Lakes are an important source of drinking water, they provide different services from fisheries to tourism, support biodiversity, and are an important component in the global carbon cycle [1
]. Monitoring the water quality and understanding the physical, chemical, and biological status of inland waters is hard to achieve without using remote sensing [4
]. However, there are many obstacles in the way to achieve sufficient accuracy of inland water remote sensing products. Some of them are related to optical complexity of the waters, some to the methodology (e.g., atmospheric correction), and some to the technology (radiometric, spatial, and spectral resolution of sensors) used [4
Only the visible part of electromagnetic radiation can potentially provide us information about the water constituents in most waterbodies as water itself absorbs light strongly at other wavelengths [5
]. The exceptions here are waters with high concentrations of suspended matter [8
] or phytoplankton [10
] where the water reflectance in the near infrared (NIR) part of spectrum can also provide us useful information. However, there may be extreme environments where the water-leaving signal is negligible also in the visible part of the spectrum, automatically preventing the use of most current remote sensing algorithms and methods. For example, in this study we investigated lakes where the CDOM concentrations are so high that the water-leaving signal is practically zero at all visible wavelengths and the above water measured signal consisted predominantly of sun and sky glint. One may assume that the number of such extreme lakes is small as the only reflectance data we were able to find from almost as dark lakes was published only recently [7
]. However, the global inventory of lakes [12
] shows that majority of lakes on the Earth are between 55N and 75N, meaning boreal and arctic lakes, are the most abundant. Many boreal lakes have high CDOM concentration [7
]. Most of the arctic lakes are actually permafrost thaw ponds that should be rich in dissolved organic carbon, DOC, and its coloured component, CDOM, although Sobek et al.
] have shown that the lake DOC pattern at higher latitudes is quite complicated. Thus, at present there is very fragmented information on the possible abundance of CDOM-rich lakes as most of them are probably in inhabited and hardly accessible regions.
It was shown recently [17
] that the iron bound to DOC makes lake water absorbance higher and variable iron to carbon ratio makes remote sensing retrieval of CDOM and DOC concentrations complicated. It means that the number of lakes in which remote sensing is challenging due to low water leaving signal, caused by high absorbance, should be relatively high globally. Retrieving the lake CDOM and DOC concentrations is important from both a drinking water perspective [19
] and the global carbon cycle studies point of view [2
]. The drinking water industry needs this information also in near real time as sudden heavy precipitation may increase the amount of carbon quickly and require modifications in water treatment processes. Consequently, there is a strong need to study optical properties of CDOM-rich lakes and the potential for retrieval of water quality parameters by means of remote sensing in such lakes.
Field radiometers have become remote sensing instruments on their own right rather than being just calibration and validation devices of satellite measurements. For example, routine reflectance measurements carried out from ferries (ferriscope.org) and hand-held devices have been developed for quick monitoring of lake water quality [20
]. Many of the black lakes are small. Therefore, in this study, we focused mainly on field radiometry rather than satellite remote sensing.
The main aim of the study was to investigate black lakes with nearly negligible water leaving signal in the visible part of the spectrum and to estimate whether remote sensing retrieval of water constituents in such extreme CDOM-rich lakes is feasible by means of hyperspectral sensors. The next step was applying the results obtained in black lakes on all other lakes for which we had field radiometry data. Satellites with sufficient spatial resolution for small lake studies, like the Landsat 8 and Sentinel-2, became available recently. Therefore, it was reasonable to evaluate are these satellites suitable from their band configuration point of view for remote sensing of CDOM-rich lakes. This evaluation was performed by recalculating hyperspectral field radiometry data into spectral bands of Landsat 8 and Sentinel-2 as well as using actual satellite imagery.
The optical properties of the studied lakes were variable as seen from the Table 1
. For example, chlorophyll-a concentration varied between 2.14 mg·m−3
and 203.31 mg·m−3
whereas TSS varied between 0.75 mg·L−1
and 63.33 mg·L−1
. All the lakes were relatively CDOM-rich—the aCDOM
(400) varied between 3.23 m−1
and 63.05 m−1
Reflectance spectra of the extreme CDOM lakes are shown in Figure 2
and reflectance spectra of all studied lakes are shown in Figure 3
. It is seen in the Figure 2
that in the extreme lakes water reflectance (red spectrum) is negligible in almost the entire visible part of the spectrum and the only usable signal is in the form of two peaks, which have maxima near 710 nm and 810 nm. It is also seen in the Figure 2
that significant part of the remote sensing signal measured above the black lakes (blue spectrum) is light reflected from the water surface. In the 350–600 nm spectral range the whole signal is glint. The only chance to get information about the water constituents in such lakes is to use these two reflectance peaks. We calculated the height of the two (710 nm and 810 nm) peaks in order to try to estimate concentrations of optically active substances chlorophyll-a, TSS (total suspended solids), SPIM (suspended particulate inorganic matter), SPOM (suspended particulate organic matter), and CDOM. Besides the peak heights themselves, we also used differences, sums, and ratios of these two peak heights.
Reflectance spectra of all studied lakes together are presented in Figure 3
. Most of the spectra have been collected in lakes with significant cyanobacterial biomass as there is a phycocyanin absorption feature at 620 nm and a peak at 650 nm (Figure 3
) typical to only cyanobacteria [11
The peak near 710 nm, caused by combined effect of absorption by water molecules and very high reflectance of phytoplankton in the infrared part of spectrum, is often used for chlorophyll-a retrieval in many waterbodies [35
]. Not surprisingly, there was also good correlation between the peak height, P1
, and chlorophyll-a in the lakes studied by us (Figure 4
The height of the 810 nm peak is clearly higher in reflectance spectra of black lakes than the height of the 700–720 nm peak (Figure 2
). Therefore, we decided to test whether the height of this peak, P2
, is in correlation with the concentrations of optically active substances. Figure 5
illustrates the correlation between P2
and chlorophyll-a and Figure 6
the correlation between the P2
and total suspended matter. The correlation was good for both parameters when data from all Estonian and Swedish lakes was used.
It was not surprising that the height of the 710 nm peak was in good correlation with chlorophyll-a concentration—R2
was 0.74 and 0.72 for above water (Figure 4
) and glint free reflectance respectively. However, it was surprising that the 810 nm peak height correlated even better with the chlorophyll-a concentration—R2
= 0.77 (Figure 5
). This result was obtained for glint-free spectra. R2
was just 0.37 if the peak height was calculated from the above water reflectance spectra. This stresses the importance of removing glint from aquatic reflectance spectra. There have been studies relating the elevated NIR reflectance values to high mineral suspended matter concentration [8
]. Therefore, the good correlation between the 810 nm peak and TSS was expected to certain extent.
In order to test theoretical performance of Sentinel-2 and Landsat 8 sensors in picking up the two peaks containing information about the water properties in the case of black lakes we took the in situ
measured glint-free spectrum of Nohipalu Mustjärv (Figure 2
a) and recalculated it using spectral response functions of Sentinel-2 MSI and Landsat 8 OLI sensors. The results are given in Figure 7
. It is clearly seen that Landsat 8 band configuration does not allow detection of either of the peaks. Sentinel-2 MSI does not have narrow spectral bands near the 810 nm peak. However, the 783 nm centered band 7 allows to detect the peak to certain extent. Especially, because the bands 6 and 8a are located at wavelengths where the lake reflectance values are low. The 705 nm band 5 of Sentinel-2 is almost perfectly located for detection of elevated biomass in waterbodies as we have also shown in our study focusing on using Sentinel-2 imagery in lake research [28
The results obtained from actual satellite imagery resemble those obtained from field measurements spectra as can be seen in Figure 8
. Both satellite reflectances are slightly elevated (not zero) in the blue to green part of spectrum where the water leaving signal is practically zero as can be seen in Figure 2
. This indicates that the satellite signal also contains glint from the water surface which may be significant compared to the water leaving signal as is clearly seen in Figure 2
b. Another potential source of the non-negligible reflectance is the adjacency effect as the black lakes are small and water leaving signal very low compared to the potential signal contamination from the adjacent land.
As was mentioned earlier our main aim was to study the extreme CDOM lakes with field radiometers in order to understand is it possible to retrieve water quality parameters of lakes where the water leaving signal is close to zero in visible part of spectrum. The first question that arises is quite subjective—what is a black or extreme CDOM lake? For example, Duan et al.
] investigated black water blooms in Lake Taihu where the CDOM absorption at 443 nm reached up to 1.68 m−1
at 400 nm). Such waters seem black compared to turbid, highly backscattering, waters of the rest of the lake. We found so low CDOM values only in a few Lake Peipsi stations (minimum value 3.23 m−1
). In the lakes we would call black the aCDOM
(400) varied between 41.45 m−1
and 63.05 m−1
. Black water lakes have been studied also in the USA. For example, Brezonik et al.
] studied a few lakes where CDOM absorption at 440 nm reached up to 25.1 m−1
at 400 nm). Thus, the terms black, CDOM-rich, extreme CDOM lakes are quit arbitrary and depend on the background CDOM levels nearby rather than absolute absorption values.
The relativeness of water colour is clearly seen also in the Figure 9
. The lake Võrtsjärv shown in the left part of the image has nearly twice as high mean CDOM concentration (Table 1
) than the black water blooms in Lake Taihu [38
]. Nevertheless, the Võrtsjärv water looks bright green compared to the Mustjärv in the same scene. There are two reasons for that. First of all the Võrtsjärv water contains relatively high amounts of particulate matter (both organic and inorganic) and is therefore a relatively bright object. On the other hand the visual appearance of all objects in processed satellite imagery depends also on the image stretch and brightness of other object in the scene.
In most lakes the absorption by CDOM is negligible in red and near infrared parts of spectrum. However, in the three black lakes studied by us the absorption of CDOM and water molecules are equal at 700 nm or the CDOM absorption is even higher (Figure 10
). Thus, the light backscattered from phytoplankton has to overcome both water and CDOM absorption in order to form detectable signal in reflectance spectra. Absorption by water molecules increases almost exponentially with increasing wavelength after 690 nm [5
]. Therefore, the elevated signal forms a relatively narrow peak near 700 nm in the case of high biomass (or benthic vegetation in shallow water) and the maximum of the peak is moving towards longer wavelength with increasing biomass. In the lakes where CDOM absorption is still strong near 700 nm it first of all causes the decrease in the height of the peak often used to estimate phytoplankton biomass in water, but it also causes slight shift of the maximum in reflectance spectra towards NIR as the CDOM absorption decreases exponentially with increasing wavelength.
The reason why the 810 nm peak occurs in reflectance spectra is a small decrease in water absorption coefficient approximately between 770 nm and 860 nm (see Figure 10
) with the lowest absorption coefficient at 810 nm. Presence of the peak is obvious in all the reflectance spectra collected by us in different lakes (Figure 3
) not only in the black lakes.
Brezonik et al.
] presented a few reflectance spectra from CDOM-rich lakes in the USA. These spectra were similar to our black lakes—nearly negligible reflectance in the visible part of spectrum and a peak near 710 nm. Unfortunately, the graphs in their paper did not show reflectance beyond 800 nm. Therefore, we do not have reflectance data from very CDOM-rich lakes in other parts of the world to compare with our black lakes. Reflectance spectra with the second near infrared peak (around 810–850 nm) have been published in several papers [10
]. However, the reason and magnitude of this peak was not discussed or even mentioned in any of these papers.
Doxaran et al.
] attributed the high reflectance values in NIR part of spectrum to high concentration of suspended matter and used different band ratios that included NIR band of different satellites (SPOT 790–890 nm, Landsat 750–900 nm, and SeaWiFS 845–885 nm) for retrieving concentrations of suspended matter. There are several publications [8
] showing that the elevated signals in the NIR part of the spectrum is a good predictor of TSM. All these studies were carried out in waters very rich in mineral particles. Therefore, the elevated NIR signal has been attributed to high mineral particle load in the water, not high phytoplankton concentration (chlorophyll-a).
This may seem contradictory to our results, but it is not. The peak near 810 nm is caused by high amount of scattering particles and local dip in absorption coefficient of water molecules. If the scattering material in the water is of mineral origin, like in previous studies [8
], then the peak height is in good correlation with the concentration of mineral particles. In lakes studied by us (and many other lakes) the dominating scattering material in water is phytoplankton. Consequently, the height of the peak at 810 nm has to be in correlation with all parameters describing phytoplankton abundance in the water like chlorophyll-a, TSS, and—its organic component—SPOM. In the lakes studied by us, there was no correlation (R2
= 0.15) between the concentration of mineral particles, SPIM, and the 810 nm peak height. On the other hand, the analysis of the in situ
data showed that there was good correlation between chlorophyll-a and SPOM (R2
= 0.74) and TSS and SPOM (R2
= 0.87) indicating that majority of the suspended particles were organic and significant fraction of them were living phytoplankton cells.
Backscattering coefficient values were relatively high in the extreme CDOM-rich lakes. For example, the backscattering coefficient at 595 nm, bb(595), varied between 0.05 and 0.15 m−1 in all studied lakes, whereas it was between 0.08 and 0.15 m−1 in the extreme CDOM lakes. Thus, the reflectance in these lakes was low not because of low backscattering, but because the high CDOM absorption masks the backscattering signal.
The concentration of phytoplankton (chlorophyll-a) was relatively high in the extreme CDOM lakes (Table 1
) and caused the appearance of the peaks at 710 nm and 810 nm. One may assume that in these lakes the amount of light available for photosynthesis is very low limiting the growth of phytoplankton, but this was not the case. The explanation here is species composition of phytoplankton. Cyanobacteria are the most dominant group in Estonian lakes and Lake Mälaren in Sweden during summer season. Many species of cyanobacteria can regulate their buoyancy and in calm weather conditions can choose the water depth most optimal for their growth. The extreme CDOM-rich lakes studied by us are relatively small and surrounded by forest. This means that the wind speed and water mixing are usually low and cyanobacteria can stay close to the surface where light is available for primary production. This also has an effect on the water reflectance. We have shown [41
] that vertical distribution of phytoplankton biomass has significant impact on the remote sensing signal and the biomass close to the surface has quite different reflectance than the same biomass uniformly mixed in the water column. Therefore, it is not surprising that the biomass located in a thin surface layer (as there is no light at depths of a few decimetres) is producing a strong remote sensing signal and spectral features typical to cyanobacteria (Figure 3
There are studies [42
] showing that the 810 nm peak is suitable for mapping water depth in very shallow (less than 1 m deep) waters. This is reasonable as benthic vegetation has high reflectance in NIR part of spectrum [43
] and there is a decrease in water absorbance at 810 nm making the bottom signal detectable in this spectral region. All reflectance measurements of this study were carried out in optically deep waters with no bottom contribution. Therefore, we are sure that the height of the 810 nm peaks is only due to water constituents and there is no contribution from benthic vegetation.
Our results show that the 810 nm peak height is relatively sensitive to glint. The glint-free reflectance spectra (measured with the radiance sensor just below the water surface) produced better results that the “normal” reflectance spectra measured above the water surface. The glint removal method developed by us [23
] performed well in the case of most measurements except for the most extreme Nohipalu Mustjärv. Most probably, the cause of the failure of the glint removal method was cyanobacterial biomass floating on the water surface producing high values of reflectance in the NIR part of spectrum. The glint removal procedure is not applicable when the NIR signal is higher than the UV signal.
Our results show that both the 710 nm and 810 nm peaks are very useful for retrieving chlorophyll-a and total suspended matter concentrations not only in the CDOM-rich lakes, where there is no measurable signal in the visible part of spectrum, but also in a much wider variety of lakes. The 710 nm peak continues to be the most useful spectral feature for retrieving phytoplankton biomass in productive waters. However, our study shows that the 810 nm peak is more useful in the extreme lakes where the CDOM absorption is still strong at 710 nm.
These two peaks can be used in the interpretation of remote sensing data in the cases where hyperspectral instruments are used (airborne and hand held devices). The only spaceborne instrument sufficient for lake studies spectral (10 nm) and spatial resolution (30 m) was Hyperion on board the EO-1. It was an experimental sensor that did not provide global coverage. The launch of Landsat 8 and Sentinel-2 opened great potential for lake remote sensing from a spatial and radiometric resolution point of view. The spectral resolution of Landsat is not very good from a water quality monitoring perspective. For example, it does not have spectral bands near the 700–720 nm peak, which is most often used to estimate chlorophyll-a concentration in coastal and inland waters [32
]. Landsat series satellites have been used for mapping lake chlorophyll content [45
] for several decades. However, it has been done mainly in eutrophic lakes where biomass is high and the total suspended matter (causing the changes in lake reflectance) is mainly phytoplankton. It is clearly seen comparing the Figure 3
and Figure 7
that the band configuration of Landsat 8 is not optimal for lake water quality monitoring.
Sentinel-2 spatial resolution is finer than that of Landsat 8, but more important for lake studies is its spectral resolution and band configuration. The narrow 705 nm band opens great opportunities in lake chlorophyll-a remote sensing studies as we have demonstrated for black lakes in this study and for a wider variety of lakes in our Sentinel-2 lake remote sensing study [28
]. We showed that, although the band 7 of Sentinel-2 OLI sensor is not positioned optimally to capture the 810 nm peak, the 783 nm band is still useful for this purpose. The suitability of this band in lake remote sensing has to be tested in the future.
We have shown with field reflectance data that, in black lakes, the water leaving signal may be very close to zero in most of the visible part of the spectrum. The measured visible part of spectrum remote sensing reflectance consists mainly of glint in such lakes.
We showed that the height of the 810 nm peak in reflectance spectra is in correlation with the parameters describing phytoplankton biomass (Chlorophyll-a, TSS, SPOM) in a wide variety of lakes. This is especially useful in black lakes where the 700–720 nm peak, normally used in retrieval of chlorophyll-a, is still affected by CDOM absorption.
Previous studies have shown that the NIR peak is caused by large amount of mineral particles in water. Our results show that the 810 nm peak is caused by combined effect of decreased water absorption between 760 nm and 860 nm and scattering by particles in the water column. If the particles in the water are primarily phytoplankton (like in the lakes studied by us), then the height of the 810 nm peak is in good correlation with chlorophyll-a and other parameters describing phytoplankton biomass (SPOM and TSS). If the scattering material in water is mainly of mineral origin (like in previous coastal and river studies) then the 810 peak is in correlation with SPIM and TSS concentration.
Landsat 8 bands are not suitable for detecting the two peaks occurring in reflectance spectra of many lakes. On the other hand, Sentinel-2 band 5 (705 nm) is almost perfectly located for mapping phytoplankton biomass (chlorophyll-a) and the band 7 (783 nm) also allows detection of the 810 nm peak in water reflectance spectra. This is especially useful in the case of black lakes as CDOM absorption may still affect the peak at 705 nm.