ABSTRACT
The role of freshwater lakes in providing water resources and supporting ecosystems is essential. Monitoring water quality using remote sensing (RS) technologies is crucial for sustainable management practices. A study on Loktak Lake was done using RS algorithms to predict post-monsoon water quality. The multiplication band model (B1 × B6) demonstrated a moderate correlation with dissolved oxygen (DO) values (mg/l) with (coefficient of determination, R2 = 0.47, root mean square error, RMSE = 0.23, and standard error of estimation, SEE = 0.23). The band combination (B2/B4) was strongly correlated with electrical conductivity (EC) values (μs/cm) (R2 = 0.60, RMSE = 9.44, and SEE = 9.69). For total dissolved solids (TDS) (mg/l), with an R2 = 0.61, RMSE = 5.95, and SEE = 6.09, Band 2 demonstrated a strong correlation between field values and satellite imagery. The post-monsoon water quality map of the lake indicates lower concentrations of DO, EC, and TDS on the western side and elevated values on the eastern side. The research concluded that RS algorithms can be effectively used to predict water quality parameters in Loktak Lake, specifically DO, EC, and TDS. The findings suggest that effective pollution management is needed on the western side of the lake.
HIGHLIGHTS
Frequent and continuous water quality monitoring has been problematic in Loktak Lake, Manipur.
A linear regression can be used to develop algorithms for retrieving water quality data.
Water quality prediction model for DO, EC, and TDS.
Remote sensing (RS) and GIS provide rapid information on water quality and spatial variability.
Spatial water quality map can be generated using RS and GIS.
INTRODUCTION
Water, the vital substance that sustains the existence of plants, humans, and various other organisms in both the land-based and underground realms, is indisputably the primary and essential renewable resource in nature (Sutradhar et al. 2021; Ruidas et al. 2023). This resource is invaluable to various life forms due to high transportation costs, purification challenges, and the impracticality of finding an alternative (Ruidas et al. 2021). According to Najafzadeh & Niazmardi (2021), the crucial importance of surface water quality is apparent in its impact on the long-term sustainability of ecological systems. Freshwater lakes, which make up the bulk of lakes on earth, offer a variety of environmental and ecological services, including irrigation water, food production, drought and flood mitigation, biodiversity preservation, power generation, and other socio-economic advantages (Laishram et al. 2022). Recent studies have pointed to a significant increase in contamination in rivers and other water bodies highlighting a major source of pollution (Ruidas et al. 2023). Surface water contamination has emerged as a widespread environmental issue on a global scale as a result of the infiltration of harmful substances and eutrophication in rivers and lakes (Ouyang et al. 2006). Sewage discharge from anthropogenic activities, such as urbanization, population explosion, and industrialization, has a high impact on overall water quality (Ji et al. 2021). Lake water quality is influenced by various physical, biological, and chemical factors. And also, the water resource system is vulnerable to stressors, climatic changes, and increased demand (Anandhi & Kannan 2018). Other scholars (Nagaraju et al. 2016; Akhtar et al. 2021) have also reported that the existing surface water has been subjected to various human-induced activities such as urbanization and agricultural practices, along with natural occurrences such as climate variations and geological factors, as well as increased usage, leading to significant degradation of water quality globally. In recent times, Ramsar sites have been increasingly subjected to significant detrimental effects arising from anthropogenic activities in conjunction with environmental factors (Salihoglu & Karaer 2004; Bhatti et al. 2019; Borah et al. 2020; Man et al. 2020). Researchers such as Mabwoga & Thukral (2014) and Nisari & Sujatha (2021) have established that numerous crucial Ramsar sites within India, along with their ecosystems, surroundings, and water sources, are at risk. Laishram et al. (2023) also discussed the impact of uncontrolled discharges and the establishment of the Loktak Multipurpose Project on the increased presence of pollutants in the Loktak Lake which holds a Ramsar designation.
Loktak Lake, the freshwater wetland of Manipur in the northeastern region of India, holds great significance for the people of Manipur due to its cultural, socio-economic, and ecological importance. This lake is a globally recognized wetland (Ramsar site) and has been listed in the Montreux Record (List 2020). The lake is widely recognized for its floating vegetation, commonly referred to as ‘Phumdis’. Nearby, there exists a portion of Phumdis measuring approximately 40 km2, which is famously known as Keibul Lamjao National Park (KLNP), and holds the distinction of being the sole floating national park in the world (LDA & WISA 1999). KLNP serves as the exclusive habitat for the imperilled and native species of Manipur's brow-antlered deer, more commonly referred to as Sangai. The lake sustains a diverse range of living organisms with a total of 132 plant species and 428 animal species (Singh et al. 2011). Additionally, it plays a vital role in providing irrigation to around 32,400 ha of agricultural land, serving as a source of income for approximately 8,700 fishermen of nearby Loktak Lake, and enabling the generation of (105 MW) hydro-power (LDA & WISA 2002). The annual benefits obtained from Loktak Lake amount to approximately 600 million Indian Rupees, accounting for nearly 2% of the state's gross domestic product (LDA & WISA 2010).
Nine rivers, namely Thongjaorok, Awang Khujairok, Khuga, Nambol, Imphal, Kongba, Iril, Nambul, and Thoubal, discharge into the lake, which has a crucial impact on the natural surroundings and the lake's overall ecological health. One of the major rivers, the Nambul, brought garbage and plastic polythenes from the city of Imphal on its route through the urban areas, increasing lake water pollution (Devi & Singh 2021). Research done by Khwairakpam et al. (2021) has identified a decline in the water quality of the lake, which is linked to the introduction of sewage, soil sediments, and agricultural fertilizers through river discharges. Due to jhum (shifting) cultivation, deforestation, and irrational land use in the catchment areas, the rate of siltation has increased (Khoiyangbam 2021). Also, the anthropogenic pressures and urbanization on Loktak are a major concern for its degradation rapidly (Tuboi et al. 2018; Laishram 2021). The deterioration of water quality in Loktak Lake is further intensified by the heightened utilization of fertilizers, resulting in adverse effects on the aquatic fauna and the distinctive phumdis within the lake (Gupta et al. 2024). Concern has been raised about the lake's water quality gradually deteriorating as a result of pollution from different sources (Kangabam et al. 2015; Das Kangabam et al. 2017). A major initial step towards creating efficient conservation and management plans for the entire Loktak Lake ecosystem is the monitoring of water quality at the catchment scale (Khwairakpam et al. 2019). This delicate ecosystem faces multiple challenges, including the direct entry of pollutants from its surrounding area, the rapid growth of urbanization, widespread fishing activities, and more. Additionally, the lake's water retention time has significantly lengthened since the construction of the Ithai barrage in 1983, which was intended for hydroelectric power (Singh et al. 2010). Over time, the quality of water has declined, leading to a decrease in biodiversity (Trisal & Manihar 2002).
Traditionally, water quality information is gathered through on-site measurements that yield accurate outcomes. However, on-site data collection is laborious and demanding in terms of time and resources, and it is constrained to geographical and temporal extents, notably in cases where the examination pertains to extensive water masses (Song et al. 2012; Bonansea et al. 2015). Moreover, the current monitoring techniques lack the capability to offer a timely, spatially, and temporally comprehensive analysis of water quality necessary for effectively managing the integrity of aquatic ecosystems and addressing public health issues (Mushtaq & Nee Lala 2017). Several previous studies on the water quality monitoring of Loktak Lake were carried out using the conventional methodology. Physiochemistry, geochemical analysis, socio-economic, and water quality studies of Loktak Lake were carried out by many researchers where the assessment of the portability, irrigation suitability, and pollution status of surface sediments were carried out (Devi et al. 2015; Mayanglambam & Neelam 2022). Roy & Majumder (2019) evaluated the water quality trends in Loktak Lake, India, through the use of statistical analyses and water quality indices. Das Kangabam et al. (2017) developed a water quality index (WQI) for Loktak Lake, India, revealing that the water is unfit for drinking due to high nitrite levels and other parameters beyond permissible limits. Water quality monitoring and modelling were performed in Loktak Lake's catchment by utilizing hybrid SHE-SWAT and MIKE 11 and MIKE 21 ECO Lab to model for simulating water quality parameters (WQPs) and assess the ecological status for effective conservation and management (Khwairakpam et al. 2019, 2021). The evaluation of water quality in water bodies using traditional approaches is restricted to a specific number of in situ water sampling locations and subsequent analysis in a laboratory. Such techniques are both time-consuming and expensive, and limited to understanding the spatial and temporal aspects of surface water quality (Liu et al. 2010).
With the advent of satellite remote sensing (RS) technology, it has allowed for the synoptic and multitemporal observation of water quality for over 25 years (Dekker et al. 1995; Cracknell et al. 2001; Wu et al. 2009; Mishra et al. 2019). Satellite sensors measure reflected solar radiation from surface water to assess water quality and estimate parameters. This approach offers advantages such as providing synoptic estimates, estimating water quality in remote regions, and maintaining lengthy records of archived imagery (Hellweger et al. 2004; Hossain et al. 2010, 2014, 2021). RS techniques have become a favourable and economically efficient option for monitoring the parameters of water quality (Kaplan et al. 2019; Pu et al. 2019). Satellite RS provides a systematic and reliable approach for consistently evaluating the water quality of lakes (Ren et al. 2018). The use of RS technology has shown a strong level of accuracy in the estimation of water quality. This is particularly evident in the monitoring of inland lakes and smaller bodies of water (Avdan et al. 2019; Afidah et al. 2020). Also, the integration of satellite data with field observations provides a more efficient and structured approach to monitoring changes in surface water quality across broad spatial and temporal dimensions (Masocha et al. 2017). The primary advantage of RS through satellites for assessing water quality is the generation of synoptic views without the need for expensive on-site surveys. Furthermore, the fusion of RS and GIS has been highlighted as an efficient approach to water quality monitoring (Afidah et al. 2020; Júnior et al. 2023). Moreover, RS can assist in obtaining water quality information more regularly to monitor the rapid fluctuations in aquatic environments, enabling prompt policy responses (Dekker et al. 2001; Gholizadeh et al. 2016). At present, there is a lack of RS algorithms for the analysis of surface water quality in Loktak Lake. Despite the numerous advantages offered by the RS and GIS technology, limited research is available on the water quality monitoring studies of Loktak Lake and is mostly confined to the conventional approach. This study aims to develop an RS-based linear algorithm for assessing the WQPs of DO, EC, and TDS, as well as generating spatial distribution maps of water quality in Loktak Lake. This study may offer to support the management of Loktak water resources by offering a more practical and efficient method for monitoring the status of water.
MATERIALS AND METHODS
Study area
The Loktak Lake is located in a central valley that covers 28% of the entire catchment area, which spans 4,947 km2. The region displays features of a tropical to semi-tropical climate, with a shift to a semi-temperate to temperate climate at higher altitudes. The elevation within the catchment varies from 800 m above mean sea level (a.m.s.l.) in the valley to over 2,500 m a.m.s.l. in the surrounding mountainous regions. The lake's climate is heavily influenced by the southwestern monsoon of the Indian subcontinent, with its rainy season lasting from June through September and contributing to 63% of annual precipitation at an average amount of 1,409 mm (Singh et al. 2010). In this region, the average yearly temperature recorded amounts to 20.5 °C, while during summer months, it goes up to 24 °C, and when winter arrives, it drops down to 14 °C. The catchment of the lake has an annual potential evapotranspiration of about 1,063 mm.
Roy & Majumder (2019) conducted a comprehensive 3-year study analysing 12 WQPs on a monthly basis in Loktak Lake. They found that the lake generally had satisfactory water quality, with slight declines during winter months. The water was slightly turbid due to organic matter but remained relatively stable throughout the study period with minimal fluctuations in WQPs. However, some samples showed lead concentrations exceeding safe drinking water limits. A study by Das Kangabam et al. (2017) also indicated poor water quality in Loktak Lake, making it unsuitable for drinking. Various pollutants such as nitrite, dissolved oxygen (DO), electrical conductivity (EC), and chemical oxygen demand contributed to high levels of pollution as reflected in the WQI values ranging from 64 to 77 with seasonal variations observed.
Sampling sites and analysis
Characteristics of sampling sites
Sampling site . | Sampling no. . | Latitude . | Longitude . | Elevation (m) . | Description of site . |
---|---|---|---|---|---|
Khoijuman Khunou | A1–A15 | 24°33′52″ E | 93°49′19″ N | 785 | Runoff from Imphal river |
Ningthoukhong | B1–B15 | 24°32′34″ E | 93°47′33″ N | 784 | Aquaculture and water runoff from hydroelectric project as well as runoff from agricultural activities |
Thanga and Karang | C1–C15 | 24°31′33″ E | 93°50′9″ N | 784 | Surrounding areas of Thanga and Karang (island) |
Naransena | D2–D3 | 24°31′38″ E | 93°46′38″ N | 783 | The runoff from the west side of Loktak Lake is primarily influenced by small tributaries and agricultural activities. |
Phubala | E2–E4 | 24°31′59″ E | 93°46′33″ N | 783 | Runoff primarily from the west side of Loktak Lake, with assistance from tiny tributaries and farming operations. |
Thinungei | F1–F5 | 24°32′10″ E | 93°46′34″ N | 783 | Small tributaries and agricultural activities produced the majority of the runoff from the west side of Loktak Lake. |
Mid | A*–F* | 24°32′6″ E | 93°48′33″ N | 784 | Middle portion of Loktak Lake |
Sampling site . | Sampling no. . | Latitude . | Longitude . | Elevation (m) . | Description of site . |
---|---|---|---|---|---|
Khoijuman Khunou | A1–A15 | 24°33′52″ E | 93°49′19″ N | 785 | Runoff from Imphal river |
Ningthoukhong | B1–B15 | 24°32′34″ E | 93°47′33″ N | 784 | Aquaculture and water runoff from hydroelectric project as well as runoff from agricultural activities |
Thanga and Karang | C1–C15 | 24°31′33″ E | 93°50′9″ N | 784 | Surrounding areas of Thanga and Karang (island) |
Naransena | D2–D3 | 24°31′38″ E | 93°46′38″ N | 783 | The runoff from the west side of Loktak Lake is primarily influenced by small tributaries and agricultural activities. |
Phubala | E2–E4 | 24°31′59″ E | 93°46′33″ N | 783 | Runoff primarily from the west side of Loktak Lake, with assistance from tiny tributaries and farming operations. |
Thinungei | F1–F5 | 24°32′10″ E | 93°46′34″ N | 783 | Small tributaries and agricultural activities produced the majority of the runoff from the west side of Loktak Lake. |
Mid | A*–F* | 24°32′6″ E | 93°48′33″ N | 784 | Middle portion of Loktak Lake |
Satellite imagery processing
The Landsat 9 imagery Operational Land Imager (OLI) imageries were used in this study. The image of the study area was downloaded from the Landsat imagery archive which is hosted by the United States Geological Survey (USGS) (EarthExplorer (usgs.gov)). The satellite image covering the study area was retrieved on the 28th of January 2023; from the dataset that belongs to path 135 and row 043. The Landsat 9 imagery (cloud-free or with cloud cover of <10%) over the area selected for analysis was processed in ArcGIS software. Processing of Landsat 9 imagery involved performing radiometric and atmospheric corrections. The radiometric calibration process converts the pixel values to precise radiation measurements per light wavelength unit or reflectance. Radiometric calibration to convert digital number (DN) values to physical units, at sensor spectral radiance (Watts/(m2/srad/μm)) and then to top-of-atmosphere (TOA) reflectance was done in ArcMap (Version 9.3) (Kapalanga et al. 2021) using the equations described by Mondejar & Tongco (2019). Subsequently, the atmospheric correction was implemented in order to remove the impact of atmospheric variables on the dynamics of surface reflections, and to convert the radiometric data into either radiation or surface reflectance. The process of atmospheric correction plays a pivotal role in the steps of image processing. Hence, in order to acquire precise and reliable quantitative information through RS technology, the execution of atmospheric correction becomes imperative (Liang et al. 2001; Chander et al. 2009; Tyagi & Bhosle 2011; Lillesand et al. 2015). In this study, the atmospheric correction was performed using the equations described by Mondejar & Tongco (2019). The nearest-neighbour resampling technique was utilized to perform geometric correction on the image rastermaps. Additionally, the correct georeferenced (WGS 1984UTM Zone 48 N) was assigned to match that of the study area. The Landsat 9 Operational Land Imager (OLI) sensor is made up of nine bands that each serve a specific purpose. Bands 1, 2, 3, 4, 5, 6, and 7 were used to build the model in this work (Hossain et al. 2021). Table 2 shows the spectral specifications of the Landsat 9 bands.
Landsat 9 spectral specifications
Band . | Minimum lower band edge (nm) . | Maximum upper band edge (nm) . | Center wavelength (nm) . | Maximum spatial resolution at Nadir (m) . |
---|---|---|---|---|
Coastal/Aerosol | 433 | 453 | 443 | 30 |
Blue | 450 | 515 | 482 | 30 |
Green | 525 | 600 | 562 | 30 |
Red | 630 | 680 | 655 | 30 |
NIR | 845 | 885 | 865 | 30 |
SWIR 1 | 1,560 | 1,660 | 1,610 | 30 |
SWIR 2 | 2,100 | 2,300 | 2,200 | 30 |
Band . | Minimum lower band edge (nm) . | Maximum upper band edge (nm) . | Center wavelength (nm) . | Maximum spatial resolution at Nadir (m) . |
---|---|---|---|---|
Coastal/Aerosol | 433 | 453 | 443 | 30 |
Blue | 450 | 515 | 482 | 30 |
Green | 525 | 600 | 562 | 30 |
Red | 630 | 680 | 655 | 30 |
NIR | 845 | 885 | 865 | 30 |
SWIR 1 | 1,560 | 1,660 | 1,610 | 30 |
SWIR 2 | 2,100 | 2,300 | 2,200 | 30 |
Algorithm development and validation
The collected water sample data were separated into 80% training and 20% testing datasets. The samples for testing were chosen at random from the complete dataset and solely employed for validating the model. The study utilized the Landsat 9 OLI bands as the independent variables. These bands were employed in the prediction of WQPs such as EC, DO, and TDS. The estimation models for WQPs, as formulated in the present study, have been constructed employing an empirical methodology that relies on establishing simple linear regression relationships between in situ observations and surface reflectance data obtained from Landsat 9 satellite imagery. Various researchers have developed a variety of algorithms aimed at estimating different WQPs. González-Márquez et al. (2018), Kc et al. (2019), Qi et al. (2020), Al-Shaibah et al. (2021), and Imran et al. (2022) have developed empirical model for DO and TDS parameters in their studies. Using Microsoft Excel, the data were statistically analysed. The selection of these variables was based on a thorough review of the available literature (Alparslan et al. 2007; He et al. 2008; El Saadi et al. 2014).
A prescreening was also done to select bands (or band combinations) for model development with high R2 values. Different combinations of bands and single bands were examined to assess the degree of their associations with WQPs. Among the various models that were tested for this purpose, the models that showed the highest R2 values were considered the most optimal models (Hossain et al. 2021). Many researchers have utilized almost all optical bands for this purpose, but the visible and infrared bands were more commonly used for the development of the WQ model. The bands used in previous studies are given in Table 3. El Ouali et al. (2021) employed Sentinel-2 satellite imagery provided by the European Space Agency to conduct research focusing on the modelling and spatiotemporal mapping of water quality within the Hassan Addakhil dam reservoir. Ferdous et al. (2019) employed Landsat 8 Operational Land Imager (OLI) images for the identification of TDS in the coastal regions of Bangladesh. In another study by González-Márquez et al. (2018), Landsat 8 satellite imagery was used to evaluate water quality indicators and depth levels in the El Guájaro Reservoir located in Colombia. Many other researchers (Japitana & Burce 2019; Kc et al. 2019; Qi et al. 2020; Hossain et al. 2021; Kapalanga et al. 2021; Imran et al. 2022; Omondi et al. 2023) in their studies have also used Landsat data to conduct a variety of assessments and analyses related to water quality. These include evaluating water pollution, monitoring trends in water quality over time, mapping water contaminants, and estimating surface WQPs.
Bands reported for developing water quality models for DO, EC, and TDS
Dependent variables . | Model performance . | . | References . |
---|---|---|---|
Predictors . | R2 . | ||
DO (mg/l) | B1 (435–450 nm), B6 (1,565–1,651 nm) | 0.47 | El Ouali et al. (2021) |
EC (μs/cm) | B2 (452–512 nm), B4 (636–672 nm) | 0.60 | González-Márquez et al. (2018) |
TDS (mg/l) | B2 (452–512 nm) | 0.61 | Ferdous et al. (2019) |
Dependent variables . | Model performance . | . | References . |
---|---|---|---|
Predictors . | R2 . | ||
DO (mg/l) | B1 (435–450 nm), B6 (1,565–1,651 nm) | 0.47 | El Ouali et al. (2021) |
EC (μs/cm) | B2 (452–512 nm), B4 (636–672 nm) | 0.60 | González-Márquez et al. (2018) |
TDS (mg/l) | B2 (452–512 nm) | 0.61 | Ferdous et al. (2019) |
For this study, the spectral bands and their band combinations used were Coastal/Aerosol (band 1, 435–450 nm), Blue (band 2, 452–512 nm), Green (band 3, 532–589 nm), Red (band 4, 636–672 nm), NIR (band 5, 850–879 nm), SWIR 1 (band 6, 1,565–1,651 nm), and SWIR 2 (band 7, 2,105–2,294 nm) (Masek et al. 2020).
The ‘Extract Multi-Values to Points’ geoprocessing features of the ArcMap software were used to get the spectral reflectance values at the appropriate pixel for each sample's coordinate position within each dataset. In the attribute database connected to every pertinent feature class at every sample site, the tool stores the spectral reflectance for every spectral band. When extraction was complete, the conversion tools were used to import the spectral reflectance values of each sample's coordinate position from every data set into Excel (Hossain et al. 2021). The training dataset was processed using single bands and various combinations of linear regressions. Plotting the spectral reflectance for each of the bands on the x-axis and the in situ measurements of each WQPs on the y-axis required creating scatter plots. The models that had the highest coefficient of determination (R2) were the ones chosen.
The next step is to compare the model that performed the best in the validation stage for each parameter using the testing datasets. Validation is considered an approach and justification for ‘creating the proper model’ (Tsioptsias et al. 2016). The selected regression models were verified and validated by calculating their R2, root mean square error (RMSE), and standard error of estimation (SEE) using the testing samples' data. Spatial variations map of DO, EC, and TDS WQPs of the study area were generated using the regression model equations. In several fields, including geology, environmental science, and agriculture, spatial interpolation is a technique used to determine values at unknown places based on known data points (Gedeon et al. 2003). Water quality has been assessed and mapped by several research using various interpolation methods. The inverse distance weighted (IDW) method of spatial interpolation has been widely utilized by several authors to estimate values at unsampled places (Yurembam et al. 2015; Yang et al. 2020; Sharma et al. 2021, 2022; Srilakshmi et al. 2022; Ruidas et al. 2023; Jaydhar et al. 2024). Another common geostatistical method for mapping water quality data is kriging interpolation. Kriging has been used in studies like Ahmadi & Sedghamiz (2008) and Omondi et al. (2023) to assess and improve groundwater level observation networks with an emphasis on enhancing data quality. Furthermore, Seyedmohammadi et al. (2016) evaluated the spatial relationships of groundwater quality measures like NO3− and Cl− and water quality variables such as TDS using kriging techniques. Strong spatial dependencies were found in the results of their geospatial analysis using kriging interpolation, which made it possible to create maps of the spatial distribution of water quality metrics. In this study, the kriging interpolation technique was used for mapping estimated WQPs. Kriging is a powerful spatial interpolation technique, particularly for data points with uneven spacing. Kriging is commonly recognized as the ‘BLUE’ acronym, which represents the best linear unbiased estimator (Sidler & Holliger 2003).
RESULTS AND DISCUSSION
Post-monsoon field data
The DO concentration was observed to be 4.57 ± 0.35 mg/l (3.64–5.50). The lowest DO value of 3.64 mg/l was observed at Khoijuman Khunou location (A1) and a slightly higher DO value of 5.34 mg/l was observed at Thanga and Karang (C4). The lower value of DO denotes higher pollution with lower water quality. Khoijuman Khunou location where the lowest value of DO was observed is influenced mostly by the agricultural land use on the northwestern sides of the lake, which drains into the Loktak. The Nambul catchment of which agriculture is the major land use covering about 47% (91 km2) of the total area (Khwairakpam et al. 2021) has also an impact on pollutant discharge. Paddy cultivation is prominent in the basin that uses agricultural pesticides (nitrogen and phosphorous-based fertilizers). This leads to the contribution of loadings rich in fertilizers which may further affect the water quality of the lake. Moreover, the Nambul River traverses urbanized regions such as Imphal, Singjamei, and Sagolband, where there is a substantial production of household refuse. These household wastes are discharged untreated directly into the river. The high organic load associated with the discharge is expected to increase the microbial load, thereby causing a low DO in the particular area. Habitation and urban settlements within the Nambul sub-catchment encompass an area of approximately 20 km2, accounting for 11% of the total land area. Singh et al. (2016) in their assessment of the WQI of Nambul River observed that the DO range of the river was 4.5–5.6 mg/l. Also majority of the rivers such as Nambol, Awang Khujairok, and Thongjairok discharging into the western sub-catchment are found to have DO concentrations of lower than 4 mg/l. The western sub-catchment has a composition of mostly agriculture (36%) and water bodies (26%). A major influence of the land use systems on the water quality was also observed by Khwairakpam et al. (2019).
The EC value of the study area was observed to be 122.5 ± 16.01 μs/cm (93–152). The EC of the water indicates the presence of biogenic and abiogenic impurities in the water (Upadhyay et al. 2012). A similar observation was also reported by Mayanglambam & Neelam (2022) in their physiochemistry study of Loktak Lake. The lowest value of EC was 93 μs/cm observed at Thinungei (F5) near the Sendra resort, as also reported by Das Kangabam et al. (2017) for the post-monsoon season. The presence of Phumdis may have impacted the lower value of EC in this region. The highest concentration of EC (155 μs/cm) was recorded at the Khoijuman Khunou (A1) area where the Nambul River and other rivers of the western sub-catchment enter the Loktak Lake. Agriculture and other anthropogenic activities carry many dissolved inorganic materials in the lake which may have enhanced the EC concentration in Khoijuman Khunou sampling area. TDS is a quantitative assessment of all the dissolved particles in water, including both organic and inorganic constituents (Sharma & Tiwari 2018). Also, the TDS concentration in the study area was observed to be 81.25 ±9.82 mg/l (60.45–102.05) for the post-monsoon season. Since most aquatic ecosystems with combined aquatic life can tolerate TDS levels of 1,000 mg/l, the TDS ranges of the study are almost similar to findings by Mayanglambam & Neelam (2022), indicating that the water is safe for aquaculture (Boyd 2019). The lowest value of TDS was 60.45 mg/l which was observed at Thinungei (F5) and the highest value of TDS was 102.05 mg/l which was observed at Khoijuman Khunou (A6). A positive correlation was also observed between the lowest and highest values of EC and TDS for the study area. A similar positive correlation was also found between the TDS and EC (r = 0.906, p < 0.01) by Sharma & Tiwari (2018) for Nachiketa Tal, Garhwal Himalaya. The EC and TDS values observed in the post-monsoon season for the study area are well below the acceptable limits of the Bureau of Indian Standards (BIS). The range of DO in this study is 4.57 ± 0.35 mg/l (3.64–5.50), which is suitable for maintaining an aquatic ecosystem in good health. According to the Central Pollution Control Board (CPCB) and BIS guidelines, a DO range of 4–6 mg/l is appropriate for aquatic life survival (Bhuyan et al. 2017; Gupta et al. 2017; Gurumayum et al. 2021). But it is important to keep DO levels above 6 mg/l since lower amounts can signal poor water quality, which can harm aquatic life and render the water unsafe to drink (Das Kangabam et al. 2017).
Regression models
List of linear models for predicting (A) DO concentration, (B) EC concentration, and (C) TDS concentration
Model No. . | Equations . | R2 . |
---|---|---|
(A) DO concentration | ||
1 | DO = 17.56 (B1 + B6) + 1.2781 | 0.45 |
2 | DO = 186.26 (B1 × B6) + 2.9316 | 0.47 |
3 | DO = 2.1473 (B5/B4) + 2.4335 | 0.45 |
4 | DO = 2.1907 (B5/B3) + 2.5841 | 0.46 |
5 | DO = −17.601 (B4 − B5) + 4.5808 | 0.44 |
(B) EC concentration | ||
1 | EC = 2,464.4 (B2 − B4) + 147.12 | 0.58 |
2 | EC = 201.66 (B1/B4) − 42.227 | 0.59 |
3 | EC = 302.99 (B2/B4) − 154.88 | 0.60 |
4 | EC = −135.51 (B4/B1) + 289.24 | 0.59 |
5 | EC = −1,601.4 (B4 − B1) + 157.08 | 0.55 |
(C) TDS concentration | ||
1 | TDS = 1,060.5 (B2) − 34.4 | 0.61 |
2 | TDS = 521.76 (B1 + B3) − 37.919 | 0.57 |
3 | TDS = 541.4 (B1 + B4) − 36.021 | 0.56 |
4 | TDS = 4,632.5 (B1 × B3) + 22.018 | 0.60 |
5 | TDS = 910.9 (B1) − 7.5861 | 0.59 |
Model No. . | Equations . | R2 . |
---|---|---|
(A) DO concentration | ||
1 | DO = 17.56 (B1 + B6) + 1.2781 | 0.45 |
2 | DO = 186.26 (B1 × B6) + 2.9316 | 0.47 |
3 | DO = 2.1473 (B5/B4) + 2.4335 | 0.45 |
4 | DO = 2.1907 (B5/B3) + 2.5841 | 0.46 |
5 | DO = −17.601 (B4 − B5) + 4.5808 | 0.44 |
(B) EC concentration | ||
1 | EC = 2,464.4 (B2 − B4) + 147.12 | 0.58 |
2 | EC = 201.66 (B1/B4) − 42.227 | 0.59 |
3 | EC = 302.99 (B2/B4) − 154.88 | 0.60 |
4 | EC = −135.51 (B4/B1) + 289.24 | 0.59 |
5 | EC = −1,601.4 (B4 − B1) + 157.08 | 0.55 |
(C) TDS concentration | ||
1 | TDS = 1,060.5 (B2) − 34.4 | 0.61 |
2 | TDS = 521.76 (B1 + B3) − 37.919 | 0.57 |
3 | TDS = 541.4 (B1 + B4) − 36.021 | 0.56 |
4 | TDS = 4,632.5 (B1 × B3) + 22.018 | 0.60 |
5 | TDS = 910.9 (B1) − 7.5861 | 0.59 |
Bold value indicates the selected model.
Scatter plot of the relationship between reflectance and in situ value, respectively.
Scatter plot of the relationship between reflectance and in situ value, respectively.
WQPs (DO, EC, and TDS) prediction model
The R2 value was employed as the basis for selecting the regression models that were used to establish the relationship between WQPs and spectral reflectance (Hossain et al. 2021). According to the algorithmic regression model (Table 4(A)), numerous combinations of diverse bands (B1 + B6, B1 × B6, B5/B4, B5/B3, and B4 − B5) have demonstrated a moderate correlation between surface reflectance value and in situ measurements of DO parameters, with R2 values ranging from 0.43 to 0.47. Out of the five models developed for DO, the multiplication band model (B1 × B6) with a better R2 (0.47) value was selected. Most of the sampling points are taken from the northern and western catchment of Loktak Lake, where most of the effluent discharge has been reported. This could have been attributed to the moderate correlation. Also, DO is optically inactive, and no sensor has been established to perform consistent accurate measurements of remotely sensed DO (Gholizadeh et al. 2016). The moderate R2 value is expected to lower the DO model's accuracy. However, a study by El Ouali et al. (2021) also observed a nearly similar R2 value of 0.56 for the DO regression model in Hassan Addakhil Dam, Morroco.
Similarly for the EC parameter, the best model was chosen with the criteria of highest R2 among the six developed models (Table 4(B)). The different band combination has shown an R2 value of B2 − B4 (0.58), B2/B4 (0.60), B1/B4 (0.59), B4/B1 (0.59), and B4 − B1 (0.55), respectively, out of the five models of EC. Among these B2/B4 which exhibited a high correlation (R2 > 0.5) and therefore was selected to predict the WQPs. A similar nearly R2 value (0.69) was also observed by González-Márquez et al. (2018). The difference in EC within an aqueous environment is connected to changes in the concentration of TDS present within the water (González-Márquez et al. 2018). The rise in the salinity of the water leads to alterations in the quantity of radiation that are reflected within the visible and infrared range bands (Khattab & Merkel 2014; Theologou et al. 2015).
The regression models (Table 4(C)) based on (Band 2, 0.61), (Band 1, 0.59) and different band combinations (B1 + B3, 0.57), (B1 + B4, 0.56), and (B1 × B3, 0.60) of TDS models best explained the relationship between the surface reflectance and the measured TDS values. Band 2 which shows a strong correlation between the in situ WQP and surface reflectance value was selected for the prediction of the WQP. Coastal/Aerosol (B1), Blue (B2), Green (B3), and Red (B4) bands used for this TDS model were reported to have produced good results (R2 > 0.6) in similar previous studies (El Saadi et al. 2014; Mushtaq & Nee Lala 2017; Abdelmalik 2018; Japitana & Burce 2019).
Validation of models
The regression and testing analysis results of the selected model for each parameter
Water quality parameter . | Equation . | Regression . | Testing . | ||||
---|---|---|---|---|---|---|---|
R2 . | RMSE . | SEE . | R2 . | RMSE . | SEE . | ||
DO | y = 186.26 (B1*B6) + 2.9316 | 0.47 | 0.23 | 0.23 | 0.45 | 0.49 | 0.47 |
EC | y = 302.99 (B2/B4) − 154.88 | 0.60 | 9.44 | 9.69 | 0.56 | 17.56 | 11.06 |
TDS | y = 1,060.5 (B2) − 34.4 | 0.61 | 5.95 | 6.09 | 0.54 | 6.98 | 7.18 |
Water quality parameter . | Equation . | Regression . | Testing . | ||||
---|---|---|---|---|---|---|---|
R2 . | RMSE . | SEE . | R2 . | RMSE . | SEE . | ||
DO | y = 186.26 (B1*B6) + 2.9316 | 0.47 | 0.23 | 0.23 | 0.45 | 0.49 | 0.47 |
EC | y = 302.99 (B2/B4) − 154.88 | 0.60 | 9.44 | 9.69 | 0.56 | 17.56 | 11.06 |
TDS | y = 1,060.5 (B2) − 34.4 | 0.61 | 5.95 | 6.09 | 0.54 | 6.98 | 7.18 |
Scatter plot of the relationship between the predicted and observed values.
In testing, the RMSE (0.49) and SEE (0.47) values computed for DO parameter after applying the tested and developed equations for the post-monsoon season show a moderate correlation between the observed and predicted values (R2 = 0.45). The results of the model show that it can be used for the estimation of WQP. For EC, the RMSE and SEE values for the post-monsoon season were 17.56 and 11.06, respectively, with an R2 value of 0.56. The value obtained was lesser than the results obtained by González-Márquez et al. (2018) with an R2 value of 0.69. The model shows a moderate correlation between the predicted and the observed values. Similarly, the validation results of the TDS show RMSE, SEE, and R2 values of 6.98, 7.18, and 0.54, respectively, for post-monsoon testing data. Similar studies by Ferdous et al. (2019) obtained an R2 value of 0.73 (slightly higher) from the blue band (Band 2) of Landsat 8 OLI (Level 2). In the Testing phase of the model, there was a slight reduction in the R2 value (Table 5) for the three parameters (DO, EC, and TDS). Along with it, the RMSE and SEE values were also observed to decline in the testing phase. This reduction may be improved by increasing the number of samples as water quality estimation through satellite imagery involves the complex relationship between the reflectance value and the in situ measurements. The performance of these models in the testing phase may be affected and influenced by the quality and quantity of accessible data, subsequently impacting the precision of statistical metrics. The models developed to forecast water quality displayed a decent linear connection for DO, signifying the necessity for enhancing the model with a more extensive dataset. The EC and TDS models exhibited a better linear relationship, implying its suitability for estimating the parameters. Overall, enhancing the accuracy of the regression models mandates additional sampling points, hinting that the existing data might not suffice for a thorough water quality assessment throughout the entirety of Loktak Lake. The study also suggests the refinement of the model with the use of non-linear models and neural network models to monitor and estimate the water quality in Loktak Lake to protect the ecosystems effectively.
Spatial water quality map
Spatial water quality map of DO (mg/l), EC (μs/cm), and TDS (mg/l) concentration.
Spatial water quality map of DO (mg/l), EC (μs/cm), and TDS (mg/l) concentration.
Roy & Majumder (2019) also reported a similar lower range of EC values in their observations of the Loktak Lake. They also found a lower EC value in the winter period in their studies. Since the data were gathered in the post-monsoon, dilution from precipitation from the monsoon may also have contributed to an overall lower average EC during the post-monsoon (Laishram et al. 2022). The higher range of EC value (131–146 μS/cm) was seen distributed more on the eastern side, with some peripheral distribution towards the northern and southern sides of the lake.
A relatively lower value of EC is also reported in similar studies. The southwest side of the lake was observed with a lower range of TDS values. The occurrence of phumdi in the central and southern parts of the lake may have contributed to the lower values of TDS in these regions. The spatial distribution shows a gradual increase in the concentration of TDS value from the northern to the southern shoreside of the lake area. The relatively higher TDS values seen on the eastern side of the lake may be due to the agricultural activities from the catchment of the lake. Significant relations between EC and TDS were observed from the spatial distribution map.
CONCLUSIONS
The study was conducted on the Loktak Lake of Manipur to present the relation between the satellite image of Landsat 9 and the measured water quality data of the lake. Regression models generated from the imagery can be useful tools for predicting the WQPs. The calculated parameters from the Landsat 9 reflectance bands can predict the parameters from the linear regression model generated. DO exhibits a moderate correlation, while EC and TDS parameters exhibited a good correlation between the field measurement data and the Landsat 9 imagery. In addition, more study including the seasonal fluctuations on the levels of radiance and/or reflectance is suggested to carry out. More sampling points and refinement of the model with the use of non-linear and neural network models to monitor and estimate the water quality in Loktak Lake for protecting the ecosystems effectively may be incorporated which is crucial to enhance the accuracy and precision of making absolute predictions. The complex nature of the ecological behaviour of a wetland system may be further studied using advanced technology to understand the dynamics of complex relationships. The study also demonstrates the use of the kriging spatial technique for generating a spatial distribution map of the WQPs. The generated spatial map can be used for water resource management including aquaculture which is prevalent in the Loktak Lake.
DATA AVAILABILITY STATEMENT
Data cannot be made publicly available; readers should contact the corresponding author for details.
CONFLICT OF INTEREST
The authors declare there is no conflict.