Mapping and estimating water quality parameters in the Volta Lake’s Kpong Headpond of Ghana using regression model and Landsat 8

Abstract

Sub-Saharan Africa faces a number of essential issues, including water quality. As such, evaluating the surface water quality of lakes and reservoirs is a crucial part of environmental monitoring and management. Especially in a region where these water bodies serve as a source of livelihood for communities living around them. Water quality parameters (WQPs) are usually taken from the site and sent to the laboratory for measurement and analysis. However, this traditional method is time-consuming, costly, and labor-intensive. Combining geographic information system and remote sensing (RS) allows researchers to analyze WQPs more conveniently. This study, therefore, used RS technology to map and estimate WQPs and correlated it with in-situ measurement. Using the empirical regression model and Landsat 8, WQPs such as chlorophyll-a (Chl-a), total suspended solids (TSS) and turbidity were estimated. The results from RS were correlated with the in-situ measurements of water quality. The results showed that the in-situ Chl-a levels varied from 0.206 to 13.5 mg/L, averaging 5.1 mg/L. The Chl-a values estimated from Landsat 8 had R² of 0.883 and 0.853, respectively, for both periods (17 December 2022 and 16 March 2023). The green band (B3) was more instrumental in detecting Chl-a. The in-situ measurement for TSS ranged between 18 and 48 mg/L, with a mean value of 28.7 mg/L. These readings were low and within tolerable bounds of 50 mg/L. High TSS concentrations were found near farms and communities with a significant influx of silt into the surrounding lake. The comparison of in-situ water quality and the reflectance from satellite data showed that the turbidity estimated from the sensor from the two periods has R² > 0.65. The study showed that the combination of the Landsat image and in-situ measurement offers great ways to provide timely and affordable estimation from WQPs.

Keywords:

• Chlorophyll-a
• total suspended solids
• Landsat 8
• water quality parameters
• empirical regression model

REVIEWING EDITOR:

• Sanjay Kumar Shukla, Edith Cowan University, Australia

SUBJECTS:

• Environmental Studies
• Climate Change
• Ecology - Environment Studies

1. Introduction

Freshwater is a precious resource, whose value has increased over the years due to expansion in the world’s population and change in climatic conditions (Vörösmarty et al., 2000). These resources, including lakes and reservoirs provide essential social and ecological services for the benefit of humans and the environment. They are used for domestic, industrial and agricultural activities (Khan & Ansari, 2005). However, in sub-Saharan Africa, freshwater is both scarce and of exceptionally poor quality; thus, only a limited fraction of water is available for human consumption due to pollution (Freitas, 2013). According to Carpenter et al. (1998), man-made and natural factors contribute to a decline in water quality. Water quality can be degraded by substances introduced into a water body from either point or nonpoint sources (Ritchie et al., 2003). Anthropogenic activities, however, have, in varied degrees from place to place and through time, changed the quality of most fresh waterways and reduced their value (Boyd, 2020). Sewage from households, urban runoff, effluents from industries, and waste from farms are the primary sources of excessive amounts of organic compounds and metal ions in freshwater (Li et al., 2017). There are several water quality factors, but only a select handful are often relevant for a given water usage. For instance, the water used for animal production does not necessarily have to be of the same level for human consumption, but it must also not harm or kill animals. Freshwater resource faces the challenge of pollution and subsequent degradation, and decision-makers are asking for newer methods to monitor water quality (Bonansea et al., 2015). This situation has necessitated the monitoring of water resources at the Kpong Headpond.

Water quality indicators are normally determined by taking samples from sites and measuring them in the laboratory. However, according to Duan et al. (2013), this traditional method is costly, labor-intensive, and time-consuming. Furthermore, traditional point sample methods fail to detect both geographical and temporal fluctuations in water quality, critical for a thorough evaluation of water bodies. This has created a barrier to proper water quality monitoring and management (Gholizadeh et al., 2016). Without precise, focused, and extensive data collection, it is impossible to assess the condition of any water resource with any degree of accuracy. Dube et al. (2015) further stressed that although traditional methods of monitoring water quality have been used extensively because of their accuracy, it is almost impossible to use these methods at larger scales. Again, these in-situ measurements offer precision for specific points and locations in time but do not offer a spatial or temporal perspective of water quality across a huge area (Usali & Ismail, 2010). These restrictions mean that sampling efforts usually misrepresent the genuine state of an entire water body. By using remote sensing (RS) and geographic information system technology to collect data that offer a synoptic picture that would otherwise be hard to obtain, these difficulties can be overcome, according to Liu et al. (2003). Improving the assessment and monitoring of water resources requires using satellite images to map water quality characteristics (Chu et al., 2021).

The relationship between the in-situ water quality and the data from RS has been established through both empirical evaluations and statistical regression models (examples include Bonansea et al., 2015; Chu et al., 2021; Ouma et al., 2020; Wang et al., 2018). Alparslan et al. (2007) utilized Landsat images and in-situ water quality measurements to thoroughly examine the water quality in Istanbul’s Omerli Dam Lake. This approach enabled the creation of maps within the Omerli Dam, which provided comprehensive information on critical parameters, including chlorophyll-a (Chl-a) concentrations, suspended solid matter, Secchi disc transparency, and total phosphate content. Hua (2017) contributed significantly to the field by demonstrating the efficacy of RS in tracking land use land cover changes through time. This action was undertaken with a specific focus on the deterioration of water quality in Malacca, Malaysia. This enabled the researchers to gain insights into the complex dynamics influencing water quality. Although RS has been effective at monitoring resources related to water, there are some drawbacks to doing so. For instance, due to its low flexibility, using aerial RS for the determination of water quality parameters (WQPs) has been limited. Another disadvantage is the requirement to measure ground truth measurements concurrently with aircraft measurements (Hakvoort et al., 2002). However, with even these limitations, RS has been very useful in managing water resources globally. The focus of this research is, therefore, to use techniques from RS to evaluate water quality on the Kpong Headpond of the Lower Volta Basin and to compare how RS and on-site measurements of WQPs relate to one another.

2. Materials and methods

2.1. Study area

The study area includes a 12 km section of the Kpong Headpond, located around 25 km downstream from the Akosombo dam (Figure 1). The dam provides mobility, serves as an important base for aquaculture, and is also used for rice irrigation (Ntiamoa-Baidu et al., 2017). The area experiences bi-modal rainy seasons with a mean annual rainfall of 870.4 mm and a mean annual potential evapotranspiration of 1600 mm (Nyamekye et al., 2021). The average relative humidity is between 70% and 80%, during the rainy season and about 55%–60% in the dry season (Gampson et al., 2014). The average annual temperature is around 27.9 °C with stream flows and groundwater being the main sources of water in the basin (Logah et al., 2017). The creation of both the Akosombo and the Kpong dams has influenced the flow regime of the Lower Volta River. The highest flows have drastically decreased, whereas downstream low levels of the Kpong Dam have increased. This has resulted in a continuous flow for the entire year instead of the natural high and low seasonal flows of the river. Additionally, there has been little sediment load in the Lower Volta River and outflow to the sea due to the sediment trapping in the lake behind the dam. The main rock type that can be found in this location is gneiss rock (Logah et al., 2017).

Figure 1. Study area showing the portion of the lake.

2.2. Water quality parameter and in-situ measurement

The data for the water quality were collected from the Lower Volta Basin as shown in Figure 1. The measured physicochemical parameters for this study are listed in Table 1. Twenty water samples were collected using a local canoe together with Garmin handheld Global Positioning System (GPS) to determine the coordinates of each sample taken. The Horiba U51 Water Quality Probe was used to measure WQPs such as temperature, pH, electrical conductivity (EC), turbidity, dissolved oxygen (DO) and total dissolved solids. At each sampling station, water samples for Chl-a were collected into 1 L clean plastic containers and kept in the dark (ice boxes to prevent the green chlorophyll pigment from quickly degrading in sunlight). A volume of 1 L of water was filtered through Whatman GF/C filter paper. Ninety percent acetone was used to extract Chl-a into solution and centrifuged at 3200 rpm for 10 min and the supernatant poured off and measured at 663, 645 and 630 nm, respectively, using the spectrophotometer. In the laboratory, the samples were kept in a refrigerator at 4 °C until the analyses were completed. The method of analysis was done based on those described in the ‘Standard Methods for the Examination of Water and Wastewater’ (APHA-AWWA-WEF, 1998). Hydrazine reduction for NO₃-N; direct nesslerization for ammonia nitrogen; stannous chloride for phosphate; and the Azide modification of Winkler for DO.

Table 1. Summary statistics of the physical and chemical parameters of the study area.

Parameter	Min	Max	Average	SD	WHO
pH	5.90	7.80	6.90	0.38	6.5–8.5
Alkalinity	24.40	34.70	27.90	8.60	500.00
Temperature	26.80	28.00	27.50	0.30	–
TSS	18.00	48.00	28.70	5.08	05.00
EC	0.026	0.17	0.06	0.02	1200.00
Turbidity	5.56	24.90	7.90	3.00	5
PO4-P	0.03	0.07	0.04	0.01	–
NO3-N	0.40	1.00	0.60	0.16	50
NH4-N	0.010	0.08	0.05	0.02	35
Chl-a	0.21	13.50	5.10	4.30	–
DO	2.02	6.60	5.60	0.67	8–10
Cl^−	0.00	12.00	2.18	1.21	200–300

All the measurements are in mg/L with the exception of pH (pH units); EC (µS/cm); turbidity (NTU).

The water quality data and satellite images used in this study were mostly collected when cloud cover was less. The sampling stations selected for the collection of water samples are shown in Figure 1. Following the standard protocols, a total of 20 samples were collected within the duration and the concentration of turbidity, total suspended solids (TSS) and Chl-a were determined. Two water samples were collected at each of the 20 points for each of the two sampling dates (17 December 2022 and 15 March 2023). Out of the 20 points, 75% was used for predicting and 15% for validation of the model. The average of the two samples for turbidity, TSS and Chl-a was recorded after it was tested.

2.3. Satellite data

The study acquired Landsat 8 operational land imager (OLI) images between 17 December 2022 and 16 March 2023. In-situ sampling dates were selected to coincide with the overpass time of +1 day. The Landsat 8 image has onboard the OLI and the thermal infrared sensor (TIRS). The OLI image data have a spatial resolution of 30 meters for bands 1–7, while the panchromatic band (band 8) has a spatial resolution of 15 meters. However, TIRS collects data for two longwave thermal bands (band 9) at 100 meters at a revisit time of 16 days, which are later resampled to 30 meters to be equal to OLI multispectral bands (Ouma et al., 2020). The radiometric resolution of Landsat 8/OLI is well improved with little noise in the image and difference in spectral bands as compared to Landsat 7 ETM+. This is essential for retrieving correct surface water and water quality information (Nguyen et al., 2019).

2.4. Radiance and reflectance estimation

The determination of the radiance and reflectance parameters from the satellite sensors was performed using these three steps: (i) calculating the total top of atmosphere (TOA) reflectance from scaled digital number values for Landsat OLI; (ii) conversion of TOA reflectance to surface reflectance; (iii) conversion of the surface reflectance to the equivalent RS reflectance (R) at these bands. The details of these processes can be found in Ouma et al. (2020).

2.5. Regression model for water quality parameters

The empirical regression model (EMRM) was adopted in this research to draw the correlation between the reflectance of the image from RS and the measured in-situ WQPs as explained in Ouma et al. (2020). It is a multivariate regression that determines the measurable relationships between the estimated in-situ WQP and the reflectance from the satellite spectral data. Out of the 20 sampling points, 15 of the sampling point data were used for calibrating the regression model in the development of the EMRM, and five data points were used in validating the model. The highest R² values between the estimated model and in-situ measurements of the WQPs were used to select the best-fit model. In order to choose the optimal equations, the R² was employed as a means to obtain the equations’ accuracy.

2.6. Performance analysis

The empirical models from the sensor used in the retrieval of WQPs were tested for their performance. This was done by comparing the results of the regression models with in-situ laboratory measurements using these error matrices: Pearson correlation coefficient R, coefficient of determination R², mean absolute error or bias, root mean square error (RMSE) as described in Omondi et al. (2023) and Ouma et al. (2020).

3. Results

3.1. Physical and chemical parameters

The results of the physical and chemical parameters are described in Table 1. The pH of the study area has values between 5.9 and 7.8 pH units, with values of 6.9 ± 0. 38 as the average and standard deviation (SD), respectively. The Alkalinity of the lake was between 24.4 and 34.7 mg/L, with 27.9 ± 8.6 as the average and SDs, respectively. The EC ranged between 0.026 and 0.174 µS/cm, with an average and SD of 0.06 ± 0.018. The TSS, which shows the number of suspended solids in the water organic or mineral had values between 18 and 48 mg/L, with average and SD of 32.7 ± 5.08, respectively. Additionally, the DO which determines whether the water is polluted or not had values ranging from 2.02 to 9.6 mg/L. These values fall outside the accepted range of 8–10 mg/L, which is an indication that the water is slightly polluted. The turbidity which defines the clear nature of the water ranged between 5.57 and 24.9, with an average and SD of 7.9 ± 3.0, respectively. These values are generally >5 NTU as indicated by the World Health Organization (2011). Also, the Chl-a levels ranged between 0.21 and 13.5 µg/L, with an average and SD of 2.18 ± 1.21, respectively.

3.2. Estimation of water quality from in-situ and Landsat 8

3.2.1. Retrieval of Chl-a parameters

The in-situ Chl-a values for the two sampling periods ranged between 0.21 and 13.5 mg/L with average of 5.1 mg/L. The observed concentration of Chl-a from the in-situ measurement may be attributed to the effluence from settlements near the lake and the inflow of fertilizer from nearby farms. The regression model for the two sampling seasons is presented in Figure 2(a,b). A linear model was found to be appropriate for both periods in the retrieval of Chl-a from the Landsat 8 with R² of 0.883 and 0.853, respectively, from both the blue (B2) and green(B3) bands (Figure 2). However, the green band (B3) is more instrumental in the detection of Chl-a from both sampling periods. The Chl-a values estimated from Landsat 8 for both periods have RMSE which are lower than the observed average in-situ value of 5.1 mg/L. The performance of the regression models for the two different periods is shown in Table 2.

Figure 2. (a) Remote sensing reflectance of Landsat band and in-situ measured Chl-a concentration for 17 December 2022. (b) Remote sensing reflectance of Landsat band and in-situ measured Chl-a concentration for 15 March 2023.

Table 2. Equations for regression models for Chl-a, turbidity, and TSS (using the blue, green and red bands, i.e. B2, B3 and B4, respectively).

Date	Parameters	Regression equation	Band combination	R	R^2	Bias	RMSE
17 December 2022	Chl-a	8176.2x + 1255.9	B3/B2	0.701	0.883	0.540	2.340
	Turbidity	62.98x + 4.67	B3	0.601	0.668	2.341	3.142
	TSS	8434.3x + 9379.2	B4/B1	0.683	0.753	3.213	3.542
15 March 2023	Chl-a	774.3In(x) + 1886.8	B2 + B3	0.772	0.853	0.256	1.573
	Turbidity	105.21x − 7.230	B3 + B4	0.598	0.701	2.975	3.682
	TSS	5324In(x) + 1433.2	B3/B4	0.618	0.712	1.971	3.276

3.2.2. Retrieval of TSS parameters

The in-situ measurement for TSS for the two sampling periods ranged between 18 and 48 mg/L. These values were low with a mean value of 0.0327 mg/L, which is within acceptable range. The highest concentrated area is close to settlements and farmlands where inflow of sediments to the lake is very high. In using the EMRM algorithm, a linear regression model of the first period (17 December 2022) gave the best results for the TSS estimation using band ratio between the red band (B4) and the blue band (B2) with R² = 0.753 (Figure 3(a)), compared to R² = 0.712 (Figure 3(b)) of the second period (15 March 2023) with band ratio between the green band (B3) and the red band (B4) for the Landsat 8. The RMSE and bias of the TSS concentration are <28.7 mg/L, which is the average for the in-situ TSS measurement. The performance of the Landsat 8 sensor for the two periods in the retrieval of the TSS for the lake is found in Table 2. The optimal bands in the retrieval of TSS concentration are bands (B3) and (B4), however, the accuracy of estimation of TSS for the first period is slightly higher than that of the second period in terms of R². The remaining parameters R, RMSE and the bias also follow the trend of higher values for the first period when compared to the second period.

Figure 3. (a) Remote sensing reflectance of Landsat band and in-situ measured TSS concentration for 17 December 2022. (b) Remote sensing reflectance of Landsat band and in-situ measured TSS concentration for 15 March 2023.

3.2.3. Retrieval of turbidity parameters

The values of the in-situ turbidity for the two sampling periods are between 5.56 and 24.9 NTU, with an average of 7.9 NTU. From the values, the turbidity of the lake was low. This can be attributed to the periods when the sampling was collected, i.e. December and March which are dry seasons. Thus, the sediment loads from surrounding areas were reduced since there were no rains to discharge sediments into the lake. From the EMRM, a comparison of the in-situ water quality and the satellite reflectance data shows that the turbidity estimated from the sensor from the two periods has R² > 0.65 (Figure 4). The green (B3) and red (B4) bands were stronger in the retrieval of turbidity from the Landsat 8 for the two periods. The RMSE and the bias error for both periods were less than the average and minimum of the in-situ turbidity value, which is an indication the linear models perfectly estimated the concentration of turbidity from the Landsat 8. The summary of the regression models used in estimating the turbidity is shown in Table 2.

Figure 4. (a) Remote sensing reflectance of Landsat band and in-situ measured turbidity concentration for 17 December 2022. (b) Remote sensing reflectance of Landsat band and in-situ measured turbidity concentration for 15 March 2023.

3.3. Validation of model with in-situ and predicted water quality measurement

The EMRM used for predicting the WQPs was adopted from (Najafzadeh et al., 2021), who used the model to evaluate the water quality index of streams. The regression model was validated using five sampling points out of the 20 collected samples. Table 3 shows the results of the validation, and the statistics from stations that were used in the model calibration is also shown. From the results, turbidity had the lowest variation in concentration followed by Chl-a and TSS (highest), respectively.

Table 3. Statistics of validation results for in-situ and estimated water quality parameters.

Date	Parameters	Data source	Number of samples (n)	Min	Max	Mean	SD	CV (%)	SE
17 December 2022	Chl-a	In-situ measurement	20	0.226	12.365	6.252	3.826	40.23	1.73
		Landsat8 OLI	20	1.40	12.833	6.721	3.375	37.23	1.65
	Turbidity	In-situ measurement	20	6.800	24.4	17.233	1.767	36.34	0.59
		Landsat8 OLI	20	5.71	23.623	19.386	1.865	35.89	0.65
	TSS	In-situ measurement	20	20.01	48.00	42.565	8.092	11.35	7.45
		Landsat8 OLI	20	23.0	52.912	41.218	9.623	13.45	6.38
15 March 2023	Chl-a	In-situ measurement	20	0.206	13.235	5.987	2.626	36.63	0.54
		Landsat8 OLI	20	0.854	12.785	6.045	3.871	39.56	1.32
	Turbidity	In-situ measurement	20	5.567	22.786	18.123	1.986	36.43	0.67
		Landsat8 OLI	20	4.71	24.371	20.588	2.017	34.64	0.725
	TSS	In-situ measurement	20	20.876	48.651	41.865	10.752	13.542	8.653
		Landsat8 OLI	20	18.072	50.652	43.199	9.326	11.965	7.23

Additionally, the graphical representation of in-situ and Landsat 8 sensors is the validation results for Chl-a, turbidity and TSS shown in Figures 5–7. The error bars on the graph indicate the coefficient of variation (CV) values between the in-situ and Landsat-predicted results for the Chl-a values. From Figure 5, it can be noted that the Landsat-predicted measurement (P4, P5, P6, P16 and P17) matched the in-situ measurement for both periods of data collection. However, the other remaining points showed slight variation by either underestimating or overestimating the predicted values with the CV <30% for all points.

Figure 5. (a) Trends of observed and sensor-estimated Chl-a concentration for the first period. (b) Trends of observed and sensor-estimated Chl-a concentration for the second period.

Figure 6. (a) Trends of observed and sensor-estimated TSS concentration for the first period. (b) Trends of observed and sensor-estimated TSS concentration for the second period.

Figure 7. (a) Trends of observed and sensor-estimated turbidity concentration for the first period. (b) Trends of observed and sensor-estimated turbidity concentration for the second period.

The results from Figures 6 and 7 show that there was substantial variation between the predicted and in-situ values for both TSS and turbidity at each collection period. There were situations of overestimating and underestimating both TSS and turbidity at each period. However, the CV for TSS was <15% as compared to that of turbidity which is <35%.

3.4. Spatial extent of Chl-a, TSS and turbidity

The spatial distribution of the WQPs was performed to help envision changes in water quality over the lake from the sample points. Both the spatial maps for the in-situ and estimated WQP from Landsat 8 were generated using spline (Figures 8–10) to enhance further analysis of the model. The results from the Chl-a distribution (Figure 8) show that both the in-situ and estimated WQP are highly correlated with the high and low Chl-a concentrations mostly predominant in the same area. Low concentrations of Chl-a were mostly predominant in the southern part of the lake, with high concentrations dominating in the middle to the northern part of the lake. However, Figure 8(b) of Landsat 8 shows large areas with high concentrations of Chl-a over the lake as compared to Figure 8(a) and in-situ measurement of Figure 8(b). This is an indication of the overestimation of the Chl-a in the areas. The results in Figure 9 show the TSS distribution of the measured and estimated measurement of in-situ and Landsat 8. The spatial maps of TSS in Figure 9(a) show much correlation between the in-situ measurement and estimated Landsat 8 measurement. However, in Figure 9(b), there is a little difference between the spatial map of the in-situ TSS measurement and the estimated Landsat 8 measurement. The highest concentration of TSS was more pronounced in the southern part of the lake for both in-situ and estimated measurements in Figure 9(a). However, for Figure 9(b), the in-situ measurement shows greater area of high TSS, with estimated measurement from Landsat 8 showing low TSS in most parts of the lake. This depicts a fair coincidence between the measured values and the satellite-based model. Figure 10 shows the distribution of turbidity for in-situ measurement and Landsat 8 estimated measurement. From the results, it can be noted that the in-situ measurement and Landsat 8 estimated measurement for Figure 10(a) are highly correlated, with much concentration of turbidity mostly in the northern part of the lake and some parts of the southern sector. However, there is much difference in the spatial distribution of turbidity of the in-situ measurement and the estimated Landsat 8 measurement (Figure 10(b)). The Landsat 8 estimated measurement overestimated the turbidity when compared to the in-situ measurement.

Figure 8. (a) In-situ and estimated Chl-a distribution on 17 December 2022. (b) In-situ and estimated Chl-a distribution on 15 March 2023.

Figure 9. (a) In-situ and estimated TSS distribution on 17 December 2022. (b) In-situ and estimated TSS distribution on 15 March 2023.

Figure 10. (a) In-situ and estimated turbidity distribution on 17 December 2022. (b) In-situ and estimated turbidity distribution on 15 March 2023.

The radar chart (Figure 11) shows the extent of coverage of WQPs (Chl-a, TSS and turbidity) categorized by land use sites (Agriculture, Akosombo textiles, Settlement, Kpong landing site and Fish farm). It is a disaggregation of the respective sites in which samples were used as validation points to determine the water quality status using the EMRM. The area under the chart at the current conditions indicates degradation of water quality at the five (5) sites which can be as a result of the land use activities in the area. The large area under the chart for TSS and turbidity for the first and second sample periods compared to Chl-a at both periods suggests a general degradation in TSS and turbidity for all the sites. The results from Figure 11(a) show that Chl-a was dominant in the agriculture site, followed by the fish farm and Kpong landing site, with the Akosombo textile site recording the lower content of Chl-a; this trend is the same for Chl-a in the second period of sampling. For the TSS and turbidity, they were more pronounced in the settlement site, followed by Kpong landing site and agricultural site, respectively, with the Akosombo textile site recording the lowest values; this trend of dominance is same for both the first and second periods. From these results, it can be noted that land use activities contribute to the concentration of a particular WQP in an area.

Figure 11. (a) Spatial extent of Chl-a, TSS and turbidity at five different land use sites. (b) Spatial extent of Chl-a, TSS, and turbidity at five different land use sites.

4. Discussion

The physicochemical parameters of the water such as pH, EC, turbidity, TSS and NH₄ were assessed. The major use of the water within the study area is for electricity generation, fishing, and farming at different localities within the lake. The pH of the lake ranges from 5.9 to 8.8 with an average and SD of 6.9 and 0.38 pH units, respectively, which makes the water in the lake more alkaline. This result corroborates the results of Tay (2021) who recorded the pH of the lake ranging between 6.3 and 8.2 pH units, with an average and SD of 7.1 and 0.41, respectively. The result is also an indication that the lake is buffered by bicarbonate system which is consistent with the results of Claude and Craig (1998). The turbidity values of the lake were high, i.e. between 5.56 and 24.9 NTU with average and SD of 7.9 and 3.0 NTU (Table 1). These values are >5 NTU, which is deemed the standard value for natural water (Tay, 2021; World Health Organization, 2011). This could be due to the ongoing activities such as irrigation farming, fish farming and the transportation of good and services on the lake. Additionally, the growth of algae due to the high availability of nutrients by natural decay influences the turbid nature of the lake. This result is in line with earlier research conducted on the lake which indicated that the turbidity on the lake is >5 NTU which is stated by WHO to be the standard value for normal water (Tay, 2021). It also corroborates the results of studies in other areas which showed turbidity value >5 NTU (Amoateng, 2016; Banunle et al., 2018). The importance of assessing turbidity is to get essential information that is related to the ecological functioning of the water masses and their interaction with the environment. The in-situ measurement of Chl-a ranges between 0.206 and 13.5 mg/L, with an average and SD of 5.1 and 4.3 mg/L, respectively. Chl-a measurement is mostly performed to determine the trophic status of a water body (Michaud, 1991), though there are no specific standard to measure Chl-a. The results show that the lake qualifies to be described by the three states, i.e. oligotrophic when Chl-a <7 mg/L, mesotrophic when Chl-a values are between 7 and 12 mg/L and eutrophic when Chl-a >12 mg/L (Tay, 2021), since the values fall within all the three states. So, it can be said that the lake is changing from oligotrophic to eutrophic based on the values from the Chl-a measurement which corroborates the results of Tay (2021). Thus, high concentration of Chl-a is an indication of high levels of nutrients which indicate eutrophication of aquatic systems. The TSS ranged between 18 and 48 mg/L with average and SD of 28.7 and 5.08 mg/L, respectively. The TSS indicate the number of solids suspended in the water being it organic or mineral and is mostly related to the cloudiness of water (turbidity). The high TSS value results in the murky nature of the water, which prevents the penetration of sunlight into the water. This condition deprives the benthic plants and algae from getting sunlight for growth.

The in-situ Chl-a measurement has average and SD of 5.1 and 4.3 mg/L, respectively. Since the lake has no restrictions and open to public use, a lot of external influence contributes to the Chl-a concentration. For example, the leachate of fertilizers from neighboring farms into the lake can contribute significantly to the concentration of Chl-a into the lake. Additionally, other activities such as fish farming on the lake also influence the Chl-a concentration resulting from the feeds that are used in feeding the fish. The reflectance of Landsat 8 based on the blue and green bands gave the best results in the estimation of the Chl-a concentration. This result contradicts the results of Omondi et al. (2023), who indicated that the reflectance that dominated in the estimation of Chl-a was green and red bands, and that of Watanabe et al. (2018), which stated that the red and Near Infrared (NIR) bands are suitable for the retrieval of Chl-a. However, similar results were obtained by Ouma et al. (2020) where retrieval algorithm of Chl-a concentration was mainly from the green and blue bands, due to the high reflectance of green algae in the water. This research compared Landsat 8 reflectance with the in-situ measurement for Chl-a prediction on the lake. The results showed that the model with the green and blue band combination produced the best results with R² > 0.85. Lailia et al. (2015), estimated Chl-a from Landsat 8 using the blue, red and SWIR 2 bands in six natural and five artificial lakes in Greece, with an R² of 0.75. Comparing the outcome of this study with earlier studies in Chl-a retrieval, it can be said that the visible bands are suitable for the prediction of Chl-a in water bodies.

The results from the in-situ TSS concentration have an average and SD of 28.7 and 5.08, respectively. The TSS concentration is low on the lake, due to the less inflow of sediments from the surrounding areas on the lake. It can also be attributed to the fact that most activities on the lake do not lead to high inflow of sediments into the lake, however, the few sediments that flow into the lake generally settle at the bottom resulting in the low TSS value. From the linear regression algorithm, the best estimation of the TSS concentration from the Landsat 8 was obtained using the blue, green and red bands with R² > 0.70. The R² is comparable with the results of González-Márquez et al. (2018) and Mushtaq and Nee Lala (2016) who obtained R² between 0.637 and 0.950. The overall performance of the satellite sensors in the estimation of TSS for lake is shown in Tables 2 and 3. The green band was optimal in the retrieval as in the case of Chl-a. This result corroborates that of Yepez et al. (2018) who used Landsat OLI in the retrieval of suspended sediments in the Orinoco River in Venezuela. The TSS results in Tables 2 and 3 demonstrate that the presence of visible bands in Landsat 8 is instrumental in the estimation of TSS. From the results of this study, it is evident that the green and red bands are important in the estimation of TSS in the lake. The average turbidity value was 7.9 NTU with a SD of 3.0 NTU. The low turbidity values may result from less inflow from the surrounding areas into the lake. This situation can also be attributed to the depth of the lake which makes it possible for the sediments to settle at the bottom of the lake with less re-suspension by waves and currents of the water. The results from the regression model show that the turbidity estimated for the two different periods has R² > 0.65. In the turbidity estimation, the green and red bands were observed to perform better in the retrieval. This result supports the work of Quang et al. (2017) who stated that the red band produces the best correlation with measured turbidity with R² of 0.84. Generally, it can be said that Landsat 8 is very effective in the prediction of WQPs with minimum variation in the SD and CV. From the results, Landsat 8 is effective and less costly as a tool for assessing WQPs on the lake. From the spatial map of Chl-a in Figure 8(a), it is observed that the highest concentrations are seen around the Akosombo dam at P18, P16 and P7 (8.74–10.47 mg/L) for the in-situ measured, with Landsat 8 overestimating the Chl-a concentration concentrations with values between 10.47 and 13.48 mg/L. However, in Figure 8(b), the high concentration of the in-situ Chl-a areas is closely related to that of the estimated Landsat 8 Chl-a concentration. This similarity provides satisfactory evidence as shown in Table 3.

The spatial map for turbidity concentration in Figure 10(a) shows a good correlation between the in-situ measurement and the estimated measurement from Landsat 8. The highest concentration points were located at P5 and P15 for both in-situ and Landsat 8 measurements. Additionally, most of the places on the turbidity spatial map show low turbidity concentration. However, in Figure 10(b), there is high discrepancy in the spatial distribution of the turbidity concentration for the measured in-situ concentration and the Landsat 8 concentration. The TSS distribution in Figure 9 is observed to have high disparity between the measured in-situ and estimated Landsat 8 measurement for the two different periods. The Landsat 8 estimated concentration overestimated the TSS concentration resulting in low correlation between the in-situ measured TSS and the estimated TSS concentration for the periods. It is evident from this study that the models used in this research can be used to map Chl-a, turbidity and TSS concentration in other lakes or rivers in the country.

5. Conclusion

The study investigated the WQPs of the study area using the physical and chemical parameters as potential indicators of pollution. Additionally, the performance of Landsat 8 in predicting Chl-a, turbidity, and TSS was evaluated in this study. The alkalinity of the river is very high with value of 27.9. The turbidity values are >5 NTU which is considered as the standard value for normal water. This is an indication that the lake water unsafe for human consumption. With the Chl-a ranging from 0.0 to 12.4 µg/L, the lake water is seen to be changing from oligotrophic to eutrophic, i.e. the lake is seen to be getting greener over time.

In terms of the retrieval of WQPs from the Landsat 8 image, the study has demonstrated that, the integration of the Landsat image and in-situ measurement offer excellent means of providing cost-effective and timely estimation of WQPs. For the three parameters, turbidity was estimated using a linear regression model with both R² and Pearson correlation coefficient (r) >0.66 and 0.60, respectively. TSS was best estimated by logarithm and linear regression models with respective mean R² and r of 0.70 and 0.618. Chl-a was best estimated using logarithm and linear regression models with mean R² and r of 0.853 and 0.701, respectively. Considering the three WQPs assessed in this study, the results from the spatial map can provide a thorough understanding of the dynamics of the water qualities in the study area, resulting in awareness creation of the changes in the environment in relation to anthropogenic activities which will help in providing better management policies. The availability of satellite images and water quality data aids in developing more integrated and robust predictive models in future that will help in effective monitoring of the water resources.

Disclosure statement

No potential conflict of interest was reported by the authors.

Data availability statement

The datasets generated during/or analyzed during the current study are available from the corresponding author upon reasonable request.

Additional information

Notes on contributors

Linda Appiah Boamah

Linda Appiah Boamah is a lecturer in Environmental Management Technology Department at Koforidua Technical University. Her research interest includes water quality analysis, climate change, waste management, and remote sensing.

Clement Nyamekye

Clement Nyamekye is an Associate Professor of Remote Sensing and Climate change in Civil Engineering at Koforidua Technical University.

Charles Gyamfi

Charles Gyamfi is a trained Civil Engineer with specialization in Hydrology and Water Resources Management. In 2017, he earned a PhD in Civil Engineering (Hydrology and Water Resources specialty) from the Tshwane University of Technology, Pretoria, South Africa.

Jonathan Quaye Ballard

Jonathan Arthur Quaye-Ballard is a Professor at the Department of Geomatic Engineering, Kwame Nkrumah University of Science and Technology (KNUST), Kumasi, Ghana.

Geophrey Kwame Anornu

Geophrey Kwame Anornu is a professor of water resources engineering and hydraulics at the Department of Civil Engineering, KNUST and he is a member of the Ghana Institution of Engineers (GhIE). With years of experience in teaching and research, Professor Anornu has supervised the research projects and directly contributed to the academic development of students at both the undergraduate and graduate levels.