Remote sensing analysis of the Land Use and Land Cover change in the Senegal river valley

(1)

(2)

Abstract

In 1986 and 1988 the construction of the Diama and the Manantali dams where finished. These dams located on the Senegal river where built by the OMVS (Organisation de Mise en Valeur du fleuve Sénegal). One of the purposes of the Manantali dam is making irrigation possible by dampening the inter-annual fluctuations in the water level of the River, the main focus of the Diama dam is blocking saltwater intrusion. The aim of this research is to analyse the change in land use and land cover change from before the dam was built until now. This has been done by computing a remote sensing analysis of Landsat 5 and Sentinel 2 imagery from 1986, 1994 and 2018. All the imagery data that has been used was collected during the dry season to be able to classify the land use and land cover as good as possible. In an accuracy assessment, it was concluded that the accuracy of the separate classifications in each of the three different time series was high. When comparing the three different time series, several trends in land use and land cover change are visible. From 1986 until now there has been a decrease in forest density but an increase in total amount of trees, due to a large increase in open areas with trees. Furthermore, there has been a rather small increase in irrigated agriculture, especially compared to the increase that was expected by the OMVS. Beside these trend, an explorational research on classification methods has been conducted. High resolution imagery, for instance Sentinel 2 should be classified using object based classification in eCognition to achieve the highest accuracy. Lower resolution imagery on the other hand, such as Landsat 5 imagery gets classified more accurate using pixel based classification in Erdas Imagine.

(3)

Content

Abstract...2 Introduction...4 Results...14 Discussion...22 Conclusion...24 Literature list...25 Acknowledgements...26 Appendix...27 Accuracy assessments...28

(4)

Introduction

Physiography of the area

The Senegal River originates in the Futa Jalon hills located in Guinea (Figure 1.). After flowing north into increasingly drier lands, the River is turning into a western direction and flows towards the Atlantic Ocean. As the surrounding lands shift to more arid states, the life in the River delta becomes more depended on the river (Adams, 2000). The international organisation that manages the Senegal River is the OMVS (Organisation de Mise en Valeur du fleuve Sénegal). This organisation, consisting of members from Mali, Mauretania and Senegal was founded in 1972 (Venema et al., 1997). The Senegal River basin consists of three parts, the upper basin, the valley and the delta. The total area of the basin is 289,000 km² (Uhlrir, 2003). This research will be focussing on the part of the valley between Bakel and Podor. This has been decided in discussion with dr. I. S. Mettrop and is based on his field visit and knowledge about the area.

Figure 1. Overview of the Senegal River basin. Modified from

(5)

Land use and Land cover in the past

Historically there were two agricultural methods that were used in the valley; rainfed agriculture on the Jeeri uplands and flood recession agriculture in the Waalo lowlands. In both areas there is also pasture for cattle (Adams, 2000). Estimations of the OMVS state that about 300,000 hectares where flooded in the valley every year between 1946 and 1971. Of these 300,000 hectares, 108,000 were cultivated (OMVS-IRD, 1999).

The flow rate of the Senegal River hinges largely on the precipitation that falls in the upper basin area. Floods occur during 4 months each year, from July up until October, and from November until June the water level is decreasing steadily. The discharge levels are the highest from the end of August until the beginning of September (Uhlrir, 2003). There is a significant difference in discharge levels between the dry and the wet season. In the flood season discharges at Bakel peak around 2000 m3/s at Bakel, in the dry season the lowest discharge is usually less than 40 m³/s.

Construction of the dams

The flood-recession based agriculture in the lowlands is bounded to a few months a year. Irrigated agriculture can be practiced the whole year, which is one of the reasons plans were made by the OMVS to build two dams in the Senegal River (Adams, 2000).

The first dam is the Manantali dam, located in Mali. The purpose of this dam was to retain the waters of the import tributary river the Bafing. This would make it possible to regulate about the half of the total regime of the Senegal River (Venema et al.,1997). The other dam, the Diama, is located at the mouth of the River. The main function of this dam is to stop the intrusion of salt water. The construction of these dams had three other reasons, namely irrigation, navigation and energy production (Adams, 2000). Both dams are visualised in (Figure 1).

Firstly, irrigation was made possible by dampening the inter-annual fluctuations of the River. Secondly, by providing a more stable regime, there is a higher potential for guaranteed agriculture production (Uhlrir, 2003). This dampening also makes it possible to navigate throughout the whole year between Saint-Louis and Kayes. Furthermore, there is hydro-electric energy production in the Manantali dam (Adams, 2000). The Manantali dam is expected to be more relevant for this research because it has a more direct influence on the water availability in the valley. The objectives of the programme were quite ambitious: The OMVS planned to irrigate 3000,000 to 400,000 hectares of land. These lands should then be able to yield two crops a year (Adams, 2000). Majorly there would be a cultivation of rice and wheat. The shift from sorghum and maize to these new crops was expected to take about 20 years. In these 20 years the floods would be manually sustained but gradually decrease so that the traditional farming had time to shift to irrigated agriculture. Expectations where that the whole population would use irrigation agriculture at the end of this 20 years period (PNUD-OMVS, 1974). The building of the Manantali dam started in 1982 and the dam was finished in 1988 (Uhlrir, 2003). There were many organisations that warned about possible effects that the construction of the dam could have. Their concerns about the negative impact on the local communities were ignored. Therefore the people the project was intended for, the inhabitants of the valley, where not included in the development of the project (Adams, 2000).

(6)

A few years after construction of the dams a social crisis occurred. This was because the governments stopped subsidising local farmers and start promoting the private agricultural sector. Therefore a large gap emerged between farmers with means and farmers that where depended on the subsidies. This created problems for access to land throughout the whole valley. In the late 1980’s there where two options for farmers to secure a living; firstly to emigrate, secondly to start working as workers in the blooming private agricultural sector. This was partly because the OMVS did not keep their promises regarding manual floods. So instead of providing a more secure and stable agricultural system, the opposite happened (Adams, 2000). The agricultural sector was not the only one affected by the dams. The dams led to a decrease in the number of ecosystems in the valley (Venema et al.,1997). All the systems, both eco- and production systems, are coping with lower production due to poor resource management (Adams, 2000). The new conditions are favourable for monocultures such as sugar cane. Furthermore they lead to plagues of aquatic weed, and waterborne diseases occur more often (Caro et al., 2011). This is because, to some extent, the diseases where held in by the dry season.

To assess what the consequences of the dams on the land use and land cover is, a time series analysis will be made. The land use and land cover will be classified for three periods: in 1986 before the construction of the dams, in 1994 after the construction of the dams and in 2018 to see the current

Relevance

This research will be computed in collaboration with dr. I. S. Mettrop form Altenburg & Wymenga, which is a ecological consultancy. They came to my supervisor dr. A. C.

Seijmonsbergen whether it was possible if land use and land cover maps could be made of the Senegal River valley. Altenburg & Wymenga is hired by the OMVS to assess the future impacts of climate change on the water management in the Senegal River Valley.

It is of great importance that more information about the valley and its vegetation comes available, because the Senegal River is vital for life in the dry Sahel area. The currently most important trend in land cover and land use is, according to (Caro et al., 2011), the fast spreading of the invasive species Typha Australis. Almost all shallow waters around the Senegal river are filled with this invasive species. The cause of this widespread plague, is the more constant water levels in the Senegal River, created by the dams (Caro et al., 2011). This is a problem because the Typha is blocking irrigation channels, which makes the River less accessible for the local communities, and also makes the water stagnate which creates conditions in which pests and diseases can thrive (Elbersen W., 2005).

This research can contribute to solutions for these problems with this problem by providing vegetation maps of the region that are not currently present. A detailed inventory of the spatial distribution of Typha will result in more insight in its spread. This is needed to

successfully cope with the problem of Typha according to dr. I. S. Mettrop from Altenburg & Wymenga. Other trends in vegetation such as deforestation are also important to assess since the area is very dry and therefore vulnerable for desertification. The vegetation cover could become even more important to the area if agroforestry becomes more important. Agroforestry could be a solution to deforestation in the Senegal River valley (Venema et al.,1997). Besides the fact that it could contribute to the solution of the problem of migration, caused by poor agricultural yields, it could potentially contribute to a better water resource management, because of a better water retention (Venema et al.,1997).

(7)

The goal of this research is to spatially analyse the vegetation dynamics in the Senegal River valley between Bakel and Podor. It will be determined whether there are trends in the area change of the different land use and land cover classes. This research will focus on the influence of the Diama and Manantali dam on the land use and land cover.

The questions that will be answered by this research is:

 How did the Land use and land cover of the Senegal river valley change over the last 32 years?

The following sub questions will be used to answer this research question.

 How did the invasive species Typha Australis spread after construction of the 1986 dam in the Senegal River valley?

 Is there a quantifiable trend of deforestation in the Senegal River valley?

 Is it possible to make a land cover land use change analysis based on imagery of both Landsat 5 and Sentinel 2a?

(8)

Methods

Classification of Landsat 5 Imagery

First, all the imagery was downloaded from the USGS data hub. From the separate images the spectral bands 1 (blue), 2(green), 3 (red) and 4(NIR) were stacked in ERDAS IMAGINE All the satellite imageries used here are taken in the period of the year where there are low water levels and can be found in the metadata tables. The used satellite images are

visualized in (table 1, 2 and 3). Due to the fact the images where taken on different dates for the Landsat 5 data no color balancing or mosaicking was performed. Next, these images were transported to ArcMap. There a cut out of the area that is affected by the river was made, and the images were cut to the area that was affected by the river. Afterwards the clipped image was transported to Erdas Imagine. In this program the satellite images were classified using supervised classification. For this classification, first training sites were identified, at least 10 per category. Based on these training sites Erdas Imagine classifies the whole area.

Figure 2. Workflow Classification of Landsat 5 imagery

Preprocessing

Download imagery from USGS data hub

Stack layers 1,2,3,4

Clip image in ArcMap

Analysis

Supervised classification in Erdas Imagine

Accuracy assessment (lower than 75% new classification)

Pixel smoothening

Deliverables

Lay out map

Calculate area

(9)

Figure 3. Overview map of the used Landsat 5 satellite images

1986

This year was one of the two years where the imagery was available from before the construction of the Manantali dam. The classes that are used for the classifications are:

 Bare soil  Forest  Agriculture  Flooded area  Water

These classes have been classified for all the four satellite images. In the class flooded area, a few of the classified areas contain Typha Australis.

1994

The classes that are used for the classifications are:  Bare soil

 Forest  Agriculture

 Flooded area with Typha  Water

These classes have been classified for all the four satellite images, with two exceptions. For the maps of 202/50 and 203/49, there has been made the decision that agriculture is left out. This is because of the fact that there are very few agricultural fields in those areas. When including agriculture it leads to a severe reduction of the accuracy of these classifications. For the class Flooded area with Typha areas that showed very little or no Typha Australis, but where solely flooded are also included.

(10)

Classification of Sentinel 2 2018

First, sentinel images were downloaded from the ESA online data hub. From the separate images the spectral bands 2 (blue), 3 (green), 4 (red) and 8 (NIR) were stacked in ERDAS IMAGINE. Afterwards color corrections where made using image dodging and the images where mosaicked. Afterwards the image was transported to ArcMap. There a cut out of the area that is affected by the river was made, and the image was cut to the area that was affected by the river.Then, field observation classification points of dr. I. S. Mettrop where converted and the locations loaded into ArcMap from google earth. These points where used to help identifying the different land use and land cover classes.

Afterwards the clipped image was transported to eCognition. In the software eCognition the clipped and mosaiced Sentinel imagery was classified using object classification. A spectral difference segmentation of 300 nm was used, which created polygons based on shape and color value. In these polygons separate land cover training sites were identified partly based on the field observation points. These training sites where used for the classification of the whole map. The different Land use and Land cover classes were reduced to the following classes:

 Water

 Typha

 Irrigated agriculture type

 Forests

 Open areas with trees

 Barren soil

 Unclassified

These are the classes that can be separated most accurately from the satellite imagery, and therefore give the most accurate representation of the LULC in the region. Flooded area is not identified as separate class, because it was not possible to differentiate flooded areas from open areas with trees and areas with forest.

Figure 4. Workflow Classification of Sentinel 2 imagery

Accuracy assessment in ArcGis

For each classified map an accuracy assessment was made in ArcMap. This was preformed to check the quality of the classification. First, stratified random points where produced as sample points, which means that the number of points is proportional to the relative area of each class. The total number of points was based on the classification made in Sentinel which contains 7 categories. The rule of thumb is to use at least 50 sample points per

Preprocessing

Download imagery from ESA data hub Stack layers 2,3,4,8

Colorbalancing and mosaicking Clip image in ArcMap

Analysis

Object based classification in eCognition

Accuracy assessment (lower than 75% new classification) Pixel smoothening

Deliverables

Lay out map Calculate area

(11)

category (Congalton, 1991). Which lead to the total number of points of 350. For all these points it was manually checked whether the classification was correct. If the classification of Erdas Imagine or eCogntion was correct a value of 1 was assigned to that point If this was not the case, a value of 0 was assigned followed by the correct classification. After this accuracy assessment was performed, an analysis was computed in Excel ,based on the methods of (Congalton, 1991). The total accuracy, the producer accuracy and the users accuracy were calculated. The total accuracy is calculated as the percentage of correct classified points divided by the total number of points. The producers accuracy is the number of correct classified points divided by the total number of points classified as that specific class. The users accuracy is calculated by correct classifications per category divided by the total number of points classified in that specific class.

Pixel smoothening

After each classification was finished and had a accuracy of 75% or higher, pixel

smoothening was performed. This deletes the value of separate pixels and assign the value of the most present neighbor class to it.

Area calculation

The area of the land use and land cover classes was calculated by multiplying the number of pixels of the classified class by the grid size. For the Sentinel imagery this meant,

pixels*10*10 /1000000= area in km². For the Landsat 5 imagery the grid size is 30 m so the formula, pixels*30*30/1000000= area in km².

(12)

Table 1. Landsat 5 imagery 1986

Landsat 5 (TM) Imagery Spectral bands Coordinate system Scale/c ell size Date Source LT05_L1TP_202050_1986010 3_20170218_01_T1 1 - Blue2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30

03-01-1986 https://landlook.usgs.gov/viewer.html (USGS)

LT05_L1TP_203049_1986112 6_20170215_01_T1 1 - Blue 2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30 15-03-1986 https://landlook.usgs.gov/ viewer.html (USGS) LT05_L1TP_204048_1986040 7_20170218_01_T1 1 - Blue2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30

LT05_L1TP_204049_1986040 7_20170218_01_T1 1 - Blue2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30

Table 1. Landsat 5 imagery 1994

Landsat 5 (TM) Imagery Spectral bands Coordinate system Scale/c ell size Date Source LT05_L1TP_202050_1994 0501_20170114_01_T1 1 - Blue 2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30

01-05-1994 https://land22look.usgs.gov/viewer.html (USGS)

LT05_L1TP_203049_1994042 2_20170115_01_T2 1 - Blue2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30

LT05_L1GS_204049_1994041 3_20170114_01_T2 1 - Blue 2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30 13-04-1994 https://landlook.usgs.gov/ viewer.html (USGS) LT05_L1GS_204049_1994041 3_20170114_01_T2 1 - Blue2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30

(13)

Table 3. Sentinel 2 imagery 2018 Sentinel-2A Imagery Spectral bands Coordinate system Scale/ cell size Date Source S2A_MSIL1C_ 20180510T112 121_N0206_R 037_T28PFC_ 20180510T150 520 2 - Blue 3 - Green 4 - Red 8 - NIR UTM, Zone 28 WGS 84 10x10 10-05-2018 https://scihub.coperni cus.eu/dhus/#/home (esa) S2A_MSIL1C_ 20180510T112 121_N0206_R 037_T28PEC_ 20180510T150 520 2 - Blue 3 - Green 4 - Red 8 - NIR UTM, Zone 28 WGS 84 10x10 10-05-2018 https://scihub.coperni cus.eu/dhus/#/home (esa) S2A_MSIL1C_ 20180510T112 121_N0206_R 037_T28QED_ 20180510T150 520 2 - Blue 3 - Green 4 - Red 8 - NIR UTM, Zone 28 WGS 84 10x10 10-05-2018 https://scihub.coperni cus.eu/dhus/#/home (esa) S2A_MSIL1C_ 20180510T112 121_N0206_R 037_T28QFD_ 20180510T150 520 2 - Blue 3 - Green 4 - Red 8 - NIR UTM, Zone 28 WGS 84 10x10 10-05-2018 https://scihub.coperni cus.eu/dhus/#/home (esa) S2B_MSIL1C_ 20180512T110 619_N0206_R 137_T28PGC_ 20180512T170 024 2 - Blue 3 - Green 4 - Red 8 - NIR UTM, Zone 28 WGS 84 10x10 12-05-2018 https://scihub.coperni cus.eu/dhus/#/home (esa) S2B_MSIL1C_ 20180512T110 619_N0206_R 137_T28PGB_ 20180512T170 024 2 - Blue 3 - Green 4 - Red 8 - NIR UTM, Zone 28 WGS 84 10x10 12-05-2018 https://scihub.coperni cus.eu/dhus/#/home (esa)

(14)

Results

Figure 5. Lands use and land

cover map

(15)

Figure 6. Zoomed classification compared (top) to

satellite image (bottom)

(16)

Figure 7. Lands use and land

cover map

(17)

Figure 8. Zoomed classification compared (top) to

satellite image (bottom)

(18)

t

Figure 9. Lands use and land

cover map

(19)

Figure 10. Zoomed classification compared (top)

to satellite image (bottom)

(20)

Table 4. Land use and land cover areas 1986,1994 and 2018

Classname

1986

1994

2018

Flooded area

1662

1066

0 km²

Forest and

vegetation

652 1342

214 km²

Bare soil

4312

4353

2494 km²

Water

195

197 234 km²

Agriculture

207

58 253 km²

Typha and flooded

vegetation

0

0 100 km²

Open area with trees

0

0 3732 km²

Table 5. Land use and land cover change between 1986 and 1994

Classname

Area change (km2)

Change (%)

Flooded area

596 -36%

Forest and vegetation

690 106%

Bare soil

41 1%

Water

2 1%

Agriculture

149 -72%

Typha and flooded vegetation

0 0%

Open area with trees

0 0%

Classname

Area change (km2)

Change (%)

Land use and land cover areas

(21)

Flooded area

1066

100%

Forest and vegetation

1128

-88%

Bare soil

1859

-43%

Water

37 19%

Agriculture

195 336%

Typha and flooded vegetation

100 100%

Open area with trees

3732

100%

Classname

Area change (km2)

Change (%)

Flooded area

1662

-100%

Forest and vegetation

438 -67%

Bare soil

1818

-42%

Water

39 20%

Agriculture

46 22%

Typha and flooded vegetation

100 100%

Open area with trees

3732

100%

Discussion

Data

The data used in this research consists of two different satellite imageries, Landsat 5 data and Sentinel 2 images. By performing this research it has been tested whether it was possible to use two

Change 1994 - 2018

(22)

images have a higher resolution. The Sentinel has a grid size of 10 meter by 10 meter, the Landsat 5 TM has a grid size of 30 m by 30 m. For the Sentinel data the classification has been made using eCognition instead of Erdas Imagine. This is because of the combination of the high resolution and the lack of difference in color between agriculture and forest. The classification in Erdas Imagine could not see difference between Agricultural field and forests. The classification method of eCognition is different since it is not only pixel based. The classification method in eCognition is an object based classification. First polygons are made based on spectral differences. A level of 350 nm was used to create these polygons. The difference with the classification method used in Erdas Imagine is that this classification is not based on pixel color but also on shape. This method gives a better classification for imagery with a high resolution. Therefore the classification of the Sentinel imagery has been performed using object based classification in eCognition. For the classification of the Landsat 5 imageries for 1986 and 1994, the eCognition classification was of a lower quality then the Erdas Imagine classification. The conclusion that object based classification only improves the quality of high resolution classification has also been made by Yan et al. (Yan et al., 2006).

Land use and land cover classes

Due to these differences, the classes that have been classified for each year, vary. For the 1986 and 1994 classifications the following classes has been classified:

 Bare soil  Forest  Agriculture

 Flooded area with Typha  Water

For the 2018 classification, the categories are:  Water

 Typha

 Open areas with trees  Irrigated agriculture  Forests

 Bare soil

The differences between those classes is that Flooded area with Typha has been changed into Typha since it was possible to classify Typha as a separate class with the higher resolution data. The areas that where only flooded has been divided over Bare soil when no trees where presents, or open areas with trees. The last category is a class that is just present in the 2018 data because of two reasons. Firstly, more trees are present in the valley. Secondly, they are easier to identify on the Sentinel 2 data.

The differences in classes have led to high accuracy assessment percentages, since the classification had high accuracies. The disadvantage of these different categories is that the area cover change is less useful.

Land use and land cover trends

One of the trends that the data shows, is a decrease in flooded area of 596 km2 between 1986 and 1994. This is not because of lower water levels during that period, because the water levels in 1994 were higher than in 1986 based on the water level and discharge data provided by (Mettrop, 2018) (appendix figure 11). When looking at specific areas that used to be flooded, many forests have occurred on former flooded areas. It is not possible to check whether this is a long term trend due to the fact that flooded area could not be classified on the Sentinel 2 classification.

Another interesting trend that the data show is a 72% decrease in agriculture. Although agriculture was not a category for almost half of the classified 1994 map, but in these areas there was almost no agriculture present before construction of the dam. Therefore the decrease shows that on the short term the goal of the OMVS to increase irrigated agriculture in the area has not been achieved.

(23)

Between 1994 and 2018 there was an 336 % increase in irrigated agriculture, meaning an increase of 46 km² between 1986 and 2018. Thus, there has been a small increase in irrigated agriculture on the long term, but the goals of the OMVS have not been achieved

Furthermore, it can be seen that between 1986 and 1994 the amount of, the area that has been classified as bare soil has not changed much. Between 1994 and 2018 there is a 43% decrease, this is mainly because of the large increase in open area with trees.

When looking at the mapping of Typha Australis, the first conclusion that can be made regarding the classes is that it is hard to quantify a potential increase in Typha. There are clearly some areas visible where there has been a vast increase in Typha cover, mainly downstream in the valley. In the higher parts of the valley there are areas with still water where Typha vegetation has not yet spread. Overall it can be concluded that there is a visible increase in Typha Australis cover, but due to the differences in classes used for the vegetation it is not possible to quantify this increase. A possible way to

distinguish flooded area and Typha could be the Normalized Difference Water Index, for Sentinel imagery it is possible to make distinguish water and Typha (Rentier, 2018). It would be interesting to check whether this distinction can also be made for Landsat 5 imagery.

In contrast to the Typha Australis, it is possible to quantify trends in forest vegetation cover.

Beforehand, the hypothesis was that there has been a decline in the area covered with forest. When looking at the classifications that where made the following has changed in forest vegetation cover. To conclude, some trends can be seen using different classification programs and methods, and it increases the quality of the separate classifications. But, solely for land use and land cover analysis, it is not ideal to use data with different grid sizes Therefore it is recommended that the newest data from Landsat 8 are compared to the Landsat 5 data from 1986 and 1994 because they have the same resolution. Potentially the higher quality of the Landsat 8 imagery could lead to other problems with classification although these are expected to be smaller due to having the same grid size

As you can see in table 5 the Forest cover has increased with 106% between 1986 and 1994, which means that on the dam have had a very positive influence on the forests in the area in the first years after construction. This could possibly be caused by the fact that the water levels are more constant since construction of the dam. This could favor the growing circumstances for the trees in the valley. More research has to be conducted to reach stronger conclusions on this. When looking at the change between 1994 and 2018 a whole different trend is visible. The forest decreases with 88% in this period, simultaneously a new category has been classified, open area with trees. This new categories covers a huge area of 3732 km². What can be concluded here is that the forest density has decreased, but the total amount of trees has increased over time. The decrease in density is probably due to logging. So overall the dams seem to have had a positive impact on the trees in the valley.

Conclusion

The computed research has led to multiple different conclusions and findings. In regard to

the question from dr I. S. Mettrop from Altenburg & Wymenga if it was possible to make a

good land use and land cover map of 2018, the conclusion can be made that it is possible to

make a reliable land use and land cover map based on satellite imagery. For the 1986 and

1994 the accuracy of the assessments is even higher. The 2018 classification is based on

Sentinel 2 imagery in contrast to this the classifications of 1986 and 1994 that is based on

Landsat 5 imagery

However, the use of these two different satellite types also has its

disadvantages, because of the differences in resolution of the two satellites there are some

differences in the classified classes. Therefore it is not possible to quantify trends in all land

use and land cover types.

The trends that the analysis and interpretation of the resulting maps focused on, were forest,

agriculture and Typha Australis. When looking at how the forested areas have changed over

the years, it can be seen that the forest cover increased with 106% from 1986 to 1994, and

(24)

decreased with 88% between 1994 and 2018. Nevertheless the total amount of trees has

increased due to the Increase in area that has been classified as open area with trees. The

conclusion that can be drawn here is that the density of the forests has decreased but the

total amount of trees has increased.

One of the goals of the construction of the Manantali dam was to greatly increase the

irrigated agriculture in the research area of the Senegal river valley. However, looking at how

the irrigated agriculture has developed throughout the years, the large upscale has not been

achieved. During the first years after construction, from 1986 to 1994 the irrigated area

decreased with 72%.From 1994 and 2018 it increased with 336% but overall the increase was

far less than the goals set by the OMVS.

This research has been a useful exploration for a good method for classifying land use and

land cover in the Senegal river valley and potentially for other areas. This research concluded

that eCognition leads to the best results for the classification of high resolution imagery. On

the other hand, for data with a lower resolution, for instance Landsat imagery, classifications

in Erdas Imagine lead to a better result.

(25)

Literature list

Adams, A. (2000). The Senegal River: Flood management and the future of the valley. International

Institute for Environment and Development, Issue Paper/Drylands Program, (93).

Bossen, R. B. (2018). An evaluation: mapping temporal land use and land cover (LULC) for the Senegal Catchment to quantify the changed area

Caro, R., de Frutos, H., Kitwana, A., & Shen, A. (2011). Typha Charcoal in Senegal: Changing a National Threat into Durable Wealth. Massachusetts Institute of Technology, Technical Report, (15.915)

Congalton, R. G. (1991). A review of assessing the accuracy of classifications of remotely sensed data. Remote sensing of environment, 37(1), 35-46.

Elbersen, W. (2005). Report on bus ticket no. B1.

Google (2018). [Google Earth overview of the Senegal River Valley]. Retrieved June 25, 2018, from Google Earth pro

Mettrop, I. (2018) Jan1980-jan2018 [Microsoft Excel spreadsheet]. Amsterdam Altenburg & Wymenga. Retrieved from Altenburg &Wymenga [accessed June 11, 2018]

OMVS, 1999. Programme d’Atténuation et de Suivi des Impacts sur l’Environnement (PASIE) de la mise en valeur du fleuve Sénégal, definitive version, Dakar, January 1999

PNUD-OMVS, 1974. Programme intégré de développement du bassin du Sénégal, 12 volumes, Norbert Beyrard, Paris, 1974

Rentier, E. S. (2018). The Senegal river valley: mapping land cover, land use and floodplains. Uhlir, P. F., & National Research Council. (2003). Scientific Data for Decision Making Toward

Sustainable Development: Senegal River Basin Case Study: Summary of a Workshop. National

Academies Press

United Nations. (2013). Overview of the Senegal River basin. Retrieved from http://www.un.org/waterforlifedecade/water_cooperation_2013/senegal_River.shtml

Venema, H. D., Schiller, E. J., Adamowski, K., & Thizy, J. M. (1997). A water resources planning response to climate change in the Senegal River Basin. Journal of Environmental Management, 49(1), 125-155

Yan, G., Mas, J. F., Maathuis, B. H. P., Xiangmin, Z., & Van Dijk, P. M. (2006). Comparison of pixel‐ based and object‐oriented image classification approaches—a case study in a coal fire area, Wuda, Inner Mongolia, China. International Journal of Remote Sensing, 27(18), 4039-4055.

(26)

Acknowledgements

I would like to thank my thesis supervisor dr. A. C. Seijmonsbergen of the Institute of Biodiversity and Ecosystem Dynamics for his help and guidance during this research. Furthermore I would like to thank dr. I. S. Mettrop from Altenburg & Wymenga for introducing this subject to me and his supervision on the classifications. Next, I would like to thank E. S. Rentier for the very pleasant cooperation on the Land use and land cover map of 2018. Lastly I would like to thank R. A. L. Bossen a lot for the pleasant cooperation on the Land use and Land cover maps of both 1986 and 2018.

(27)

(28)

Accuracy assessments

Table 8. Accuracy assessment 1986 204048

4 - Forest

6 - Bare soil

7 - Water

14 - Flooded area (typha)

CORRECT

CLASSES

4

6

7 14 Row total

4

7

0

3

10

6

3

76

0

2

81 CLASSIFIE

D

7

0

10

0

10

14

0

1

0

9

10 Column total

10

77

10

14

111 Overall accuracy = 102/11 = 92%

Producers accuracy

Users accuracy

4 - Forest = 7 / 10

70%

4 - Forest = 7 / 10

70%

6 - Bare soil = 76 / 77

99%

6 - Bare soil = 76 / 81

94%

7 - Water = 10 / 10

100 %

7 - Water = 10 / 10

100 %

14 - Flooded area

(typha) = 9 / 14

64%

14 - Flooded area (typha) = 9 / 10

90%

28

4 - Bare soil

_{7 - Water}

8 - Forest and

vegetation

(29)

Overall accuracy = 60/63 = 95%

Producers accuracy

Users accuracy

4 - Bare soil = 29 / 30 =

97%

4 - Bare soil = 29 / 30 =

97%

7 - Water = 13 / 13 =

100%

7 - Water = 13 / 13 =

100%

8 - Forest and vegetation =

11 / 11 =

100%

8 - Forest and vegetation =

11 / 13 =

85%

9 - Flooded area = 7 / 9 =

78%

CORRECT

9 - Flooded area = 7 / 7 =

_100%

CLASSES

4

7

8 9 Row total

4

29

0

1

30

7

0

13

0

13 CLASSIFIE

D

8

1

0

11

1

13

9

0

7

7 Column total

30

13

11

9

63

(30)

8 - Flooded area

13 - Agriculture

15 - Water

17 - Bare soil

19 - Forest and vegetation

CORRECT

CLASSES

8

13

15

17 19 Row total

8

59

0

2

3

64

13

4

0

1

10 CLASSIFIED

15

1

0

8

1

0

10

17

3

0

110

0

113

19

1

0

1

0

13

15 Column total

68

4

9

114

17

212 Overall accuracy = 194/212 = 92%

Producers accuracy

Users accuracy

8 - Flooded area = 59 / 68

87%

8 - Flooded area = 59 / 64

92%

13 - Agriculture = 4 / 4

100%

13 - Agriculture = 4 / 10

40%

15 - Water = 8 / 9

89%

15 - Water = 8 / 10

80%

17 - Bare soil = 110 / 114

96%

17 - Bare soil = 110 / 113

97%

19 - Forest and vegetation

(31)

3 - Agriculture

6 - Bare soil

8 - Forest

13 - Water

14 - Flooded area

CORRECT

CLASSES

3

6

8

13 14 Row total

3

5

0

10

6

0

87

0

1

88 CLASSIFIE

D

8

2

0

9

0

3

14

13

0

1

0

9

0

10

14

0

2

1

0

23

26 Column total

7

95

10

9

27

148 Overall accuracy = 110/148 = 74%

Producers accuracy

Users accuracy

3 - Agriculture = 5 / 7

71%

3 - Agriculture = 5 / 10

50%

6 - Bare soil = 87 / 95

92%

6 - Bare soil = 87 / 88

99%

8 - Forest = 9 / 10

90%

8 - Forest = 9 / 14

64%

13 - Water = 9 / 9

100%

13 - Water = 9 / 10

90%

14 - Flooded area =

(32)

6=Agriculture 7=Water 9= Bare soil 10=forest

20=Flooded area with typha

CORRECT CLASSES 6 7 9 10 20 Row total 6 4 1 0 0 5 10 7 0 10 0 0 0 10 CLASSIFIED 9 0 0 42 0 0 42 10 1 0 0 8 1 10 20 0 0 0 1 16 17 Column total 5 11 42 9 22 89 Overall Accuracy= 80/89 90%

Producers accuracy Users accuracy

Agriculture 4/5 80% Agriculture 4/10 40%

Water 10/11 90,91% Water 10 /`10 100%

Bare soil 42 / 42 100% Bare soil 42 / 42 100%

Forest 8/9 88,89% Forest 8/10 80%

Flooded area with

Typha 16/22 71,43%

Flooded area with

(33)

Table 13.. Accuracy assessment 1994 204049

3= Bare soil

4= Flooded area with Typha 6=Forest 10= Agriculture 12=Water CORRECT CLASSES 3 4 6 10 12 Row total 3 9 4 0 0 0 0 94 4 1 18 0 1 0 20 CLASSIFIE D 6 0 1 19 0 0 20 10 0 0 1 9 0 10 12 0 0 0 0 10 10 Column total 9 5 19 20 10 10 154 Overall Accuracy = 150/154 97,40%

Producers accuracy Users accuracy

Bare soil 94/95 98,95% Bare soil 94 / 94 100%

Flooded area with

Typha 18/19 94,74% Flooded area with Typha 18 / 20 90,00%

Forest 19/20 95,00% Forest 19/20 95% Agricultur e 9/10 90,00% Agriculture 10./10 100% Water 10 ./10 100,00% Water 10./10 100% 3= bare soil 5=wate r 8= Forest 9=Floo ded area with Typha

(34)

Overal l Accur acy= 58/62 93.55% Producers accuracy Users accuracy Bare soil 30/31 96.77% Bare soil 30/31 96.77% Water 10 ./10 100.00 % Water 10 ./10 100.00% Forest 10 ./10 100.00 % Forest 10 ./10 100.00% Flood ed area with Typha

8/9 88.89% Flooded area with Typha 8/11 72.73%

CORRECT CLASSES 3 5 8 9 Row total 3 30 0 0 1 31 5 0 10 0 0 10 CLASSI FIED 8 0 0 10 0 10 9 1 0 2 8 11 Column total 31 10 12 9 62

(35)

9=water

15=Flooded area with Typha 16=Bare soil 21=Forest CORRECT CLASSES 9 15 16 21 Row total 9 7 1 1 1 10 15 0 29 0 0 29 CLASSIFIED 16 0 1 115 0 116 21 0 8 1 41 50 Column total 7 39 117 42 205 Overall Accuracy = 192/2 05 93.66% Producers accuracy Users accuracy Water 7./7 100% Water 7/10 70.00% Flooded area with Typha 29/39 74.36% Floode d area with Typha 29/29 100.00 % Bare soil 115/1 17 98.29% Bare soil 115/11 6 99.14% Forest 41/42 97.62% Forest 41/50 82.00%

(36)

Table 16. Accuracy assessment 2018

1 - Typha

2 - Open area with trees 3 - Water 4 - Bare soil 5 - Forest 6 - Irrigated agriculture 7 - Unclassified CORRECT CLASSES 1 2 3 4 5 6 7 Row total 1 5 2 0 1 1 1 0 10 2 2 152 0 31 0 1 0 186 CLASSIFI ED 3 0 2 9 0 1 0 0 12 4 1 13 1 105 1 0 3 124 5 1 0 0 1 7 0 2 11 6 0 0 0 0 2 11 0 13 7 0 0 8 0 0 0 2 10 Colum n total 9 169 18 138 12 13 7 366 Overall Accuracy = 291/36 6 79.51 % Producers

accuracy Users accuracy

Typha 5/9 55.56 % Typha 5/10 50.00 % Open areas with trees 152/16 9 89.94 % Open areas with trees 152/1 86 82% water 9./18 98.29 % water 9./12 75.00 % Bare soil 105./1 38 97.62 % Bare soil 105/1 24 85% Forest 7/12 58% Forest 7/11 64% Agricultur e 11/13 85% Agricultur e 11/13 85%

(37)

Unclassifi

ed 2/7 29%

Unclassifi

ed 2/10 20%

Water levels at Podor

Figure 11. Water levels in Podor from 1980 to 2018 (Mettrop, 2018)

1/28 /198 0 5/18 /198 1 9/6/ 1982 12/2 6/19 83 4/15 /198 5 8/4/ 1986 11/2 3/19 87 3/13 /198 9 7/2/ 1990 10/2 1/19 91 2/8/ 1993 5/30 /199 4 9/18 /199 5 1/6/ 1997 4/27 /199 8 8/16 /199 9 12/4 /200 0 3/25 /200 2 7/14 /200 3 11/1 /200 4 2/20 /200 6 6/11 /200 7 9/29 /200 8 1/18 /201 0 5/9/ 2011 8/27 /201 2 12/1 6/20 13 4/6/ 2015 7/25 /201 6 11/1 3/20 17 -1000 100 200 300 400 500 600 700

Podor

cm