africover map shows an overestimation of the
rainfed cereal class in this area.
– Forest cover is overestimated in the Africover map, especially on the eastern side of Mount Kenya.
The new map shows that this area contains large areas of tea, coffee and maize.
– The new land use map will be used to improve hydrological and erosion modelling. This will lead to a more
accurate estimation of the current situation regarding water resources and land degradation, and will also
lead to improvements in the choice of GWC target areas.
– Unresolved uncertainty in the new land use map involves the distinction between bare/degraded lands and
rainfed agriculture in dry areas. The occurrence of rice in areas outside the Mwea scheme also requires
further investigation.
– The SVM map is based on satellite images from 2000; however land use changes have occurred between
2000 and 2009. The Green Water Credits Pilot Implementation phase will require an updated detailed land
use map.
– The observations made in May 2009 confirm that accelerated erosion is a serious issue in the Upper Tana
catchment. The main contributing factor to accelerated erosion is inappropriate soil and water
conservation in farmland, in particular within maize and coffee fields.
6 Green Water Credits Report 9
Green Water Credits Report 9 7
Contents
Foreword 3
Key Points 5
Acronyms and Abbreviations 9
1 Introduction and objectives 11
2 Upper Tana catchment 13
2.1 Location 13
2.2 Climate 13
2.3 Hydrology 15
2.4 Geology 15
2.5 Land use 16
2.6 Population 16
3 Methods 19
3.1 Fieldwork 19
3.2 Remote sensing analysis 20
4 Results 23
4.1 Fieldwork 23
4.2 Maximum Likelihood classification 25
4.2.1 Validation 28
4.2.2 Comparison with Africover 2000 land use map 29
4.2.3 Comparison with the ML classification of the Landsat image 30
4.3 SVM classification 32
4.4 Extrapolation: classification of Landsat images 33
4.5 Final Land use map Tana 36
4.5.1 Forest area 37
4.5.2 Tea zone 39
4.5.3 Coffee zone 41
4.5.4 Irrigated areas 44
4.5.5 Rainfed cereal and maize 46
4.5.6 Rangeland 48
4.6 Comparison with Agro-Ecological Zone maps 48
5 Conclusions and recommendations 51
References 53
Green Water Credits Report 9 9
Acronyms and Abbreviations
ASTER Advanced Spaceborne Thermal Emission and Reflection Radiometer
AEZ Agro-Ecological Zone
Africover Africa’s Land Cover Map (2000)
ENVI Environment for Visualising Images
FAO Food and Agriculture Organisation
GEF Global Environment Facility
GWC Green Water Credits
IFAD International Fund for Agricultural Development
KARI Kenya Agricultural Research Institute
KenGen Kenya Electricity Generating Company Ltd.
ML Maximum Likelihood (classification)
NASA National Aeronautics and Space Administration
NWC Nairobi Water Company
POC Proof-of-Concept
ROI Region of Interest
SVM Support Vector Machine
SWAT Soil and Water Assessment Tool
UNEP United Nations Environment Programme
WOCAT World Overview of Conservation Approaches and Technologies
10 Green Water Credits Report 9
Green Water Credits: the concepts
Green water, Blue water
, and the GWC mechanism
Green water
is moisture held in the soil. Green water flow refers to its return as vapour to the atmosphere through transpiration
by plants or from the soil surface through evaporation.
Green water
normally represents the largest component of precipitation,
and can only be used
in situ
. It is managed by farmers, foresters, and pasture or rangeland users.
Blue water
includes surface runoff, groundwater, stream flow and ponded water that is used elsewhere - for domestic and stock
supplies, irrigation, industrial and urban consumption. It also supports aquatic and wetland ecosystems.
Blue water
flow and
resources, in quantity and quality, are closely determined by the management practices of upstream land users.
Green water
management comprises effective soil and water conservation practices put in place by land users. These practices
address sustainable water resource utilisation in a catchment, or a river basin.
Green water
management increases productive
transpiration, reduces soil surface evaporation, controls runoff, encourages groundwater recharge and decreases flooding. It
links water that falls on rainfed land, and is used there, to the water resources of rivers, lakes and groundwater:
green water
management aims to optimise the partitioning between
green
and
blue
water
to generate benefits both for upstream land users
and downstream consumers.
Green Water Credits
(GWC) is a financial mechanism that supports upstream farmers to invest in improved green water
management practices. To achieve this, a GWC fund needs to be created by downstream private and public water-use
beneficiaries. Initially, public funds may be required to bridge the gap between investments upstream and the realisation of the
benefits downstream.
The concept of green water and blue water was originally proposed by Malin Falkenmark as a tool to help in the understanding
of different water flows and resources - and the partitioning between the two
(see Falkenmark M 1995 Landwater linkages. FAO
Land and Water Bulletin 15-16, FAO, Rome).
Green Water Credits Report 9 11
1 Introduction and objectives
Green Water Credits (GWC) is an environmental reward system that promotes sustainable land and water
management by farmers, so that land and water degradation diminish and both water quantity and quality
increase. Farmers on rainfed land will receive investment support to apply simple, but effective, soil and water
conservation measures, which lead to an increase in the amount of
green water
upstream and
blue water
downstream (see “GWC – the concepts” on page 10).
Phase I of the GWC programme, the Proof-of-Concept
(POC), finished in December 2007 and work for phase II,
the
Pilot Operation
, was started. The POC identified the benefits of the GWC programme and explored the
feasibility of the programme in the Middle and Upper Tana catchment in Kenya.
For the Pilot Operation, an up-to-date and high resolution land use map is required, both for hydrological
modelling and for the implementation of the Green Water Credits programme itself. The POC used the
Africover map (FAO 2000), a 1:250,000 land use map based on Landsat images from 1999. In recent years,
however, there have been major land use changes in the Upper Tana catchment. Therefore, it was decided to
develop a new land use map based on remote sensing analysis and supported by fieldwork.
The main goal of this study is to come up with a higher resolution, and more up-to-date, land use map
compared to the Africover 2000 map. To reach this goal, a remote sensing analysis was performed, using
ASTER and Landsat satellite images. The two image analysis methods used are 1) Maximum Likelihood (ML)
classification and 2) Support Vector Machine (SVM) classification. The main question addressed by this study
is:
What is the current land use in the Upper Tana?
This report provides the results of the research; it describes the methods used and displays the final land use
map of the Upper Tana.
The report is constructed as follows. Chapter 2 describes the Tana basin in Kenya. In chapter 3 the methods
of both the fieldwork and the remote sensing analysis are outlined. In chapter 4 the results of the fieldwork and
of the Maximum Likelihood classification are presented and interpreted. Chapter 5 discusses the validation of
the land use map, and looks towards future research actions, then reaches conclusions.
12 Green Water Credits Report 9
Green Water Credits Report 9 13
2 Upper Tana catchment
2.1 Location
The Upper Tana catchment is located 50 km northeast of Nairobi and covers an area of approximately
16,000 km
2
(Figure 1). There are 11 districts in the catchment: Thika, Maragua, Murang’a, Nyeri, Kirinyaga,
Embu, Mbeere, Meru South, Meru Central, Meru North and Tharaka (World Bank 2007; World Resources
Institute 2007).
Figure 1
Location and elevation of the Upper Tana catchment, Kenya
2.2 Climate
The Upper Tana area experiences two rainy seasons a year as a result of the Inter-tropical Convergence Zone;
the
long rains
last from around March to June and the
short rains
from September to December. The rainy
seasons vary considerably from year to year in their duration and rainfall totals. Figure 2 shows the total
precipitation in a wet year (2006). Rainfall patterns in the mountainous catchments are very heterogeneous.
Average annual precipitation increases from 400 mm in the savannah to 2300 mm on the windward south-
14 Green Water Credits Report 9
eastern side of Mount Kenya and drops to 800 mm in the summit region (IFAD/UNEP/GEF 2004; Notter
et al.
2007).
Figure 3 shows the variability of precipitation over twelve years, as measured in Embu (located at 1493 m) and
Meru (1554 m) for 1996-2008
1
.
Potential evapotranspiration ranges from around 1700 mm in the low elevation savannah zone to less than
500 mm in the summit region. All areas below the forest zone have a rainfall evapotranspiration deficit. As a
consequence, the high elevation forest and moorland zones provide most of the discharge of the rivers in the
dry periods (Notter
et al.
2007).
Figure 2
Precipitation in a wet year (2006)
1
Data source: www.tutiempo.net
Green Water Credits Report 9 15
Figure 3
Precipitation as measured in Meru and Embu, Kenya
2.3 Hydrology
The main river in the catchment is the Tana, which supplies water to 17 million people, about 50% of the
country’s population (IFAD/UNEP/GEF 2004). The Upper Tana river receives its water from the higher elevation
regions, in particular from the Aberdares range and Mount Kenya. Rivers originating from Mount Kenya are: the
Thingithu, Rutugi, Ena, Rupingazi, Nyandi and Thiba. Mathioya, Maragua and Sagana drain from the Aberdares.
The Nairobi Water Company (NWC), that delivers water to the municipality of Nairobi, extracts about 75% of its
demand from the Thika river through the Ndakaine reservoir.
The water resources of the Upper Tana catchment provide water for 1 million ha of rainfed agriculture and
68,700 ha of irrigated land (Hoff and Noel 2007), which accounts for over 75% of total water demand
(IFAD/UNEP/GEF 2004). There is increasing demand for irrigation water on the slopes of Mount Kenya,
particularly to support horticulture production. Water usage in the upstream areas however affects water
availability in the lower drier areas. Water is also important for electricity generation, industry (3% of total
demand) and livestock (4%) (IFAD/UNEP/GEF 2004; Mogako
et al.
2006).
KenGen, Kenya’s power company, has 7 hydropower stations in the Upper Tana, of which Masinga Dam,
holding up to 1560 million cubic metres, is the largest. This so called 7-Forks cascade delivers up to 65% of
the country’s electricity (Droogers
et al.
2006).
2.4 Geology
The Upper Tana can be divided into two main geological structures: volcanic rocks of the Cenozoic Era are
found in the west while in the east the bedrock consists of metamorphic rocks of the Mozambique belt. Mount
Kenya, an extinct volcano formed between 100-400 million years ago, is located in the west of the catchment
(IFAD/UNEP/GEF 2004).
Mean annual precipitation in Meru and Embu
(1996 - 2008)
0
500
1000
1500
2000
2500
1996 1998 2000 2002 2004 2006 2008
Meru Embu
16 Green Water Credits Report 9
2.5 Land use
Land use in the Upper Tana can be divided into three main classes:
natural vegetation
(forest, grassland and
wetlands),
rainfed and irrigated agriculture
(tea, coffee, maize and rice) and
rangeland
. Figure 4 shows the
Africover 2000 land use map. More detailed information on land use will be given in Chapters 3 and 4.
2.6 Population
Approximately 3.1 million people live in the Upper Tana (World Resources Institute 2007). The largest towns
are Thika and Nyeri, with respectively about 90,000 and 100,000 inhabitants
2
. Figure 5 shows the population
density of the Upper Tana. Population density declines with elevation, due to decreasing rainfall and soil
fertility.
Figure 4
Africover 2000 land use map of the Upper Tana
2
According to Kenya’s last official census in 1999, but both the towns are estimated to have grown by at least 50,000.
Green Water Credits Report 9 17
Figure 5
Population density in the Upper Tana (1999)
18 Green Water Credits Report 9
Green Water Credits Report 9 19
3 Methods
3.1 Fieldwork
The purpose of the fieldwork was twofold:
– to ground truth points of different land use types for the land use classification;
– to observe erosion features (results are given in Annex 1).
The field observations were collected at 364 sites. Observation areas were selected in advance based on
ASTER satellite images, the KenSOTER soil map and the Africover land use map, in order to get a
representative sample of the area. In the field, local conditions, such as accessibility by car and the presence
of fences determined whether or not to sample a plot.
Coordinates were recorded using a GPS receiver on land plots larger than 20 m x 20 m. For each site its land
use class, land management practice, vegetation cover and erosion severity was described and photographs
were taken to complement the observations.
The fieldwork focused on an area covered by ASTER image 1 (Figure 6).
Figure 6
ASTER images. Image no. 1 is used for the initial remote sensing analysis
20 Green Water Credits Report 9
The central ASTER image (no. 1 in Figure 5, see also Figure 8) was chosen because it covers a large part of
the
rainfed agricultural land
. The image dates from 26 August, 2004. The most preferable image would have
been one dating from April or May 2008 or 2009, since these months coincide with the cropping season of
most important crops
3
, but from this period no cloud-free (< 20% cloud in the Upper Tana) images are
available between 2000 and 2009. Figure 7 shows the Landsat images that were used for the analysis of the
entire Upper Tana catchment. These images date from February 2000.
Figure 7
Two Landsat images (2000) covering the Upper Tana
3.2 Remote sensing analysis
ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) is an imaging instrument flying on
a satellite launched in December 1999 as part of NASA’s Earth Observing System. It has a spatial resolution of
15 m in the visible and near-infrared bands. The swath width
4
is 60 km.
Landsat 7 has eight spectral bands, with resolution ranging from 15 to 60 meters. It has a swath width of
185 km. Combined, two images dating from 2000 cover the Upper Tana catchment almost entirely; only the
3
This is most important for maize and beans, as tea and coffee are grown year-round.
4
This is the width of the covered area on the earth that is sensed with the satellite image.
Green Water Credits Report 9 21
upper east is excluded. The first Landsat satellite was launched in 1972. Landsat 7 experienced a failure in
2003; therefore correct images are only available until then.
ASTER was chosen for the first detailed analysis, because of its high resolution, its recurrence interval of
48 days, and the availability of cloud-free images between 2000 and 2009.
Landsat was chosen for the construction of the land use map of the whole Upper Tana catchment, because it
is the only satellite with images available that cover such a large surface area; thus only 2 images are
necessary to make the land use map. The main drawback of the Landsat images are that they date from
2000, but the advantage is that two Landsat images combined cover almost the whole Upper Tana (Figure 7).
ASTER could have been used as well, but there are disadvantages when satellite images of different
acquisition dates are combined. Differences between the images that occur due to season, time of day,
meteorological conditions and environmental factors, make it more difficult to form consistent classes and
perform a classification. Pre-processing of the images is then necessary, which is a demanding job.
The analysis of the ASTER image consists of three steps:
1. Construction of Regions of Interest (ROIs)
5
for the different land use types. This is done on basis of the
ground truth points collected in the field.
2. Classification with a
– Maximum Likelihood (ML) classifier in ENVI
6
(see Box 1); and
– Support Vector Machine (SVM) (Box 1).
3. Validation of the outcome with
– Rule images
7
, which show the probability that a pixel belongs to a certain class;
– Google Earth images; and
– Field observations.
These three steps are repeatedly performed in order to achieve the best result. In this process land use
classes are combined or split, ROIs are changed or added and the probability threshold for the classes is
increased or decreased.
After the analysis of the ASTER image, the same method was applied to the Landsat images. The classification
of the Landsat images was compared to the result of the ASTER classification.
5
ROIs (Regions of Interest) are regions selected from the satellite image, that represent a certain class, such as
maize
or
water
.
6
ENVI is a software product used for processing and analysing geospatial imagery.
7
Rule images are automatically generated when a classification is performed. The images show for each class the probability (0-1)
that a pixel belongs to this class, or in other words, they show the levels of classification confidence for each class. From these
images one can, for example, determine which areas cannot be classified with high accuracy.
22 Green Water Credits Report 9
Box 1
Automatic Classification methods
Maximum Likelihood classification method
The Maximum Likelihood (ML) classification is based on the assumption that the cells in each class sample in the multi-
dimensional space are normally distributed. A class can be characterised by a mean vector and covariance matrix. Given these
two characteristics for each cell value, the statistical probability is computed for each class to determine the membership of the
cells to the class. The result is a map in which each cell has been assigned to its most likely class.
Support Vector Machine Method
SVM is a method used for classification. A SVM seeks to fit an optimal hyper plane between the classes and uses only some of
the training samples that lie at the edge of the class distribution in feature space (support vectors) (Foody 2002; Mathur and
Foody 2008; Oommen et al. 2008). The advantage of this method is that, unlike the Maximum Likelihood method, only a few
training samples are necessary. The disadvantage of this method is the long computation time..
Data sources:
http://www.pmel.noaa.gov/tao/elnino/el-nino-story.html
http://en.wikipedia.org/wiki/El_Ni%C3%B1oSouthern_Oscillation#Causes_of_El_Ni.C3.B1o
Figure 8
The ASTER image used for the remote sensing analysis
Green Water Credits Report 9 23
4 Results
4.1 Fieldwork
In total 364 sites were visited and characterised. Figure 9 shows the groundtruth points.
Figure 9
Groundtruth land use observations
Many different land use types were observed during the fieldwork. They can be grouped into four main
classes: rainfed crops, irrigated crops, natural vegetation and semi-natural vegetation. These classes can be
further subdivided as follows:
1. Rainfed crops:
a. Coffee
i. Monocropped plantations
ii. Combinated with banana and
Grevillea robusta
trees
iii. Intercropped with beans, passion fruit, napier grass or maize
b. Tea
24 Green Water Credits Report 9
c. Maize
i. Maize only
ii. Maize and beans (the most common combination)
iii. Combined with trees, potatoes or napier grass
d. Beans
e. Potatoes
f. Banana plantations
g. Trees
i. Macadamia
ii. Avocado
iii. Mango
iv. Coniferous plantations
v. Eucalyptus plantations
2. Irrigated crops:
a. Rice
b. Tomatoes
c. Passion fruit
3. Natural vegetation:
a. Forest
b. Moorland
c. Shrubs
d. Wetland
e. Riverbank vegetation
4. Semi-natural vegetation:
a. Grassland/Rangeland
Figure 10 shows the distribution of ground truth observations. Some land use types, such as
tomatoes
, have
only few observations; these classes cannot be used in the remote sensing analysis, because their spectral
signal is not distinctive enough. Sub-classes, such as the four types of coffee plots, are grouped into one class
for the classification, because spectrally they cannot be distinguished from each other.
Green Water Credits Report 9 25
Figure 10
Distribution of groundtruth observations in the Upper Tana
4.2 Maximum Likelihood classification
For the spectral information based land use classification, it is necessary to group the land use types into
classes based on spectral separability. The following classes have been used for the classification:
coffee
,
tea
,
forest
,
rice
,
rangeland
(grassland and/or shrubs),
urban
,
bare/degraded land
,
rainfed agriculture in dry areas
,
rainfed agriculture on black soils
,
water
and
cloud
.
The
rainfed agriculture in dry areas
class represents the lower elevation and low rainfall areas, where mostly
maize and sorghum are grown. In these areas, the crops are widely spaced on the fields because of the water
deficit. The spectral signature therefore is different from the signature of
maize
.
Rainfed agriculture on black soils
is agriculture (dominated by maize and sorghum; though millet and cotton
are also grown
8
) on black clay-rich soils, which are classified as vertisols (FAO 1988). In this area there are
also irrigated tomatoes. The image below (Figure 11) shows the result when the probability threshold for all
classes is set to 0.4.
8
Confirmed by Dr P. Macharia, Kenyan Soil Survey.
Field observations (%)
in the Upper Tana basin
25.
24.
23.
22.
21.
20.
19.
18.
17.
16.
15.
14.
13.
12.
11.
10.
9.
8.
7.
6.
5.
4.
3.
2.
1.
1. coffee (6 %)
2. coffee+maize/grass/beans (7 %)
3. coffee and bananas (1 %)
4. coffee and trees (1 %)
5. passion fruit&beans/coffee (1 %)
6. tea (9 %)
7. maize (16 %)
8. maize&beans (15 %)
9. maize&beans+potatoes/trees (8 %)
10. maize&bananas/grass/potatoes (2 %)
11. beans (1 %)
12. tomatoes (1 %)
13. mango trees (2 %)
14. bananas (3 %)
15. potatoes (1 %)
16. rice (1 %)
17. napier grass (1 %)
18. grassland (5 %)
19. grassland with shrubs (4 %)
20. shrubs (7 %)
21. eucalyptus trees (2 %)
22. riverbank vegetation (2 %)
23. forest (2 %)
24. coniferous trees (1 %)
25. other (2 %)
26 Green Water Credits Report 9
Figure 11
Land use map made with a ML classification, with a probability threshold of 0.4
The black pixels in this map represent pixels that cannot be classified with a probability higher than 0.4.
Especially in the areas south, east and west of Mwea rice fields, pixels are difficult to classify (indicated with
red circles). Apart from a shortage of groundtruth data, one difficulty in these areas is the heterogeneous
character of present soil types, specifically the vertisols. Vertisols are black, fertile soils that have a distinct
spectral signal that overwhelms the signal of the land use type.
Apart from the Mwea rice scheme, rice is rarely grown in the catchment. Therefore, the pixels of
rice
west of
Embu and west of Murang’a are probably incorrectly classified (indicated with blue circles). Annex 2 shows the
separability between the classes. It indicates that the classes
maize
and
coffee
, and
maize
and
urban
have
spectral signatures that are overlapping.
Green Water Credits Report 9 27
Figure 12
Land use map made with a ML classification, without applying a probability threshold
Figure 12 shows the ML result when no probability threshold is applied. To further improve this image, the
satellite image was classified again, without the
rice
ROIs. The
rice
fields at Mwea were extracted from the
original image and added to this new classification. In the resulting image (Figure 13) the
rice
pixels have
mostly been replaced by
coffee
or
maize
.
The cloud in the north of the image has been removed by replacing it by
forest
. Forest is the dominant land
cover in this zone (Jaetzold and Schmidt 1983), which can be seen on high resolution images released by
Google Earth. The result is shown in Figure 13.
28 Green Water Credits Report 9
Figure 13
Final land use map made with Maximum Likelihood method and improved using Google Earth. Rice only appears at the Mwea rice
scheme in the south east of the image and the cloud in the north has been replaced by forest
4.2.1 Validation
The overall accuracy of the image is 80.4 %, with a Kappa Coefficient
9
of 0.87. The producers’ accuracy was
lowest in the following classes:
rice
,
potatoes
and
coffee
. The users’ accuracy was lowest for
maize
,
coffee
and
potatoes
(for more information on the validation method, see Box 2).
9
The Kappa Coefficient (0-1) is a statistical measure of agreement, beyond chance, between the ground truth data and the output
of the classification.
Green Water Credits Report 9 29
Box 2
Validation method
To investigate the accuracy of the classified image, an error matrix has been constructed. An error matrix shows for each class
the amount of pixels that belong to the class (based on groundtruth) and the number of pixels that have been classified to each
class.
95% of the ROIs (selected stratified randomly) has been used for the ML classification. The remaining 5% was used to validate
the classified image. As these 5% are located within or close to the ROIs that have been used for the classification, the outcome
is biased: the overall accuracy is overestimated. Nevertheless, it gives an indication of the heterogeneity within the land use
classes and it assesses whether the ROIs have been chosen well.
From the error matrix an overall accuracy, producer accuracy and user accuracy can be calculated.
Producer’s accuracies are calculated from dividing the number of correctly classified pixels in a category, by the number
of training set pixels used for that category. This figure indicates how well training set pixels of the given cover type are
classified (Lillesand and Kiefer 1987).
User’s accuracies are computed by dividing the number of correctly classified pixels in a category by the total number of pixels
that were classified in that category. This figure is a measure of commission error and indicates the probability that a pixel
classified into a given category actually represents that category on the ground (Lillesand and Kiefer 1987).
4.2.2 Comparison with Africover 2000 land use map
Compared to the Africover 2000 map (Figure 14), the new classified image (Figure 13) shows more detail. The
main differences are:
– The
Forest
area in the Africover 2000 map is 10% higher;
– The Africover map shows homogenous
coffee
and
tea
zones; however the
coffee
and
tea
zones have
mixture of various crops which is shown on the newly classified map;
– The Africover map shows “irrigated unspecified”, however the main crops in this area are maize, sorghum
and tomatoes; and
– The new map shows also areas that are classified as bare land, which is often seriously degraded
(
Bare/degraded land
class).
30 Green Water Credits Report 9
Figure 14
Land use according to the Africover (2000) map
4.2.3 Comparison with the ML classification of the Landsat image
To assess whether or not the Landsat image shows comparable outcomes to the ASTER image, a
ML classification has been performed on the Landsat image for the same extent of the ASTER image
(Figure 15).
Green Water Credits Report 9 31
Figure 15
ML classification on a Landsat (30 m resolution) image from 2000
The main difference is the presence of
maize
: there is 5.5% more
maize
in the ASTER classification (see for
comparison of all values Table 1). The
maize
is mainly substituted by
water
,
coffee
,
rice
and
rangeland
. One
other remarkable difference between these images is the occurrence of
coffee
: in the Landsat (2000) image
there is 1.5% more
coffee
, which is more widely spread and extends to lower elevations than in the ASTER
classification. Another difference occurs as a result of the
cloud
in the ASTER image. The Landsat image is
cloud-free and it is likely that the image therefore has a higher reliability in the centre northern part of the
image. This part will be added to the final result. In general the maps show correlating land use patterns.
It can therefore be concluded that although the Landsat images date from 2000, they are suitable for the land
use classification. The Landsat images cover the entire Upper Tana catchment and are therefore useful for the
construction of a new land use map.
32 Green Water Credits Report 9
Table 1
Land use in percentages as result of ML classification of an ASTER (2004) and Landsat (2000) image, SVM classification on an
ASTER image (2004) and the Africover land use map
Land use
ML Aster
(%)
ML Landsat
(%)
Africover
(%)
SVM
(%)
Rock
-
0.6
-
-
Water
1.2
4.2
1.2
2.9
Urban
3.3
5.1
0.1
10.8
Forest
24.6
18.1
34.6
29.9
Tea
4.0
4.2
4.5
3.2
Coffee
18.9
20.4
24.9
7.9
Maize
22.1
14.6
4.4
9.9
Rainfed cereal
-
-
17.6
-
Potatoes
0.0
-
-
0.0
Mango
1.0
-
-
0.9
Rice
2.8
3.6
3.4
6.3
Rangeland
13.6
15.6
6.8
22.1
Rainfed agriculture on black soils
3.4
6.2
-
2.7
Rainfed agriculture in dry areas
2.8
3.5
-
0.7
Bare/degraded land
2.0
3.7
-
2.4
Pineapples
- 0.1 - -
4.3 SVM classification
The ASTER image has also been classified using a Support Vector Machine (SVM; Figure 16). The main
difference of SVM compared to the ML classification of the ASTER image is the change from 14.6%
maize
in
the ML image, to 9.9% in the SVM image. The
maize
has been replaced by
rangeland
, which covers 13.6% in
the ML image and 22.1% in the SVM image. This change from
maize
to
rangeland
occurs particularly in the dry
southeast part of the image. Validation on Google Earth shows that most of the south-east, east of the Mwea
rice scheme, comprises a significant amount of
rangeland
, although not as much as is displayed on the SVM
classification. According to Africover, only a small part consists of
rangelands
, and part of this area is forest,
which can neither be seen on the ML, the SVM, or on Google Earth. The large amount of
rice
outside of the
Mwea rice scheme is probably due to a classification error, as happened with the ML classification. When the
image is classified without including
rice
, the pixels west of the Mwea rice scheme are classified mostly as
water
; indeed the pixels are located close to rivers. The pixels to the north of the rice scheme are classified as
urban
, which cannot be correct. The overall accuracy of the image is 88.9%, with a Kappa Coefficient of
0.88%.
Green Water Credits Report 9 33
Figure 16
SVM classification on ASTER image (2004). The red circles indicate areas with rice pixels
4.4 Extrapolation: classification of Landsat images
The Green Water Credits programme focuses on the Upper Tana catchment. Therefore, it is useful to use
Landsat images for the construction of the final land use map, as explained under Methods (3.2).
The Landsat images were mosaiced into one image that was classified twice: using the ML (Figure 17)
and SVM method (Figure 18). An
afro-alpine flora & rock
and
pineapple
class were added to the ROIs. These
ROIs were selected using Google Earth.
34 Green Water Credits Report 9
Figure 17
The land use map made with a ML classification on two Landsat 2000 images
The results of the both methods are comparable. The ML classification gave an overall accuracy of 76.1%,
with a Kappa Coefficient of 0.7; the SVM classification has an overall accuracy of 77.7% with also a Kappa of
0.7. It should be noted that for the classification of the SVM only 10% of the data was used, and 90% for the
validation. The SVM is therefore an excellent method when only few observations have been made.
The main differences between the land use images lie in the change of the
rainfed agriculture dry areas
with
bare/degraded land
.
Coffee
fields are denser in the ML classification; they occupy 5% more area in the
ML image.
Green Water Credits Report 9 35
Figure 18
SVM classification of Landsat images (2000)
36 Green Water Credits Report 9
Table 2
Land use as calculated using ML and SVM for the Upper Tana catchment
Land use
ML
(ha)
ML
(%)
SVM
(ha)
SVM
(%)
AFR
(ha)
AFR
(%)
Afro
-alpine flora and rock
697,829 4.0 8196 5.2 1823 0.1
Water
648,967 3.7 107,259 6.8 11,428 0.7
Urban
571,472 3.3 73,485 4.7 1862 0.1
Forest
211,697 11.9 208,393 11.7 409,741 25.7
Tea
803,990 4.6 890,918 5.1 83,611 5.3
Coffee
2,388,228 13.7 133,459 8.5 174,420 11.0
Maize
1,806,829 10.3 176,612 11.2 88,690 5.6
Rainfed cereal
- - - - 390,759 24.5
Rice
574,713 3.3 37,119 2.4 23,232 1.5
Rangeland
4,451,349 25.5 376,663 23.9 362,511 22.8
Rainfed agriculture on black soils
810,700 4.6 63,727 4.1 - -
Rainfed agriculture in dry areas
1,418,829 8.1 74,252 4.7 - -
Bare/degraded land
909,591 5.2 153,355 9.7 - -
Pineapple
45,071 0.3 6,785 0.4 8,470 0.5
Wetlands
- - - - 8,919 0.6
Irrigated unspecified
- - - - 24,564 1.5
TOTAL
1,573,178 100 1,573,208 100 159,0031 100
4.5 Final Land use map Tana
To produce the final land use map of the Tana, the SVM classification of the Landsat (Figure 18) was chosen
because of slightly higher overall accuracy and the possibilities of using this method when only few field data
have been collected. Statistics about the land use, in comparison to Africover data can be found in Table 2.
It should be noted that a large part of the map is based on Landsat images of 2000. The
coffee
zones have
declined since then and possibly the
tea
zone has expanded.
Green Water Credits Report 9 37
Figure 19
The Africover 2000 map
Now the major changes between the Africover (Figure 19) and SVM land use map (Figure 18) will be
discussed.
4.5.1 Forest area
The extent of the
forest
in the Upper Tana is overestimated on the Africover map, the
forest
is almost twice as
extensive as on the SVM land use map. This is shown in Figure 20. The major differences are seen in particular
on the east and south-west side of the Mount Kenya forest.
Figures 21A and B show zooms of Google Earth, which concur with the forest areas on the SVM land use map.
It should be mentioned that some small patches of
forest
that are visible on Google Earth, are not displayed on
the SVM.
38 Green Water Credits Report 9
Figure 20
Africover forest extent superimposed on the SVM land use map. “A” and “B” indicate the areas for which a zoom has been made of
Google Earth in Figure 21
Figure 21
Zooms of Google Earth showing coffee and tea fields. “A” is located to the east of Mt Kenya and south of Meru, “B” is located to
the southwest of Mt Kenya
B
A
Green Water Credits Report 9 39
4.5.2 Tea zone
Figure 22 shows the
tea
zones from the Africover superimposed on the SVM classification. There is
1.8% more tea in the SVM map compared to the Africover map. The tea is spread over a larger area, and the
tea
zone includes also
coffee
fields. East of the Aberdares, the
tea
zone extends further to the north according
to the SVM classification and groundtruth observations on Google Earth. In the Africover tea zone the SVM land
use map consists of 37.1% tea, 15.8%
coffee
, 14.6%
rangeland
, 10.7%
bare/degraded land
, and 3.8%
maize
.
The large amount of
rangeland
could be an overestimation, because in these zones there is not much
rangeland; however, grasslands do occur.
Bare/degraded land
accounts for a high percentage, probably
because of pruned tea fields. The
tea
west of the Aberdares seems to be an overestimation (see blue arrow in
Figure 22); Google Earth images confirm this.
Figure 22
Tea zones as defined by Africover displayed on the SVM land use map. “A” and “B” indicate zoom-ins showed in Figure 23. The red
arrow east of the Aberdares indicates the extension of the tea zone in this area to the north. The blue arrow indicates the area
where tea is likely to be overestimated by the SVM method
40 Green Water Credits Report 9
Figure 23
Zoom of the tea zone (indicated as "A" in Figure 22). The black line is a border of the tea zone according to Africover
(right-side tea, left-side rain fed cereal). As can be seen, on the left part of the image there is a lot of tea (green box indicates an
example of a tea field). Coffee is also present in the tea zone (an example is shown by the red box)
100 m
Green Water Credits Report 9 41
Figure 24
This image is a zoom in of the area east of the Aberdares (B). There are many coffee fields (example indicated by red box), but only
few tea fields (indicated by green box), or they are all bare
4.5.3 Coffee zone
The Africover map has 1.3 times more
coffee
as the SVM map (Figure 25). However,
coffee
on the SVM map
is spread more; into the
tea
zone and to the east of Mt Kenya (as explained in 5.5.1).
"A" and "B" indicate areas for which a view of Google Earth is shown. “A” shows a
rangeland
area, where
Africover indicated
coffee
; “B” shows an area where no
coffee
was indicated but is indeed present. “C”
(as shown on Figure 25) indicates an area where there are only some large-scale
coffee
plantations.
In between these
coffee
plantations, there are shrubs and grassland. At the location of “D”, Africover indicates
some more areas as
coffee
. Although there are some
coffee
plantations, the area here is mostly covered by
rangeland
and partly by
irrigated pineapple
.
In the
coffee
zone as defined by Africover, SVM indicates the following distribution of land use types:
coffee
29%,
maize
22%,
urban
14%,
rangeland
11% and
water
9%.
100 m
42 Green Water Credits Report 9
It is concluded that the SVM shows a more realistic image of the
coffee
area. The
coffee
zone is broader than
the Africover indicates, as it extends also to the
tea
zone. The area indicated by “B” does contain
coffee
, as
does the area east of Mt Kenya. The extension of the
coffee
zone up to “D” (Figure 25) is questionable, since
the main land use in that area seems to be
rangeland
. The
coffee
zone as shown on the Africover map does
not consist of
coffee
alone; there are a variety of land use types in the
coffee
zone, of which
coffee
and
maize
are dominant.
Figure 25
Coffee as indicated by Africover, superimposed on the SVM land use map
Green Water Credits Report 9 43
Figure 26
Zoom of the coffee area. "A" shows an area where there is no coffee present; “B” shows an area outside the Africover coffee zone,
where there is considerable coffee
Figure 27
A. Zoom of a rangeland area, which was indicated as coffee by Africover
B. Zoom of an area outside the coffee zone. There are many coffee fields in this area
100 m
B
A
100 m
100 m
44 Green Water Credits Report 9
Figure 28
C. Coffee plantations in a relatively dry area
D. This zoom in of Google Earth shows rangeland; there are hardly any coffee fields
4.5.4 Irrigated areas
The irrigated area is overestimated on the Africover map (Figure 30). The SVM land use map show that a part
of this area consist of dominantly rainfed cereals, which is confirmed by field and Google Earth observations
(Figure 29A and B).
Figure 29A shows a zoom of
rainfed agriculture
on Vertisols. There are, however, also some irrigated tomato-
fields in this area, but the majority consists of rainfed maize or sorghum
10
. In the “irrigated unspecified areas”
as indicated by Africover, the SVM map consists of 67%
Rainfed agriculture on black soils
, 6.5% consist of
maize
and 6.1% is classified as
rangeland
.
10
This was confirmed by P. Njuguna, MKEPP Officer (personal communication, 2009).
C
100 m
D
100 m
Green Water Credits Report 9 45
Figure 29
A. Zoom of an area with rainfed cereals on Vertisols
B. Zoom of rice fields at Mwea rice scheme
A B
46 Green Water Credits Report 9
Figure 30
Irrigated unspecified as classified by Africover superimposed on the SVM land use map. In the SVM classification the irrigated
unspecified areas are classified as rainfed agriculture (mainly maize/sorghum) on black soils. “A” and “B” indicate areas on which
has been zoomed in on Google Earth
4.5.5 Rainfed cereal and maize
The map below (Figure 31) shows the
rainfed cereal
and
maize
classes from the Africover map. The maize
patches compare well with the SVM map, although
maize
is also distributed widely over the
coffee
zone and in
the lower zones on the SVM map. The
rainfed cereal
area however, includes mainly
rangeland
on the SVM
map: 29%. According to the SVM, there is also a lot of
maize
and
bare/degraded land
, respectively 16% and
13%, which concurs with field and Google Earth observations.
Figure 32A and B show zooms of the
rainfed agriculture
and
maize
area on Google Earth. Both images show
there is a lot of rangeland in these areas.
Green Water Credits Report 9 47
Figure 31
Maize and rainfed cereal from the Africover land use map
48 Green Water Credits Report 9
Figure 32
A. This is a zoom of the rainfed cereals area, where there is in fact also a lot of rangeland (grassland/shrubs)
B. This is zoom of the maize area in the north-east, where there are indeed maize fields, but a large part of the area consists of
rangeland
4.5.6 Rangeland
Rangeland
comprises almost 24% of the catchment according to the SVM. The main difference with the
Africover is that Africover defined the high elevation parts of Mt Kenya and the Aberdares as
rangeland
. On the
SVM map this zone was defined as
afro-alpine zone and rock
, because literature resources indicate that the
vegetation is a combination of moorland (between 3500 and 3800 m) and grasses and lobelias between 3800
and 4500 m (UNEP 1997; World Bank 2007).
4.6 Comparison with Agro-Ecological Zone maps
In Figure 33 the land use map is shown with the agro-ecological zones as proposed by Jaetzold (Jaetzold and
Schmidt 1983). The zones are defined in Table 3.
A B
100 m 100 m
Green Water Credits Report 9 49
Table 3
Definition of Agro-ecological zones (Jaetzold and Schmidt 1983)
Agro
-ecological zone
Climate
Key crop or land use
I
Humid
Tea and forestry
II
Sub
-humid
Coffee, maize
III
Semi
-humid
Coffee, maize, cotton
IV
Semi
-humid to semi-arid
Maize, cotton
V
Semi
-arid
Rangeland
VI
Arid
Rangeland
VII
Very arid
Rangeland
Figure 33
SVM land use map, with the AEZ zones as defined by Jaetzold and Schmidt (1983)
As can be seen, most AEZ zones do not match the current land use situation.
50 Green Water Credits Report 9
Green Water Credits Report 9 51
5 Conclusions and recommendations
The first steps of the GWC programme are defined as follows:
1. Perform hydrological and erosion modelling with the SWAT model, using meteorological, hydrological,
land use and land management data
11
.
2. Define focus areas and
green water
management packages.
3. Model the planned improvements in
blue water
due to changes in land management
4. Implement the most economically viable and effective
green water
management techniques via farmer
groups.
Green Water Credits focuses on rainfed agriculture. Coffee and maize are the dominant crops; there is much
scope to improve their land management (more information on land uses and their management in the Upper
Tana is given in Annex 1). The distribution and extent of
coffee
, m
aize
and
rainfed agriculture
is significantly
different in the new SVM land use map.
Coffee
and
maize
are more widely distributed and the
rainfed
agriculture
class consists of mainly
rangeland
and
maize
.
The new land use map should be used to improve the modelling results. This will lead to a more accurate
estimation of the current situation regarding water resources and land degradation, and also to the choice of
GWC target areas.
In summary the following are the main observations and conclusions:
– The SVM land use map provides a more reliable land use map (accuracy 78%, resolution 30 m) than the
Africover and the ML classification map;
– The classes as proposed by the Africover map are in reality a mix of land cover types;
– The distribution of the
tea
and
coffee
zones extend over a larger area on the SVM map;
– The Africover map overestimates the
rainfed cereal
in the low elevation drier areas;
rangeland
dominates
on the SVM map
– The SVM derived map is suitable for hydrological modelling
– Uncertainties of the SVM are:
– the distinction of bare/degraded lands and rainfed agriculture in dry areas
– the occurrence of rice outside the Mwea scheme
– The SVM map is based on images of 2000; however land use changes have occurred between 2000 and
2009. The Green Water Credits Implementation phase will require a detailed updated land use map.
Updates can be carried out using recent ASTER images, covering part of the Upper Tana.
11
This has been done by FutureWater using the Africover land use map. In the implementation phase of GWC, the modelling will be
out carried by WRMA.
52 Green Water Credits Report 9
Recommendations:
– A land use map should be updated every 5 to 10 years.
– It is recommended that a procedure should be followed such as that described in Methods (Chapter 3).
This procedure includes:
1. Sampling
2. Classification of up-to-date cloud-free satellite images using a Support Vector Machine approach
3. Validation
– Remote sensing can be used to monitor changes in land use and land degradation as a result of
green
water
management.
Green Water Credits Report 9 53
References
Clampett WS, Nguyen VN and Tran DV 2002. Sustainable rice production for food security. The development
and use of Integrated Crop Mangement for rice production. In: FAO (Ed.) 20th Session of the International
Rice Commission. FAO, Bangkok.
Droogers P, Kauffman JH, Dijkshoorn JA, Immerzeel W and Huting JRM 2006. Green Water Credits: Basin
identification. Green Water Credits Report 1, ISRIC - World Soil Information, Wageningen.
Facon T 2000. Water Management in Rice in Asia: Some Issues for the Future. In: FAO (Ed.) Regional Office for
Asia and the Pacific, FAO.
Falkenmark M and Rockström J 2004. Balancing water for humans and nature. Earthscan, London.
FAO 1988. FAO/Unesco Soil Map of the World, Revised Legend, with corrections. FAO, Rome.
FAO 2000. East Africa Module Land Cover mapping based on satellite remote sensing. Food and Agriculture
Organisation of the United Nations, Rome.
Foody GM 2002. Status of land cover classification accuracy assessment. Remote Sensing of Environment
80: 185-201.
Hoekstra AY 2003. Virtual Water Trade. Proceedings of the International Expert Meeting on Virtual Water
Trade, pp 244.
Hoff H and Noel S 2007. Quantifying water usage and demand in the Tana River basin: an Analysis using the
Water and Evaluation and Planning Tool (WEAP). Green Water Credits Report 4, ISRIC - World Soil
Information, Wageningen.
IFAD/UNEP/GEF 2004. Mount Kenya East Pilot Project for Natural Resource Management, Project Brief
Report.
Jaetzold R and Schmidt H 1983. Farm management handbook of Kenya. Ministry of Agriculture, Nairobi.
Kauffman JH, Droogers P, Odada E, Macharia P, Gicheru P and Dijkshoorn JA 2007.
Green
and
blue
water
services and assessment of improved soil and water management scenarios using an integrated
modelling framework. Green Water Credits Report 3, ISRIC - World Soil Information, Wageningen.
Lillesand and Kiefer 1987. Concepts and foundations of remote sensing. In: Remote sensing and image
interpretation, J Wiley and Sons.
Mathur A, Foody GM 2008. Crop classification by support vector machine with intelligently selected training
data for an operational application.
International Journal of remote Sensing
29: 2227-2240.
Mogako H Gichere S, Davis R and Hirji R 2006. Climate variability and water resources degradation in Kenya:
Improving water resources development and management
. World Bank, Washington DC.
Notter B, MacMillan L, Viviroli D, Weingartner R and Liniger HP 2007. Impacts of environmental change on
water resources in the Mt. Kenya region.
Journal of Hydrology
343: 266–278.
Oommen T, Misra D, Twarakavi NKC, Prakash A, Sahoo B and Bandopadhyay S 2008. An objective analysis of
support vector machine based classification for remote sensing.
Mathematical Geosciences
40: 409-
424.
Pratt DJ, Le Gall F and De Haan C 1997. Investing in Pastoralism. Sustainable Natural Resource Use in Arid
Africa and the Middle East. World Bank, Washington.
UNEP 1997. World Heritage Sites. Protected Areas and World Heritage. Mount Kenya National Park/Natural
Forest Kenya. In: UNEP (Ed.). UNEP pp 7.
UNEP/SEI 2009. Rainwater harvesting: a lifeline for human well-being. A report prepared for UNEP by
Stockholm Environment Institute. In: Barron J (Ed.) SEI/UNEP pp 80.
Ward RC, Robinson M 2000. Principles of hydrology. McGrawhill (p.450).
54 Green Water Credits Report 9
WOCAT 2007. Where the land is greener - case studies and analysis of soil and water conservation initiatives
worldwide. Editors Liniger HP and Critchley WRS. CTA, FAO, UNEP and Centre for Development and
Environment, Bern.
World Bank 2007. Project appraisal document on a proposed credit in the amount of sdr 46.0 million
(US$68.5 million equivalent) to the government of Kenya for a natural resource management project.
World Bank pp 176.
World Resources Institute 2007. Nature's benefits in Kenya: an atlas of ecosystems and human well-being.
World Resources Institute, Washington DC and Nairobi.
Green Water Credits Report 9 55
Annex 1
1. Land and water degradation in
the Upper Tana
1.1 Introduction
Land and water degradation are core problems in Kenya. The Upper Tana catchment provides water for
about 50% of the population of Kenya. It is therefore of crucial importance that water and soil resources are
maintained or improved. This Annex gives an overview of the signs of land degradation as observed in the field
in the period 13 to 27 May 2009.
Soil erosion in the Upper Tana is in most cases the result of a lack of soil cover by crop or mulch. Increasing
the covered area will result in a higher infiltration rate, lower evaporation, lower runoff values and
consequently, less soil erosion.
In order to decrease soil erosion and to increase yields, the general objective is therefore to increase soil
cover. In this report, recommendations will be given per land use on how to reach this objective.
1.2 Soil erosion
Soil erosion is defined as the detachment and movement of the topsoil by wind and water. Soil erosion is a
natural process, but can be accelerated by human influence.
With soil erosion, the quality and quantity of the soil decreases; this has a negative effect on agricultural yields.
Soils are of vital importance to the growth of crops, because they contain important minerals and micro-
organisms. Another important function of the soil is its filtering effect on water. Without soil filtration,
agricultural pollutants will move faster into the groundwater, or will flow downstream. The lower infiltration
capacity of eroded soils causes an increase in runoff, resulting in the occurrence of more and severe flooding
events (Ward and Robinson 2000). Sediment deposited on roads and the silting up of reservoirs are yet other
consequences of soil erosion.
1.3 Land degradation in the Upper Tana
During the fieldwork in May 2009, observations on soil erosion were carried out to estimate the degree of
erosion. The main question was: where and in what form does land degradation occur in the catchment?
Figure 34 shows the slope and the sample sites in the Upper Tana catchment. At each observation point, the
erosion severity was estimated. Signs of erosion are: splash erosion, rills, gullies, sediment deposition and
56 Green Water Credits Report 9
landslides. A steep slope, low vegetation cover, erosive soil, high intensity rainfall and improper soil, crop and
water management are factors leading to, or increasing, erosion.
It should be noted that during the long rains in 2009 (March-May) relatively little rain fell. The average rainfall
in the period March-May in Embu is 357 mm, rain fell. The average rainfall in the period March-May in Embu
is 357 mm, calculated for the years 1996-2009. Over this period in 2009 there was only 247 mm of
precipitation, mostly consisting of low intensity rain (Mrs A.N. Muchira: pers. comm.). During years with more
and higher intensity rainfall, erosion increases. Gullies are often permanent features; rills can be recovered.
Figure 34
Locations where rills and gullies where observed
1.3.1 Signs of erosion
Rills were observed in 9% of the total observations; gullies in only 1%. Erosion is widespread in the Upper
Tana catchment, but some regions showed more severe erosion than others.
Table 4 gives an overview of the percentage of gullies and rills per land use type. It should be kept in mind
that the total number of observations is limited and therefore the percentages are only a first indication of
the degree of erosion per land use type.
Green Water Credits Report 9 57
Table 4
Overview of field observations per land use type
Land use
Number of observations Percentage gullies/rills per land use Average vegetation cover (% (+σ))
Coffee
56 8.9 53 (± 24)
Tea
32 3.1 93 (± 15)
Maize (total)
135 11.9 60 (±27)
Maize
72 11.4 54 (±29)
Maize & Beans
63 12.3 68 (± 24)
Mangos
9 33.3 51 (±29)
Bananas
11 0.1 54 (± 14)
Rangeland
66 12.1 70 (±23)
Total
444
Although these numbers indicate that erosion is widespread in the Upper Tana, there are too few observations
to make any statistically valid statements. The occurrence of erosion is probably mostly dependant on
vegetation cover: rills or gullies occur at 29% of the observation sites that have a vegetation cover equal to or
lower than 50%. At sites where the vegetation cover is higher than 50%, rills or gullies occur only in 6.8% of
the cases. This indicates the importance of maintaining a high vegetation cover throughout the year. This can
be done by conventional biomass or by synthetic mulching. Also, the use of manure and/or fertilizer
significantly increases the plant density, and therefore the vegetation cover.
The erosion observations were also plotted on the soil map of the Upper Tana, to be able to view possible
links between erosion and soil type in this area (Figure 35). Worth mentioning is the occurrence of erosion in
the area where regosols are the dominant soil type (indicated by black circle). In this area all the sites visited
contained rills or gullies.
58 Green Water Credits Report 9
Figure 35
Soil type map of the Upper Tana with erosion features
Table 5
Erosion features per soil type
Soil type
Total Number of observations Amount of gullies/rills Percentage of gullies/rills
Andisol
5 0 0
Acrisol
1 0 0
Arenosol
3
1
33.3
Cambisol
38 2 5.3
Ferralsol
24 5 20.8
Leptosol
6 2 33.3
Luvisol
8 0 0
Nitisol
247 17 6.9
Phaeozems
8
0
0
Regosols
5 5 100
Vertisols
15 2 13.3
Total
360 34
Green Water Credits Report 9 59
1.4 Land use & land and water management
1.4.1 Tea
In tea fields hardly any erosion occurs, because the vegetative cover of tea is dense (Figure 36). The rivers in
this zone are nearly free of sediment. However, when the soil is left uncovered, erosion will increase rapidly
due to the steep slopes (Figure 37).
Figure 36
The vegetative cover of tea is nearly 100%; therefore the soil is well protected
60 Green Water Credits Report 9
Figure 37
In the tea zone, erosion easily occurs when the soil is left uncovered
1.4.2 Coffee
Considerable erosion occurs in coffee fields. In 9% of the coffee fields signs of erosion are found. Coffee
plantations are often poorly managed, because of the decline in coffee prices (Figure 38). This can be noticed
from the poorly maintained
fanya juu
terraces and low amount of mulch on the fields. The intercropping of
beans and maize in the coffee fields and the presence of weeds and grass however limits the soil somewhat
from eroding. This is, however, a short-term solution. In the long-term, farmers will decide whether to continue
with growing coffee or not. If they continue to grow coffee, it will be important to develop stable and vegetated
terraces in the fields and to mulch between the coffee plants. Intercropping in coffee is often not permitted,
because coffee quality will decrease.
If farmers decide to switch from coffee to another crop, it is important to choose a suitable crop for the area
and to incorporate, for example, agroforestry techniques. The Green Water Credits programme should
anticipate changes in the
coffee
zones and be able to provide sustainable solutions to coffee-farmers that
decide to change their crop and management.
Green Water Credits Report 9 61
Figure 38
Poorly maintained terraces in a coffee plantation
1.4.3 Maize (and beans)
Gullies or rills occurred in 12% of the maize fields visited (Figure 39). This often occurs in fields where there is
mono-cropping of maize and when the maize is widely spaced. When no fertilizer or manure is applied, maize
is often widely spaced because of a lack of nutrients. In many fields, maize is intercropped with beans;
together they provide a good vegetation cover. The soil remains covered which prevents erosion and the
farmers reduce the risk of a failed harvest: if the maize fails, the beans may still survive. The combination of
maize and beans is only suitable for manual cultivation and not for machines. Before the planting season, the
maize fields lack vegetation cover and the soil is therefore subject to erosion.
To meet the objective of increasing soil cover (a combination of) the following actions can be taken:
– Application of fertilizer or manure
–
Green water
management
– Mulching
– Intercropping
– Agroforestry
– Vegetative strips
Another point Green Water Credits could consider in the drier areas, is the recommendation of replacement
with more drought resistant crops like sorghum.
62 Green Water Credits Report 9
Figure 39
Rills in a maize field
1.4.4 Rice
Rice is cultivated in the Mwea rice scheme approximately 20 km southwest of Embu and on a small scale
along rivers and streams west of Embu. Evaporation is very high on rice fields. According to Hoekstra (2003),
rice has a virtual water content of 2656 m
3
/tonne, compared to 450 m
3
/tonne for maize
12
. Rice thus is a
highly water-intensive product. Because the cropping of rice demands a lot of water, this has a negative effect
on the quantity of water downstream (Hoekstra 2003).
Improvement of water management in rice production can be achieved using techniques to reduce evaporation
or evapotranspiration, losses through seepage and percolation, and surface runoff. Practices and strategies to
improve rice water productivity include development of improved rice varieties, changing the planting/sowing
schedule, making more effective use of rainfall and developing better water distribution strategies (Clampett
et al.
2002; Facon 2000). Especially the choice of an optimal planting and sowing period could, in particular,
be a relatively easy to implement strategy to save water.
Another option is to replace rice with rainfed maize, as suggested in GWC report 3 (Kauffman
et al.
2007).
12
Virtual water is defined as the volume of water required to produce a commodity or service. Water use efficiency at global scale
can be achieved in a water scarce region by adopting a policy to grow and export products with relatively low virtual water
content and import products having higher virtual water content (Hoekstra 2003).
Green Water Credits Report 9 63
Figure 40
Rice fields at Mwea rice scheme
1.4.5 Eucalyptus and other firewood trees
The planting of eucalyptus trees is quite common in the Upper Tana catchment and should be changed for
other types of trees, since eucalyptus trees use a lot of water. The most suitable alternative is
Grevillea
robusta
(WOCAT 2007)
13
.
1.4.6 Rangeland
In rangeland, erosion often develops as a result of overgrazing. Figure 41 illustrates a gully complex in
grassland near Murang’a. On the opposite side of the road, where soil type, slope gradient and land use are
the same, but only land management is different (no overgrazing), hardly any signs of erosion were observed.
This illustrates that erosion can be prevented when the appropriate land management techniques are applied.
Eroded rangelands are common near the large reservoirs of the catchment. A lot of sediment ends up in the
reservoirs, causing the productivity of the hydrological power station to decline. Green Water Credits phase I
focused on upstream rainfed farmers. The degradation of rangelands is an important problem (Pratt
et al.
1997) that will be included in phase II analysis for the management of the Upper Tana. For suggestions for
sustainable management is referred to WOCAT (Pratt
et al.
1997; WOCAT 2007).
13
For other suitable agroforestry options refer to WOCAT 2007 “Where the land is Greener” and the WOCAT database:
www.wocat.net
64 Green Water Credits Report 9
Figure 41
Gullies to the south of Murang’a
1.4.7 Fruit trees
In mango tree plantations, rills and gullies occur often, because of a lack of vegetation cover between the
mango trees. In banana plantations, if adequately mulched, there is hardly any erosion. Mango trees are
suitable for dry, hot areas. Mulching between the trees will decrease evaporation, erosion and surface runoff.
1.4.8 Urban
Road erosion is a common problem in the Upper Tana (Figure 42). Other problems in urban areas include
water pollution, poor sanitation and forest degradation for the use of fuel wood. Although these issues are
related to land and water management, the solutions are the responsibility of public administration agencies.
Some recommendations for improvement could be to use rainwater roof harvesting systems, which would
reduce surface runoff in villages and towns and therefore decrease road erosion and improve fresh water
quantities for household water use (Centre for Science and Environment 2009
14
; UNEP/SEI 2009).
14
Centre for Science and Environment 2009: Rainwaterharvesting.org
Green Water Credits Report 9 65
Figure 42
Road erosion occurs on many roads in the Upper Tana catchment
1.4.9 River bank and wetland management
Another problem is wetland or riverbank cropping. Although forbidden by law, agricultural fields can often be
found on sides of the river. The natural ecosystem is hereby disturbed; soil flushes away from the land into
the river and evaporation is increased (Figure 43).
66 Green Water Credits Report 9
Figure 43
Farming on the riverbank
Green Water Credits Report 9 67
2. Conclusions: possible solutions to land
and water related problems in
the Upper Tana
The field observations made in May 2009 confirm that man induced accelerated erosion is a serious issue in
the Upper Tana catchment. The main contributing factors to accelerated erosion are:
– Inappropriate soil and water conservation in farmland, in particular in maize and coffee fields;
– Soil type;
– Inappropriate rangeland management; and
– Road erosion.
The following actions are proposed to improve the use of land and water in the Upper Tana and move towards
sustainable land use:
1. Application of appropriate
green water
management techniques on agricultural land;
2. Riverbank farming and wetland cropping should be put to a halt by giving farmers other options;
3. The main land use types Green Water Credits should focus on are maize (and beans) and coffee;
4. Enhance the use of fertilizer or manure; and
5. Irrigation should be used effectively by for example drip-irrigation
68 Green Water Credits Report 9
Green Water Credits Report 9 69
Annex 2 Pair Separation (least to most),
ML ROIs
– Coffee 591 points and Maize 1260 points - 1.54579436
– Urban 2889 points and Maize 1260 points - 1.78828290
– Maize 1260 points and Rangeland 7251 points - 1.80152150
– Rangeland 7251 points and Bare land/degraded land 1365 points - 1.81590637
– Rangeland 7251 points and Rainfed agriculture in dry areas - bare land - red soil 741 points - 1.86642367
– Bare land/degraded land 1365 points and Rainfed agriculture in dry areas - bare land - red soil 741 points -
1.86680009
– Urban 2889 points and Rangeland 7251 points - 1.87137657
– Coffee 591 points and Tea 700 points - 1.87364291
– Coffee 591 points and Natural Forest 3856 points - 1.88031517
– Maize 1260 points and Rainfed agriculture in dry areas - bare land - red soil 741 points - 1.88718130
– Tea 700 points and Natural Forest 3856 points - 1.94957050
– Maize 1260 points and Bare land/degraded land 1365 points - 1.95190155
– Urban 2889 points and Coffee 591 points - 1.95370203
– Coffee 591 points and Rangeland 7251 points - 1.95476560
– Rangeland 7251 points and Mango trees 227 points - 1.95561881
– Urban 2889 points and Bare land/degraded land 1365 points - 1.96070334
– Rice 2731 points and Water 1115 points - 1.96337093
– Natural Forest 3856 points and Coniferous trees 1178 points - 1.96817620
– Natural Forest 3856 points and Rice 2731 points - 1.97424687
– Rangeland 7251 points and Natural Forest 3856 points - 1.97506193
– Rangeland 7251 points and Rice 2731 points - 1.97592320
– Urban 2889 points and Rainfed agriculture in dry areas - bare land - red soil 741 points - 1.97609383
– Maize 1260 points and Rainfed agriculture on black soils - mostly maize 1390 points - 1.97630207
– Rangeland 7251 points and Rainfed agriculture on black soils - mostly maize 1390 points - 1.97933796
– Maize 1260 points and Rice 2731 points - 1.98023047
– Coffee 591 points and Rice 2731 points - 1.98254014
– Maize 1260 points and Natural Forest 3856 points - 1.98385683
– Urban 2889 points and Rainfed agriculture on black soils - mostly maize 1390 points - 1.98532323
– Mango trees 227 points and Bare land/degraded land 1365 points - 1.98629578
– Coffee 591 points and Rainfed agriculture in dry areas - bare land - red soil 741 points - 1.98749295
– Rainfed agriculture on black soils - mostly maize 1390 points and Bare land/degraded land 1365 points -
1.98775091
– Maize 1260 points and Mango trees 227 points - 1.98846554
– Mango trees 227 points and Rice 2731 points - 1.98906748
– Coffee 591 points and Bare land/degraded land 1365 points - 1.98985852
– 3856 points and Water 1115 points - 1.99163319
– Urban 2889 points and Rice 2731 points - 1.99279970
– Tea 700 points and Rice 2731 points - 1.99302101
– R
ice 2731 points and Bare land/degraded land 1365 points - 1.99413291
– Rainfed agriculture on black soils - mostly maize 1390 points and Rainfed agriculture in dry areas - bare land -
red soil 741 points - 1.99450236
70 Green Water Credits Report 9
– Maize 1260 points and Water 1115 points - 1.99491613
– Urban 2889 points and Mango trees 227 points - 1.99506176
– Mango trees 227 points and Rainfed agriculture in dry areas - bare land - red soil 741 points - 1.99533917
– Tea 700 points and Maize 1260 points - 1.99599632
– Natural Forest 3856 points and Moorland 1051 points - 1.99604556
– Urban 2889 points and Natural Forest 3856 points - 1.99636447
– Coffee 591 points and Water 1115 points - 1.99641462
– Natural Forest 3856 points and Bare land/degraded land 1365 points - 1.99755822
– Rangeland 7251 points and Water 1115 points - 1.99785226
– Tea 700 points and Rangeland 7251 points - 1.99790363
– Water 1115 points and Bare land/degraded land 1365 points - 1.99810366
– Coffee 591 points and Mango trees 227 points - 1.99833145
– Urban 2889 points and Water 1115 points - 1.99876714
– Natural Forest 3856 points and Rainfed agriculture on black soils - mostly maize 1390 points - 1.99877772
– Rice 2731 points and Rainfed agriculture in dry areas - bare land - red soil 741 points - 1.99884850
– Coffee 591 points and Rainfed agriculture on black soils - mostly maize 1390 points - 1.99914738
– Rice 2731 points and Rainfed agriculture on black soils - mostly maize 1390 points - 1.99927451
– Maize 1260 points and Potatoes 59 points - 1.99932099
– Mango trees 227 points and Natural Forest 3856 points - 1.99938088
– Urban 2889 points and Tea 700 points - 1.99940145
– Rangeland 7251 points and Potatoes 59 points - 1.99955422
– Urban 2889 points and Potatoes 59 points - 1.99961317
– Tea 700 points and Bare land/degraded land 1365 points - 1.99969352
– Rainfed agriculture on black soils - mostly maize 1390 points and Water 1115 points - 1.99970369
– Rangeland 7251 points and Coniferous trees 1178 points - 1.99981316
– Mango trees 227 points and Rainfed agriculture on black soils - mostly maize 1390 points - 1.99989842
– Tea 700 points and Water 1115 points - 1.99990114
– Mango trees 227 points and Water 1115 points - 1.99992322
– Water 1115 points and Rainfed agriculture in dry areas - bare land - red soil 741 points - 1.99993516
– Natural Forest 3856 points and Rainfed agriculture in dry areas - bare land - red soil 741 points - 1.99994256
– Coffee 591 points and Potatoes 59 points - 1.99995575
– Potatoes 59 points and Bare land/degraded land 1365 points - 1.99996143
– Tea 700 points and Moorland 1051 points - 1.99997784
– Coffee 591 points and Moorland 1051 points - 1.99997899
– Coniferous trees 1178 points and Moorland 1051 points - 1.99998353
– Coffee 591 points and Coniferous trees 1178 points - 1.99998920
– Coniferous trees 1178 points and Water 1115 points - 1.99999025
– Tea 700 points and Rainfed agriculture in dry areas - bare land - red soil 741 points - 1.99999026
– Rangeland 7251 points and Moorland 1051 points - 1.99999107
– Potatoes 59 points and Rainfed agriculture in dry areas - bare land - red soil 741 points - 1.99999417
– Tea 700 points and Mango trees 227 points - 1.99999423
– Coniferous trees 1178 points and Bare land/degraded land 1365 points - 1.99999477
– Potatoes 59 points and Natural Forest 3856 points - 1.99999899
– Bare land/degraded land 1365 points and Moorland 1051 points - 1.99999907
– Tea 700 points and Potatoes 59 points - 1.99999934
– Rice 2731 points and Coniferous trees 1178 points - 1.99999941
– Urban 2889 points and Coniferous trees 1178 points - 1.99999952
– Maize 1260 points and Coniferous trees 1178 points - 1.99999988
– Potatoes 59 points and Rainfed agriculture on black soils - mostly maize 1390 points - 1.99999992
– Tea 700 points and Rainfed agriculture on black soils - mostly maize 1390 points - 1.99999994
Green Water Credits Report 9 71
– Tea 700 points and Coniferous trees 1178 points - 1.99999997
– Water 1115 points and Moorland 1051 points - 1.99999998
– Potatoes 59 points and Rice 2731 points - 1.99999999
– Mango trees 227 points and Potatoes 59 points - 1.99999999
– Potatoes 59 points and Water 1115 points - 2.0
– Maize 1260 points and Moorland 1051 points - 2.0
– Urban 2889 points and Moorland 1051 points - 2.0
– Cloud 704 points and Bare land/degraded land 1365 points - 2.0
– Cloud 704 points and Water 1115 points - 2.0
– Coniferous trees 1178 points and Rainfed agriculture in dry areas - bare land - red soil 741 points - 2.0
– Potatoes 59 points and Coniferous trees 1178 points - 2.0
– Mango trees 227 points and Coniferous trees 1178 points - 2.0
– Cloud 704 points and Rainfed agriculture on black soils - mostly maize 1390 points - 2.0
– Cloud 704 points and Coniferous trees 1178 points - 2.0
– Cloud 704 points and Coffee 591 points - 2.0
– Rainfed agriculture on black soils - mostly maize 1390 points and Moorland 1051 points - 2.0
– Cloud 704 points and Mango trees 227 points - 2.0
– Cloud 704 points and Moorland 1051 points - 2.0
– Potatoes 59 points and Moorland 1051 points - 2.0
– Mango trees 227 points and Moorland 1051 points - 2.0
– Cloud 704 points and Rangeland 7251 points - 2.0
– Cloud 704 points and Natural Forest 3856 points - 2.0
– Cloud 704 points and Maize 1260 points - 2.0
– Rice 2731 points and Moorland 1051 points - 2.0
– Cloud 704 points and Potatoes 59 points - 2.0
– Cloud 704 points and Rice 2731 points - 2.0
– Coniferous trees 1178 points and Rainfed agriculture on black soils - mostly maize 1390 points - 2.0
– Cloud 704 points and Urban 2889 points - 2.0
– Moorland 1051 points and Rainfed agriculture in dry areas - bare land - red soil 741 points - 2.0
– Cloud 704 points and Rainfed agriculture in dry areas - bare land - red soil 741 points - 2.0
– Cloud 704 points and Tea 700 points - 2.0
72 Green Water Credits Report 9
Green Water Credits Report 9 73
GWC Reports Kenya
GWC K1
Basin identification
Droogers P and others 2006
GWC
K2
Lessons learned from payments for environmental services
Grieg Gran M and others 2006
GWC
K3
Green and blue water resources and assessment of improved
soil and water management scenarios using an integrated
modelling framework.
Kauffman JH and
others 2007
GWC
K4
Quantifying water usage and demand in the Tana River basin:
an analysis using the Water and Evaluation and Planning Tool
(WEAP)
Hoff H and Noel S 2007
GWC
K5
Farmers' adoption of soil and water conservation: the potential
role of payments for watershed services
Porras I
T and others 2007
GWC
K6
Political, institutional and financial framework for Green Water
Credits in Kenya
Meijerink G
W and others 2007
GWC
K7
The spark has jumped the gap. Green Water Credits proof of
concept
Dent DDL
and Kauffman JH 2007
GWC
K8
Baseline Review of the Upper Tana, Kenya
Geertsma R, Wilschut LI and
Kauffman JH
2009
GWC
K9
Land Use Map of the Upper Tana, Kenya:
Based on Remote Sensing
Wilschut
LI 2010
GWC K10
Impacts of Land Management Options in the Upper Tana,
Kenya:
Using the Soil and Water Assessment Tool - SWAT
Hunink
JE, Immerzeel WW,
Droogers P, Kauffman JH
and
van Lynden GWJ
2011
GWC K11
Soil and Terrain Database for the Upper Tana, Kenya
Dijkshoorn JA,
Macharia PN,
Huting JRM, Maingi PM and
Njoroge CRK 2010
GWC K12
Inventory and Analysis of Existing Soil and Water Conservation
Practices in the Upper Tana, Kenya
Muriuki
JP and Macharia PN
2011
GWC K13
Estimating Changes in Soil Organic Carbon in the Upper Tana,
Kenya
Batjes NH 2011
GWC K14
Costs and Benefits of Land Management Options in the Upper
Tana, Kenya:
Using the Water Evaluation And Planning system - WEAP
Droogers P, Hunink J
E,
Kauffman
JH and
van Lynden GWJ 2011
GWC K15
Cost-Benefit Analysis of Land Management Options in the
Upper Tana, Kenya
Onduru DD and Muchena FN
2011
GWC K16
Institutes for Implementation of Green Water Credits in the
Upper Tana, Kenya
Muchena FN and Onduru DD
2011
GWC
K17
Analysis of Financial Mechanisms for Green Water Credits in
the Upper Tana, Kenya
Muchena FN, Onduru DD
and
Kauffman JH
2011
74 Green Water Credits Report 9
ISRIC - World Soil Information
Ministry of Agriculture
Water Resources Management Authority
Kenya Agricultural Research Institute
Ministry of Water and Irrigation
International Fund for Agricultural Development
Future Water
ISRIC – World Soil Information has a mandate to serve the international community as custodian of
global soil information and to increase awareness and understanding of soils in major global issues.
More information: www.isric.org
ISRIC – World soil Information has a strategic association
with Wageningen UR (University & Research centre)
Green Water Credits Report 9
L.I. Wilschut
Land Use Map of the Upper Tana, Kenya
Based on remote sensing
