An evaluation: mapping temporal land use and land
cover (LULC) for the Senegal River Basin to analyze
the changed area
Author: R. A. L. Bossen (11032685)
Supervisor: A.C. Seijmonsbergen
BSc Earth Sciences Thesis
Cover image: Bobst, C. (n.d.). Farmers in Yonoféré [Photograph]. Retrieved from https://christianbobst.photoshelter.com/image/I0000abfa6.j4Ux8.
Abstract
The Senegal River Basin (SRB) is dominated by a flood-recession crop culture, in which the farmers rely on the annual flood. The Organisation pour la Mise en Valeur du fleuve Sénégal (OMVS) constructed two dams in the Senegal river in 1986 to regulate floodwater and develop an irrigation culture. This research aims to investigate what the influence of the construction of the dams is on the land use and land cover (LULC) in the region by quantifying the changed areas. Remote sensing techniques for both Sentinel-2 and Landsat imagery are used for the classification of the LULC in 1986, before the construction of the dams, 1999 after the construction of the dams, and in 2018, at present. Also three thematic maps representing solely the vegetation for these three years are created. This research includes an investigation of the possibility to compare the different sets of Sentinel and Landsat imagery with post-classification methods. A trend of an overall increase in vegetation and agriculture was quantified from the LULC and thematic maps, which can be associated with higher water levels in the SRB since the construction of the dams, and the development of an irrigation culture. Comparing maps that are classified from Sentinel and Landsat imagery brings along difficulties due to differences in spatial and spectral resolution, which can lead to a wrong indication of the changed area. Multiple post-classification methods had to be used for an accurate classification of the imagery.
Content
Abstract ...1
List of Figures ...4
List of Tables...5
Abbreviations ...6
1 Introduction ...7
1.1 General Introduction ... 7 1.2 Theoretical framework ... 7 1.2.1 Study area... 81.2.2 History flooding management of the OMVS ... 9
1.3 Hypothesis and aim ... 10
1.4 Research questions ... 10
2 Methods ... 11
2.1 LULC map to support climate change scenarios ... 12
2.2 Temporal LULC maps ... 13
2.3 Accuracy Assessment ... 14
2.4 Quantification of the LULC change ... 15
2.4 Thematic vegetation maps... 15
3 Results ... 16
3.1 LULC map to support climate change scenarios ... 16
3.2 Temporal LULC maps ... 18
3.3 Accuracy assessment ... 21
3.4 Quantification of the LULC change ... 21
3.5 Quantification of vegetation change ... 22
4 Discussion ... 23
4.1 Methodological Discussion ... 23 4.2 Interpretation of Results ... 24 4.3 Further Research ... 255 Conclusion ... 27
Acknowledgements ... 28
Literature list ... 29
A.1 Metadata imagery ... 31
A.2 Specifications Landsat and Sentinel Satellites ... 34
A.3 Pre-processed imagery ... 36
A.4 River data ... 39
A.5 Accuracy sampling points ... 39
A.6 Criteria for Thematic Maps ... 40
B Appendix Results ... 42
B.1 Overall accuracy ... 42
B.2 Area per class of LULC maps ... 43
B.3 Thematic maps ... 44
B.4 Area per class of Thematic maps ... 47
List of Figures
1.1 Research area……….9 2.1 Workflow………..11 3.1 LULC map 2018………17 3.2 LULC map 1986………19 3.3 LULC map 1999………20A.1 Pre-processed imagery 1986……….36
A.2 Pre-processed imagery 1999……….37
A.3 Pre-processed imagery 2018………38
A.4 Water level Podor……….39
B.1 Thematic map 1986………..44
B.2 Thematic map 1999………..45
List of Tables
2.1 LULC classes 2018………..12
2.2 LULC classes temporal maps………....14
2.3 Accuracy sampling points……….15
3.1 Accuracy LULC map 2018………...21
3.2 Producer’s and user’s accuracy LULC map 2018……….21
3.3 Changed LULC area………..22
3.4 Changed vegetation………...22
A.1 Metadata Landsat imagery 1986………...31
A.2 Metadata Landsat imagery 1999………..32
A.3 Metadata Sentinel imagery 2018………..33
A.4 Specifications Landsat 5………...34
A.5 Specifications Landsat 7………...34
A.6 Specifications Sentinel-2………..35
A.7 Accuracy sampling points 1986………...39
A.8 Accuracy sampling points 1999………40
A.9 Pixel criteria SAVI 1986………..40
A.10 Pixel criteria SAVI 1999……….41
A.11 Pixel criteria SAVI 2018……….41
B.1 Overall accuracy 1986………..42 B.2 Overall accuracy 1999………..42 B.3 Area LULC 1986………..43 B.4 Area LULC 1999………..43 B.5 Area LULC 2018………..43 B.6 Area vegetation 1986………47 B.7 Area vegetation 1999………47 B.8 Area vegetation 2018………47
Abbreviations
ETM+
Enhanced Thematic Mapper Plus
LULC
Land Use and Land Cover
OBIA
Object Based Image Analysis
OMVS
Organisation pour la Mise en Valeur du fleuve Sénégal
SAVI
Soil Adjusted Vegetation Index
SRB
Senegal River Basin
TM
Thematic Mapper
1 Introduction
1.1 General Introduction
The Senegal River Basin (SRB) suffered from environmental degradation and severe droughts in the second half of the 20th century. The region was dominated by a flood-recession crop culture, in which farmers cultivating at small-scale relied on the annual flood that occurred in high water season from July to October. However, in the years 1972-1973 no significant rainfall occurred and the drought was devastating for the population and local economy (Uhlir & National Research Council, 2003). In the 1960’s the Organisation pour la Mise en Valeur du fleuve Sénégal (OMVS), the supranational authority that regulates the basin was formed by Senegal, Mauritania and Mali. After the years of severe drought they decided to construct two dams in the Senegal river in the 1980s to serve three purposes: irrigation, navigation and energy (Venema et al., 1997). Whether this state-managed irrigation system turned out to be successful remains ambiguous. According to Adams (2000), the OMVS prioritized commercial economic development in their flood management, instead of ensuring reliability of incomes of the local population and reducing the vulnerability of the region to climatic and external factors (Venema et al., 1997). As a consequence the local farmers in the middle valley of the basin were excluded from commercial irrigated farming and could no longer depend on the annual floods for their food production (Adams, 2000). Degeorges & Reilly (2006) stated ‘Both the land and the people involved in traditional agriculture will cease to be as a result of the dams and irrigation’. Besides, according to I. Mettrop (personal communication, April 30, 2018) salt water intrusion at the delta increases even further as a consequence of erosion of the coast, and the river banks deteriorate due to a decrease in sediment deposition. The two dams that were built, do have a high potential to not only provide the local people with a sufficient food production, but also to conserve the ecosystem services and vegetation in the area. A re-arrangement of the floodwater management could secure the Senegal river basin with a prosperous future. For a rearrangement of the flood management to succeed, it is important that it is investigated how the dams influenced and still influence the land use and land cover (LULC) in the region. Therefore this research evaluates the influence of the dams on the agricultural production and vegetation in the SRB.
1.2 Theoretical framework
The OMVS has asked the Dutch ecological consultancy Altenburg & Wymenga to conduct a study about the effects of climate change on the ecosystem services for the local population in the SRB (Altenburg & Wymenga, 2018). In this study it is of specific importance how the ecosystem services
can be preserved in the future. This research provides the ecological consultancy Altenburg & Wymenga LULC maps of the Senegal basin, which will be used in their study of climate change scenarios for the OMVS.
1.2.1 Study area
The SRB consists of an upper basin, valley and delta, distributed over three regions with distinct environmental conditions (Uhlir & National Research Council, 2003). The Senegal river flows through Senegal, Mali, Guinea and Mauritania and constitutes the border between these four countries (Uhlir & National Research Council, 2003; Venema et al., 1997). The total drainage area that is occupied is 290,000 km2 and include all tributaries that origin in Mali and Guinea (Venema et al., 1997). The flow of the Senegal river depends on the annual rainfall in the upper basin. Most of the rain falls from July to October in high-water season, after which it decreases substantially from November to May (Uhlir & National Research Council, 2003). The traditional agricultural production is subject to the seasonal rainfall and floods that occur at the end of the high-water season (Adams, 2000). Mainly maize and sorghum are cultivated by small-trade farmers, who start growing their crops in the dry season after the waters of the flood have receded (Adams, 2000; Gaye et al., 2013). Since the construction of the dams however, an irrigation culture is developed, in which maize and sorghum are replaced for large-scale rice and wheat cultivation (Venema et al., 1997).
Dutch consultancy Altenburg & Wymenga designated the research area to the valley of the SRB (Figure 1.1) covering 7028 km2 of the total catchment. This region constitutes the alluvial valley in which most of the agriculture and population is situated (Gaye et al., 2013). Consultant I. Mettrop visited the research area near Podor and made field observations of the LULC in the area. These field observations are used in this research as a reference for the classification of the LULC in the Senegal catchment and can be found in the Digital Appendix C.
Figure 1.1 The Senegal catchment. The research area is delimited to the valley of the SRB, outlined in red. The
country borders are outlined in yellow and is also the river in the research area. South of the river lies Senegal, North of the river Mauritania and the border between Mali and Senegal is situated South-East of Bakel.
1.2.2 History flooding management of the OMVS
An average area of 459,000 ha was flooded every year before the construction of the dams on both sides of the bank, of which between 15,000 and 150,000 ha could be farmed depending on the magnitude of the flood (Degeorges & Reilly, 2006; Uhlir & National Research Council, 2003). The valley was divided in a Senegalese side of the river and a Mauritanian side (Gaye et al., 2013). With a potential area of 108,000 ha for cultivation after a flood, approximately 65,000 ha lied on the Senegalese side and 43,000 ha on the Mauritanian side (Degeorges & Reilly, 2006). In dry years, when the flood did not exceed 108,000 ha, the agricultural yields were not sufficient. In 1973 the OMVS announced the construction of the two dams that would be built in the Senegal river (Adams, 2000). The Diama Dam was built at the mouth of the river at Saint-Louis and completed in 1986 (Uhlir & National Research Council, 2003). This dam was built to stop the saltwater intrusion in the Delta and lower valley to make the land suitable for agriculture (Adams, 2000; Uhlir & National Research Council, 2003). The second dam, the Manantali dam, was built upstream in Mali and completed in 1987. The dam would serve multiple purposes, including water retention of the tributaries, regulation of river flows, development of an
irrigation culture in the valley, river navigability and energy production (Uhlir & National Research Council, 2003). The total area that would be brought under irrigation was aimed between 300,000 and 400,000 ha of land and the main crops would be rice and wheat (Adams, 2000; Woodhouse, 2012). The annual flood of the river would be artificially maintained to sustain flood-recession farming of maize and sorghum, but would be reduced to 190,000 ha by 2028 (Adams, 2000; Degeorges & Reilly, 2006).
1.3 Hypothesis and aim
According to McCartney (2009) dams can affect riverside and floodplain vegetation due to a changing magnitude and extent of floodplain inundation and land-water interaction. Plant reproduction of original species can be disrupted as a result and besides lead to an encroachment of upland plant species, an invasion which was previously prevented by frequent flooding (McCartney, 2009). In Senegal the water level and hydrology are also changed by the construction of the dams, which lead to the invasion of the species Typha Australis mainly in the lower valley and delta of the SRB (Dumas et al., 2010). A change in vegetation due to the construction of the dams is therefore expected for this research. Current studies mainly focus on flood management of the OMVS based on a literary approach, but lack an investigation of the quantification of vegetation change as a result of the construction of the dams. Therefore this research aims to quantify the changed LULC before and after the construction of the dams and at present. Remote sensing classification techniques applied to Sentinel and Landsat imagery form the basis for a detailed LULC change analysis. This quantifies the spatial and temporal effects of the construction of the dams on vegetation and agricultural land. Such data contributes to a better insight in the effects of dam construction on LULC.
1.4 Research questions
1. How has the land use and land cover (LULC) in the Senegal River Basin changed since the construction of the two dams?
(a) What is the current LULC in the Senegal Basin?
(b) How has the LULC changed before and after the construction of the dams? (c) What trends can be identified from the LULC maps?
2 Methods
Remote sensing techniques are used to create LULC maps and quantify the LULC change. Data processing of the Sentinel and Landsat imagery will be done with the software programs ‘ArcMap 10.6’ (ESRI, https://www.esri.com/en-us/home), ‘ERDAS IMAGINE’ (Hexagon, https://www.hexagongeospatial.com/) and ‘eCognition’ (eCognition, https://www.ecognition.com). Open data sources are used to download the Sentinel-2 and Landsat imagery. Only imagery with a cloud cover of less than 5% are selected for this research. All metadata of the individual images are given in Appendix A.1. Figure 2.1 shows a generalized workflow for obtaining the temporal LULC maps. The methods will be explained in further detail here.
Figure 2.1 Workflow for this research, in which remote sensing techniques are used for constructing LULC maps
of the SRB. The Sentinel and Landsat imagery are collected and pre-processed, after which the training sites for the different LULC classes can be initialized. Classification is subsequently done with the supervised classification method for the Landsat imagery and object based image analysis (OBIA) for the Sentinel imagery to digitize the LULC maps. Small erroneous classified pixels are filtered out using the majority filter tool in ArcMap. An accuracy assessment is produced to indicate the accuracy of the maps. Subsequently, the change in area can be calculated using the field calculator in ArcMap. The thematic vegetation maps are generated using the soil adjusted vegetation index (SAVI). No accuracy assessment is required for these maps and vegetation change can be calculated directly.
2.1 LULC map to support climate change scenarios
The LULC map of 2018 is made in cooperation with the BSc projects of C. Juijn (Juijn, 2018) and E. Rentier (Rentier, 2018). Sentinel-2 imagery are used for the classification of the LULC, because of their high resolution size. In Appendix A.2 a table is given with a detailed description of the specifications of Sentinel-2 imagery. The most recent Sentinel-2 imagery data of May 2018 is downloaded. Pre-processing (Figure 2.1) is done in ArcMap and ERDAS IMAGINE. The spectral bands 2 (blue), 3 (green), 4 (red) and 8 (NIR) are stacked in ERDAS and the colors are corrected using the color correction tool in ERDAS. Subsequently the different images are mosaicked and clipped in ArcMap to display solely the research area. An image of the pre-processed map is given in Appendix A.3.
The coordinates of the field observation points (Digital Appendix C) of I. Mettrop are converted from decimal degrees to degrees and decimal minutes. The point locations are then displayed in Google Earth and saved as layer file and opened in ArcMap. With the coordinates of the field observations the different classes can be identified on the satellite imagery and subsequently classified for the full area of the SRB.
In the software program eCognition the pre-processed Sentinel-2 image is classified using object based image analysis (OBIA). In OBIA, image pixels are first aggregated into spectrally homogenous image objects using an image segmentation algorithm and then the individual objects are classified (Liu & Xia, 2010). The algorithm aggregates the pixels based on their features, such as color, texture, shape and size (Sun, 2016). A spectral difference segmentation of 300 is used, in which the individual land cover areas are identified most precisely. The LULC areas are subdivided to the classes in table 2.1.
Table 2.1 Classes of the LULC map of 2018
LULC classes 2018 Irrigated agriculture Bare soil
Forest Water
Open areas with trees Typha
These are the classes that can be separated most correctly from the satellite imagery using OBIA and therefore give the most accurate representation of the LULC in the region.
2.2 Temporal LULC maps
Landsat imagery dates back to 1972 (https://landsat.usgs.gov/) and will therefore be used for the classification of LULC change. The specifications of Landsat 5-7 are given in Appendix A.2. The imagery of before the construction of the dam is obtained from Landsat 5 Thematic Mapper (TM) and after the construction the imagery is obtained from Landsat 7 Enhanced Thematic Mapper plus (ETM+). The imagery is downloaded from the website of USGS (https://landlook.usgs.gov/viewer.html), but due to limited data, the individual imagery of the catchment area are obtained from different dates. See Appendix A.1 for a specification of the different dates per image. Pre-processing of the imagery is done in ArcMap, where the individual imagery are clipped to display the research area. Appendix A.3 shows the pre-processed imagery, including the outline of the catchment and Landsat image boundary.
The years 1986 and 1999 are selected for classification of the LULC. 1986 shows the LULC of the SRB before the construction of the dams and 1999 after the construction of the dams. 1986 is the only year before the construction of the dams with image data available for the full catchment. The specific year 1999 was selected after an evaluation of the water level and discharge data provided by I. Mettrop (Appendix A.4 & Digital Appendix C). In 1999 water levels and discharge of the Senegal river show above average values. Evaluating the area that is flooded in this year, provides information on the seasonal change in LULC and the change in flooded area after the construction of the dams.
The individual imagery are classified separately, for the reason factors such as weather conditions and daytime influence spectral reflection (Walter, 2004). In the software program ERDAS the pre-processed Landsat imagery are classified using a supervised classification method with maximum likelihood. In this method the pixels are grouped into clusters based on statistical analysis, using the K-means algorithm (Al-doski et al., 2013). The clusters are then identified with the specified land cover classes. The different LULC classes are adjusted to the classes in table 2.2.
Table 2.2 Classes of the temporal LULC maps in 1986 and 1999
LULC classes 1986 LULC classes 1999
Agriculture Agriculture
Bare soil Bare soil
Forest and vegetation Forest and vegetation
Water Water
Flooded area Flooded area
Typha and flooded vegetation
Unclassified
These are the classes that can be separated most correctly from the satellite imagery, and therefore give the most accurate representation of the LULC in the region. In ArcMap the tool majority filter is used to filter out small pixels that were classified incorrectly to increase accuracy and smoothen the LULC maps.
2.3 Accuracy Assessment
An accuracy assessment is performed in ArcMap for all three LULC maps. The sampling strategy ‘stratified random’ is used for the distribution of points. This method creates points that are randomly distributed within each class, where each class has a number of points proportional to its relative area (“The Image Classification Wizard,” n.d.). According to Congalton (1991) a good rule of thumb is to collect a minimum of 50 samples for each land use category. The sampling points are assigned either the value 0 or 1, indicating whether the classified area is assigned the correct class. Table 2.3 shows the amount of sampling points used for the accuracy assessment for each LULC map.Appendix A.5 gives the amount of sampling pointsper individual image of the research area for the years 1986 and 1999. The following formula are used to measure the accuracy, in which the rows are the classified classes and the columns the correct class (reference):
Overall accuracy = Total sampling points classified correct / (2.1) total sampling points * 100
Producer’s accuracy = Sampling points classified correct per class / (2.2) column total * 100
The producers accuracy, formula 2.2, indicates the probability of a reference pixel being correctly classified, while the users accuracy, formula 2.3, indicates the probability that a pixel classified on the map actually represents that class (Congalton, 1991).
Table 2.3 The total amount of sampling points taken for the accuracy assessment for each year. Appendix A.5 gives the amount of sampling points per individual satellite image for the years 1986 and 1999.
LULC map year Accuracy sampling points
1986 534
1999 524
2018 366
2.4 Quantification of the LULC change
The change in LULC between the separate years is quantified in ArcMap using the field calculator. The following formula are applied to each individual map:
Area of class (km2) = Count of pixels * resolution size (m) * (2.4) resolution size (m) / 1.000.000 km2
Percentage of class (%) = Area of class (km2) / total area (km2) * 100 (2.5)
2.4 Thematic vegetation maps
To quantify the vegetation change between 1986, 1999 and 2018, thematic maps are created for all three years using the soil adjusted vegetation index (SAVI) (eq. 2.6) in ERDAS. Criteria is set to assign pixels above a specific value a 1, depicting vegetation, and all other pixels a 0, depicting all other classes. Appendix A.6 gives the pixel criteria that is used for each individual image. Subsequently the overall vegetation change is quantified using the field calculator in ArcMap. Since the pixels of agricultural fields are assigned a 1, the agricultural area is subtracted from the total vegetation area afterwards.
3 Results
In this chapter the results of this research will be presented and analyzed shortly. Figures 3.1-3.3 present the LULC maps that were classified from the satellite imagery. The area surrounding Podor shows the most LULC classes for all three maps, therefore this part is emphasized and zoomed in to.
3.1 LULC map to support climate change scenarios
Figure 3.1 shows the classified LULC map from the Sentinel-2 imagery for the year 2018. The LULC classes were determined after a visual evaluation of the satellite imagery. The class ‘unclassified’ is an automatically generated class using OBIA. However, no significant unclassified areas are shown in the map. The class ‘open areas with trees’ constitutes the biggest area on the map. In the area surrounding Podor, irrigated agriculture is visible abundantly and also some area with typha is apparent. Remarkably, the area does not show many places of forests, but instead mainly less dense trees in open areas.
Figure 3.1 LULC map of the Senegal catchment in 2018 based on Sentinel-2 imagery in dry season. A map in
3.2 Temporal LULC maps
Figure 3.2 and 3.3 show the classified LULC map from the Landsat imagery for the years 1986 and 1999. The LULC classes were determined after a visual evaluation of the satellite imagery.
In 1986 (figure 3.2) the research area shows mainly areas of bare soil. Besides, in the area surrounding Podor, there is hardly any agriculture, mainly for the reason an irrigation culture had not yet been developed. The small segments of agriculture in the lower map are not actual agricultural areas, but were classified erroneously with the supervised classification method. Also, the class ‘unclassified’ is an automatically generated class using the supervised classification method. However, no unclassified pixels are shown in the map. Flooded areas were classified at places where the reflection of the ground showed a darker color than the surrounding soil. These areas were investigated using the software program Google Earth (Google, https://www.google.nl/earth/download/gep/agree.html), after which it could be determined that these areas indicate a wet soil. However, there was no actual water on these areas. Besides, no typha could be identified from the imagery.
Figure 3.3 shows a very different LULC in comparison with 1986. This can mainly be attributed to the time of the year from which the imagery was selected. A small increase in agriculture is notifiable from the area near Podor and the class ‘typha and flooded vegetation’ is added as a separate class. Further, forest and vegetation seems to have increased since 1986. The class ‘unclassified’ is visible South-East in the research area near Bakel and depicts clouds in this map.
Figure 3.2 LULC map of the Senegal catchment in 1986 based on Landsat 5 imagery in dry season. A map in
Figure 3.3 LULC map of the Senegal catchment in 1999 based on Landsat 7 imagery in wet season. A map in
3.3 Accuracy assessment
Table 3.1 and 3.2 show the overall, producers and users accuracy of the LULC map of 2018. The overall accuracy is 79% with no class less than 50% accurate. The low producers and users accuracy probabilities for the class ‘typha’ are mainly a result of the color reflection of the species that is difficult to recognize. Open areas with trees and irrigated agriculture show the highest accuracy rates. These are the most important LULC classes for this research.
The Digital Appendix C contains the overall, producer’s and users accuracy for each individual image for the years 1986 and 1999.
Table 3.1 The overall accuracy of the 2018 LULC map based on 366 sampling points.
Overall Accuracy (%) LULC map Senegal catchment 2018 79
Table 3.2 The producer’s and user’s accuracy of the 2018 LULC map based on 366 sampling points.
Class Producers accuracy (%) Users accuracy (%)
Typha 56 50
Open area with trees 90 82
Water 50 75
Bare soil 76 85
Forest 58 64
Irrigated agriculture 85 85
3.4 Quantification of the LULC change
Table 3.3 shows the LULC change per class and the corresponding percentage. Appendix B.2 shows the quantified area per class for all three years. The results of this table will be discussed in the next chapter.
Table 3.3 The changed LULC area’s between 1986, 1999 and 2018. An increase of 100% is given when for the
subsequent year this class was not identified. A 0% is given when the class is not identified for both years when the change is calculated.
Class Area change (km2) 1986 - 1999 Change (%) Area change (km2) 1999 - 2018 Change (%) Area change (km2) 1986 - 2018 Change (%) Unclassified 20 100 0 0 0 0 Flooded area 293 -18 0 0 0 0 Forest and vegetation 596 48 1034 -83 438 -67 Bare soil 1999 -46 181 7 1818 -42 Water 1436 88 1397 -86 39 16 Agriculture 5 2 41 16 46 18
Typha and flooded
vegetation 235 100 135 -57 100 100
Open area with trees 0 0 3732 100 3732 100
3.5 Quantification of vegetation change
Table 3.4 shows the vegetation change between all three years and their corresponding percentage. The vegetation change was quantified using the thematic maps (Appendix B.3). Appendix B.4 gives the quantified vegetation for all three years, from which the agricultural area has been subtracted to represent solely the vegetation. The results of this table will be discussed further in the next chapter.
Table 3.4 The changed total vegetation area minus the agricultural area between 1986, 1999 and 2018.
Class Area Change (km2) 1986 - 1999 Change (%) Area Change (km2) 1999 - 2018 Change (%) Area Change (km2) 1986 - 2018 Change (%) Vegetation 136 17 269 25 405 37
4 Discussion
The aim of this research was to quantify the changed LULC before and after the construction of the dams and at present using remote sensing techniques. The results show a broad variation in LULC between the three specific years and to quantify solely the vegetation change, thematic maps were made in addition. Here, the LULC maps and their corresponding methodologies are discussed and analyzed in more detail and compared to current literature. Besides, some recommendations for future research are given.
4.1 Methodological Discussion
The methods that were used to obtain the LULC maps proved to be sufficient. However, several difficulties were encountered during classification:
1) Due to a limited availability of Landsat data, the individual imagery of the catchment were taken of different dates for the years 1986 and 1999. The classification had to be done separately for each image due to differences in color reflection and LULC classes. The images were however taken in the same season, to make the classification as accurate as possible.
2) For the classification of the Sentinel-2 imagery, a supervised classification method with maximum likelihood in ERDAS was used initially to create a LULC map. However, using this method the LULC was not assessed accurately. Therefore OBIA was used in eCognition for the classification of the Sentinel-2 imagery. In contradiction with the Sentinel-2 imagery, the supervised classification method in which pixels are classified directly, did classify the LULC of the Landsat imagery most accurately. According to Liu & Xia (2010), this could be due to the difference in resolution size between the Sentinel-2 and Landsat imagery. Although OBIA reduces within-class spectral variation in comparison to pixel based classification, the specific features size, texture and shape, cannot be identified as accurate in imagery with high resolution sizes or larger scales (Liu & Xia, 2010; Myint et al., 2011). This leads to under-segmentation and a less accurate classification. This proves to hold for this research, and therefore the supervised classification method was used for the Landsat imagery, while OBIA was used for the Sentinel-2 imagery.
3) The classes of the LULC differ between the three years. For this reason, comparing the LULC between these years is difficult and the quantification of the change is not exact. Especially comparing LULC maps based on both Sentinel-2 and Landsat imagery brings along complications due to differences in spatial and spectral resolution. This refers back to the investigation whether it is possible to compare different sets of Sentinel-2 and Landsat imagery.
With post-classification methods similar results can be created. However, during quantification of the LULC classes, higher resolution sizes lead to a higher quantification of some classes. This can lead to an erroneous indication of the change in LULC area.
4) The criteria that was used for creating the thematic maps is based on a visual evaluation. Visual evaluation however, is not as accurate as computational processing. Also, in these maps the resolution size influences the classification of the vegetation. Possibly more trees are classified in the LULC map based on the Sentinel-2 imagery, for the reason smaller objects can be detected from imagery with a resolution size of 10x10. This could lead to an inaccurate quantification of the total area of vegetation in the catchment.
4.2 Interpretation of Results
The LULC maps show a highly differing LULC between the three years. The change in LULC before and after the construction of the dams can be analyzed from the maps (figures 3.1-3.3). Table 3.3 shows a gradual increase of agriculture. Where from 1986 to 1999 the agriculture only increased with 2%, from 1999 to 2018 it did with 16%, with a total increase of 18% since 1986. The aim of the OMVS was to increase the area under cultivation to approximately 375,000 ha before 2028 (Uhlir & National Research Council, 2003), but according to Degeorges & Reilly (2006) and Uhlir & National Research Council (2003) the area actually brought under irrigation is estimated at 100,000-131,000 ha in 2006, with only half of this area being cropped. In 1995 the area actually farmed that year was 29,792 ha, with an annual increase of 2000 ha for the whole catchment (Degeorges & Reilly, 2006). Quantified from the LULC maps there was a total area cultivated of 21,000 ha in 1999 (Appendix B.2) for the research area of 7028 km2. In 2018 the agriculture increased to 25,300 ha, which assumes only an increase of 226 ha/year between 1999 and 2018. However, from the maps it appears that most of the cultivated area is situated more South-West towards the delta, while the boundary of the research area lies slightly West of Podor. Besides, although an increase in irrigated agriculture is measured, flood-recession agriculture more to the East in the valley decreased after the construction of the dams. The flood was reduced every year after the construction of the Manantali dam and large-scale commercial rice production forced peasant farmers into sharecropping arrangements with wealthy outsiders (Degeorges & Reilly, 2006) or emigration as a consequence of exclusion from the irrigated farming (Adams, 2000).
The LULC of 1999 was classified in a period of high water season. Water availability increases with 88% in the research area during flooding after the construction of the dams, with a total flooding area of 163,100 ha (Appendix B.2). As stated by Degeorges & Reilly (2006) the flooding area would be gradually reduced from 459,000 ha to 190,000 ha by 2028. However, in this research it appears that in 1999, the flooding area in the valley was already reduced to more than half of the flooding area before
Considering the imagery of 1999 is selected from a period of high water season, comparison of the classes bare soil and forest and vegetation provides information on the seasonal change in LULC. Between 1986 and 1999 a decrease of 46% in bare soil is quantified and at least 23,500 ha of vegetation is flooded. Surprisingly, between 1999 and 2018, bare soil increases with only 7%. This is associated with the increase in open areas with trees. Typha and flooded vegetation decreases 57%, while a decrease of 100% of flooded vegetation would be expected. However, after the construction of the dams the water levels were kept higher, which resulted in the rapid growth of Typha Australis throughout the area.
Table 3.3 shows a decrease of 42% for bare soil between 1986 and 2018. This decrease of bare soil is a result of a 100% increase in open area with trees. This class was not identified in 1986 and 1999 for the reason there was no reflectance of open area’s with trees in the Landsat imagery. Instead, area’s with dense vegetation were classified as ‘forest and vegetation’. This class increases between 1986 and 1999 with 48%, but decreases with 83% between 1999-2018, while an increase is expected. Possibly, the difference in resolution size between Sentinel-2 and Landsat influences the classes ‘open area’s with trees’ and ‘forest and vegetation’, since in the Sentinel-2 imagery smaller trees can be detected.
To overcome the complication of differing classes, the three thematic maps (Appendix B.3) using the SAVI were created. Table 3.4 shows the increase in vegetation between 1986, 1999 and 2018. This increase in vegetation after the construction of the dams is likely the result of the increased water levels. Since 1992 the water level is kept above approximately 190 cm in Podor and the annual flood levels have decreased (Appendix A.4). Evident from the water level and discharge data (Digital Appendix C) is that the water levels at nearly all places in the research area have not fallen to and below 0 cm since the construction of the dams. As was already stated by McCartney (2009), a change in floodplain inundation and land-water interaction can lead to a change in vegetation. This appears valid for the Senegal River Basin, since the vegetation has increased with 37% since 1986. Besides, from the LULC maps a clear increase in the plant species Typha Australis can be detected, as expected.
4.3 Further Research
This research is part of a bigger research that is performed by the Dutch consultancy Altenburg & Wymenga for the OMVS. They investigate the effects of climate change on ecosystem services for the local population in the Senegal basin, in which they run flooding models for the area to report climate change scenarios. The LULC map of 2018 can be used to indicate which LULC’s will be flooded under different scenarios. Subsequently an improved flood management report can be composed.
In order to do this, it is of importance to analyze the influence of flooding on the LULC in the region. Therefore the LULC maps of 1986 and 1999 were created in this research. However, to increase accuracy, the LULC of more years should be analyzed. The BSc thesis of C. Juijn (Juijn, 2018) also provides a LULC map of 1994 in dry season, which can be used for further analysis. Besides, a recent LULC map based on Landsat imagery could be created to increase the accuracy of comparing the temporal maps.
The research area covers a great part of the catchment at a very large scale. In further research smaller areas could be classified separately to increase accuracy and include small-scale agricultural fields of local farmers. Besides, evaluating the influence of the dams on other factors, such as increased erosion due to less sediment deposition, or increased salt water intrusion at the delta can improve flood management in the future.
5 Conclusion
This research has investigated how the LULC has changed in the Senegal river basin by classifying the LULC before and after the construction of the dam and at present. Therefore, three LULC maps and three thematic maps presenting vegetation were created for the years 1986, 1999 and 2018. 3 main conclusions can be drawn from this research:
1) An investigation of the changed area between the three years, showed a trend of an overall increase in vegetation and agriculture after the construction of the dams and a decrease in bare soil. The increase in cultivated area is gradual and has not reached the development goal of the OMVS. A quantification of the LULC showed that the agricultural area between Bakel and Podor has increased with solely 4,600 ha between 1986 and 2018. Besides, in 1999 the flooding area was already reduced to more than half of the flooding area before the construction of the dams, which lead to a decrease in flood-recession agriculture.
2) The increase in vegetation was quantified from the thematic maps. This increase can be associated with higher and more constant water levels in the region. However, these higher water levels also resulted in the invasion of the species Typha Australis.
3) Difficulties are encountered when different sets of Landsat and Sentinel-2 imagery are compared. Different post-classification methods, OBIA and supervised classification, had to be used for the imagery to digitize the LULC classes most accurately. Besides, differences in spectral and spatial resolution could lead to an erroneous indication of the change in LULC area.
From this research the construction of the dams seems to have had a positive effect on the SRB. However, this research only provides a large scale representation of the LULC of the SRB. Consequences as a result of the construction of the dams such as increased erosion due to a decrease in sediment deposition and consequences such as emigration of local farmers with the rise of a large scale irrigation culture are not included in this research. Further research is therefore necessary. However, the maps do give a better insight in the effects of the dam construction on the LULC and vegetation in the Senegal catchment.
Acknowledgements
It would not have been possible to write this bachelor thesis without the support and involvement of Harry Seijmonsbergen, for his supervision and specialized knowledge, Ivan Mettrop, who instructed us on the subject and made time to visit the Science Park a couple of times to review the results, and lastly Eline Rentier and Casper Juijn, with whom I closely worked together while creating the LULC maps.
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).
Al-doski, J., Mansorl, S. B., & Shafri, H. Z. M. (2013). Image classification in remote sensing. Department of Civil Engineering, Faculty of Engineering, University Putra, Malaysia.
Almutairi, A., & Warner, T. A. (2010). Change detection accuracy and image properties: a study using simulated data. Remote Sensing, 2(6), 1508-1529.
Altenburg & Wymenga (2018) Bachelor studentenopdracht UvA i.s.m. ecologisch adviesbureau Altenburg & Wymenga [Assignment handout]. Department of Earth Sciences, University of Amsterdam, Amsterdam, the Netherlands.
Congalton, R. G. (1991). A review of assessing the accuracy of classifications of remotely sensed data. Remote sensing of environment, 37(1), 35-46.
Davis, D. S. (2017). The Applicability of Short-Wave Infrared (SWIR) Imagery for Archaeological Landscape Classification on Rapa Nui (Easter Island), Chile. Alpenglow: Binghamton University Undergraduate Journal of Research and Creative Activity, 3(1), 7.
Degeorges, A., & Reilly, B. K. (2006). Dams and large scale irrigation on the Senegal River: impacts on man and the environment. International journal of environmental studies, 63(5), 633-644.
Dumas, D., Mietton, M., Hamerlynck, O., Pesneaud, F., Kane, A., Coly, A., ... & Baba, M. L. O. (2010). Large dams and uncertainties: the case of the Senegal River (West Africa). Society and Natural Resources, 23(11), 1108-1122.
ESA (2018, 8 May). Spatial Resolution. Retrieved from https://earth.esa.int/web/sentinel/user-guides/sentinel-2-msi/resolutions/spatial
Foody, G. M. (2002). Status of land cover classification accuracy assessment. Remote sensing of environment, 80(1), 185-201.
Gatti, A., & Bertolini, A. (2013). Sentinel-2 products specification document. Available online (accessed February 23, 2015) https://earth.esa.int/documents/247904/685211/Sentinel-2+ Products+ Specification+ Document. Gaye, C. B., Diaw, M., & Malou, R. (2013). Assessing the impacts of climate change on water resources of a West African trans-boundary river basin and its environmental consequences (Senegal River Basin). Sci. Cold Arid Reg, 5(1), 0140-0156.
GISGeogrpahy (2018, 17 February). Landsat Program: Satellite Imagery Data and Bands. Retrieved from https://gisgeography.com/landsat-program-satellite-imagery-bands/
Huete, A. R. (1988). A soil-adjusted vegetation index (SAVI). Remote sensing of environment, 25(3), 295-309. Juijn. C. M. (2018). Remote sensing analysis of the land use and land cover change in the Senegal River Valley. BSc thesis.
Liu, D., & Xia, F. (2010). Assessing object-based classification: advantages and limitations. Remote Sensing Letters, 1(4), 187-194.
Mettrop, I. (2018). Jan1980-jan2018 [Microsoft Excel spreadsheet]. Amsterdam: Altenburg & Wymenga. Retrieved from: Altenburg & Wymenga [accessed 08 May 2018]
Montoya, S. (2017, 7 April). How many Spectral Bands have the Sentinel 2 images?. Retrieved from https://www.hatarilabs.com/ih-en/how-many-spectral-bands-have-the-sentinel-2-images
Myint, S. W., Gober, P., Brazel, A., Grossman-Clarke, S., & Weng, Q. (2011). Per-pixel vs. object-based classification of urban land cover extraction using high spatial resolution imagery. Remote sensing of environment, 115(5), 1145-1161.
Nasa (2018, 25 April). Landsat Image Gallery. Retrieved from https://landsat.visibleearth.nasa.gov/
Rentier, E.S. (2018). The Senegal river valley: mapping land cover, land use and floodplain. BSc thesis.
Sun, D. W. (Ed.). (2016). Object Classification methods. Computer vision technology for food quality evaluation (pp. 81-107). Academic Press.
The Image Classification Wizard. (n.d.). Retrieved from https://pro.arcgis.com/en/pro-app/help/analysis/image-analyst/the-image-classification-wizard.htm
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.
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.
Walter, V. (2004). Object-based classification of remote sensing data for change detection. ISPRS Journal of photogrammetry and remote sensing, 58(3-4), 225-238.
Woodhouse, P. (2012). Foreign agricultural land acquisition and the visibility of water resource impacts in Sub-Saharan Africa. Water Alternatives, 5(2), 208.
A Methodology Appendix
A.1 Metadata imagery
Table A.1 Metadata of the imagery that was used for the classification of the 1986 imagery. Spectral bands,
coordinate system, cell size, the dates and source are given. An image with the outlines of the individual imagery can be found in the Digital Appendix C.
Landsat 5 (TM) Imagery Spectral bands Coordinate system Scale/cell
size Date Source
LT05_L1TP_202050_ 19860103_20170218_ 01_T1 1 - Blue 2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30 03-01-1986 https://landlook. usgs.gov/viewer. html LT05_L1TP_203049_ 19861126_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 LT05_L1TP_204048_ 19860407_20170218_ 01_T1 1 - Blue 2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30 07-04-1986 https://landlook. usgs.gov/viewer. html LT05_L1TP_204049_ 19860407_20170218_ 01_T1 1 - Blue 2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30 07-04-1986 https://landlook. usgs.gov/viewer. html
Table A.2 Metadata of the imagery that was used for the classification of the 1999 imagery. Spectral bands,
coordinate system, cell size, the dates and source are given. An image with the outlines of the individual imagery can be found in the Digital Appendix C.
Landsat 7 (ETM) Imagery Spectral bands Coordinate system Scale/cell
size Date Source
LE07_L1TP_202050_ 19991030_20170216_ 01_T1 1 - Blue 2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30 30-10-1999 https://landlook. usgs.gov/viewer. html LE07_L1TP_203049_ 19991021_20170216_ 01_T1 1 - Blue 2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30 21-10-1999 https://landlook. usgs.gov/viewer. html LE07_L1TP_204048_ 19991113_20170216_ 01_T1 1 - Blue 2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30 13-11-1999 https://landlook. usgs.gov/viewer. html LE07_L1TP_204049_ 19991113_20170216_ 01_T1 1 - Blue 2 - Green 3 - Red 4 - NIR UTM, Zone 28 WGS 84 30x30 13-11-1999 https://landlook. usgs.gov/viewer. html
Table A.3 Metadata of the imagery that was used for the classification of the 1999 imagery. Spectral bands,
coordinate system, cell size, the dates and source are given.
Sentinel-2A Imagery Spectral bands Coordinate system Scale/cell size Date Source S2A_MSIL1C_201805 10T112121_N0206_R 037_T28PFC_201805 10T150520 2 - Blue 3 - Green 4 - Red 8 - NIR UTM, Zone 28 WGS 84 10x10 10-05-2018 https://scihub.co pernicus.eu/dhus /#/home S2A_MSIL1C_201805 10T112121_N0206_R 037_T28PEC_201805 10T150520 2 - Blue 3 - Green 4 - Red 8 - NIR UTM, Zone 28 WGS 84 10x10 10-05-2018 https://scihub.co pernicus.eu/dhus /#/home S2A_MSIL1C_201805 10T112121_N0206_R 037_T28QED_201805 10T150520 2 - Blue 3 - Green 4 - Red 8 - NIR UTM, Zone 28 WGS 84 10x10 10-05-2018 https://scihub.co pernicus.eu/dhus /#/home S2A_MSIL1C_201805 10T112121_N0206_R 037_T28QFD_201805 10T150520 2 - Blue 3 - Green 4 - Red 8 - NIR UTM, Zone 28 WGS 84 10x10 10-05-2018 https://scihub.co pernicus.eu/dhus /#/home S2B_MSIL1C_201805 12T110619_N0206_R 137_T28PGC_201805 12T170024 2 - Blue 3 - Green 4 - Red 8 - NIR UTM, Zone 28 WGS 84 10x10 12-05-2018 https://scihub.co pernicus.eu/dhus /#/home S2B_MSIL1C_201805 12T110619_N0206_R 137_T28PGB_201805 12T170024 2 - Blue 3 - Green 4 - Red 8 - NIR UTM, Zone 28 WGS 84 10x10 12-05-2018 https://scihub.co pernicus.eu/dhus /#/home
A.2 Specifications Landsat and Sentinel Satellites
Table A.4 Designation of Landsat 4-5 bands, central wavelength and resolution (GISGeopgraphy, 2018).
Landsat 5 Bands Central Wavelength (µm) Resolution (m)
Band 1 – Blue 0.45 to 0.52 µm 30 Band 2 – Green 0.52 to 0.60 µm 30 Band 3 – Red 0.63 to 0.69 µm 30 Band 4 – NIR 0.76 to 0.90 µm 30 Band 5 - SWI 1.55 to 1.75 µm 30 Band 6 - Thermal 10.4 to 12.3 µm 120 Band 7 - SWI 2.08 to 2.35 µm 30
Table A.5 Designation of Landsat 7 bands, central wavelength and resolution (GISGeopgraphy, 2018).
Landsat 7 Bands Central Wavelength (µm) Resolution (m)
Band 1 – Blue 0.45 to 0.52 µm 30 Band 2 – Green 0.52 to 0.60 µm 30 Band 3 – Red 0.63 to 0.69 µm 30 Band 4 – NIR 0.76 to 0.90 µm 30 Band 5 - NIR 1.55 to 1.75 µm 30 Band 6 - Thermal 10.4 to 12.3 µm 60 Band 7 – Mid-infrared 2.08 to 2.35 µm 30 Band 8 - Panchromatic 0.52 to 0.90 µm 15
Table A.6 Designation of Sentinel-2 bands, central wavelength and resolution (ESA, 2018; Montaya, 2017).
Sentinel-2 Bands Central Wavelength (µm) Resolution (m)
Band 1 – Coastal aerosol 0.443 60
Band 2 – Blue 0.490 10
Band 3 – Green 0.560 10
Band 4 – Red 0.665 10
Band 5 – Vegetation Red Edge 0.705 20
Band 6 – Vegetation Red Edge 0.740 20
Band 7 – Vegetation Red Edge 0.783 20
Band 8 - NIR 0.842 10
Band 8A – Vegetation Red Edge 0.865 20
Band 9 – Water vapour 0.945 60
Band 10 – SWIR – Cirrus 1.375 60
Band 11 – SWIR 1.610 20
A.3 Pre-processed imagery
Figure A.1 Pre-processed imagery of 1986. The imagery is clipped to represent solely the research area of the
Senegal catchment. The red line represents the outline of the catchment and the Landsat imagery boundary, in which from the upper-left to down-right the order is: 204048-204049-203049-202050.
Figure A.2 Pre-processed imagery of 1999. The imagery is clipped to represent solely the research area of the
Senegal catchment. The red line represents the outline of the catchment and the Landsat imagery boundary, in which from the upper-left to down-right the order is: 204048-204049-203049-202050.
Figure A.3 Pre-processed imagery of 2018. The imagery is clipped, color balanced and mosaicked to represent
A.4 River data
Figure A.4 Water level data at Podor, obtained from I. Mettrop (2018). The full data of water levels and discharge
of the river is available in the Digital Appendix C.
A.5 Accuracy sampling points
Table A.7 Accuracy sampling points per individual image of 1986.
Landsat Image specification Accuracy sampling points LT05_L1TP_202050_19860103 _20170218_01_T1 63 LT05_L1TP_203049_19861126 _20170215_01_T1 212 LT05_L1TP_204048_19860407 _20170218_01_T1 111 LT05_L1TP_204049_19860407 _20170218_01_T1 148
Table A.8 Accuracy sampling points per individual image of 1999.
Landsat Image specification Accuracy sampling points LE07_L1TP_202050_19991030 _20170216_01_T1 62 LE07_L1TP_203049_19991021 _20170216_01_T1 209 LE07_L1TP_204048_19991113 _20170216_01_T1 106 LE07_L1TP_204049_19991113 _20170216_01_T1 147
A.6 Criteria for Thematic Maps
Table A.9 Pixel criteria that was used for each individual image of 1986.
Landsat Image specification Criteria LT05_L1TP_202050_19860103 _20170218_01_T1 > 0 LT05_L1TP_203049_19861126 _20170215_01_T1 > 0 LT05_L1TP_204048_19860407 _20170218_01_T1 > -0.3 LT05_L1TP_204049_19860407 _20170218_01_T1 > 0.1
Table A.10 Pixel criteria that was used for each individual image of 1999.
Landsat Image specification Criteria LE07_L1TP_202050_19991030 _20170216_01_T1 > 0.05 LE07_L1TP_203049_19991021 _20170216_01_T1 > 0.08 LE07_L1TP_204048_19991113 _20170216_01_T1 > 0.05 LE07_L1TP_204049_19991113 _20170216_01_T1 > 0.13
Table A.11 Pixel criteria that was used for the image of 2018.
Criteria LULC map Senegal catchment 2018 > 0.25
B Appendix Results
B.1 Overall accuracy
Table B.1 Overall accuracy per individual image of 1986. The producer’s and user’s accuracy for each individual
image are given in the Digital Appendix C.
Landsat Image specification Overall Accuracy (%) LT05_L1TP_202050_19860103 _20170218_01_T1 95 LT05_L1TP_203049_19861126 _20170215_01_T1 92 LT05_L1TP_204048_19860407 _20170218_01_T1 92 LT05_L1TP_204049_19860407 _20170218_01_T1 74
Table B.2 Overall accuracy per individual image of 1999. The producer’s and user’s accuracy for each individual
image are given in the Digital Appendix C.
Landsat Image specification Overall Accuracy (%) LE07_L1TP_202050_19991030 _20170216_01_T1 77 LE07_L1TP_203049_19991021 _20170216_01_T1 90 LE07_L1TP_204048_19991113 _20170216_01_T1 94 LE07_L1TP_204049_19991113 _20170216_01_T1 93
B.2 Area per class of LULC maps
Table B.3 Area in km2 per class map of 1986.
Classname Area (km2) Percentage (%)
Unclassified 0 0
Flooded area 1662 24
Forest and vegetation 652 9
Bare soil 4312 61
Water 195 3
Agriculture 207 3
Typha and flooded vegetation 0 0
Open area with trees 0 0
Table B.4 Area in km2 per class map of 1999.
Classname Area (km2) Percentage (%)
Unclassified 20 < 1
Flooded area 1369 19
Forest and vegetation 1248 18
Bare soil 2313 33
Water 1631 23
Agriculture 212 3
Typha and flooded vegetation 235 3
Open area with trees 0 0
Table B.5 Area in km2 per class map of 2018.
Classname Area (km2) Percentage (%)
Unclassified 0 0
Flooded area 0 0
Forest and vegetation 214 3
Bare soil 2494 35
Water 234 3
Agriculture 253 4
Typha and flooded vegetation 100 1
B.3 Thematic maps
Figure B.1 Thematic Map using the SAVI on the imagery of 1986. Black represents vegetation and agriculture,
Figure B.2 Thematic Map using the SAVI on the imagery of 1999. Black represents vegetation and agriculture,
Figure B.4 Thematic Map using the SAVI on the imagery of 2018. Black represents vegetation and agriculture,
B.4 Area per class of Thematic maps
Table B.6 Area in km2 per class map of 1986. The area of agriculture has been subtracted from the total area of
vegetation and agriculture.
Classname Area (km2) Percentage (%)
Other 6139 99
Vegetation 889 1
Table B.7 Area in km2 per class map of 1999. The area of agriculture has been subtracted from the total area of
vegetation and agriculture.
Classname Area (km2) Percentage (%)
Other 5998 99
Vegetation 1030 1
Table B.8 Area in km2 per class map of 2018. The area of agriculture has been subtracted from the total area of
vegetation and agriculture.
Classname Area (km2) Percentage (%)
Other 5688 98
