Long-term spatial-temporal monitoring of eutrophication in Lake Burdur using remote sensing data

Freshwater lakes serve as critical water resources for many areas of the world and have various purposes. However, eutrophication threatens the sustainability of these lakes. Eutrophication trends in lakes are more pronounced in recent decades under the influence of climate change due to increasing anthropogenic pollutant inputs such as ammonium, nitrate, organic nitrogen species, and orthophosphates (Smith 2003; Le et al. 2010). Although eutrophication is a natural event of water body enrichment, it can cause significant environmental nuisances and be a problem for water authorities. Long-term datasets of water quality variables causing lake eutrophication prove useful in developing effective water management strategies (Cao et al. 2022).

Traditionally, the water quality of a lake is determined with in-situ measurements and laboratory analysis. This approach leads to more or less accurate ground-based measurements that reflect the actual water quality of the lake. However, it requires the allocation of time, monetary and human resources. Furthermore, in-situ measurements cannot provide a large-scale spatial and temporal overview of the water quality (Gao et al. 2015; Li et al. 2017). Lakes in potentially stressed watersheds are frequently ungauged, meaning they are not monitored. Additionally, they may only have a limited number of site measurements that do not provide enough information to assess their level of eutrophication or water quality. In other cases, water quality data may be classified information that is not shared with the public. On the other hand, remote sensing of lakes with satellites offers some possibilities that can be used to estimate water quality variables that are related to eutrophication. It has the benefits of working with a much broader spatial and temporal coverage and being cost-effective and less time-intensive compared with ground measurements. For example, Landsat images have a spatial resolution of 30 m and a revisit cycle of 16 days, while meeting the requirements of image quality and spectral range. The launch of the Landsat-5 Thematic Mapper (TM, 1984–2012), the Landsat-7 Enhanced Thematic Mapper Plus (ETM + , active since 1999), and the Landsat-8 Operational Land Imager (OLI, active since 2013) provided nearly 40 years of satellite imagery, enabling the extraction of historical water quality information (He et al. 2021). Landsat images can be used to obtain Chl-a concentrations (Cardall et al. 2021; Cao et al. 2022), nutrient concentrations (Li et al. 2017), transparency or turbidity (Butt & Nazeer 2015; Liu et al. 2019), and eutrophication status (Chen et al. 2020) in lakes. Among these variables, Chl-a is related to phytoplankton biomass and indicates the lake's trophic status. The trophic state of water bodies can be quantitatively evaluated with the trophic state index (TSI), which was first proposed by Carlson (1977). Several studies for lakes worldwide have shown the potential for TSI to be an effective indicator of lake water quality (e.g., Iwashita et al. 2004; Ha et al. 2017; Hu et al. 2021). It provides a practical lake classification system concerning its trophic state using single or multiple water quality parameters, including either Secchi disk depth, Chl-a concentration, total nitrogen, total phosphorus or as a function of all of them together.

Remote sensing imagery data is widely used to effectively monitor the spatial and temporal dynamics of eutrophication over large areas. In this study, the long-term (1984–2021) and high spatial resolution algae concentrations of Lake Burdur were obtained based on the application of an empirical model on the reflectance time series of Landsat imagery using Google Earth Engine (GEE) (Gorelick et al. 2017) as a cloud computing interface. Lake Burdur is under significant environmental stress and the lake's shrinking rate is alarming (Elçi 2019). TSI was also calculated based on Landsat-derived Chl-a concentrations. Inter-annual and seasonal patterns of the TSI and Chl-a concentrations were identified.

Organic carbon and nutrient influx originating from diffuse sources, and numerous discharges of industrial and urban wastewater threaten the lake's ecosystem (GDWM 2020). Based on a water quality modeling study by GDWM (2020), the average hydraulic retention time is 33.8 years. This modeling study involved the setup and execution of a basin-scale hydrological model that runs in conjunction with a receiving medium water quality model. In this case, the receiving media was defined as all flowing streams and all lakes in the Burdur Lake watershed. Several scenarios representing actions to improve water quality and conserve water quantity in the basin were simulated. As a result of the modeling study, changes in lake water volumes, nutrient, priority pollutant, and specific pollutant concentrations were determined, which consequently aided decision-making.

This study is executed in three steps: (1) remote sensing data retrieval, (2) remote sensing data pre-processing, and (3) determining spatiotemporal distributions of Chl-a concentrations using the empirical model with Landsat images from 1984 to 2021. The starting date of the study period is defined by the beginning of Landsat data availability, and the ending date marks the end of the calendar year before the time when the study was conducted. Surface reflectance images of Lake Burdur were obtained from the Landsat-5 TM, Landsat-7 ETM + , and Landsat 8 OLI sensors. Landsat data have the longest records, and the highest consistency spectrum among the current satellites, which makes it suitable for the remote analysis of any surface water quality variable. With the launch of the Landsat-5 Thematic Mapper (TM) in 1984, followed by the Landsat-7 Enhanced Thematic Mapper Plus (ETM+) in 1999 and the Landsat-8 OLI in 2013, nearly 40 years of remote sensing data are available for extracting historical water quality information (Wulder et al. 2019; He et al. 2021). Although Landsat 5 was decommissioned on 5 June 2013, after 29 years in space and was largely superseded by Landsat 7 and 8 due to equipment failures near the end of their mission, Landsat 7 and Landsat 8 continue to operate. Landsat 5 and 7 have the same band definitions and numbered wavelength ranges in the electromagnetic spectrum, while Landsat 8 has different band definitions. It is important to note that the band names were changed while the Landsat 8 satellite was used. Simultaneously, some cloud computing systems were set up to handle petabytes of geospatial data. GEE is an example of a cloud computing platform that can efficiently provide access, manage, and process big data due to the capabilities of high-performance computing. All data were collected and processed on the GEE platform.

After retrieval of all images for the study period, the data were pre-processed. First, images were cropped to fit the image coverage to the study area, reducing data processing loads. Retrieved images represented data from January 1984 until December 2021 at 16-day intervals. A total of 1,275 images were accessed, although not all images are usable due to cloud, ice, and other issues. GEE commands were used to calculate the Modified Normalized Difference Water Index (MNDWI) (Xu 2006) to mask off the pixels that do not represent water body pixels. Also, pixel quality masking was performed to exclude pixels with issues that affect data accuracy and/or are affected by cloud cover or shadowing. The pixel quality is represented as an integer number stored in a separate band of the Landsat data. After masking, the remaining pixels represent the extent of the visible water body at the time of the image.

In the next step, algae concentrations were computed. Algae concentrations are typically represented as Chl-a concentrations in water. Surface reflectance data from satellite images are used to estimate Chl-a concentrations by using a previously developed empirical model that relates surface reflectance to the unique spectral signature of Chl-a (Hansen & Williams 2018). The empirical model equation developed by Hansen & Williams (2018) was implemented in this study to calculate the Chl-a concentration in each image pixel over the entire 38-year dataset in GEE. This equation represents the empirical relationship between Chl-a and various band surface reflectances from Landsat image data (Equations (1) and (2)). Chl-a spatial distribution maps and time series charts are created with R studio 3.5.2.

After applying these equations to each pixel of the bands in an image, Chl-a concentration time series in mg/m³ are calculated for each pixel. A formal outlier analysis with the IQR (interquartile range) method was performed to identify unusually high Chl-a values that stem from bad-quality image pixels. Consequently, outlier values comprising 16% of all data are excluded from the analysis, which corresponds to roughly a total of six years of data over the 38-year observation period. Finally, by appending the time series obtained from Equations (1) and (2), spatially averaged annual, seasonal, and monthly Chl-a concentrations were calculated to obtain algae concentration time series from 1984 to 2021. Also, the distribution of algae concentrations over the lake is evaluated by computing a median concentration for each month.

Figure 2(b) presents the calculated TSI values changing over time. The highest TSI (82.40) was calculated for the year 2010 and the lowest TSI (70.55) for 1992. The TSI value reached 82.40 in 2010, suggesting excessive growth of biomass leading to an algae bloom problem. Overall, the TSI values indicate that Lake Burdur's hypertrophic status from 1984 to 2021 did not change.

There are several possible explanations for the inter-annual variations of Chl-a concentration in Lake Burdur. The lake receives a moderate amount of water from both rainfall and intermittent flowing streams. Since the lake is in a hydrologically closed basin, therefore having no stream outlet, water loss occurs only through evaporation and its hydraulic connection with karstic aquifers. Changes in the difference between precipitation and lake evaporation, as well as a decrease in inflowing stream discharge because of upstream constructions of dams, resulted in a declining lake level (Girgin et al. 2004). Furthermore, increasing water demand for agriculture in the lake catchment exacerbates the problem. Davraz et al. (2019) conducted a study to evaluate climate and human effects on the hydrology and water quality of Lake Burdur. According to this study, the lake level has been in a state of continuous decline, with a steady shrinking of the lake area. According to this study, 1995 was extremely dry, while moderately dry years were 1986, 1989, 1990, 1992, 1999, and 2008. Conversely, a soaked year was observed in 2003, while moderately wet years were recorded in 1998, 2001, 2009, and 2015. It is worth noting here that the highest Chl-a concentration derived with the remote sensing approach coincides with the aforementioned dry years. It can be concluded from this study that in essence, factors related to the hydrological budget of the lake can directly affect inter-annual variations in Chl-a concentrations and TSI.

The variability of TSI can be caused by several factors. For example, Girgin et al. (2004) found that surface water temperature and salinity were crucial factors affecting the stratification in the lake, hence affecting the hydrodynamics and consequently the distribution and abundance of chlorophyll. Analysis of hydrologic, climatic, and human activity in previous studies suggested that anthropogenic factors were more influential in lake-level changes. In particular, water surpluses originating from high rainfall events in the winter were stored in upstream dams and ponds rather than in Lake Burdur, which led to low water levels and lower TSI values in this season.

Figure 5 also shows that the spatial distribution patterns of temporally averaged Chl-a concentration are similar for all months of the year. Although the Chl-a levels showed absolute differences between different months, the spatial variations displayed a similar pattern. A relatively stable and homogeneous distribution of the highest and lowest values is observed over the lake. Median Chl-a values were highest during October and November, with Chl-a concentrations in the 450–500 mg/m³ range. Concentrations lower than 200 mg/m³ could be observed during summer. Overall, it appears that algae blooms in Lake Burdur are typically occurring from November to February and early spring (March) rather than during the late spring (April and May) and summer (June to August), and early fall (September).

Due to the limited availability of in-situ measurement data, it was impossible to validate our results entirely. Nevertheless, the results were compared with the limited amount of measured Chl-a concentration data (Table 1) that was obtained for the years 1997 (Girgin et al. 2004) and 2018 (GDWM 2020). All measured Chl-a concentrations reflect Chl-a content in water samples obtained from the water surface of the lake. Error bars for the measured concentration data were not available. Based on the field sampling data, Chl-a concentrations varied between 1.17 mg/m³ (in October) and 14.74 mg/m³ (in April) during different months of the 1997 seasons. According to the study, the samples were taken from the surface in the middle of the lake; however, exact sampling coordinates were unknown. The remote sensing monitoring data was consistent with the measured data in April, but significant differences were observed in July and October 1997. Measurements from the GDWM (2020) study provided more comparable results with our Landsat-derived Chl-a concentrations. In 2020, Chl-a concentrations were measured between 291 mg/m³ (in April) and 750 mg/m³ (in October). These were in line with spatially averaged estimated concentrations for January, April, and October. However, our results suggested lower Chl-a concentrations for July, unlike the estimated results for the same month in 1997.

Furthermore, it was evident that Chl-a concentrations were lower in 1997. Assuming that measured concentrations are accurate, results from Landsat data overestimated Chl-a levels at the lake surface. It should be noted that considerable uncertainty is included in the in-situ data because of timing, location, and method of water sampling. Measurements provided by GDWM (2020) show that Chl-a concentrations were significantly higher in 2018, which was partly confirmed by estimations from Landsat data. It should be noted, however, that estimations given in Table 1 are spatially averaged values and that concentrations over the lake area can vary considerably, as it is shown in Figure 5. A point-by-point comparison of measured and calculated Chl-a concentrations was not possible for two reasons: first, sampling coordinates of the 1997 study were missing. Second, results obtained from Landsat data were available every 16 days, and in some instances, bad-quality pixel clusters that were masked off coincided with the sampling locations.

The novelty of using remote sensing data to determine the long-term trophic state values of lakes is the capability to continuously track and monitor changes over large areas of the lake in time, thereby providing a comprehensive view of the spatial and temporal patterns of eutrophication. As a result, researchers can detect trends and patterns in the lake's trophic state by scrutinizing and contrasting remote sensing data time series pertaining to different time frames. These patterns may consist of surges and downturns in algal blooms as a consequence of alterations in the lake's nutrient concentrations. In this study, long-term satellite remote sensing-based Chl-a concentrations and related TSI index values were obtained for Lake Burdur using empirical equations representing the relationship between optical reflectance and Chl-a concentrations. A procedure was developed using GEE that retrieves raw data, applies the empirical equations to convert information about light reflectances at different wavelengths, and post-process results for interpretation and presentation of the resulting information. Using this procedure, 38 years of reflectance data was processed to determine Chl-a concentrations and explore the spatial-temporal patterns of algae concentrations and the evolution of eutrophication in the lake. The time series of Chl-a concentrations estimated from 1984 to 2021 over Lake Burdur indicated apparent spatial-temporal differences. Spatially, higher Chl-a concentrations were observed in the middle of the lake, possibly associated with factors related to the water depth, and showed a clear decreasing gradient in nearshore regions.

Further studies are needed to investigate the causes of the spatial variability of Chl-a. Anthropogenic organic carbon and nutrient influxes interacting with meteorological factors likely play a key role in the variability of the lake's trophic state. From a seasonal perspective, Chl-a appeared to be higher during the late fall and winter and lower during the summer and spring. Further studies should be conducted to understand the driving factors in particulate climate variables affecting the trophic level by season in the lake. Also, the study can be extended to explore relationships of spatial-temporal trophic state with lake bathymetry, changes in water level, water temperature, point source locations, discharge timings, and the influence of wind and temperature.

The datasets generated in our study provide critical information at a high spatial and temporal resolution, which can be used to identify hotspots of eutrophication and prioritize water management actions. Our research highlights the effectiveness of utilizing multi-Landsat satellite observations in water quality monitoring by expanding the study's temporal and spatial scope. However, due to the lack of an adequate dataset of site measurements, estimated Chl-a concentrations could not be fully validated. Therefore, the estimated concentration values might differ from actual historical concentrations, although the fluctuations and trends are expected to be consistent with actual observations. Conclusively, this non-invasive remote sensing approach can also be applied to other lakes to support strategies for effective watershed management and mitigate problems related to eutrophication.

The data on surface reflectance that was used in this study are openly available in the Earth Engine data catalog at: https://developers.google.com/earth-engine/datasets/catalog/landsat.

The authors declare there is no conflict.