Assessment of a Non-Optical Water Quality Property Using Space-based Imagery in Egyptian Coastal Lake

Several studies in literature have addressed retrieval of water quality parameters using remote sensing techniques. Significant correlations have been found between specific water quality parameters and reflectance measured with satellite sensors. These parameters cause change to the spectral properties of reflected light and, hence, are remotely detectable (Gholizadeh et al., 2016). Recent research by Swain and Sahoo (2017) argued that certain conservative pollutants can be distinctively detected with different reflectance received in the electromagnetic spectrum because no biochemical reactions or ionic exchange are experienced.

Retrieving properties such as water clarity; turbidity, and Total Suspended Solids (TSS) concentrations using earth observation imageries have been tackled in applied research studies worldwide (Kloiber et al. 2002; Zhang, 2002; Bilge et al., 2003; He et al., 2008; Sravanthi et al., 2013; Dona et al., 2014; Dorji and Fearns 2016; Abayazid and El-Gamal 2017). In 2008, He et al. presented water quality retrieval models with proven successful results for optical nitrogenous and phosphorous components. Other parameters such as chlorophyll-a (chl-a) and Colored Dissolved Organic Matter (CDOM)) have also been covered in various studies (i.e. Brezonik et al., 2005; Thiemann and Kaufmann, 2000; Li et al., 2002; Dona et al., 2014).

Remotely deriving weak and/or non- optical water quality characteristics, that have no directly-detectable reflection, is challenging. Consequently, early studies are mostly focused on water physical and biogeochemical components that are considered optically active (Giardino et al., 2014). However, limitation to water quality characteristics that are related to Inherent Optical Property (IOP) narrows down the parameters that can be assessed by remote sensing techniques.

The Dissolved Oxygen (DO) concentration is considered a crucial indicator of water system healthiness, and governs recovery capability (UNESCO, 2005). Yet, being a non-optically active parameter, DO levels cannot be directly retrieved using remote sensing technique. This research study aims to present an approach to detect Dissolved Oxygen concentrations in an inland shallow coastal lake, using space-based imageries.

Based on grounds of early DO modeling theories, as well as regional conditions, the study investigates the potentiality of deducing DO levels from optically detectable water quality parameters that affect, and be affected by, Oxygen presences in water.

With growing population and development activities, the Nile Delta of Egypt experience challenging conditions. Lake Edku is located within the active Northwestern coastal zone of the Delta, between longitudes 30°8’ & 30°23’E and latitudes 31°10’ & 31°18’N (Fig. 1). The lake is characterized of having systematically shrinking free open water, altered ecosystem and deteriorating water quality state (Abayazid, 2015). Edku lake serves an active agriurban basin, and bordered by dense aquaculture practices. Accordingly, the lake receives wastewaters with different pollution degree from fish farming therapeutic drugs, nutrient flux from agricultural drainage network (e.g. Edku, El-Boussili, Khairy and Bearsik drains), in addition to effluents from municipal WasteWater Treatment Plants (WWTPs) and industrial facilities (Siam and Ghobrial, 2000). The lake is connected to the Mediterranean Sea with single opening “Boghaz Al-Maadia”, which allows temporal tidal inflows and localized saline water interaction. Discharges with heavy nutrient levels, as well as the deceased salinity inputs, have encouraged excessive unwanted aquatic vegetation. That, in turn, disturbed natural circulation; flow dynamic and sediment transport, and hence self –purification within the lake (Hossen and Negm, 2017).

Modeling of Dissolved Oxygen in water bodies has been initiated in 1925 by Streeter and Phelps through an application in the Ohio River of the United States of America (Chapra, 1997). Simulation studies were based on the fact that the rate at which DO fluctuates in waters reflect the rate of Oxygen demand and release. Their modified model set foundation of DO sinks and sources through inclusion of factors proved affecting the Dissolved Oxygen depletion and recovery in a water body. Beside the initially considered coefficients that represent reaeration as well as settling/ decay processes, the model extension added representative components of aquatic flora role in the Oxygen production and exhaustion with photosynthetic activity. Furthermore, sediment consumption of DO has been added as an effective factor to be employed in the modified model for DO prediction. Equations 1 and 2 state the early model and modefied version; respectively. More details can be found in the text book of Chapra (1997)(1)Dt=Do EXP(−Kat)+[KdLoKa−Kc×{EXP(−Kct)−EXP(−Kat)}]

Figure 1.

Where; Dt is the predicted dissolved oxygen deficit concentration, t is the travel time, L is the BOD level at point of interest, Lo is the ultimate BOD level, Ka is the reaeration rate, Kd is the decomposition rate, Ks is the settling removal rate, Kc is the CBOD decay coefficient, and Do is the initial value of the oxygen deficit.(2)Dt=DoEXP(−Kat)+[KdLoKa−Kc×{EXP(−Kct)−EXP(−Kat)}]+Sb(1−e−Kat)Ka+Doe−Kat

The modified model has added factors as; P the photosynthetic oxygen production rate, R the algal respiration rate, Sb the sediment oxygen demand rate, No the initial Nitrogenous BOD (NBOD), and Kn the NBOD decay coefficient.

Ground truth data used were obtained from published research study by Okbah et al. (2017). Authors presented data collected in ten sampling locations distributed throughout the Edku Lake. Spatial distribution of field measurement locations reflects variability in the lake water quality, with regard to boundary interaction as well as flow movements within the lake (Fig. 2). Further, sampling campaigns have been carried out during four seasons; spring, summer, fall and winter of year 2016, which reflected the variable conditions that the coastal lake experince. Statistics of the field measurements show that in summer time DO levels reach the lowest concentrations, ranging from 1.6 to 9.4 mg/L, and experience wide variability within the lake with standard deviation of 3 among the ten investigated locations. Meanwhile, the highest DO levels occur in winter, ranging from11.3 to 18.1 mg/L, with standard deviation of 2.3. The lake water DO range from 7.5 to 14.0 mg/L in spring, whereas the fall season has slightly less concentrations, ranging from 5.0 to 13.1 mg/L. Maximum measured DO concentrations were mostly found in zones “C” and “D”, as illustrated in figure (3). On the other hand, minimum levels occur in locations within the eastern zone “A”, where most of direct wastewater discharges reach the lake water, especially in summer season.

Figure 2.

Figure 3.

In 2013, Ganoe and DeYoung presented theoretical basis for DO retrieval with the use of air-borne Raman spectroscopy instead of ship-based technology that customary required direct contact with the waterbody. Authors argued the advantages of air-based technique in measuring DO when compared to time consuming as well as limitation in detecting variability in changing water conditions during field trips. The research concluded promising success of remote sensing retrieval of the temporal and spatial dynamics of dissolved gas distributions in coastal ecosystems. Yet aerial arrangements are costly and not always readily available, while space-borne sensors can have more frequent revisits and reasonably spatial coverage with advancing spectral resolution.

The imageries used in this study are the freely available Landsat 8 Operational Land Imager (OLI) from the Unitd States Geological Survey (USGS) Earth- Explorer website. The Landsat 8 (OLI/TIRS) is the most recent satellite that was launched in 2013 under the Landsat program, with swath width of 170 km and 16 days’ revisit interval. Since the in situ DO data have been collected during spring, summer, fall and winter of year 2016, images used in this study were acquired on nearest corresponding overpass dates to match the sampling data timing Table (1). Table (2) states the spectral range considered in this study, covering visible and Near-Infrared as well as Thermal Infrared bands. The necessary image processing and result analysis were carries out in Geographic Information System (GIS) environment.

Satellite imageries were processed for carrying out multiple regression technique and reaching best algoritm model for DO retrival in Lake Edku. Analysis have been performed with the spatially distributed field measurements of Dissolved Oxygen as related to the spectral reflectance values derived from Landsat images at corresponding dates in year 2016. The available field data were divided into two groups for building algorithm models. Then performance has been tested with the reserved second group of data for validation process.

Based on the previously mentioned grounds of Streeter and Phelps DO modeling, as well as regional conditions defining water quality in coastal lakes of Egypt, the trophic and sediment-related properties were considered key factors in selection. Primary, the parameters included for developing DO derivative algorithms were Turbidity, Total Suspended Sediments (TSS), and Chlorophyll-a. Also, Temperature was added in the DO retrieval process as an important driver affecting Oxygen level in water, especially with the thermal anthropogenic releases and flow dynamic irregularity within Lake Edku. Cutomarily, DO concentration in water is inversely related to the temperature. Therefore, zonation of thermal property is expceted to have direct reflection on DO retrieved distribution. In process, alternate combinations of the considered parameters have been investigated for optimal model results.

Turbidity and TSS levels were deduced using findings of the recent research study of Abayazid and El-Gamal (2017). Authors concluded regional algorithm models for remotely sensed Turbidity and TSS in the Nile delta coastal zone, in terms of Landsat 8’s reflectance from spectral bands; Band2ref, Band4ref and Band5ref, as presented in Equations 3 and 4, respectively.(3)LN TURBIDITY =−1.2247+[0.08112 Band4ref]+[2.944⁢Ln (Band2ref/Band4ref)](4)LNTSS=0.0496[3.325+13.222 Band5ref]3.2214

While literature applications showed various retrieval algorithms for Chlorophyll-a (e.g. Dona et al., 2014; Akbar et al., 2010; Brivio et al., 2001), best agreeable results for quantifying Chlorophyll-a in Lake Edku were found with the ratio between reflectance from spectral bands 2 and 4 of landsat 8 (Eq. 5).(5)Chlorophyll-a = Band2ref/Band4ref

Thermal spectral data have been converted to Temperature “T”, using the conversion formula presented in Equation 6 (USGS, 2015)(6)T=K2Ln⁡(K1Lλ+1)

Where Lλ is the spectral radiance, and K1 and K2 are the thermal conversion constants found in Landsat imagery metadata files. Surface water temperature levels are calculated using averaged values of Thermal Infrared Bands (10) and (11).

Sensitivity analysis was performed to select best combination of the considered parameters, with different seasonal conditions experinced in the lake. Accordingly, the optimal DO retrieval algorithm model with best fitting predictions, as well as least data requirement and calculation efforts, has been selected for result demonstration.

The four parameters initially considered as inputs for DO modeling have been spatially derived for spring, summer, fall and winter seasons of year 2016, with corresponding Landsat imageries. Example illustrations are found in figures 4, 5, 6 and 7 for Temperature, Turbidity, TSS and Chlorophyll-a levels within Lake Edku, respectively.

Figure 4.

Figure 5.

Figure 6.

Figure 7.

For developing satellite-based Dissolved Oxygen retrieval model, analysis has been carried out with various combinations of input parameters versus DO field measurements. The selected ground truth data is distributed throughout the lake zones. Furthermore, the data covers the four seasons so that a wide range of water quality levels experienced in the lake is considered in the developed model. Table (3) presents model predictive capacity and goodness of fitting while considering selective inputs as well as seasonal variations of DO presence in the lake waters.

Sensitivity analysis showed that TSS and Turbidity have similar effect in detecting Oxygen consumption in the waterbody under consideration. It was also found that the least influential factor is the Chlorophyll-a. Minor change with least effect in predictive capacity of the developed algorithm occurs when adding Chlorophyll-a to the input parameters.

Optimal derivative capacity for DO levels in Lake Edku, with least input parameter requirements, hence less processing works and costs, was found by using only Turbidity and natural logarithm of Temperature. The developed algorithm model, stated in Equation 7, proved reasonable fitness with regression coefficient of 0.79 (Fig. 8).(7)DO=36.27+0.19 Turbidity−10.36Ln (Temperature)

The proposed algorithm was then applied to the second group of data reserved for testing, and satellite-based derived DO concentrations were compared with the corresponding field measurements. Validation results proved acceptable predictive capacity of the developed algorithm model, with R2 of 0.66 (Fig. 9). Descriptive statistics for observed versus modeled yearly average DO levels within Lake Edku are presented in table (4). The developed model shows highly agreeable predictions with field measurements. However, the model failed to represent the very low DO concentration occurrence in the lake during summer season.

Figure 8.

Figure 9.

Once validated, the developed model has been applied to a set of landsat imageris to obtain retrieved DO spatial distribution in Lake Edku in time steps. Figure 10 illustrates comparison for observed versus derived DO concentrations during four seasons in Lake Edku. For the purpose of demonstration, figure (11) shows an example mapping of DO concentrations throughout the lake zones in winter time.

Figure 10.

Figure 11.