Mapping land use, land cover and the spatial-temporal
floodplain extent of the middle valley of the Senegal river
E.S. Rentier
Bachelor Future Planet Studies, University of Amsterdam, Netherlands
Supervisor: Dr. A.C. Seijmonsbergen
12th July 2018
Abstract
The middle valley of the Senegal river is an extensive floodplain where many traditional Senegalese farmers practice flood recession ag-riculture. After the Manantali dam was com-missioned in 1988, the downstream river re-gime changed drastically. On top of this, cli-mate change is negatively affecting the dis-charge and is expected to continue to do so. Therefore, the aim of this research is to con-tribute to the knowledge base of the middle valley of the Senegal river and develop meth-odologies to produce two types of maps. These maps include a Land Cover and Land Use map (LULC) and Normalized Difference Water In-dex (NDWI) maps for a dry (May) and wet (September) period. These maps were success-fully created using remote sensing analysis of Sentinel-2A imagery and a work flow for fu-ture research was developed. The LULC map clearly shows the variation and distribution of six different types of land cover and land use. The NDWI maps show the spatial-temporal extent of water bodies and after a quantit-ative analysis, a total change of water area of 965.6km2between May and September was calculated. The generated maps and work flow contribute to the knowledge base of the Senegal river catchment and can be used to localize land use and land cover types and the extent of flooding areas.
Keywords— Senegal river catchment, land cover and land use, LULC, spatial-temporal extent, floodplain, remote sensing, Normalized difference water index, NDWI, classi-fication, Sentinel-2A.
1
Introduction
1.1
Theoretical framework
The Senegal river flows from southeast to northwest in the upper African Sahel region. It originates in the high-lands of Guinea and then flows through Mali to Senegal and Mauritania, where it forms a natural border between both countries (Fig 1) (Varis & Fraboulet-Jussila, 2002). The middle valley of the Senegal is an extensive floodplain where many traditional Senegalese farmers practice flood recession agriculture (Man´e & Fraval, 2001). The prevail-ing climate is tropical with a dry period from November to May and a rainy season from mid-June to mid-October due to the African monsoon (Uhlir & Council, 2003). In the late 1970s, the aforementioned countries formed the OMVS, which stands for: l’Organisation pour la mise en valeur du fleuve S´en´egal (OMVS). The main goals of the OMVS are to focus on economic growth and infrastructure develop-ment (Uhlir & Council, 2003). In 1986 and 1988, under the auspices of the OMVS, the Maka Diama dam and the Man-antali dam were built. The former was built near the coast of Senegal, where the river enters the Atlantic Ocean and the latter on the Bafing river in Mali (OMVS, 2009). Due to its far upstream location, the Manantali dam had major consequences for the downstream river valley. The purpose of the dam was threefold: enabling irrigated agriculture, producing electricity and making the river more navigable (OMVS, 2009). The OMVS stated that the key objective was to ”guarantee and improve farmer’s incomes” (Adams, 2000, p.6), but the benefits of the dam turned out to be less favourable than expected and the farmers were the ones to pay the price (Horowitz & Salem-Murdock, 1993).
1.2
Managing the dam
When the dam was finished, the river’s regime changed drastically and abruptly. As a result, the low-flow months had an increased flow and the high-flow months a de-creased flow (Sambou, Di´em´e, Tour´e, Badji & Malanda-Nimy, 2009). Although this theoretically means that there is water all year long, there are also fewer floods through-out the year. When managed properly, the dam could still function as an artificial flood producer and a safety net for dry periods. Unfortunately, the OMVS has not always been successful with proper management; the year after the dam was commissioned, there were no floods due to the filling up of the dam’s reservoir (Adams, 2000). Then in 1989, they allowed two floods, causing major losses as farmers had started to sow their flood- recession crops when the second flood wiped everything clean. Most of the farmers did not have enough money or seeds in storage to sow an-other crop. The next year, when a good yield was needed more than ever, the OMVS decided not to discharge any water so they could test the reservoir’s storage capacity (Adams, 2000). This poor management continued for an-other couple of years and was destructive for the farmers in the Senegal river valley. On top of this, climate change is strongly affecting the river’s annual discharge. Sambou et al. (2009, p.110) concluded after studying the effects of the Manantali dam on the river’s regime that: “the effects of climate change on the river’s regime is so strong that it masks the effects of the dam”. J.-C. Bader and Albergel (2015, p.12) conducted a research on artificial flood support on the Senegal river. In their discussion, they state that: “Since the complete installation of its turbines in 2003, the Manantali Dam has not achieved any flood support and has only released discharges through the turbines (. . . ) signi-ficant discharges are released through the turbines during the dry season to the meet energy demand, which causes a strong lowering of the level in the reservoir. This res-ults mostly in a low level in the reservoir, which could be considered insufficient for flood support”.
1.3
Research aim and questions
The research of J.-C. Bader and Albergel (2015) shows that there is a knowledge gap when it comes to deciding the volume of the outflow of the dam and Sambou et al. (2009) point out that the effects of climate change need to be considered when it comes to the future of the traditional farmers in the Senegal river valley. Therefore, the aim of this research is to contribute to the knowledge base of the middle valley of the Senegal river and develop methodolo-gies to produce two types of maps. These maps include a Land Use and Land Cover (LULC) map to show the vari-ation of land cover and land use in the middle valley and a map which indicates where along the
Figure 1: The Senegal river catchment. The red squares indicate the cities where the study area starts (Bakel) and ends (Gani) (Musser, 2010,.sig).
river overflowing occurs during a flood, to what extent and what the total change of water area is between a wet and a dry period. The latter map will be computed using the Nor-malized Difference Water Index (). Once these maps have been generated, they can be analyzed to determine how much water should be allowed though the dam, and equally important: when. The developed workflow will allow future research to replicate these maps for different months. The results of this research will contribute to the knowledge base on the case study of the Senegal river catchment.
The research question is twofold:
1. What is the distribution of land cover types in the Senegal river valley?
2. What is the spatial-temporal extent of flooding in the Senegal river valley?
In order to answer these research questions, a remote sens-ing analysis will be performed on Sentinel-2A satellite im-agery. The LULC map will be created for May 2018, which was the most recent available imagery at the sime this re-search was conducted. The NDWI maps will be created for May and September 2016. The former of the NDWI map being a period of drought and the latter a period of floods. With the aid of these maps, the total change of the surface area of water can be calculated. Subsequently, the results will be analyzed and interpreted. Any used abbreviations are explained in the glossary at the end of this dissertation.
2
Methods
For this research, high resolution and accuracy is de-sired. Thus, satellite imagery from Sentinel-2A was used. Sentinel-2A imagery has thirteen different spectral bands and a 10-20 meter resolution (European Space Agency, 2018). Currently this is the most recent and detailed im-agery available in our study area. The study area covers the middle valley of the Senegal. More specific, it stretches from Bakel in the southeast to Gani in the northwest and is up to 30 kilometers wide Figure 1. The general workflow of this research was as follows: first, all necessary data was downloaded. Then, the data was prepared for classifica-tion of the land cover types and NDWI calculaclassifica-tion. This is called pre-processing. After pre-processing, the images were classified (for the LULC) or used to calculate the NDWI. Lastly, an accuracy assesment was performed on the LULC map and statistics on the surface area of water bodies were computed for the NDWI maps. The workflow is visual-ized in Figure 2 and also shows the programs in which the process was performed in the right column.
Figure 2: General workflow of the research. Processes are visualized in the left column, the programs that were used for the processes are shown in the right column.
All the used satellite imagery, data-sets and generated maps are compiled in a digital appendix. This digital appendix is accessible via the GIS studio of the University of Ams-terdam. The LULC map was made in collaboration with BSc students C. Juijn and R.A.L. Bossen, who were at the time conducting research in the same river valley (Bossen, 2018; Juijn, 2018).
2.1
Data
The LULC map is based on satellite images from 2018-05-10. The NDWI maps are based on satellite images from 2016-05-10 and 2016-10-27. The used coordinate system of these images is UTM, Zone 28 WGS 84 and the resolution 10x10 meters. In total, 24 images were downloaded from the European Space Agency Science Hub (ESA scihub), six for every year (European Space Agency, 2018). An over-view of all used data, including the names of each individual image and their tile number, can be found in Appendix A. There are thirteen available spectral bands, but only band B2 (blue), B3 (green), B4 (red) and B8 (NIR) were used for this research (Table 1).
2.2
Pre-processing
For the LULC map, spectral bands B2, B3, B4 and B8 (Table 1) were stacked for each individual image and the colors were corrected using the color correction tool. For these two processes, ERDAS Imagine 2015 was used. After layer-stacking was completed, all the individual tiles were mosaicked and color-balanced to create a raster file that covered the entire river catchment. Subsequently, this ras-ter was clipped to the shapefile of our study area in Ar-cMap. The process for the NDWI maps was slightly dif-ferent. Solely band 3 and 8 were needed for these maps. So instead of layer stacking, the individual tiles of each band were mosaicked (without color-balancing to preserve original values). This raster was then, just like the LULC, clipped to the shapefile of our study area. The final raster images were saved as an Imagine (.img ) file.
2.3
Classification
The next step for the LULC map was to classify the pre-processed raster of 2018-05-10 in eCognition. This was done using Object Based Image Analysis (OBIA). When per-forming OBIA, pixels are being grouped into objects based on their spectral resemblance. This is called segmentation.
Table 1: Spectral bands and their characteristics from the Sentinel-2 mission. Band B3 and B8 will be used for the NDWIcalculation. Band B2, B3, B4 and B8 will be used for the LULC map. VNIR stands for Visible Near Infrared and SWIR for Short Wave Infrared (European Space Agency, 2018).
The research of Darwish, Leukert and Reinhardt (2003) was used as a guideline to obtain the optimal scale parameter for this segmentation. After some trial-and-error, this value was set to 300 so the polygons of the segmentation fitted the shape of the land cover and land use types best. Consequently, the different seg-ments were assigned to the following classes:
• Water • Typha
• Irrigated agriculture • Forests
• Open areas with trees • Barren soil
The class ’Water’ and Barren soil are self-explanatory. Typha is an invasive species that thrives in wetland habitats. Irrigated agriculture in this area consists mainly of rice, but also at times of tomatoes, auber-gines, unions and others. Forests in this valley consist mainly of acacias and have a high biodiversity. In the rainy season, the water can rise up to a meter in these areas. Open areas with trees have a low biodiversity and a low tree density. They are often located close to or at abandoned agricultural fields. These are the
classes that could be separated most correctly 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 in May (ac-quisition date of the used satellite image), the flooding is minimal. During the classification, eCognition auto-matically adds the class ’unclassified’ for the areas that could not be classified. In order to run the classifica-tion, training sites had to be appointed to each class. These training sites were based on field observations of locations within the study area that were provided by ecological consultancy Altenburg & Wymenga. The coordinates of the field observation points of Alten-burg & Wymenga had to be converted from decimal degrees to degrees and decimal minutes. The full con-version method and coordinates of these observation points can be found in Appendix B. The point loca-tions were then displayed in Google Earth and saved as layer file (.lyr) to visualize the locations in ArcMap. With the coordinates of the field observations, Sentinel-2A imagery and the help of Google Earth, the different classes could be identified on the satellite imagery and subsequently classified for the full study area. After the classification, small pixels were filtered out using the “majority filter” tool in ArcMap.
2.4
Accuracy assessment
There are various methods of accuracy assessment which have been described in remote sensing literature (e.g. (Aronoff, 1985; Kalkhan, Reich & Czaplewski, 1995; Koukoulas & Blackburn, 2001). However, the most widely used method is derived from a confusion matrix ((Congalton, 1991). The accuracy assessment was performed in ArcMap. The sampling strategy for the distribution of points was “stratified random”. This method generates points that are randomly dis-tributed within each class, where the number of points in each class is proportional to its relative area (Buja, 2012). According to Congalton (1991) a good rule of thumb is to collect a minimum of 50 samples for each land use category. For this research this sums up to a total minimum of 350 sampling points for the accur-acy assessment. Eventually, 366 points were assessed. During the assessment, each sampling point was given the value 0 or 1, indicating whether the classified area was assigned to the correct class (1) or not (0). In case of a false classification, the correct class (ground truth) was also assigned to the sampling point. At the end of the accuracy assessment the total accuracy, producer’s accuracy and user’s accuracy were computed using re-spectively the following three formulas (1, 2 and 3):
Total accuracy(%) =
100 ∗ number of correctly classified points
total number of points (1)
Producer’s accuracy(%) =
100 ∗ number of correctly classified points in class column total of class
(2)
User’s accuracy(%) =
100 ∗ number of correctly classified points in class row total of class
(3) The total accuracy(1), or average accuracy, is the frac-tion of the accurately classified points with regard to the total number of points. When looking at a con-fusion matrix, the producer’s accuracy (2) is the frac-tion of the correctly classified points in a column with
regard to the total number of points in that column. The producer’s accuracy is from the point of view of the producer of the map and shows how often the real land cover is classified as that certain land cover. The user’s accuracy (3) is the fraction of the correctly classified points in row, with regard to the total number of points in that row. The user’s accuracy is from the point of view of a map user and tells you how often the land cover you see on the map, will actually be seen in reality (Congalton, 1991). Usually, the average-, producer’s-and user’s accuracy are not the same. These accuracy’s were therefore compared to evaluate how appropriate the map is.
2.5
NDWI calculation
For the NDWI map, the pre-processed raster’s of band 3 and band 8 were used to calculate the NDWI values. These values are calculated with the following formula (4), where X is a value between minus one and one. (Gao, 1996, p.258).
X = (B03 − B08)
(B03 + B08) (4)
This calculation was computed in ERDAS using the spatial model editor for normalized differences. Ap-pendix C shows how this formula was incorporated in the model. For this map, there were only two classes, namely water and no water (other). The NDWI values range from -1 to 1, where a value between -1 and 0 is no water and a value between 0 and 1 is water (Gao, 1996). The class width for water was set between 0 and 1 and the class width for other was set from -1 to 0. The created NDWI map was then used for cal-culating the total water area for May and September and the total change of water area between May and September. The values of the NDWI map had to be converted from stretched values to unique values and from non-integers to integers before the attribute table could be build. Subsequently, the following formulas were used to calculate the total area (5), relative area of water (6) and total change of area (7):
Total water area(km2) =number of pixels ∗ resolution 106
(5)
Relative water area(%) =100 ∗ total water area total study area (6)
Area change(km2) =
total water area September − total water area May (7) In the formula for the total area of water( 5), the num-ber of pixels is yet to be determined, the resolution of one pixel is 10x10 meter and this is divided by one mil-lion to convert to square kilometers. Formula 6 and 7 are self-explanatory.
3
Results
The generated maps are shown in Figure 3, 4 and 5. Tables 2 and 3 provide the corresponding statistics of these maps. Figure 33.a shows the LULC map of the entire study area for May 2018 and Figure 33.b shows a magnified view of the upper left area of the valley. Blue indicates water, brown indicates Typha, light green in-dicates open area with trees, red inin-dicates irrigated agriculture and grey indicates bare soil. Table 3 con-tains the obtained values of the accuracy assessment of the LULC map in a confusion matrix. The matrix should be interpreted as follows: typha (first column) has correctly been classified as typha five times. It has
also been wrongly classified as open areas with trees two times and as bare soil and forest once. In total (column total) Typha has been classified nine times, of which five times correctly.There is no correspond-ing column of ’unclassified’, because this is not an as-signed class. The producer will assign every pixel into a class and only the program assigns a pixel to ’un-classified’ when it can not classify it as anything else. Hence, there is also no user’s or producer’s accuracy of this class. When looking at all classifications, the pro-ducers accuracy is higher than the users accuracy for typha and open area with trees, lower than the users accuracy for water, bare soil and forest and equal to the users accuracy for irrigated agriculture. The total accuracy of the LULC map is 79% (286/366). Figure 4 and 5 show the NDWI maps. Figure 44.a shows the NDWI map of the entire study area for May 2016 and Figure 44.b shows a magnified view of the upper area of the valley. Figure 55.a and 55.b show respectively the same areas, only for September 2016 when there was a major flood. On both maps, black areas are wa-ter bodies and grey/white areas are considered ’other’ or at least not water. Table (2) shows the computed statistics of these maps. The total area of water bod-ies in May was 145.2km2and in September 1110.8km2. This indicates an absolute increase of 965.6km2 water area and a relative increase of 13.6%.
Table 2: Statistics on the NDWI map. The number of pixels have to be multiplied by 1000, the total area is in square kilometers and the relative area in percentage. The numbers have been completed.
Date Pixels×1000 Total area (km2) Relative area (%)
May Water 1452 145.2 2.1 (2016-05-10) Other 68833 6883.3 97.9 Total 70285 7028.5 100 September Water 11108 1110.8 15.8 (2016-10-27) Other 59177 5917.7 84.2 Total 70285 7028.5 100
Table 3: Confusion matrix of the results of the accuracy assessment. The rows represent the classified results and the column the ground truth (or reality).
Ground Truth Typha Open area
with trees Water Bare soil Forest Irrigated agriculture Row total users accuracy(%) Classification resul ts Typha 5 2 0 1 1 1 10 50
Open area with trees 2 152 0 31 0 1 186 81.7
Water 0 2 9 0 1 0 12 75 Bare soil 1 13 1 105 1 0 124 84.7 Forest 1 0 0 1 7 0 11 63.6 Irrigated agriculture 0 0 0 0 2 11 13 84.6 Unclassified 0 0 8 0 0 0 Column total 9 169 18 138 12 13 366 Producers accuracy(%) 55.6 89.9 50 76.1 58.3
(3.a)
(3.b) LULC map of the entire study area in May 2018 (a) and a magnification to show the detail of the map (b).
(4.a)
(4.b) NDWI map of the entire study area in May (a) and a magnification to show the detail (b). Black areas are water, all other areas varying from grey to white are considered not water or other
(5.a)
(5.b) NDWI map of the entire study area in September (a) and a magnification to show the detail of the map (b). Black areas are water and grey or white areas are considered not water or other.
4
Discussion
4.1
Methods
Before switching to eCognition, ERDAS was used to perform a supervised classification with maximum like-lihood. Unfortunately, irrigated agriculture was clas-sified too often as typha and most of the forest was classified as city (this was a land cover type in the beginning of this research. Later, it was decided to leave this type out as it caused too many errors due to its complex spectral reflection). When leaving Typha out of the classification, the maps were more accurate. However, Typha is an important land cover type in the valley and therefore the switch to eCognition was made. Since there was time nor money for field work in this area, the field observations from Altenburg & Wymenga were used. These were however limited in quantity and variety. Together with a resolution of just 10x10 meters, it became very hard to see the difference between flood recession agriculture and trees or open area with trees. Initially, an objective of this research was to incorporate a Digital Elevation Model (DEM) into the NDWI maps. The highest resolution available at the time of this research was 30x30 meter of the AS-TER satellite. Figure 6 shows the created DEM-map. As you can see, there are no distinct patterns visible. It was concluded that the DEM was unfitting, even after trying to exaggerate the differences with for example the hillshade tool in ArcMap. Another objective was to reproduce the NDWI maps for the months in between May and September. Unfortunately, there were too many clouds on all the available imagery, limiting the options to May and September. The final NDWI map is gray-scaled. This was preferred over a map with blueish colors for aesthetic reasons.
Figure 6: DEM of a section of the catchment. Blue indicates a relatively low area and red a relatively high area.
4.2
Interpretation of results
The generated maps meet the requirements set prior to this research. They are detailed, cover the entire study area and contain all classes The LULC map clearly shows the distribution of different types of land cover and land use throughout the valley. The confusion matrix of the results of the accuracy assessment (Table 3) demonstrates that open area with trees, barren soil and irrigated agriculture can be classified very precise. A notable occurrence is the confusion of open area with trees with barren soil (thirteen times) and barren soil with open area with trees (thirty-one times). This im-plies that these areas are very much alike in shape and spectral reflection. This was confirmed by Dhr. Dr. I. Mettrop from Altenburg and Wymenga (2018a) who has been to the study area for field work. Figure 7a-c, shows photo’s of the study area for respectively barren soil, open area with trees and forest.
As you can see, these three types of land cover have many similarities. In reality, open area with trees is of-ten simply barren soil in a further transition towards forest.
The producers accuracy is higher than the users ac-curacy for typha and open area with trees, lower than the users accuracy for water, bare soil and forest and equal to the users accuracy for irrigated agriculture. This means that even though for example typha has correctly been as such 55.6 times, only 50% of the areas identified as typha were actually typha. The users and producers accuracy are relatively close to one another, resulting in a relatively high overall accuracy. Only the users and producers accuracy of water and forest differ a lot. For the former, the NDWI maps can be used if a higher accuracy is desired. For the latter, more field observation would be recommended. Ac-cording to Landis and Koch (1977) a value greater than 0.80 (i.e., 80%) represents strong agreement; a value between 0.40 and 0.80 (i.e., 40-80%) represents moderate agreement; and a value below 0.40 (i.e., 40%) represents poor agreement. The accuracy of the LULC map was 79%, which represents moderate agreement, but leans toward a strong agreement. For compar-ison, the global land cover map of the IGBP has an area weighted accuracy of 66.9%. Trodd (1995), found that the mean of the reviewed producer’s accuracy was merely 59%. Foody (2002) states that the accuracy often decreases with increased study area. Therefore, when considering the size of the classification area, this is an acceptable result and answers the first research question: What is the distribution of land cover types in the Senegal river valley?
The second research question: “What is the spatial-temporal extent of flooding in the Senegal river val-ley?” can be answered using the NDWI maps of May 2016 and September 2016. In May, there had been a dry period for months and the water level was at a low point. In September, floods occurred under the in-fluence of ongoing monsoons (Altenburg & Wymenga, 2018b; Gueye et al., 2011). When comparing these two maps and statistics 2, the absolute increase of wa-ter area was 965.6km2and the relative increase 13.6%. The total water area in May (145.2km2) could be con-sidered as the minimal water area that corresponds with the base flow of the river. If you then extract the minimum from the maximum (total water area September: 1110.8km2) you get the area (965.6km2) that is subject to flooding and “unflooding” and should
be suitable for flood recession agriculture. This, how-ever is just a speculation. The results have to be valid-ated in the field and more parameters, like soil quality, should be taken into account before an area can be considered “suitable” for flood recession agriculture.
Altenburg & Wymenga also provided a database containing gauge levels and discharge values of mul-tiple measurement stations throughout the catchment. In theory, it should now be possible to determine which water level corresponds with which amount of flooded area. However, in an attempt to do so, several implic-ations emerged. First, the locimplic-ations of the measuring stations were not ideal. For example, there was no sta-tion located at or near the dam. This made it very hard to determine how much water was flowing through the dam, because the more downstream, the more chan-nels convene. Second, Dhr. Dr. I. Mettrop remarked that the gauge levels are measured by hand, due to lim-ited equipment. Much research (Andersen, Refsgaard & Jensen, 2001; J. Bader, Cauchy, Saura & Duffar, 2014; J.-C. Bader & Albergel, 2015; Lamagat & Bader, 2001) has already been conducted on determining the rivers dicharge and also in more depth. Therefore, the decision was made to let it be for this research.
4.3
Further research
The recommendations for further research are fourfold. First, it would be beneficial to link the flooding area to the water levels. For this, an in-depth quantitative analysis of the river flow is needed. When you know which water level corresponds to which flood extent, the OMVS can adjust the outflow of the dam to the water levels they measure downstream. Also, no wa-ter would be wasted on floods that are too high or at the wrong time. The second recommendation concerns the flood areas. For the farmers, it would be profitable to know which areas are suitable for flood recession agriculture. The NDWI maps provide information on where these floods are, but more research is needed to determine whether those areas are sufficient. For example the soil type, chemical composition and sur-roundings have to be analyzed. The LULC map could be helpful for determining the current land cover and land use types, although it could not classify the flood recession crops. This brings me to the third recom-mendation: increasing the accuracy of the LULC map. With more field observations, it might be possible to achieve a higher accuracy of the LULC map. Perhaps
it would then also be feasible to include more classes. This was not possible to do during this research due to limited time and budget. Lastly, the NDWI could not be calculated for the months July, August and Oc-tober 2016 (even though this was desirable). Perhaps in future research, other imagery than Sentinel (e.g. Landsat) could be used to perform an NDWI analysis. This way, the development of floods could be mapped and analyzed to see if there are any trends.
5
Conclusion
The aim of this research was to contribute to the know-ledge base of the middle valley of the Senegal river and develop methodologies to produce two types of maps. The LULC map of the middle valley of the Senegal and the NDWI maps showing the spatial-temporal extent of water bodies in May and September 2016 were
cre-ated successfully. The users and producers accuracy of the LULC map are relatively close to one another, resulting in a relatively high overall accuracy, namely 79%. After comparing this result to the results of sim-ilar research, it can be concluded that this is sufficient. The absolute increase of water area was 965.6km2 and the relative increase 13.6%. Unfortunately,, there was no sufficient data to link these maps with corres-ponding gauge levels. Therefore, a quantitative ana-lysis is needed to determine the exact amount of wa-ter flowing through the catchment at times of flooding as well as an analysis on the flooding areas that were computed for the NDWI map . Also, more field obser-vations could result in a higher accuracy of the LULC map. Nonetheless, it can be concluded that the maps are a valuable contribution to the knowledge base of the Senegal river valley and the methodology can be used for future research to create similar maps.
Acknowledgments
It is with great pride that I hereby present my bachelor thesis. For the past two months I have layer stacked, mosaicked, segmented, classified and accuracy assessed for weeks. During this research I have gained a lot of knowledge which was not just beneficial for my thesis, but will be for my future studies and career. I would like to express my gratitude to my supervisor Dr. A.C. Seijmonsbergen for his encouragement, enthusiasm and expertise. Working with such a dedicated supervisor is a delight. Furthermore, I would like to thank Ivan Mettrop from Altenburg & Wymenga for his time and for sharing his knowledge and datasets about the current situation in Senegal. Our collaboration was pleasant during the research and has proved to be fruitful in the end. Lastly, I would like to thank my fellow BSc student R.A.L Bossen and C. Juijn for their excellent cooperation during the process of making the LULC map.
Glossary
DEM Digital Elevation Model. A DEM is created from elevation data and gives a 3D representation of the Earth’s surface (or even another planet’s surface)..
ESA scihub European Space Agency Science Hub. The European Space Agency (ESA) has an open access hub. This hub provides free access to Sentinel-1, 2 and 3 products. This hub can be accessed via: https://scihub.copernicus.eu/.
LULC Land Use and Land Cover. Land cover is the physical material, such as water, bare ground and trees, at the Earth’s surface. Land use describes the utilization of the surface of the earth, such as agriculture.. NDWI Normalized Difference Water Index. NDWI is a very accurate way of mapping water bodies, due to the strong absorbability and small range of wavelengths of water. For mapping the change of water (e.g. floods), the green band (3) and the Near Infra-red band (NIR/8) are used. The values range from -1 to 1, where a value between -1 and 0 is no water and a value between 0 and 1 is water. Liquid water bodies however, have a value ¿ 0.5, which is significantly higher than for example grasslands (around 0.2)(Gao, 1996).
OBIA Object Based Image Analysis. Object based image analysis is a type of analysis used in remote sensing. When performing OBIA, pixels are being grouped into objects based on their spectral resemblance. The larger the scale parameter, the wide the range under which pixels are called ”similar” and the larger the segregates..
OMVS l’Organisation pour la mise en valeur du fleuve S´en´egal. The OMVS is an intergovernmental organ-isation of four countries: Guinea, Mali, Mauritania and Senegal. The OMVS was formed in March 1972 and has the objectives to increase food security, produce energy and reduce poverty in the Senegal river catchment and improve navigability of the river..
References
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).
Altenburg & Wymenga. (2018a). Bachelor studentopdracht uva i.s.m. ecologisch adviesbureau altenburg & wymenga.
Altenburg & Wymenga. (2018b). Jan1980-jan2018[microsoft ecxel spreadsheet]. Retrieved from: Altenburg and Wymenga [accessed 28 May 2018.
Andersen, J., Refsgaard, J. C. & Jensen, K. H. (2001). Distributed hydrological modelling of the senegal river basinmodel construction and validation. Journal of Hydrology, 247 (3-4), 200–214.
Aronoff, S. (1985). The minimum accuracy value as an index of classification accuracy. Photogrammetric Engineering and Remote Sensing, 51 (1), 99–111.
Bader, J., Cauchy, S., Saura, P. & Duffar, L. (2014). Monographie hydrologique du fleuve s´en´egal. IRD, Montpellier.(sous presse).
Bader, J.-C. & Albergel, J. (2015). Artificial flood support on senegal river: a challenge to protect natural resources in the valley. Montpellier : L’Institut de Recherche pour le Dveloppement(IRD).
Bossen, R. (2018, 7). An evaluation: mapping temporal land use and land cover (lulc) for the senegal catchment to quantify the changed area. (University of Amsterdam (dissertation))
Buja, K. (2012). Arcgis: Sampling design tool documentation. NOAA´s Biogeography Branch. http:// www.arcgis.com/home/item.html?id=f7289cfc69204aa688e8c7c739fc0901.
Congalton, R. G. (1991). A review of assessing the accuracy of classifications of remotely sensed data. Remote sensing of environment , 37 (1), 35–46.
Darwish, A., Leukert, K. & Reinhardt, W. (2003). Image segmentation for the purpose of object-based classific-ation. In Geoscience and remote sensing symposium, 2003. igarss’03. proceedings. 2003 ieee international (Vol. 3, pp. 2039–2041).
European Space Agency, E. (2018). Copernicus open acces hub (dataset). https://scihub.copernicus.eu/ dhus/#/home. (Accessed: 2018-06-12)
Foody, G. M. (2002). Status of land cover classification accuracy assessment. Remote sensing of environment , 80 (1), 185–201.
Gao, B.-C. (1996). Ndwia normalized difference water index for remote sensing of vegetation liquid water from space. Remote sensing of environment , 58 (3), 257–266.
Gueye, A., Janicot, S., Niang, A., Sawadogo, S., Sultan, B., Diongue-Niang, A. & Thiria, S. (2011). Weather regimes over senegal during the summer monsoon season using self-organizing maps and hierarchical ascendant classification. part i: synoptic time scale. Climate dynamics, 36 (1-2), 1–18.
Horowitz, M. M. & Salem-Murdock, M. (1993). Development-induced food insecurity in the middle senegal valley. GeoJournal , 30 (2), 179–184.
Juijn, C. (2018, 7). Remote sensing analysis of the land use and land cover change in the senegal river valley. (University of Amsterdam (dissertation))
Kalkhan, M. A., Reich, R. M. & Czaplewski, R. L. (1995). Statistical properties of five indices in assessing the accuracy of remotely sensed data using simple random sampling. In Proceedings acsm/asprs annual convention and exposition (Vol. 2, pp. 246–257).
Koukoulas, S. & Blackburn, G. A. (2001). Introducing new indices for accuracy evaluation of classified images representing semi-natural woodland environments. Photogrammetric Engineering and Remote Sensing, 67 (4), 499–510.
Lamagat, J. & Bader, J. (2001). Programme doptimisation de la gestion des rservoirs, phase iii, syhthese. Projet dOptimisation de la Gestion des Rservoirs (POGR).
Man´e, L. & Fraval, P. (2001). Suivi par teledetection des cultures de decrue dans la vallee du fleuve senegal en saison froide 2000/2001. in frenchestimating crop yield in flood recession agriculture in the senegal river valley using remote sensing.
Musser, K. (2010,.sig). Senegal river map (elevation data from srtm, drainage basin from gtopo, all other features from vector map). Retrieved from https://commons.wikimedia.org/w/index.php?curid=9822604 OMVS. (2009). Le barrage de manantali. rapport de synthese, 1-48. (retrieved 22 May 2018 from: http://
www.eib.org/attachments/ev/ev manantali rapport de synthese fr.pdf)
Sambou, S., Di´em´e, Y., Tour´e, A. K., Badji, A. M. & Malanda-Nimy, E. N. (2009). Effet du barrage de manantali sur les modifications du r´egime hydrologique du fleuve s´en´egal dans le bassin amont: une approche statistique. Science et changements plan´etaires/S´echeresse, 20 (1), 104–111.
Trodd, N. (1995). Uncertainty in land cover mapping for modelling land cover change. Proceedings RSS95: Remote Sensing in Action, 1138–1145.
Uhlir, P. & Council, N. R. (2003). Scientific data for decision making toward sustainable development: Senegal river basin case study: summary of a workshop. National Academies Press.
Varis, O. & Fraboulet-Jussila, S. (2002). Water resources development in the lower senegal river basin: con-flicting interests, environmental concerns and policy options. International Journal of Water Resources Development , 18 (2), 245–260.
Appendices
A
Satellite data
Specifics: Sentinel 2A, UTM, zone 28 WGS 84, 10x10 resolution. Source: https://scihub.copernicus.eu/ dhus/#/home Remarks: For the first six satellite images, the each product was a different tile. For the other twelve images, some products contained several tiles, hence the occurrence of double product names.
B
Field observations
The coordinates were converted from decimal degrees to degrees and decimal minutes: Example: convert from 16.53746562 N to degrees and decimal minutes
Multiply the decimal part (53746562) times the number of minutes in a degree (60) and replace the decimal degrees with the minutes.
53746562x 60 = 32.24794 So the answer is 32 24794 N
C
NDWI calculation
These values are calculated with the following formula (8), where X is a value between minus one and one. (Gao, 1996, p.258).
X = (B03 − B08)
(B03 + B08) (8)
In the spatial model editor of ERDAS this formula is constructed as follows (Figure 8):
Figure 8: Spatial model of normalized difference calculation in ERDAS. “Raster input” is band 3 and band 8. These are first subtracted (3-8) and then divided by the sum (3+8). The result is merged with a correction factor “stretch” and then saved as a raster file named: ndwi map.img
