Mesocosm Experiment to Evaluate Relations between Chlorophyll-a Concentration and Water Surface Reflectance in an Anthropogenic Reservoir

Abstract

This paper presents the results of a mesocosm experiment for the evaluation of remote sensing chlorophyll-a (chl-a) concentration estimations in an anthropogenic water reservoir. The chl-a presence in the water causes changes in the water surface reflectance spectrum, especially in the green and red part, but many factors could affect the remote measurements of chl-a content. The in situ mesocosm method of the experiment was used for investigating the spectral reflectance of the inland water surface in a wide range of chl-a concentrations. Eight specially designed measurement boxes were placed into the water. In half of the boxes, the devices to support the development of the submerged water plant were installed. During the experiment, simultaneously, spectral data from the water surface were gathered and physical–chemical analyses of water were carried out. The obtained results confirm the usefulness of the mesocosm experiment for the remote sensing chl-a concentration algorithms being developed. The concentration of dissolved organic carbon was identified as a key factor that interfered with remote chl-a estimations in the analyzed reservoir.

1. Introduction

Chlorophyll-a concentration is a key parameter indicating an undesirable disturbance in inland waters caused by eutrophication processes. Nutrient enrichment affected the development of microscopic plants in water columns (phytoplankton). Chlorophyll-a (chl-a) concentration as an indicator of phytoplankton biomass is a widely used method for estimations of eutrophication in water ecosystems. Using remote sensing for measuring the eutrophication of inland water is very useful for water management which allows one to make the right and quick decisions to limit the negative impact of water reservoir eutrophication, especially with recreational use. Automated systems for phytoplankton monitoring are part of the technical arsenal needed to achieve the Water Framework Directive goals [1]. The ability to use remote sensing for monitoring water quality has been studied since the 1970s [2,3,4]. Based on CHRIS mode 2 and MERIS satellite images and field data, the development of algorithms for chlorophyll concentration monitoring was conducted. These algorithms allow the determination of ecological states of water ecosystems and are used to monitor the ecological quality ratio (EQR) in Spain’s inland water [5]. The application method for monitoring water quality using Landsat imagery on lakes in Oklahoma shows a significant correlation between band combinations and turbidity, but chlorophyll is not significantly correlated with any bands or band combinations during summer [6]. High correlation coefficients (0.944) for chlorophyll using the Landsat 3 band values were generated by Allan et al. 2011 [7] for a lake in New Zealand. These and other studies [8,9,10,11,12] have shown that relationships between remotely sensed surface reflectance and water parameters in inland waters are often unique to a particular geographic region and have less applicability in another area. Phytoplankton is made up of organisms belonging to a number of systematic groups of species, with variable size, shape, and environmental requirements in relation to light, temperature, and water mobility. The gradient of altitude and temperature gradient of nutrients and anthropogenic activity has the greatest impact on phytoplankton variability [13]. Radiance or reflectance signals leaving the water ecosystem surface contain phytoplankton pigment information that can be related to community structure and size classes [14,15]. Most of the remote methods of monitoring have relied on the surface water spectrum to retrieve chlorophyll concentrations in phytoplankton.

Therefore, spectral-based methods for chlorophyll-a estimation have been mostly developed for natural water ecosystems [16,17,18,19]. Due to a disturbance in bio-physical–chemical interactions and specific morphological conditions (e.g., shallow depth, limited share of banks and bottoms suitable for vegetation development), anthropogenic reservoirs are more subjected to eutrophication [20,21]. In the case of anthropogenic reservoirs, the diversity of phytoplankton species composition depends on the availability of genetic information and the age of the reservoir, as well as the type and intensity of anthropogenic pressure. These reservoirs are most often characterized by a lower diversity of species composition and greater susceptibility to the dominance of a single species. It often causes human risk and a loss of biodiversity. In addition, the negative effects of nutrient enrichment like harmful algal blooms could be enhanced by climate change [22]. To decrease health risks and increase the effectiveness of rehabilitation work, the continuous monitoring of anthropogenic water ecosystems has to be implemented. Due to the complex correlation between water quality parameters and water optical properties, remote sensing for inland water quality monitoring is still a grand challenge [23]. An insufficient number of replicates and the uniqueness of every single reservoir mostly represent limitations for developing and implementing inland water remote sensing monitoring methods. The duration of the experiment is another important factor affecting the inconvenience of applying this method. Mesocosm approaches could solve most of these limitations in developing algorithms for remote in situ chlorophyll-a concentration estimations in the anthropogenic reservoir. It is an intermediate option between laboratories and in situ studies and gives the ability to control environmental variables jointly with the replicability [24,25]. In these experiments, reflection measurements are taken directly from above the water surface using handheld spectrometers [26]. The significant negative relations between periphyton growth on a mesocosm wall and phytoplankton abundance observed by Chen et al. (1997) [27] gives opportunities to observe water ecosystems in a wider range of chl-a concentrations in a water column. Due to this process, a shorter time is needed to analyze the optical properties of a water surface with a wide range of environmental factors.

The study’s objectives are to verify if a mesocosm experiment could support developing algorithms for remote chlorophyll-a concentration estimations in a selected anthropogenic water ecosystem and help to identify the water physio-chemical parameters that determine water surface spectral reflectance and affect remote chlorophyll-a concentration retrieval.

2. Materials and Methods

2.1. Study Area

The research was carried out in a pond located in the central part of the Silesian Voivodship (Bytom city). A reservoir appeared as a result of open mining activity in the 20th century. Currently, it is located in the southeastern part of 43 ha urban green spaces (Franciszek Kachl Park). The direct catchment (approx. 6.5 ha) covers mostly low vegetation, with a small share of impervious elements (pedestrian and bicycle paths). The pond reserves also rainwater from the sewage system that dewaters about 5 ha of the urban area. The maximum depth of the reservoir does not exceed 2 m, and the average depth is about 1.1 m. The area of the water reservoir is 1.23 ha. It was revitalized between 2012 and 2015. The contaminated sediments were removed and the bottom was covered with an isolating layer to decrease water infiltration. The reservoir is a place of rest and recreation for residents, it is also an element increasing the biodiversity of urban spaces. The function of the pond is limited by water quality. Low water transparency (<1 m) during the growing season is caused by the high biomass of phytoplankton in the water column. High water turbidity causes the absence of submerged vegetation. The mean physio-chemical parameters of reservoir waters calculated, based on data gathered during one vegetation season, are presented in

Table 1. Physio-chemical parameters of reservoir waters.

Parameater	Mean Value	Standard Deviation
Chlorophill a [µg/L]	35.27	8.02
Secchi disk [m]	0.76	0.09
Chemical oxygen demand [mg/L]	28	2.53
Total organic carbon, TOC [mg/L]	7.44	0.90
Dissolved organic carbon, DOC [mg/L]	3.82	1.34
Total phosphors [mg/L]	0.026	0.004
Total nitrogen [mg/L]	0.79	0.29
Conductivity [μS/cm]	596.50	9.65
pH	8.34	0.19
Oxygen [mg/L]	8.57	2.26

. The method of physio-chemical water parameter determination is indicated in Section 2.2.

2.2. Description of the Experiment

An in situ mesocosm approach was used. The study point was located within the littoral zone at a distance of over 60 m from the reservoir’s inlet and outlet. Six boxes made of wooden construction and sides of the boxes covered with impermeable material (greenhouse foil) were placed in the pond to limit the water exchange (Figure 1 and Figure 2) during the experiment. The size of the boxes was the following: length: 1 m; width: 1 m; and height: 1 m.

In half of the boxes, the device for supporting submerged water plant development was placed. Myriophyllum spicatum was used as a supporting species. The device was made of plastic mesh with a circular buoy inside (with a diameter of 20 cm), anchored to the bottom by a link and weight. The mesh with vegetation inside was kept near the water surface (Figure 2). No additional experimental treatments on the boxes were carried out. The significant improvement in water quality in boxes without devices for supporting submerged plants is a well-known process resulting from periphyton growth on a mesocosm wall (Chen et al. 1997) [27]. At all monitoring points, the bottom was covered by dark gravel and stones without submerged vegetation.

The spectral reflectance from each box and the reservoir was collected seven times. Measurement intervals were maintained approximately on a weekly basis. The measurement device utilized to collect radiance from the water surface was the UV/NIR spectrometer connected with optic fiber and an optical tube with a lens. The UV/NIR spectrometer with diffraction grating in a Czerny–Turner configuration was manufactured by OPTEL Opole. The aperture, chosen in this way, allows aberration reduction while maintaining high incident light energy on the grating. The monochromators are equipped with slits that are infinitely adjustable in width from 0 to 3 mm, with the best specular resolution of 0.01 mm. During the experiment, the spectral resolution of the measuring device was 1.8 nm. The correct operation of the measuring device was demonstrated by its calibration against reference light sources (standard broadband illuminators). The optical tube placement on the box, angle view, and field of view was established to reduce the influence of the mesocosm walls on the collected reflection spectrum (Figure 3). Each time, the measurements were made around noon in windless weather.

Before the reflectance curves were collected from each box, sunlight spectral reflectance was measured to ensure stable measurement conditions. The reflectance standard—a white panel was used in this case (Figure 4).

Water from the surface layer was collected in a dark glass bottle immediately after spectral measurements from each study site (n = 49). The samples were protected from extensive heat and transported to a laboratory. The chl-a concentration was measured using spectrophotometry according to the procedure of PN-ISO 10260:2002 [28]. The water samples were filtered on glass fiber with a pore diameter of 1.0 µm and then extracted using the ethanol method. The additional physicochemical water parameters that could have had significance on surface water spectral reflectance [29] were also determined according to standard laboratory methods:

• Total phosphorus—by optical emission spectrometry according to the standard PN ISO 11885:2009 [30];
• Total nitrogen—by high-temperature combustion with IR detection according to the standard PN-EN 12260:2004 [31];
• Chemical oxygen demand—spectrophotometric method according to the standard PN-ISO 15705:2005 [32];
• Conductivity—conductometric method according to the standard PN-EN 27888-1999 [33];
• pH—potentiometric method according to the standard PN-EN ISO 10523:2021 [34];
• Oxygen—electrochemical method according to the standard PN-EN ISO 5814:2013-04 [35];
• Secchi depth—measuring the optical properties of the water column using a disc in accordance with the standard EN ISO/IEC 17025 [36];
• Turbidity (NTU)—using nephelometers according to standards (PN-EN ISO 7027-1:2016-09) [37];
• Total organic carbon (TOC) and dissolved organic carbon (DOC)—by oxidation according to PN-EN 1484:1999 [38];
• Particulate organic matter (POM) by filtration using a cellulose filter and combustion at 600 °C (PN-EN 872:2007 and PN-C-04559-02:1972) [39,40].

Only water quality parameters that were implemented in the current practice of the Central Mining Institute accredited laboratory were used in the research.

2.3. Methodology of Data Analyzing

One of the most common approaches to the remote sensing of chlorophyll-a concentration is the collection of the spectral reflectance value within wavelengths where a reflection of the chl-a occurs. The ratios of the two reflectance values were calculated for the known concentration of chl-a. The correlation between the ratios, chl-a concentration, and other physical–chemical parameters is determined [41,42].

Two ranges of wavelength were selected as characteristics for chlorophyll-a concentration changes. One was in the blue and green part of the reflectance spectrum and included the reflectance ratio at 450 and 531 nm (R450/531). The second included the reflectance ratio in the red part of the optical spectrum at 695 and 674 nm (R695/674). The normal distribution of data was confirmed by the Shapiro–Wilk test. The Pearson correlation between the ratios, chlorophyll-a concentration, and other physical–chemical parameters was determined.

Firstly, the analysis was conducted for each box separately. In the next step, all measurement data were treated as one measurement series and the correlation was tested for different concentration ranges.

3. Results

Applied experimental treatments cause significant changes in phytoplankton density in the water (Figure 5 and Figure 6). Especially in 4 and 5 (boxes without submerged vegetation), lower chl-a concentrations were observed. A wider range of chlorophyll was observed in box 5. On the 57th day of the experiment, chl-a concentration in boxes where devices for the support growth of submerged vegetation were introduced (boxes 2, 3, and 6) was maintained between 29.8 µg/L and 39.5 µg/L, while in the boxes without additional maintenance, the concentration was between 2 µg/L and 8.7 µg/L (boxes 1, 4, and 5). At this phase of the experiment, the concentration of chl-a in the pond reached its highest value (43.6 µg/L).

The analysis of the reflectance curves in box 1 where chlorophyll ranged from 8.7 to 37.2 mg/m³ is presented in Figure 5. The higher the content of chlorophyll-a in the water, the higher relative intensity in green observed. In lower chl-a concentrations, higher differences in the red range of the VIS spectrum are observed (Figure 7).

For each box, Pearson correlations were calculated, taking into account the spectral indexes and analyzed physical–chemical parameters (

Table 2. Pearson correlations between spectral indexes (R706/674 and R450/531) and water physical–chemical parameters for each box.

Name of Data	Index	Range of Chl-a Concentration [µg/L]	Correlation
			Chl-a	Range of Turbity [NTU]	Turbidity [NTU]	Range of POM [mg/L]	POM [mg/L]	Range of TOC [mg/L]	TOC [mg/L]	Range of DOC [mg/L]	DOC [mg/L]
Pond, n = 6	R706/674	24.8–43.6	−0.59	7.5–37	0.71	7–13	−0.34	6.1–8.8	−0.91	<0.5–5.07	−0.18
	R450/531		−0.79	7.5–37	0.46	7–13	−0.47	6.1–8.8	−0.68		0.02
Box 1, n = 6	R706/674	8.7–37.2	0.23	2.3–31	0.73	2.8–9.4	0.53	6.7–8.6	−0.71	1.83–8.28	−0.66
	R450/531		−0.01	2.3–31	0.44	2.8–9.4	0.35	6.7–8.6	−0.44		−0.42
Box 2, n = 6	R706/674	24.3–47	−0.52	6.5–36	0.76	6.8–16	0.13	6.3–9	−0.89	<0.5–6.56	−0.41
	R450/531		−0.90	6.5–36	0.31	6.8–16	−0.48	6.3–9	−0.63		0.13
Box 3, n = 6	R706/674	24.1–47	−0.49	5.7–33	0.75	6.6–16	0.56	7.4–9.6	−0.81	<0.5–6.76	−0.77
	R450/531		−0.76	5.7–33	0.43	6.6–16	0.21	7.4–9.6	−0.80		−0.37
Box 4, n = 6	R706/674	1.8–26.9	0.97	0.72–23	0.88	4–8.4	0.89	6.7–7.8	−0.48	3.84–11.24	−0.96
	R450/531		0.66	0.72–23	0.55	4–8.4	0.86	6.7–7.8	−0.08		−0.85
Box 5, n = 6	R706/674	5.8–28.4	0.43	2–27	0.78	4–7.8	0.61	6.4–8.4	−0.86	3.84–12.86	−0.73
	R450/531		0.30	2–27	0.47	4–7.8	0.59	6.4–8.4	−0.55		−0.61
Box 6, n = 6	R706/674	18.3–42.4	−0.68	5–26	0.83	5.8–9	−0.70	6.5–8.2	−0.85	3.14–6.13	−0.07
	R450/531		−0.84	5–26	0.62	5.8–9	−0.83	6.5–8.2	−0.57		0.39

The bold correlations were considered significant at a probability level of p < 0.05.

). Very strong negative and significant correlations were found between chl-a and the R 450/531 index in box 2 where the variability in chl-a ranged between 24.3 and 47 µg/L (−0.90) and in box 6, with chl-a concentrations from 18.3 to 42.4 µg/L (−0.84). Strong negative and significant correlations between these factors were also found in the pond (−0.79; 24.8 > chl-a < 43.6) in box 3 (−0.76; 24.3 > chl-a < 47). An almost perfect linear relationship between the R706/674 index and chl-a shows data from box 4, where chlorophyll-a concentration is maintained between 1.8 and 26.9 µg/L (Table 2). In the next step, the measurement data from each box were grouped into five subseries based on different chlorophyll-a concentration ranges (1–5). A strong negative and significant correlation between R450/531 and chl-a concentration was found for data range 2 and range 4. For R706/674, strong positive correlations and significant dependencies were accounted for also for data range 1 (0.63; 1.8 > chl-a < 21.5) and data range 3 (0.81; 1.8 > chl-a < 26.9). These strong dependencies did not show a correlation analysis of the full range of data (range 5; 1.8 >chl-a < 47) (

Table 3. Pearson correlations between spectral indexes (R706/674 and R450/531) and water physical–chemical parameters for distinguished measurement subseries.

Name of Data	Index	Range of Chl-a Concentration [µg/L]	Correlation
			Chl-a	Range of Turbidity [NTU]	Turbidity [NTU]	Range of POM [mg/L]	POM [mg/L]	Range of TOC [mg/L]	TOC [mg/L]	Range DOC [mg/L]	DOC [mg/L]
Range 1 (from all, n = 14)	R706/674	1.8–21.5	0.63	0.72–25	0.90	1–8.4	0.79	7–8.6	−0.34	3.84–12.36	−0.90
	R450/531		0.40	0.72–25	0.56	1–8.4	0.59	7–8.6	−0.34		−0.63
Range 2 (from all, n = 30)	R706/674	24.1–47	−0.36	4.8–37	0.70	5.4–16	0.15	6.1–9.6	−0.67	<0.5–6.76	−0.43
	R450/531		−0.79	4.8–37	0.38	5.4–16	−0.33	6.1–9.6	−0.64		0.03
Range 3 (from all, n =24)	R706/674	1.8–26.9	0.81	0.72–27	0.88	1–9	0.84	6.5–8.6	−0.67	2.66–12.36	−0.85
	R450/531		0.42	0.72–27	0.51	1–9	0.61	6.5–8.6	−0.34		−0.57
Range 4 (from all, n = 25)	R706/674	28.4–47	−0.17	4.8–37	0.72	5.4–16	0.39	6.1–9.6	−0.62	<0.5–6.76	0.22
	R450/531		−0.70	4.8–37	0.35	5.4–16	−0.14	6.1–9.6	−0.71		0.31
Range 5 (all, n = 44)	R706/674	1.8 −47	0.23	1–16	0.79	19–35	−0.40	6.1–9.6	−0.60	<0.5–12.36	−0.64
	R450/531		−0.47	1–16	0.33	19–35	−0.03	6.1–9.6	−0.58		0.01

The bold correlations were considered significant at a probability level of p < 0.05.

).

In the case where the concentration of chlorophyll-a exceeds 24 µg/L, the R_706/674 index shows a very strong negative and significant correlation with TOC (−0.91 in the pond and 0.89 in box 2). Such a correlation is not observed in lower concentrations of chlorophyll-a in box 4 (−0.48) and data range 1 (−0.34). The correlation between physical–chemical parameters accounted for the full range of data (range 5) shows that TOC is negatively correlated with the NTU and has no correlation with other parameters (POM and chl-a). TOC shows a negative correlation with the R_450/531 index when higher concentrations of chlorophyll–a are observed (−0.71 in range 4 and –0.64 in range 5). A significant correlation between TOC and the R_450/531 index is not observed when the concentration of chlorophyll-a is below 27 µg/L (box 4, range 1, range 3). A strong negative correlation between DOC and R706/674 index was found when ch-a concentration was maintained at relatively low levels (box 4, range 1, and range 3). Correlations between the NTU, POM, TOC, DOC, and chlorophyll-a were also considered. Chlorophyll-a in analyzed waters has a strong positive correlation with particulate organic matter (POM) and a strong negative correlation with DOC (−0.91) (

Table 4. Pearson correlations between water physical–chemical parameters for full range of data (n = 44).

Variables	R706/674	R450/531	Chl-a	NTU	POM	TOC	DOC
R706/674	1	0.54	0.23	0.79	0.45	−0.60	−0.64
R450/531	0.54	1	−0.47	0.33	−0.23	−0.58	0.01
Chl-a	0.23	−0.47	1	0.28	0.79	0.19	−0.76
NTU	0.79	0.33	0.28	1	0.52	−0.51	−0.69
POM	0.45	−0.23	0.79	0.52	1	0.09	−0.91
TOC	−0.6	−0.58	0.19	−0.51	0.09	1	0.24
DOC	−0.64	0.01	−0.76	−0.6	−0.91	0.24	1

The bold correlations were considered significant at a probability level of p < 0.05.

).

4. Discussion

The mesocosm type of experiment allows us to gather a wider range of chl-a variability compared to standard in situ water ecosystem observations. The variable regarding the range of chlorophyll in the reservoir without treatments during the experiment was between 24.8 and 43.6 µg/L, while chl-a in boxes placed in the same reservoir ranged between 1.8 and 47 µg/L. The occurrence of artificial conditions inside boxes termed by Carpenter (1996) [43] “wall effects”, which set limitations on the experiment’s focus on understanding ecological processes in water ecosystems, gives opportunities for the case of experiments aimed at developing methods for the remote estimation of chl-a. The relative importance of periphyton in a mesocosm increases with a decreasing volume-to-surface ratio [44]. This gives the possibility to reduce the time needed for the spontaneous reduction in phytoplankton concentration in the water depth and limit the time of the experiment. In this case, however, the risk of the mesocosm wall’s influence on the analyzed spectrum is increased. The negative relationship between periphyton growth on the mesocosm wall and phytoplankton abundance enables the study of spectral properties of the water surface at low chl-a values in the water column and the search for the relationship in a wider range of environmental factors.

The whole range of observed chl-a variability did not show relationships with selected spectral indexes (R706/674 and R450/531). This means that the wide range of chl-a concentration changes in water did not cause direct changes in water surface spectral characteristics in the range where protein structures were shown to be spectrally active. The correct estimation of chl-a in low concentrations is crucial for developing the effective remote monitoring of anthropogenic water ecosystems. The data gathered by automated remote sensing technologies can be used for planning and verifying the effectiveness of measures to improve the water quality in the reservoir, assessing the risk of cyanobacterial blooms within bathing areas (a health risk for those in direct contact with the water), and supporting the planning and conducting of surface water treatment for municipal needs [45]. The lack of this relationship could be a result of the presence of water and other substances that influence water surface reflectance curves.

In observed waters, the chl-a shows the strongest significant correlation with POM (0.78) (Table 4). Particulate organic matter (POM) normally consists of phyto- and zooplankton cells, bacteria, and detritus [46,47]. Phototrophic microorganisms are primary producers that convert carbon dioxide (CO₂) and inorganic nutrients into organic matter using sunlight. Increasing chlorophyll-a concentration indicates a level of phytoplankton biomass. A high level of algae production in the water column stimulates zooplankton and bacteria to grow. The dead biomass of such water organisms forms detritus. Phototrophic microorganisms interact with heterotrophic bacteria and other microorganisms through the microbial loop. Microbes (such as bacteria, archaea, fungi, and protozoa) first access organic nutrients, transforming them into inorganic forms, which are then taken up by phytoplankton, contributing to the recycling of nutrients and the formation of new POM [48].

Light is scattered by suspended particulate material (POM) in the water, causing increased turbidity (NTU). Positive correlations between chl-a and POM and between POM and the NTU confirm the above relationships in analyzed water ecosystems. Among the well known spectral characteristics of chl-a, water could contain other substances that absorb VIS radiation. In particular, phycobilin pigments could be strong absorbers of light from the mid-regions of the spectrum (λ∼460–650 nm) [49]. In some reservoirs, a significant component of particulate organic matter is a protein associated with bacterial activity [50]. These proteins have strong absorption in UV and blue regions (UV absorbers) [51,52]. In Baltic Sea water where POM ranged between 0.329 and 11.3 mg/L, best fit equations between the blue spectral absorption 440 nm and POM concentration were found [53]. In this study, the POM shows a strong correlation with both spectral indexes (R450/531 and R706/674) but only in the range of POM concentration between 4 and 8.4 mg/L (respectively, 0.86 and 0.89) and between 1 and 8.4 mg/L (respectively, 0.59 and 0.79). In the higher POM range, a strong relationship with spectral analyses was not indicated. The results suggest that POM variability is mostly shaped by phytoplankton abundance in the water column, but a significant part of POM consists of another form of organic matter that could interfere with the remote sensing of chl_a in analyzed anthropogenic water. In the case of the analyzed anthropogenic reservoir, POM shows a moderate positive correlation with the R706/674 index, and this means that it could, in a moderate way, interfere with the remote estimation of chl-a concentration.

In the whole range of chl-a concentration, total organic carbon (TOC) reveals a moderate negative correlation both with the R450/531 and R706/674 index. In the same chl-a range, dissolved organic carbon (DOC) shows a strong negative correlation with the R706/674 index. Dissolved organic carbon (DOC) very often contains chemical substances (like humic acids) that heavily influence the spectral water properties and therefore the results of chl-a estimation in estuaries and inland waters [54,55,56]. These substances have the highest absorbance in the ultraviolet and blue range of the electromagnetic spectrum, and the absorption declines with increasing wavelengths [57]. In our case under review, DOC shows a stronger correlation with the red range of the electromagnetic spectrum. The strong correlation between DOC and the R706/674 index was found when the ch-a concentration was maintained at low levels (box 4, range 1, and range 3) (Table 3).

While the variability in chl-a was maintained between 24.1 µg/L and 47 µg/L (range 2) and DOC concentration did not exceed 6.76 mg/L, a stronger correlation between chl-a and the R450/531 index was found. The abovementioned relationship confirms that DOC is the most important factor disturbing the remote estimation of chl-a in analyzed water. Additional research is needed to determine key factors that define the spectral properties of waters in analyzed anthropogenic reservoirs and allow the development of algorithms for higher-accuracy estimations of chl-a.

The research on bio-optical complex water showed that colored dissolved organic matter (CDOM) is independent of the chl-a water quality parameter, which leads to degrading the performance of chl-a remote retrieval [44]. The significant absorption of CDOM in red and near-infrared parts was observed in some lakes in boreal and arctic regions. This means that light backscattered from phytoplankton, the most useful spectral feature for retrieving phytoplankton biomass in optically complex waters (near at 710 nm), could be absorbed by CDOM [58]. The variability in CDOM in anthropogenic reservoirs needs to be analyzed in future research to identify the physio-chemical parameters that determine their water spectral reflectance properties. Many species of phytoplankton, especially cyanobacteria, can directly use organic nutrients, though some forms are energy-intensive to process because they require specific enzymes. The supply of organic matter to the reservoir, in addition to internal processes (primary production and the release from bottom sediments), may have external sources, some of which are influenced by anthropogenic activities (discharges of rainwater and water from sewage treatment plants, leachate from agricultural land, etc.). Human inputs from runoff or groundwater often mix with natural organic sources, creating a chemically diverse nutrient mixture. These processes may significantly affect aquatic ecosystems and their phytoplankton community structures, in particular water ecosystems under anthropogenic pressures [59,60]. For example, phycocyanin, a pigment produced by cyanobacteria but not by many other phytoplankton, causes changes in water surface spectral properties. Due to the complex correlation between water quality parameters and complex surface water optical properties, to enhance remote sensing algorithms, developing multimodal deep learning models for chl-a retrieval in anthropogenic inland water ecosystems should be applied [23]. Comprehensive studies of phytoplankton species composition and identification of all spectrally active substances must also be carried out.

5. Conclusions

In the case of anthropogenic inland waters, developing evaluation algorithms for chl-a retrieval might be a complex process. Performing a mesocosm experiment is an effective approach to evaluating relations between the water quality parameters and corresponding remote sensing reflectance. The obtained results allow for the identification of the most important factor influencing remote sensing. The result shows that the remote estimation of chl-a concentration in anthropogenic water reservoirs is possible, but interfering water compounds have to be taken into account. In situ mesocosm experiments provide valuable data in a wider range of chl-a concentrations, reducing the length of the experiment and ensuring replicability.