Many people have contributed to this piece of work and I would like to thank them all. First, many thanks to my supervisor Dr Gareth Roberts for his guidance throughout this process. Without his knowledge in 'all things remote sensing' this study certainly would not be what it is.
A very special thank you to Dr. Nikoli Shadrin from the Institute of Marine Biological Research, Russia for his correspondence and expertise on saline lakes and generously contributing thoughts from inception to critical review, your published works provided an invaluable resource for this budding researcher. My thanks also to Dr.Ragab Elsheikh (General Manager of R&D at EMISAL Co.) for your kind permission to use the in–situ measurements present in this study, without which the methodology used, would have not been possible.
Thank you also to those who have generously offered their time to proof read; your contributions have been invaluable.
Finally, a very special thank you to Kate Williams for her critical appraisal in the writing of this report. Without your expert guidance, this piece of work certainly would not be what it is.
Saline lakes are characterised as lakes with high concentrations of salt (Zheng, 1999), and represent 0.006% of global surface water (Williams, 1999) comprising a global volume of 85km 3x10 6 (Chenchouni, 2016).). Saline lakes are therefore a significant component of the biosphere, but they also have a number of critical uses and values (Williams, 1999). They are generally restricted to closed, 'endorheic' drainage basins in semi–arid and arid regions (Moore, 2016). These areas are characterised by low annual rainfall (25–500mm), high evaporation rates and currently comprise 1/3 of the total global land surface (Williams, 1993). The issues affecting saline lakes and their impact is acknowledged to have a critical influence in these regions, with an estimated population of 400million (Williams, 1999); although rapid population growth in start the 21 st century means this is likely to have increased considerably. Their global relevance is likely to increase further with the extension of drylands as a result of climate warming (IPCC, 2014; Digna et al, 2016).
Many of the issues affecting contemporary saline lakes globally are brought into sharp focus by a consideration of Lake Qarun, Egypt, which experienced a substantial increase in salinity since through the 20 th century. Since 1986, the Egyptian Salts and Mineral Company (EMISAL) have operated on the southern coast of the lake and its activities are credited with the balance regulation of salinity since (El–Shabrawy and Dumont, 2009), however levels are salinity is still rising. The lake has a variety of important uses, but its status will remain under threat unless the causes of these changes are understood.
The impact of land use and land cover change on other contemporary saline lakes has been well studied in many regions around the world (Flower et al, 2006; Li et al, 2009; Jeppesen et al, 2015). These studies have mainly drawn upon archaeological and paleolimnological evidence to determine the impact of historical land usage of lake systems. While in–situ measurements remain the most accurate method for quantifying lake water quality parameters (Digna et al 2016) understanding the mechanisms which drive variation in water quality and predicting situations where this may occur is more troublesome (Löw et al, 2013). The monitoring of lake water dynamics requires larger scale monitoring and remote sensing techniques are therefore considered desirable (Dube et al, 2015) and therefore remote sensing technology is increasingly being used for the contemporary observation land use changes (Li et al, 2009).
Though the effects of changing land use practices on water quality characteristics is well known (Williams, 1993; Necsoiu, 2013), at present, no studies in Qarun have been successful at mapping this change to determining how catchment wide processes have affected lake water salinity. The study therefore aims to evaluate the extent to which changing agricultural land cover has affected water salinity in Lake Qarun, Egypt.
The research aims and objectives are summarised in the process flow diagram (Figure 1), showing how literature review and chosen methodology are combined to resolve the research question
The exact threshold beyond which a 'lake' is to be referred to as a 'saline lake' has been debated between disciplines (Williams, 1996; Zheng, 2001; Tweed et al, 2011). For the ease of comparison, Zheng (1999) categorises saline lakes as either 'sensu lato' or 'sensu stricto'; incidentally these terms are Latin for 'the strict sense of the term' and in 'the broad sense of the term'. The lower limit for lake sensu stricto is above the average salinity of global sea water (>3.5wt% (NaCl)), and is the point at which large quantities of salts precipitate (Abdel Wahed et al, 2015). Sensu lato lakes are brackish bodies of water, characterised by salinities ≥0.3wt% (NaCl); the point at which the water is no longer suitable for supporting fresh water organisms (El–Shabrawy et al, 2016) and where salt is sensible to human taste (Zheng et al, 2004).
Saline lakes are a significant component of the biosphere, but they also have a number of critical uses and values. Saline lakes have a great economic use as a source of minerals, power and in some areas, as a source of fish of commercial importance (FWMP, 1999). Salt lakes are considered of great importance as 'sentinels' for climatic forcing (Tweed et al, 2011), to which they are very sensitive, and provide a holographic record of environmental change and the impact of hydrological stresses on a catchment, characterised by area, volume (Zheng et al, 2004) and hydro chemical properties (Tweed et al, 2011). The uses and values of saline lakes are increasingly subject to degradation from a variety of impacts, predominantly as a result of human activity.
In natural or 'primary' systems, salt accumulates in the terminal basin and is enriched by solar radiation (Williams, 1993), evolving from freshwater to brackish water bodies that are in dynamic balance with climate and the environment (Zheng, 2004). Salt is naturally lost by deflation and seepage (Williams, 1999), which results in a natural equilibrium below saturation level. In endorheic basins, lakes have a rapid response to hydrological stress (Abdel Satar et al, 2010). Fritz (1990) determines that volume and salinity are inversely related, and that changes in salinity are likely to be as a result from volume loss as salt content remains constant. Although this may come as a result of climate forcing, the salinity of many world lakes is rising to toxic levels due to human activity (Gad et al, 2011; Jeppesen, et al 2015).
Secondary salinization is caused by the disturbance of the balance present in primary systems and is an increasing issue for lakes around the world. Williams (1999) cite the main driver for secondary salinization is catchment activities which change the natural hydrological balance and mobilise underground salts; this viewpoint is well supported in subsequent literature (Löw, 2013; Yan and Zheng, 2015; Moore, 2016). It is recognised that salinity is not just associated with natural ocean derived salt inputs from precipitation, but with inputs from the land during runoff and mineral precipitation (Tweed et al, 2011). It is therefore important to consider agriculture in relation to changing salinity within saline lakes.
According to Tweed et al (2011), the greatest impact on salt lakes in the last century has come from agriculture and mineral extraction. Agricultural activities represent a major change in land use, characterised by the replacement of deep rooted native vegetation with shallow rooted crops (Tweed et al, 2011). Salinization results from two mechanisms, catchment clearance (resulting in dryland salinity) and irrigation which results in the elevation of groundwater (Williams, 1999). Saline ground water is then able to reach the surface through capillary action alone, where surface deposition resulting from evaporitic processes means it can be leached into catchment waters (Williams, 1999).
The issues affecting contemporary saline lakes maybe be brought into broad focus by through the study of Lake Quran, Egypt.
The long–term salinity change in Lake Qarun is well studied (Meshal, 1977; Abdel Satar, 2010) as a result of the site's regional importance as the main source of fish for the Fayoum Province (FWMP, 1999), its cultural value (Keatings et al, 2010) and more recently, owing to the site obtaining nature reserve status in 1989 (Abdel Wahed, 2015) and as an economic resource for salt extraction (FWMP, 1999). This comprehensive summary of the present state of knowledge on salinity change in Lake Qarun considers a historical review of salinity change through the appraisal of lake water levels, geo–archaeological and paleolimnological studies. This is followed by a review of salinity change in the instrumental record from published literature.
Lake Qarun is the remnant of a natural freshwater reservoir which previously occupied the whole of the Fayoum Depression at about 15 metres above sea level (Ball, 1939). Keatings et al, 2010 conceived through archaeological evidence that since the Pharaonic period (approximately 5000 years before present), freshwater inputs have been placed under human influence, evidenced by the damming of the communication with the Nile and the construction of the Bahr Yousef canal to form Lake Moeris. At this stage the lake was utilised as a basin for Nile flood water and as a freshwater resource for the Nile Valley (Brown, 1892; Ball, 1939). The salinity level is therefore likely to have remained low and relatively constant due to surplus freshwater inputs. Lake level variability prior to this period is likely to be as a result of natural climate changes (Hassan, 1989) which may have resulted in minor fluctuations in salinity.
Geo–archaeological evidence from Hassan (1986) indicates the drastic drainage of land for agriculture between 323–30 BC which resulting in a fall in lake level. These findings are reinforced in Keatings et al (2010), whom quantify a change from around 18metres above sea level to around 5 metres below sea level, exposing 1300km 2 of land. This severe fall in lake volume alone is expected to have resulted in a rise in salinity (cc Fritz, 1990), although the weathering of newly exposed soils would have resulted in the remobilisation of salts into the lake. This is likely to result in a further increase in salinity in a similar process to that which can be observed in Lake Qarun at present (Mohamed, 2015); as subsequent to the reclamation of the land for agriculture, drainage water from these lands was directed to flow into Lake Moeris (Meshal, 1977). Archaeological evidence from Keatings et al (2010) indicates the progressive fall in lake level up to the beginning of the 20 th Century associated with the loss of function of many canals through siltation, whereby a fall in lake volume can again be associated with increase in salt concentration (cc Fritz, 1990), although at a rate which is far less than observed in the 20 th century (Meshal, 1977; Flower et al, 2006).
The existing saline Lake Qarun is therefore the shrunken remnant of the freshwater Lake Moeris. The historical trend can be summarised as a progressive increase in salinity from natural freshwater reservoir, due to a reduction in lake level over the last 4000years predominantly as a result of anthropogenic alteration of the catchment hydrological regime.
Instrumental records for salinity of Lake Qarun exist since 1901, and direct sampling methodology demonstrates a strong increase in salinity gradient through the 20 th Century as summarized (Figure 1). The rise in salinity is of particular scientific interest, as it contradicts the paradigm that volume and salinity are inversely related (cc Fritz, 1990), as lake level is reported to have risen during this period (Baioumy et al, 2010). The considerable increase in salinity is considered to be separate from climatic induced factors and attributed to the dramatic alteration of hydrological regime through the period (see below), which has resulted in the considerable accumulation of salts (Figure 2). In–situ measurements have been corrected to Practical Salinity Units (PSU) to correspond with contemporary literature.
Figure 2 (above): Diagram showing continuous salinity gradient in Lake Qarun from measurements reported in literature, between 1901 and 2012. Salinity levels during the period of study are characterised by approaching that of sea water (or sensu stricto) Source: Figure based on data derived from existing literature
Figure 3: Average annual salt accumulation in Lake Quran during the 20 th Century quantified in existing literature
The first quantification of the water salinity in Lake Qarun was carried out by Lucas (1906), considering 'salinity' as the total dissolved salts expressed in grams per litre (g/l). Similar methodology has been adopted in studies from Ball (1939) and Fouad (1928), providing a record of average salinity levels on a near–annual basis until 1932. Meshal (1977) in a reappraisal of these measurements considered them to be and overestimation of salinity as contemporary definition considers it to be less than salt content (Harvey, 1957). Quantification of lake salinity was undertaken by Naguib (1958) using Knudsen's Hydrographical Tables (cc. Knudsen, 1901); estimating salinity from measured levels of chlorinity. This reading for average salinity is inconsistent with the observed trend (Figure 1). In a reappraisal of the existing instrumental record, Meshal (1977) considers the anomalous result to be as a result of the methods intended for use in seawater which contains different components to those which are present in the lake water column. The revised reading of 27.4% proposed is still inconsistent with the observed trend, though this correlated with the increase in volume observed at that period (Flower, et al 2006).
Monthly determination of average salinity undertaken by Meshal (1973) quantified salinity using measurements of electrical conductivity (EC), where values were converted to salinity through the use of an empirical relationship established for Lake Qarun. This method has since been adopted in a number of more recent studies (Bitelli et al, 2011; Shadrin et al, 2016). The measured salinity is almost identical to that reported 40 years previously (Ball, 1939), suggesting salinity increase slowed in this time and is again likely to be as a result of lake level, resulting from elevated Nile flood and promoted perennial water availability due to the construction of the Aswan High Dam (Flower et al, 2006).
An enhancement in average salinity observed between 1974–77 is reported to be resulting from the accumulation of salts from drainage water (Abdel–Malek and Ishak, 1980), where greatest salinity is measured to the North East decreasing to the West, though subsequent studies have found salinity to vary considerably on an inter–annual basis) resulting from seasonal variation in drainage water inputs from the El–Wadi and El–Batts drains (El–Shabrawy et al, 2014; Shadrin, et al 2016). Volume increase during 1978–1980 was twice the historic rate, increasing 20cm/yr. This led to the opening of the Wadi El Rayan diversion (Mansour and Sidky, 2003). A reduction in lake water level to 43.5 metres below sea level is largely attributed to the increase an annual increase in salinity of 0.36‰/yr to 34.48 by 1989 (Abdh–Ellah (2009) (Table 1), which is close to that of seawater 'sensu stricto' (Zheng, 1999).
This period is characterized by average salinity levels exceeding levels of global average sea water, and becoming 'sensu stricto' (cc Zheng, 1999). The trend since 1989 appears to show relative stability (Figure 1) and is the result of commercial salt extraction, which has taken place since 1986 (FWMP, 1999). The annual intake of saline water from the lake has increased to present 7.5 × 10 6 m3 in 1986 to 20 × 10 6 m3 in 1997, although this is not considered to have a significant effect on the overall salinity of the lake (Abd–Ellah, 1999). El–Shabrawy (2016) suggests that annual salt accumulation Qarun is still as much as 70–85 mln.kg/year, resulting in a mean increase in salinity (cc Table 2). It is generally considered that the average salinity of the lake has remained relatively constant throughout this period. Through increasing in–situ measurements, Ali (2002) and Sabae & Ali (2004) determined that salinity (and pollution) varied considerably through the lake. Because this study relied on in–situ methods, it could only give an approximate account of spatial variability. Bitelli, et al, 2011 combined in–situ measurements with Landsat ETM+ to salinity variability in the lake; this was shown to have a high degree of accuracy. The use of remote sensing techniques enables both the geochemical characteristics of the lake and the dynamics of catchment activities to be observed simultaneously. Both are therefore needed to be considered separately.
Few studies exist quantifying the change in land use of Lake Qarun. The evidence for historical and contemporary changes in agricultural extent in the Fayoum Area has been reviewed from a few in–situ studies of geo–archaeological and paleolimnological pollen assemblage analysis analysed from shallow cores take from the lake, and some governmental reviews of agricultural policy. The literature agrees that the expansion and intensification of agriculture through this period has been the result of population increase.
Historic accounts indicate that the area was fertile (Mehringer et al, 1979); suggesting agriculture has long been an aspect of the catchment hydrological regime. Geo–archaeological evidence shows the human influence of water inputs and the impact of drainage of land for agriculture between 323–30 BC; resulting in a fall in lake level (Flower et al, 2006). These findings are reinforced by Keatings et al (2010). Bichara and Baldwin (1985) reason that this period of drainage was short lived as a result of the relative stability of salinity levels. Subsequent studies have indicated a decline in agriculture from 150 BC evidenced through canal siltation and evidence of desertification (Flower et al, 2006). An increase in the abundance of Maize pollen from 1700 and throughout the century is considered to be indicative of an intensification of agriculture through this period (Mehringer et al, 1978). This is concurrent with the reduction in lake level observed by Flowers et al, 2006. Canal siltation is considered to have resulted in a reduction of irrigated agriculture through the 19 th century is evidenced through paleolimnological analysis (Mehringer et al, 1978), although this has not been quantified.
The increased pollen assemblages observed by Mehringer et al (1979) at the beginning of the 20 th century is associated with the renewed canal maintenance under the agricultural policy of Mohammed Ali as perennial irrigation was developed from 1873 (Flower et al, 2006). In the years 1900–1920, irrigation infrastructure was developed to its present state as a result of agricultural policy and expanding population (Bichara and Baldwin, 1985). Salt accumulation throughout this period is therefore considered to be related to the volume of Nile water brought in for irrigation (Keatings et al, 2010). Analysed pollen from a shallow core taken from the lake. Pollen assemblages indicate an increase in in the abundance of olive, dates and cereals after through the 1920s (Mehringer et al, 1979) which corresponds with the intensification of agriculture in the region after the First World War (Safei, 1940).
Period of Study
Construction of the Aswan high dam began in 1960 resulting in the increased availability of water for irrigation into the area and an intensification of cultivation between 1960 and 1970 (Flower et al, 2006). This corresponds with an increase in lake level and reduced salinity through this period (Meshal, 1977). Increased land reclamation in 1970 resulted in the opening of the Wadi–el–Rayan as drainage repository (Mansour and Sidky, 2003). Between 1985 and 1989, irrigated agriculture capacity increased from 1,428km 2 to 1500km 2 (Keatings et al, 2010) increasing the potential out washing of salts from soils.
Table 1: Reported Agricultural Extent
Although no published literature has indicated agricultural extent during the period of study, an appraisal of population growth is used to indicate land use change. During this period, the population of the Fayoum area has doubled, from 1.25million in 1980, to 3,118,078 in 2006 (Census, 2006), and is reported to have increased further by 1.76% from 2005–2010 (Abdel Wahed et al, 2015). At this stage, it is tentatively assumed that an increase in population results in an increase in intensification of land use activities–and therefore increased urban land cover, however this study aims to quantify this change using appropriate methodology (see next section). The lack of available data on agricultural extent during this period emphasises the need for the present study.
The factors for the progressive increase in salinity observed in this period can be therefore be summarised as resulting from 3 factors:
These are considered to be subsequent to the excessive modification of the catchment hydrological regime to accommodate the intensification agricultural land use activities due to increases in population. At present due to the lack of available literature, the relationship is tentative. Subsequent analysis will seek to quantify the relationship between agricultural land cover and water salinity in Lake Qarun to further consider the evidence for whether changing agricultural land cover has affected water salinity in Lake Qarun.
Figure 5: Location map of the Fayoum region, showing the sampling sites of the in situ water samples, drainage network and the location of EMISAL Salt Extraction Plant (Image acquired: 23/08/2016 Earth Explorer,2016)
Lake Qarun is located at 29°30'N 30°30'E on the Eastern margin of the Libyan Desert (Figure X). It exists in the deepest part of the Fayoum depression (Bitelli et al, 2009), at approximately 45m below sea level at its surface. The lake is shallow, with an average depth of 4m, reaching a maximum of 8m at the centre (Bitelli and Mandanici, 2010). Since 1986, the Egyptian Salts and Mineral Company (EMISAL) have operated on the southern coast of the lake and its activities are credited with the balanced regulation of salinity since (El–Shabrawy and Dumont, 2009), however, salinity levels are still rising.
Due to its proximity to the desert, the lake is situated in hyper–arid Saharan climate (Digna et al, 2016) with low annual rainfall <10mm (Bitelli et al, 2011; Flower et al, 2006). Lowest evaporation occurs in January (1.9mm/day) with highest (7.3mm/day) recorded in June (Ali and Abdel Kawy, 2013).
The irrigation and agricultural system in Fayoum relies mainly on the Nile as a source of freshwater. Qarun is a natural sink for drainage from the area, and is surrounded by cultivated lands to the south and south east, which slope steeply towards the lake (Abdel–Wahed et al, 2015a). Water inputs from the irrigation network are principally through the El–Wadi and El–Batts drains (Figure X) with a volume of about 338 x10 6 m 3/year (Abdel–Wahed et al, 2015b); in addition to about 67.8x10 6 m 3/year from groundwater which is nourished by the percolation of irrigation water from canals (Metwaly et al, 2010).
Qarun is situated in an endorheic basin and therefore water losses are entirely due to evaporation, however some excess drainage water is transferred to the nearby Wadi El Rayan depression (Figure 4). The area of the lake therefore remains relatively stable (El–Shabrawy and Dumont, 2009), with minor seasonal fluctuations (Abdel–Wahel et al, 2015). This closed hydrological setting exacerbates the impacts of human activity on the watershed (Bitelli et al, 2011), resulting in lake water deterioration and increased salinity.
Most cultivated soils in Fayoum Province comprise of Quaternary Deposits of deep alluvial loam or clay, derived from Nile flood (Figure Y). The southern shore is predominantly surrounded by lacustrine deposits, covering an area of 198km 2. These are classified as highly saline, poor productivity soils (Abdel Kawy and Betal, 2013) comprised of claystone, gypsum, and calcareous deposits (Abdel–Wahel et al, 2015). Most of these soils are further degraded by salinization, and sodification (Ali and Abdel Kawy, 2013). An intensification of agricultural activities on these deposits is therefore likely to mobilise more salts, resulting in their deposition in Lake Qarun.
Traditionally, lake water parameters in Qarun have been measured using in situ methods (See Chapter 2, 2:2). In situ techniques allow for direct observation of changes in water quality parameters at local and temporal scales (Meshal, 1977), however such techniques are costly (Yan and Zheng, 2015) which may explain the absence of data at some points throughout the 20 th century. Points based in–situ measurements do not provide data of a sufficient spatial and temporal coverage to allow effective observations of water quality parameters on their own (Dube et al, 2015).
Remote sensing techniques are considered desirable as they provide a means to monitor changes in land cover condition at a multi–year perspective (Dube et al, 2015). Remote sensing can therefore be used to observe lake water environments at a large spatial extent and longer timescales (Löw et al, 2013), in addition to acquiring information from remote or inaccessible areas (Yan and Zheng, 2015), where ground observations are lacking. The multispectral nature of remote sensing allows application to features specific to lake dynamics.
An interest in the use of remote sensing techniques for water quality determination has existed since the 1970s. Spectropic studies have demonstrated how the presence of dissolved salts changes the amplitude and intensity of some water absorption bands in the thermal and infra–red spectrum (Bowden, 1975). The presence of sodium chloride reduces absorption at 1.5 and 2.1 micrometres, where the magnitude of this variation is dependent on the concentration of salts in solution (Dube et al, 2015). Temperature of the solution has an additional effect on the intensity of the affected wavebands (Perkin and Lewis, 1980).
An empirical relationship can therefore be derived between the salt content and solution absorption at a given temperature (Bitelli, 2011), which can be observed using multispectral satellite imagery. Although the benefits of using remote sensing techniques are great (see above), the low reflectance of surface waters means received wavebands are strongly affected by atmospheric noise (Bitelli, 2011). It is therefore important to use atmospherically corrected scenes.
Vegetation indices derived from remotely sensed data are useful for indicating the presence of vegetation and monitoring its dynamics (Munyati and Ratshibvumo, 2011). Healthy vegetation is associated with high reflectance in the near infra–red (NIR) part of the electromagnetic spectrum, with strong absorbance in the red part of the spectrum, and vegetation cover can therefore be deduced from multispectral images (Lillesand et al, 2008).
Examples of vegetation indices include the simple vegetation index (VI), transformed vegetation index (TVI) and the normalised difference vegetation index (NDVI), in addition to a number of related derivative indices (Lillesand et al, 2008). Compared to the VI, the NDVI helps compensate for changing illumination conditions, surface slope, aspect and other factors present in vegetation monitoring (Lillesand et al, 2008). For this reason, it is the most commonly used vegetation index to determine productivity (Kerr and Ostrovsky, 2003; Mehrian et al, 2016).
The NDVI ranges between –1 and +1, with positive values indicative of vegetated areas. Vegetation at peak growth periods generally have high NDVI values, however, the threshold value indicative of such vegetation depends on regional setting and vegetation type (Munyati and Ratshibvumo, 2011). Pax–lenney and colleagues (1996) defined a threshold value of 0.342 when using the NDVI to map agricultural lands: below which regions were classified as uncultivated; above which lands were identified as agriculture. For the present investigation, NDVI values will be used alongside an unsupervised classification approach to quantify and map agricultural land cover.
Image classification is one of the primary steps for information extraction of remote sensing data (Sharma and Gosh, 2013). The techniques are used to assign one of various classes to each pixel based on its spectral properties. Due to the absence of ground truth data available for the present investigation, the Iterative Self Organizing Data Analysis (ISODATA) clustering method and Normalised Difference Vegetation Index (NDVI) empirical approach are preferred.
The scenes were classified using the ISODATA Clustering classification technique. This method produces a land cover classification map with no need for ground truth data (Sharma et al, 2013). The satellite data was clustered into 15 classes, with 10 iterations and a 0.95 convergence threshold. The clusters were then assigned to one of five land use categories identified (Table 2), and merged to produce an unsupervised classification (Figure 7). Although scholars have indicated the ISODATA classification approach is less accurate than the NDVI approach (Sharma et al, 2013), the approach will be combined with the NDVI Empirical Approach to explore the present research objective of considering the extent of changing agricultural land use. The accuracy assessment shows the decision tree approach significantly improves the accuracy of the classification result.
The Normalised Difference Vegetation Index (NDVI) has been shown to be a highly accurate method for mapping parameters associated with plant health and productivity (Pax–Lenney, 1996), particularly for quantifying the extent of agricultural growth in lake catchments. To calculate this index, the mean of each pixel was applied:
Since the images are all taken from the same time of year, phenological differences in vegetation greenness between images is likely to be small.
The threshold derived by Pax–Lenney et al, (1996) was applied to classify areas into cultivated and non–cultivated areas. These areas of high productivity are assumed to have a greater input of resources and therefore result in greater agricultural run–off. Areas below the threshold are classified as 'non–cultivated', which refers to land with reduced or no productivity in each image. When combined with the ISODATA classification, this includes non–agricultural features: urban areas, desert (sand) and the lake water body (Table 2).
Table 2: Class classification scheme.
An NDVI image was created for each year. In a colour composite of 3 images, regions with consistent value across the 3 dates will appear grey. Areas with variable NDVI values between dates will appear in different combination of red, green and blue. In the series of NDVI composites created from the three NDVI images, both the lake body and the sands were consistently grey and the agricultural vegetated fields are multi–coloured.
Although increasing the number of data sets may improve the accuracy for mapping agricultural area, due to the nature of the study, the number of images used in the present investigation is considered suitable to determine the extent of change in agricultural land use. Due to variations in phenology by seasonal variation, it is important that the images are from similar dates. In this study, areas of agriculture are considered as areas classified as cultivated land with a high level of NDVI.
The Normalised Difference Built–Up Index (NDBI) was considered as a method for mapping the urban regions; however this method was rejected as it has been shown ineffective at assimilating between urban areas and barren land which have similar reflectance signatures. (Zha et al, 2003). This method was therefore considered unsuitable for the region of study due to the widespread occurrence of desiccated soil and sand in the region.
This approach is based on the premise that an increase in sodium chloride in solution is correlated with reduced reflectance of bands at the wavelengths 1.5 and 2.1 µm (Lin and Brown, 1992). The following Water Absorption Salinity Index (WASI) is proposed:
where the reflectance values refers to the involved wavebands.
It therefore follows that the Landsat OLI and TM sensors, whose spectral ranges are close to those wavebands (Table 1 and 2), may be sensitive to variations in salinity, whilst far off wavebands are likely to be unaffected. The adapted Water Absorption Salinity Index (WASI) is therefore proposed:
An increase in the numeric value of the index is expected to reflect an increase in water salinity (Bitelli et al, 2011), assuming the reduced absorption results in an increase in reflectance only in bands at the numerator, whilst near infra–red (Band 4), remains unaffected. It is noted that sodium chloride is not the only dissolved salt present in the lake (Abdel Wahed et al, 2015a), but it can be used as a proxy given its abundance and spectral behavior. A calibration will be performed (from Bitelli et al, 2011) to determine a relationship between computed WASI values and salinity. This will allow for comparison to existing literature.
A mask was applied using the thermal wavebands to isolate the water body from the catchment. The WASI images are classified in–line with the classification scheme outlined by Zheng (1999) (Table 4).
Table 3: Classification Water Absorbed Salinity Index (WASI) as converted into Practical Salinity Units (PSU ‰)
The Ocean Chlorophyll 2 and Ocean Chlorophyll 4 algorithms were considered for the present salinity investigation, following the appraisal of work carried out on Lake Qarun by Bitelli and Mandanici (2010). These algorithms rely on the presence of chlorophyll to detect change in water quality parameters. These detectable wavebands are only currently available from sensors on the Advanced Land Imager and hyperspectral sensors such as the Hyperion sensor (Fange et al, 2009). Though their spectral resolution is advantageous, due to the long–term nature of the present investigation, these datasets are deemed inappropriate. This empirical method is therefore unsuitable to meet the present aims.
The suitability of the chosen methodology for the present investigation can be summarised as follows:The NDVI is considered suitable for quantifying agricultural extent. This will be combined with the ISODATA classification using a decision tree approach, which will increase accuracy of classification result and allow for the classification of 5 classes: Water, Natural Vegetation, Agriculture, Barren soil/sand and Urban areas The WASI approach has shown to correlated well with Sodium Chloride concentrations; this will be calibrated with in–situ EC measurement to quantify salinity through the study period. The methodology is appropriate for use with multi–spectral sensors of medium spatial and radiometric resolution. These will be used to measure both catchment characteristics and water quality parameters at the same time, and ascertain a quantitative association to evaluate the extent to which changing agricultural land cover has affected water salinity in Lake Qarun, Egypt.
Imagery from the Landsat Series has been preferred for the proposed application because of its radiometric and spatial resolution, and the free availability of data from the USGS Earth Explorer Website ( http://earthexplorer.usgs.gov). In the range of medium resolution sensors, ASTER imagery would have been suitable but the Short Wave Infra–Red (SWIR) sensor was no longer available for the period of study. The Advanced Land Imager was also considered but, despite higher radiometric resolution, the limited period of activity (since 2000) meant it was unsuited to detecting longer term landscape change dynamics (Bitelli, 2011). Imagery from both the Landsat 4 and 5–TM (Thematic Mapper), and Landsat 8–OLI (Operational Land Imager) is used, allowing observation of the region over a 27 year period.
Both Landsat sensors provide multi spectral wavebands, spanning from the visible blue to short wave infra–red, with a 30 metre spatial resolution and 12 bits radiometric resolution (NASA, 2016). Data of the blue to short wave infra–red wavebands are used in this study. The ultra–blue waveband from the Operational Land Imager (OLI) sensor and the coarse 120m resolution thermal waveband from the Thematic Mapper (TM) were excluded. This permitted direct comparison between images through the 27 year period. All subsequent image analyses were implemented using the standard approaches provided by the software ENVI Classic 5.3 64bit ( Exelis Visual Information Solutions, 2012).
The TM and OLI images are distributed in L–1Gst processing level, meaning that they are already radiometrically calibrated and geometrically corrected. For the present analysis, scenes supplied were of surface reflectance and had already been atmospherically corrected. Three scenes were acquired on demand under clear sky conditions from 16 th August 1989 (TM), 17 th August 1998 (TM) and 23 rd August 2016 (OLI). In addition, one Landsat ETM+ Image was acquired from 12 th January 2011 for calibration purposes (Landsat TM data was not available for the study period).
Reference data from Google Earth was used to implement an accuracy assessment for the classification. This is deemed suitable due to its greater spatial resolution, allowing for accurate determination of land cover type. Additionally, the image was acquired on 18 th August 2016, very close to the acquisition data of the Landsat–OLI scene used for the accuracy assessment and was therefore appropriate.
Field salinity measurements were acquired by means of in situ conductivity measurements for calibration purposes. Nine surface measurements were taken between October 18 th and 1 st November 2010 (Figure 1). The data was originally collected by Dr Ragab Elsheikh (EMISAL Co.) and quantified as a winter seasonal average. The instrument automatically accounts for temperature variations, thus conductivity measurements are quantified at a reference temperature of 20 oC, with a nominal accuracy of 1% (Shadrin et al, 2016). This level of accuracy is far greater than the maximum accuracy achievable by remote sensing techniques (Bitelli et al, 2011) and is therefore highly suited to the appropriate application. As the reference data available is quantified as a seasonal average the ground data acquisitions are not strictly concurrent with images acquired for calibration, but it is noted that salinity levels should not have undergone abrupt variations within the investigated time lapse (Bitelli et al, 2011).
The long–term record and free data availability of Landsat data, in addition to a large spatial coverage and appropriate resolution for the observation of changes to the lake interior changes in the wider catchment makes it suitable for use in the present study. In situ measurements will be used for calibration.
Chapter 5 outline the results from the 27 year study period, where a quantification of salinity and change in urban and agricultural cover is presented. A Spearman's rank analysis is also offered to determine the significance of the relationship between increasing agricultural and urban land cover, and increasing salinity within Lake Qarun.
A calibration was carried out between the values computed through the adapted Water Absorption Salinity Index (Equation 2) and the surface water conductivity measurements published in previous literature (Table 4).
Table 4: Lake Qarun Electrical Conductivity measurement and converted Practical Salinity Unit (PSU) values. All measurements were carried out between 18th October and 1st November 2010 (From Bitelli et al, 2011)
A good correlation (R 2= 0.7740) has been found between the computed WASI values from the 2010 image and the PSU values derived from the conductivity measurements (Figure 7). This is a weaker correlation than originally reported by Bitelli et al (2011), which is likely a result of the use of the Landsat +ETM sensor in this study in preference to the Advance Land Imager.
Figure 9: Correlation between the WASI index values obtained from the Landsat 2010 image, and the electrical conductivity (EC) in–situ measurements. Practical Salinity units (PSU) are also shown on the right vertical axis. (R=0.8798 R2= 0.7740)
WASI values are converted into Practical Salinity Units (PSU) derived from Electrical Conductivity measurements (EC) using an empirical approach outlined by Bitelli et al, 2011. A numerical relationship between computed WASI values(x) and PSU (Y) is therefore formulated:
The conversion into PSU allows for the comparison of the computed WASI values with existing literature.
There has been a gradual, but progressive increase in the mean average salinity over the 27 year study period (Figure 8), although this increase is far reduced to that observed throughout the 20 th century. Average salinity increases from 31.401 ‰ in 1989 to 31.416 ‰in 2016.
This reduced increase in salinity levels is in line with previous scholars (Abdh–Ellah, 2009; El–Shabrawy and Dumont, 2009); though the computed increase is considerably less than the 0.007 ‰ annual increase reported. This underestimation is likely to be as a result of systematic error resulting from atmospheric effects. This could be corrected at the pre–processing stage through the acquisition of integral values of water vapour and ozone content (Bitelli and Mandanici, 2010); however these are neither well defined nor well quantified at present.
Levels of salinity are highly variable across the spatial distribution of the lake (Figure 9). Thresholds were applied using the density gradient application. The present colour scheme is based on the classification approach reported by Bitelli and colleagues (2011), where red areas indicate areas where salinity is>35 PSU and measured as high (above that of seawater), and blue, green and yellow areas indicating lower salinity below 35 PSU.
The lake is predominantly brackish with a few regions of high salinity, and can therefore be classified as 'Sensu Lato' in line with the Zheng (1999) classification. The distribution of the areas of high salinity appears to reflect the distribution of the major inflows of drainage water into the lake, which are located on the east side (El–Batt drain) and the south–eastern bank (El–Wadi drain), and have resulted in a dilution of salt content in these regions. This could also be associated with outflows of salt as a result of salt extraction from EMISAL plant. All images indicate an enhancement of salinity in the west region, indicative of enrichment from evaporation.
The salinity in the lake appears to be becoming increasingly homogenous through the 27 year period (Figure 10), as indicated by the increase in green areas (30–32.5‰). The areas of salinity greater than 35% have also reduced.
A quantitative comparison for the distribution of different land use classes shows that both Urban (68.8%) and Agricultural (22.5%) areas have experienced considerable growth through the study period (Figure 13) which had resulted in the significant reduction in natural vegetation by 62.5% and reclamation of barren land and sand, which has reduced by 22.4%. These represent significant catchment changes which are likely to have had a large effect on the catchment hydrological regime and are corroborative with an increase in the intensification anthropogenic alteration of the catchment over the period.
The rate of urban increase is far greater than increases in agriculture through the period of study, suggesting that catchment alterations are dominated by urbanisation processes through the period.
Understanding the distribution of land cover types is important for the management and determining sources of water pollution (Löw, et al 2013). Figure 14 presents the spatial variability of class types through the study period.
As noted above, the decision tree approach used in this study has resulted in an underestimation of agricultural extent in comparison to existing literature (Keatings et al, 2010) however the method is still considered useful for determining spatial dynamics and indicates a spreading of agriculture into the desert lands of the west. These finding are substantiated by reports by Pax–Lenney et al, (1996) whom indicate the reclamation o f desert lands for agriculture.
Urban expansion is founded around existing conurbations, with considerable growth of areas branching out from Fayoum city, which are located in the central east of the catchment. The most notable increase takes place in area immediate to the southern coast of the lake. Due to its close proximity, this is likely to have a great impact on immediate runoff into the lake. Additionally, these are classified as highly saline, poor productivity soils (Abdel Kawy and Betal, 2013) and thus increasing disturbance as a result of agriculture is likely to liberate salts which will be deposited into the lake, and thus increase salinity.
In order to assess the accuracy of the results, 100 points were randomly selected from the 2016 NDVI image and compared with reference data from Google Earth (Table 5).
Table 5: Accuracy assessment of NDVI Land Type classification
The overall accuracy for the decision tree classification ranged from 78–96%. The error of NDVI classification is observed as resulting from the misclassification of forested areas as agricultural, though it is noted this error is unidirectional and will be equivalent for each data set in the study.
Figure 15 indicates a strong statistical correlation exists between agricultural extent and increasing salinity in Lake Qarun (R 2=0.927). Similarly, a strong relationship has been quantified between the increases in urban land cover, although this appears to have less influence on the level of salinity within the lake with a lower power correlation (R 2=0.812), which may be as a result of a lower spatial coverage of the land cover type.
This suggests that overall, anthropogenic influences have a significant influence on salinity change within Lake Qarun, where agriculture is the most significant mechanism for salinity increase.
The present analysis has been successful in quantifying the relationship between agricultural and urban land cover and water salinity. The findings from the present investigation can be summarised as follows:
A relationship has been identified between the increase in agricultural and urban land cover and the progressive increase in salinity through the period. This relationship has found to be statistically significant. A critical evaluation of how these findings related to existing literature will be reviewed in the following discussion.
The analysis of Lake Qarun demonstrates how changes to the surrounding catchment have a significant effect on lake water salinity. Previous literature indicates the study period is characterized by average salinity levels exceeding levels of global average sea water (Ali, 2002; Abdel Satar et al, 2010; Abdel Wahed et al, 2015b) and becoming 'sensu stricto' (cc Zheng, 1999). The present study has quantified a lower salinity, and indicates that the lake is predominantly brackish with a few areas exceeding the 35‰ salinity threshold.
The trend of salinity increase since 1989 appears to show relative stability in accordance with findings from Abdh–Ellah, (2009); however the quantified increase is considerably less than the 0.07‰ annual increase suggested. These findings are evidential for the effectiveness of commercial salt extraction with respect salt sequestration, and its applicability for the management of salinization in saline lakes globally, in addition to the reported economic benefits (FWMP, 1999). The reduced salinity increase observable from 1998–2016 is likely to be as a result of increasing capacity of the plant through the period (El–Shawbury, 2016). EMISAL salt extraction plant has therefore been successful in the objective of mitigating a rise in salinity, and has resulted in the increasing homogeneity of salinity in the water, however the increase in salinity indicates annual salt accumulation in Qarun is still as much as 70–85 mln.kg/year (Shadrin et al, 2016) and thus the capacity of the salt extraction process must be increased to prevent further salinity increases in the future.
The spatial pattern of salinity change is supported by existing literature (Ali, 2002; Sabae & Ali, 2004), indicating salinity has a strong spatial variability through the lake and is related to the inflow of drainage water from the catchment (Dardir and Wali, 2006), which has resulted in an increase in salinity through the period. The geochemical characteristics of the lake are observed to be significantly affected by the dynamics of catchment activities in accordance with existing literature (Williams, 1999; Abdel Satar et al, 2010).
A significant relationship has been identified between agricultural land cover and lake salinity and indicates that agricultural land cover has increased through the period. These findings are in accordance with national agricultural policies which seek to enhance food production at a national scale (FAO, 2009). Finding by Abdel Satar et al (2010) suggest that agricultural lands in Fayoum contribute to the salt budget of Lake Qarun in two ways; both indirectly through enhancing groundwater seepage as a result of irrigation water and through direct drainage water input. This increase in production is considered to the result of the increase in agricultural land cover. Shadrin and colleagues (2016) indicate the annual volume of drainage water into the lake has increased from 349.2million in 1954 to 400million m 3 in 2000, which is considered to the result of the increase in agricultural land cover.
Successful quantitative analysis indicates land cover change through the study period is dominated by the increase in agricultural land cover, with progradation into the desert land to the East of the catchment. These finding are supported by findings by Pax–lenney et al (1996) which indicate increased agronomy in desert lands to satisfy demands of a growing population, which has since doubled (Census, 2006) and is reported to be increasing further (Abdel Wahed, 2015).
Urban development represents the greatest percentage change in land use. This is consistent with existing literature which documents the rapid increase in population. (Census, 2006; Abdel Wahed, 2015). Though some growth can be observed around existing conurbations, the most significant change has taken place on the south coast of the lake where lacustrine deposits are highly saline with relatively poor productivity (Abdel Wahed, 2015). Soil disturbance in these regions is therefore likely to release significant amount of salt and serves to heighten lake salinity. Though urban flux is greatest through the study period, statistical analysis indicates that agriculture as the leading mechanism for the observed salinity change in Lake Qarun through the study period.
The remote sensing techniques and the subsequent analysis in this study have been successful in quantifying a statistically significant relationship between salinity and anthropogenic alteration of the catchment, using agricultural and urban land cover change. The decision tree classification approach has been determined to be more accurate than using vegetation indices alone and a powerful correlation has been quantified between computed Water Absorption Salinity Index values and in–situ Electrical Conductivity measurements; however there are several uncertainties associated with the land class quantification and salinity mapping.
The images taken from the Landsat TM (1989 and 1998), appear to be strongly affected by streaking, which is typical of push broom sensors (Yan and Zheng, 2015). This results in vertical line artefacts in both images which create false leaps in salinity values and has resulted in an overestimation of salinity observable in the central–west in 1989 and underestimation in the same region in the 1998 image (Figure 9). This issue may be overcome in the pre–processing stage; however, atmospheric parameters for the area have yet to be properly defined (Bitelli and Mandanici, 2010). The origin for the anomalous result in the Eastern region in 1998 is unclear: it could be due to overestimation as a result of differential atmospheric 'hue' over the image (Bitelli and Mandanici, 2010), but since all images were cloud free, this conclusion is unresolved.
The quantification of agricultural land cover is considerably less than reported by Keatings et al, (2010); whom report an agricultural area of 1500km 2 1989), which suggests that the classification approach has been inaccurate at quantifying agricultural extent. Direct comparison between existing literature is therefore not viable, however it is noted that this error is likely to be present for all the images and therefore the relationship of change between the images is likely to be relevant for considering the extent to which changing agricultural land cover has affected water salinity in Lake Qarun.
The present study in Lake Qarun has shown the value of remote sensing techniques for quantifying and observing catchment wide dynamics of larger lakes. The techniques can provide useful information which is critical for inland water management (Chawira, 2013; Kibena, 2014) and may provide a cost effective means for the routine monitoring of the state of lakes (Löw et al, 2013).
Information obtained by remote sensing at different time intervals has been presented to be effective when integrated with fixed observations to yield 'twice the results with half the effort' (Zheng et al, 2004), and at a fraction of the cost associated with large scale field campaigns (Dube et al, 2015). The present methodology may be useful to baseline information to facilitate decision making and hazard mitigation in other regions (Zheng et al, 2004); with a focus on contributing to missing long term monitoring of land cover dynamics as a result of the extended Landsat Archive.
The increasing availability of remote sensing platforms with adequate spatial and temporal resolutions allows for a near global coverage (Yan and Zheng, 2015) and the increasing availability of existing algorithms for data pre–processing and image classification permits to a high degree of automation and accuracy (Löw et al, 2013), which is suited for monitoring. The absence of push broom artefacts in the OLI is indicative of how improving technology is likely to increase the accuracy of salinity determination in future studies using the present methodology.
Water allocation to the Fayoum Agricultural province is to be considerably impacted by the construction of the Ethiopian Dam upstream, which is set to open in early 2017 (Digna et al, 2016). Remote sensing techniques will be useful in providing a multidisciplinary, interdepartmental approach for monitoring environmental impact of this development. As observed in the present study, the dynamic change of lakes is closely related to sub–environment changes. Multidisciplinary research can form the basis for establishing integrated scientific observations of catchment wide processes (Zheng et al, 2004) and can useful to guide decision making for conservation in this lacustrine environment (Löw et al, 2013), and of the wider Nile basin.
The present investigation has been successful in quantifying a progressive increase in salinity through the period of study, observed to be significantly related to the increase in agricultural land cover as a result of progression into desert regions. A strong relationship has also been quantified between the mapped urban land cover and salinity, which serves to highlight the critical impact of human catchment activity on lacustrine environments.
The sensitivity of the methods are likely to have been affected by streaking effects in the Thematic Mapper images from 1989 and 1998, however salinity values are largely in agreement with existing literature.
The present study indicates that the EMISAL salt extraction plant had been successful in bringing relative stability to the salinity budget, despite increases in both agricultural and urban land cover. The study has indicated how these methods may be used to help protect saline lakes around the world and to analyse the issue of secondary salinization. Remote sensing methods have an increasing appeal as a result of improvements in techniques and technology.
As higher spatial resolution topography data becomes available, a Digital Elevation Model may be useful to specifically determine sources of sediment and derived salts. GIS software can then be used to determine source sink relationships. Additional images may be used to determine the variation of salinity on an inter–annual scale. As this analysis was intended to quantify longer term trends over an extended period of 27years, this was not appropriate. However this may be useful to determine how agricultural phenology and crop cycles affects salinity, as it is known that water is in more in demand in the summer months than in the winter.
It is hoped the accuracy of methods used in this study may increase with improved technology. As Zheng and colleagues (2004) note, one can only effectively manage saline environments on the premise that one has 'monitored and mastered the trend of salt lake trend'. The approach used in this study can be useful to monitor and assess the causality of environmental crisis.