- Email: [email protected]
Estimating lake carbon fractions from remote sensing data
RSE-09078; No of Pages 9 Remote Sensing of Environment xxx (2014) xxx–xxx
Contents lists available at ScienceDirect
Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse
Estimating lake carbon fractions from remote sensing data Tiit Kutser a,b,⁎, Charles Verpoorter c,b, Birgot Paavel a, Lars J. Tranvik b a b c
Estonian Marine Institute, University of Tartu, Mäealuse 14, Tallinn, 12618, Estonia Evolutionary Biology Centre, University of Uppsala, Norbyvägen 18D, 75236, Uppsala, Sweden INSU-CNRS, UMR 8187, LOG, Laboratoire d'Océanologie et des Géosciences, Université de Lille Nord de France, ULCO, 32 avenue Foch, 62930 Wimereux, France
a r t i c l e
i n f o
Article history: Received 30 December 2013 Received in revised form 14 May 2014 Accepted 14 May 2014 Available online xxxx Keywords: Remote sensing Lakes CDOM DOC TOC MERIS
a b s t r a c t Issues like monitoring lake water quality, studying the role of lakes in the global carbon cycle or the response of lakes to global change require data more frequently and/or over much larger areas than the in situ water quality monitoring networks can provide. The aim of our study was to investigate whether it is feasible to estimate different lake carbon fractions (CDOM, DOC, TOC, DIC, TIC and pCO2) from space using sensors like OLCI on future Sentinel 3. In situ measurements were carried out in eight measuring stations in two Swedish lakes within 2 days of MERIS overpass. The results suggest that the MERIS CDOM product was not suitable for estimating CDOM in lakes Mälaren and Tämnaren and was not a good proxy for mapping lake DOC and TOC from space. However, a simple green to red band ratio and some other MERIS products like the total absorption coefficient, turbidity index, suspended matter and chlorophyll-a were correlated with different carbon fractions and could potentially be used as proxies to map these lake carbon fractions (CDOM, DOC, TOC, DIC, TIC and pCO2) from space. © 2014 Elsevier Inc. All rights reserved.
1. Introduction Knowing the lake carbon content is important from several aspects. For example, the increase in the amount of dissolved organic matter in lakes used as drinking water resources increases the cost of water treatment (Matilainen, Vepsalainen, & Sillanpää, 2010) and the risk of cancer (McDonald & Komulainen, 2005). Moreover, recent studies (Tranvik et al., 2009) show that lakes play an important role in the global carbon cycle. It is obvious that remote sensing has to be used in order to get spatial and temporal coverage needed for water management, and there is no other option to clarify the role of lakes in the global carbon cycle than by means of remote sensing. First attempts to map lake carbon content with remote sensing were made already in the 1980s. For example Vertucci & Likens (1989) estimated dissolved organic carbon (DOC) and absorbance at 360 nm (CDOM) in Adirondack lakes from reflectance measured with a hand held spectrometer from a boat. Different airborne and satellite sensors have been used in mapping lake dissolved organic matter or its coloured component CDOM (coloured dissolved organic matter) since then (Kallio et al., 2008, 2001; Kutser, Herlevi, Kallio, & Arst, 2001; Kutser, Pierson, Kallio, Reinart, & Sobek, 2005; Kutser, Pierson, Tranvik, et al., 2005; Kutser, Tranvik, & Pierson, 2009; Shuchman et al., 2013; Zhu et al., 2014). The sensors used in these studies (Landsat, ALI, and Hyperion) were not suitable for operative monitoring and/or global studies ⁎ Corresponding author at: Estonian Marine Institute, University of Tartu, Mäealuse 14, 12618 Tallinn, Estonia. Tel.: +372 6718947; fax: +372 6718900. E-mail address: [email protected] (T. Kutser).
due to either low spatial coverage, low temporal coverage or low radiometric sensitivity. MERIS provided global coverage and high temporal coverage, had sufficient radiometric sensitivity and was shown to provide reliable CDOM estimates in boreal lakes (Zhu et al., 2014). The only disadvantage of this sensor was 300 m spatial resolution which was suitable only for relatively large lakes while the majority of the lakes on Earth are too small to be studied with sensors with such spatial resolution. The situation with sensors potentially suitable for lake carbon studies is changing now. Landsat 8 and planned Sentinel-2 missions will provide data for smaller lakes and Sentinel-3 will provide data for larger lakes. Therefore, it is now timely to evaluate what satellite sensors can actually provide for lake carbon studies. Most of the carbon in lakes is usually in dissolved form (Wetzel, 2001). Therefore, there is often good correlation between the dissolved organic carbon and total organic carbon (TOC). Some studies have shown that the lake TOC can be mapped with remote sensing (Chang & Vannah, 2012; Song et al., 2006). Both of these studies used Landsat TM imagery and data from one lake. Although Landsat imagery has been used in some other studies for lake carbon mapping (Kallio et al., 2008; Kutser, 2012) it is generally known that the low signal to noise ratio and 8-bit radiometric resolution (256 gray levels) are suboptimal for aquatic studies because just a few of the gray levels in each band have to describe the whole possible variations in optical properties of water bodies. Stramski, Reynolds, Kahru, and Mitchell (1999) have shown that particulate organic carbon (POC) can be estimated with remote sensing in Southern Ocean waters. The algorithm was based on a good correlation between the particulate backscattering coefficient at 510 nm and
http://dx.doi.org/10.1016/j.rse.2014.05.020 0034-4257/© 2014 Elsevier Inc. All rights reserved.
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
2
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
POC. However, this relationship varied in different parts of the ocean and the correlation was good for log-transformed variables. Hadjimitsis & Clayton (2011) used hand-held spectrometer data, recalculated it to Landsat bands and developed a statistical POC retrieval algorithm for water treatment reservoirs. Backscattering was not measured in these water bodies. Therefore, it is not known whether the backscattering from organic particles made their quantitative mapping with remote sensing feasible. It must be noted that both of these studies use the term particulate organic carbon (POC), but neither of them measured actually carbon. The actual measured parameter was suspended particulate organic matter (SPOM) which was determined as the difference between the dry weight of total suspended solids (TSS) and suspended particulate inorganic matter (SPIM) that is left after combustion of the TSS filters. Sobek, Algesten, Bergström, Jansson, and Tranvik (2003) have shown that pCO2 and DOC are correlated (R2 = 0.51) during ice free season in many boreal lakes. This should allow us to estimate roughly pCO2 from space if we can estimate DOC with sufficient accuracy. We measured pCO2 in both Mälaren and Tämnaren in order to test this hypothesis. The aim of our study was to carry out a small scale experiment and test which lake carbon fractions could potentially be mapped with remote sensing. In this pilot study we tried to estimate from remote sensing data both the parameters that have direct impact on water colour (like CDOM), but we tested also the possibility of estimation of characteristics which do not have known direct impact on water colour (like pCO2). The only satellite sensor suitable for lake carbon studies that was available during our pilot study was MERIS. However, the results of this study allow us to test the potential of future OLCI sensor onboard Sentinel-3 as this sensor will be MERIS follow up. 2. Study sites and data Most of the measurements were carried out in Lake Mälaren (Fig. 1) on September 14–18, 2011 while one sampling station was in Lake Tämnaren. Lake Mälaren is the third largest lake in Sweden. Its surface area is 1140 km2. It is a gemorphologically very complicated lake and its basins behave quite differently from optical point of view (Köhler et al., 2013; Kutser, 2012). Lake Tämnaren is a much smaller (32.6 km2) and a very shallow (max 1.7 m) lake. Lake Mälaren was chosen as a study object because of its optical variability while Lake
Tämnaren was the main study site during 2011 in the Colour of Water project in the frame of which our pilot study was carried out. The lake is located about 50 km north of Uppsala and was equipped with a flux tower and underwater instrumentation for continuous carbon flux studies during 2011 ice free period. Weather conditions were quite variable throughout the fieldwork week. During the first day we experienced strong wind and rain showers while during the last days the conditions were nearly perfect from remote sensing point of view — calm and cloud free. The conditions were most extreme in Lake Tämnaren as strong wind and very shallow water depth caused resuspension of considerable amount of sediments. This, however, guaranteed that the water was optically deep as the Secchi depth was just 0.3 m in the sampling station while the water depth was 1.5 m. The in situ reflectance measurements were performed with Trios Ramses spectrometers (spectral range of 350–900 nm). The measurements were carried out in two different ways. At first we measured reflectance as a ratio of upwelling radiance to downwelling irradiance while both sensors were above the water surface. Second set of measurements was carried out having upwelling radiance sensor (which is equipped with a 5 cm black plastic tube) just below the water surface. Ten spectra were measured and averaged in both cases. The first measurements gave us reflectance with sky and sun glint while the second series gave true reflectance without glint. Absorption, attenuation and scattering coefficient measurements were carried out in situ with WetLabs AC-S, volume scattering measurements with WetLabs ECO-VSF3, and backscattering measurements with Wetlabs ECO-BB3 and ECO-VSF3 for which the wavelengths were chosen complementary to cover the visible part of spectrum (BB3 412 nm, 595 nm and 715 nm; VSF3 460 nm, 532 nm, and 660 nm). In situ pCO2 measurements were performed with SAMI by Sunburst Sensors. The carbon fractions measured in the laboratory were dissolved organic carbon (DOC), dissolved inorganic carbon (DIC), total organic carbon (TOC) and total inorganic carbon (TIC). The carbon samples were collected in acid-rinsed (10% HCl) plastic vials by filling the bottle almost to the top. The sample set was not preserved with acid at the time of collection. The samples were packed in ice and frozen gel packs and delivered to laboratory within hours where most of the analysis where performed immediately. Water chemistry samples were divided into subsamples in the laboratory and thus for analysing dissolved
Fig. 1. Landsat image of Lake Mälaren showing the sophisticated geomorphology of the lake. Sampling stations are indicated with red stars. Lake Tämnaren is located 50 km north from Uppsala and is not shown in this map.
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
organic carbon (DOC), dissolved inorganic carbon (DIC), total organic carbon (TOC), total absorbance for wavelength range from 200 nm to 1100 nm, coloured dissolved organic matter (CDOM), chlorophyll-a and total suspended solid (TSS). Chlorophyll-a was measured with ISO standard methods. TSS was measured gravimetrically and its organic and inorganic components were measured separately. The amount of suspended particulate inorganic matter (SPIM) was determined by combusting TSS filters at 550 °C for 30 minutes. The amount of suspended particulate organic matter (SPOM) was determined by subtracting SPIM from dry weight of TSS. Water samples for analysis of DOC and measuring CDOM absorbance spectra were filtered with a gentle vacuum (N0.17 atm). Precombusted Whatman glass-fiber (GF/F) filters (0.7-μm pore size) were used. DOC concentrations were measured in precombusted borosilicate glass vials (50 ml) using a Siever 900 Total Organic Carbon analyser (General Electric Analytical Instruments, Boulder, CO, USA) equipped with an autosampler. DOC was measured in triplicate maintaining an analytical error to within ±0.5%. Blanks were negligible and the coefficient of variation between injections in a given sample was ±0.5%. Samples for determination of CDOM spectral absorbance were stored in refrigerator (4 to 8 °C). Before measurements the samples were warmed to room temperature. Absorbance spectra of both total and CDOM (filtered) absorption were measured using a spectrophotometer (Lambda 40, Perkin Elmer, Waltham, MA, USA) and Suprasil quartz 1 cm path length cells with ultraviolet (UV) oxidized Milli-Q water as the blank and reference (Mitchell, Kahru, Wieland, & Stramska, 2003). The absorbance of the filtrate was then recorded at wavelengths ranging from 200 to 1100 nm with an interval of 1 nm, but the value at 420 nm was used as the measure of CDOM quantity. MERIS full resolution data were downloaded from CoastColour project archive. Both Level 2R and Level 2W data were used. MERIS imagery was available on September 16th. Previous days were cloudy. Most of Lake Mälaren was between two MERIS swaths on 17th and most of the lake was covered with clouds during MERIS overpass on September 18th. Thus, we had one good image for the 5 days of our measurements campaign. Fortunately the image was collected in the middle of our fieldwork and the time difference between water sampling and MERIS data collection was less than 2 days. MERIS standard absorption and backscattering products are retrieved by means of neural network. For example the MERIS CDOM product is estimated as CDOM absorption coefficient at 443 nm (in m−1). On the other hand the MERIS chlorophyll-a product is a conversion of pigment absorption coefficient at 443 nm and MERIS total suspended matter product is a conversion from total scattering coefficient derived with neural network. MERIS CoastColour product contains also IOP values derived by QAA method (Lee, Carder, & Arnone, 2002) besides the standard neural network approach. These products were also used in the analysis.
3
3. Results and discussion Summary of the laboratory measurements is given in Table 1. It can be seen from variability of all measured parameters that different parts of Lake Mälaren can be considered as different water bodies for the developing and validating remote sensing products. Several parameters in Lake Tämnaren were significantly different from all stations of Lake Mälaren. For example phytoplankton biomass (chlorophyll-a) was 2–15 times higher in Lake Tämnaren and the total absorption coefficient was significantly higher than in any station in Lake Mälaren. The Lake Tämnaren total absorption coefficient was mainly high due to the high amount of resuspended particles in the water column. Most of the particles were of organic origin but also the concentration of inorganic particles was twice higher in Tämnaren than in Mälaren. Usually DOC is more than 90% from TOC in lake environments. This was the case in Mälaren where the percentage was between 94% and 99%. However, it was just 78% in Tämnaren where the strong resuspension of sediments increased the amount of particulate organic material. In situ reflectance spectra were measured in two different ways as was described above. Glint free spectra are shown in Fig. 2. These were measured directly although they can be retrieved from above water measurements using the glint removal method proposed by Kutser, Vahtmäe, Paavel, & Kauer (2013). The highest reflectance was measured in sampling station 3 where the chlorophyll-a concentration was 50.8 μg/l. Chlorophyll-a concentration was twice as high in Lake Tämnaren, but the reflectance values were much lower there due to strong absorption of light by particulate material in water. On the other hand, the higher concentration of biomass is clearly visible in Lake Tämnaren reflectance spectrum as the peak at 704–706 nm is relatively much higher than in any of the Lake Mälaren stations. The height of the peak near 700 nm is often used as an indicator of phytoplankton biomass (Kutser, Arst, Mäekivi, Leppänen, & Blanco, 1997; Mittenzwey, Gitelson, & Kondratiev, 1992; Moses, Gitelson, Berdnikov, Saprygin, & Povazhnyi, 2012). It is clearly observable from the reflectance spectra that cyanobacteria were present in relatively large quantities in all sampled waters as the phycocyanin absorption feature at 624 nm is found in the reflectance spectra of all stations. Surprisingly, the cyanobacterial presence was the least obvious in reflectance spectra of station 51 where we carried out two reflectance measurements — one series of measurement was carried out in the situation where the bloom was mixed in the upper water column and one series of measurements was carried in the situation where where cyanobacterial colonies were floating on the water surface. In both cases the phycocyanin absorption was hardly detectable, chlorophyll-a absorption at 676 nm was relatively low and the peak near 700 nm was also small despite chlorophyll-a concentration was 13.5 μg/l. The absence of the peak and high reflectance values at NIR wavelengths in the case of surface floating cyanobacteria was surprising as the high NIR reflectance of cyanobacterial blooms can be observed even from
Table 1 Results of the laboratory measurements of water samples collected in lakes Tämnaren and Mälaren.
Tämnaren Mälaren 42 Mälaren 11 Mälaren 9 Mälaren3 Mälaren 34 Mälaren 51 Mälaren 17
Chl
CDOM(420)
absTOT(420)
TSS
SPOM
SPIM
DOC
DIC
TIC
TOC
CO2
114.1 10.8 39.4 26 50.8 7.1 13.5 22.4
6.31 5.71 4.82 4.21 8.06 2.16 2.43 3.31
55 7.23 11.14 9.01 19.52 2.97 4.11 9.22
64.95 21.1 39.71 29.93 40.05 21.59 18.89
40.62 10.12 25.66 19.88 31.78 10.65 7.37
24.33 10.98 14.05 10.05 11.28 10.94 11.52
18,713 12,362 10,516 9526 11,466 8202 8867 8986
17,700 25,737 7732 6842 5329 9138 13,875 7560
18,400 27,100 7796 6955 5443 9310 14,100 7779
24,116 12,950 10,837 9816 12,108 8322 9119 9546
109 145 196 111 259 106 177 105
Chl — chlorophyll-a in μg/l; CDOM(420) — absorption of filtered water samples at 420 nm in m−1; absTOT(420) — absorption of unfiltered water samples at 420 nm in m−1; TSS — total suspended solids in mg/l; SPOM — suspended particulate organic matter in mg/l; SPIM — suspended particulate inorganic matter in mg/l; DOC — dissolved organic carbon in ppb; DIC — dissolved inorganic carbon in ppb; TIC — total inorganic carbon in ppb; TOC — total organic carbon in ppb; CO2 — carbon dioxide partial pressure in water in ppm. The number in the station name indicates the number of each station in lake Mälaren monitoring program.
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
4
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
Fig. 2. Glint-free reflectance (Lu/Ed) spectra measured with Ramses spectrometers (A) and MERIS reflectance (Lw/Ed) for the same stations extracted from CoastColour L2R imagery (B). Numbers in the legend indicate the number of each sampling station in lake Mälaren monitoring program (for interpretation of the references to colour in this figure legend, the reader is referred to the web version of the article).
space (Kutser, 2004). In station 51 the cyanobacteria were present as aggregates of a few millimetres in diameter. This explains why the bloom was clearly visible and why the reflectance signal was low although the bloom was floating on the water surface — the isolated aggregates covered just a small fraction of the area sampled by Ramses spectrometers (usually a few cm2) and the signal was dominated by relatively clear water. In other stations the cyanobacteria were present in individual colonies. Although the biomass was higher in the other stations the bloom just made water greenish and turbid which is visually not so spectacular like large aggregations of colonies of cyanobacteria in relatively clear water. However, the high amount of individual colonies well mixed in the water gave much stronger remote sensing signal than the biomass that was present in larger aggregations. MERIS imagery was available for September 16 as was mentioned above. Reflectance from CoastColour L2R product was compared with the in situ reflectance data (Fig. 2). MERIS reflectance in station 42 differed significantly from the in situ measured reflectance. The reason may be small clouds and their effects in the vicinity of this station. MERIS results from this station were excluded from further analysis due to the incorrect reflectance. Comparing the spectra in Fig. 2 panels A and B gives an impression that there is something is wrong with the shape of MERIS reflectance. For example MERIS reflectance has often a peak at 619 nm where the in situ spectra have a minimum due to phycocyanin absorption. The actual reason for the unusual reflectance shape is not bad atmospheric
correction, but the absence of spectral bands around 590 nm and 650 nm. At these wavelengths are the true peaks in Mälaren and Tämnaren reflectance spectra. It can also be observed in the MERIS spectra that the relative depth of the chlorophyll-a absorption feature in red part of spectrum is smaller than in field spectra. This is partly caused by the fact that the location
Fig. 3. Correlation between CDOM measured from water samples and MERIS CoastColour CDOM product. Note that the laboratory and satellite product use different wavelengths.
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
of the peak near 700 nm is shifting towards red with increasing phytoplankton concentration. The actual peak is usually at shorter wavelengths than the MERIS 709 nm band and MERIS cannot capture the actual height of this peak. Also the maximum in chlorophyll-a absorption occurs between MERIS 664 nm and 680 nm bands. It means that the MERIS spectra are smoothed in the red part of spectrum compared to the field reflectance data. Remote sensing can provide information about water properties if these properties affect the reflectance of light or are correlated with a parameter that affects light reflectance from a water body. There is
5
good correlation between DOC and its coloured component CDOM in many lakes (Kallio, 1999; Tranvik, 1990; Zhang, Qin, Zhu, Zhang, & Yang, 2007). This was the case also in our study where the correlation between DOC and CDOM measured from water samples was strong (R2 = 0.92; p = 0.0022) in Lake Mälaren but the Lake Tämnaren station did not fit with the DOC-CDOM relationship. The first logical option was to try to estimate lake DOC by using MERIS CDOM product. However, we found no correlation between the in situ measured and MERIS CDOM estimates (Fig. 3). The correlation between the MERIS CDOM product and measured DOC was even worse. The results of other authors (Alikas &
Fig. 4. Correlation between CDOM measured from water samples and estimated from Ramses spectra using the band ratio algorithm by Kutser, Pierson, Kallio, et al., 2005 (A); between the green to red band ratio and measured DOC (B); the band ratio and TOC (C).
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
6
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
Reinart, 2008; Campbell, Phinn, Dekker, & Brando, 2011) are similar to ours — MERIS CDOM product does not perform well in inland water bodies. MERIS CoastColour products contain also CDOM produced be QAA algorithm (Lee et al., 2002). However, the QAA algorithm failed in estimating CDOM absorption for half of the stations. There are many empirical CDOM retrieval algorithms (Kallio et al., 2008, 2001; Kutser, Pierson, Kallio, et al., 2005; Kutser, Pierson, Tranvik, et al., 2005; Kutser et al., 2009; Shuchman et al., 2013) and analytical methods for retrieving lake CDOM (Kutser et al., 2001) besides the MERIS neural network approach. For example Zhu et al. (2014) have tested several of the band ratio type algorithms in optically complex inland waters. One of the band ratio algorithms that worked in boreal lakes (Kutser, Pierson, Kallio, et al., 2005; Kutser, Pierson, Tranvik, et al., 2005) and performed well over wider variety of lakes according to Zhu et al. (2014) is green to red band ratio. Before applying this ratio to MERIS data we tested it with field reflectance. We recalculated Ramses spectra into broad bands of ALI/Landsat 8 and tested how the green to red band ratio is performing in the studied lakes. There was good correlation between measured CDOM and CDOM estimated from Ramses data using the band ratio algorithm by Kutser, Pierson, Kallio, et al. (2005) as is seen in Fig. 4A (R2 = 0.78; p = 0.0037). We used Ramses data and the same band ratio for estimating DOC and TOC concentration in lakes Mälaren and Tämnaren. The results were not very good as can be seen in Fig. 4B and C. The correlation between the band ratio and measured DOC concentration was poor (R2 = 0.35, p = 0.22) and the correlation with TOC was even lower (R2 = 0.30, p = 0.32). It is clearly seen in both graphs that the one sampling point from Lake Tämnaren differs significantly from all the results from Lake Mälaren. Removing the sampling station of Lake Tämnaren improves the correlation significantly for both DOC (R2 = 0.59; p = 0.075). and TOC (R2 = 0.62; p = 0.062). Approximate equivalents of ALI/Landsat green (525–605 nm) and red (630–690 nm) bands are MERIS bands 6 (559 nm) and 8 (664 nm). The green to red ratio was calculated from MERIS reflectance data. It is seen in Fig. 5 that the b6/b8 ratio had a very good correlation with measured CDOM (R2 = 0.81, p = 0.002). Thus, this ratio works better in CDOM retrieval than the MERIS neural network based CDOM product. We tested whether or not this band ratio is suitable for estimating DOC and TOC from MERIS data, but the correlation coefficients were low (R2 = 0.43, p = 0.50 and R2 = 0.40, p = 0.45 respectively). It can be argued that using a power function is not justified and the high correlation coefficient seen in Fig. 5 is misleadingly high due to the distribution of the band ratio values. However, there are modelling studies carried out for different satellites (Landsat 7, ALI, IKONOS) and
Fig. 5. Correlation between MERIS band 6 and band 8 ratio and in situ measured CDOM absorbance at 420 nm.
for wide range of concentrations of optically active water substances (Kutser, Pierson, Tranvik, et al., 2005) as well as studies based on real satellite and in situ data (Kutser, Pierson, Kallio, et al., 2005; Kutser, Pierson, Tranvik, et al., 2005; Zhu et al., 2014) showing that the green to red band ratio and power functions, similar to that presented in Fig. 5, perform well in estimating lake CDOM. MERIS processors produce many different water quality products from total-, pigments-, and CDOM absorption coefficients to estimates of chlorophyll-a, and TSS concentrations as well as parameters like turbidity index. Before testing suitability of different MERIS products as proxies for some lake carbon parameters we analysed relationships between different parameters in our in situ data. Analysis of the in situ results showed that the total absorption coefficient at 420 nm measured from water samples is well correlated with DOC (R2 = 0.99; p = 0.000004) and TOC (R2 = 0.99; p = 0.000004). Most of the suspended matter in the studied lakes was of organic origin and there was very strong correlation between TSS and SPOM (R2 = 0.92; p = 0.0022). The MERIS TSS product is derived from suspended sediment backscattering coefficient (at 443 nm) by neural network. We do not have backscattering sensors that measure at 443 nm, but one of the six wavelengths, where we measured backscattering coefficient, is 460 nm. We can confirm that there is also very strong correlation between the backscattering coefficient at 460 nm and the TSS concentration (R2 = 0.96). From the carbon parameters under investigation TIC was slightly correlated with the TSS concentration, and DIC was correlated with chlorophyll-a (correlation coefficients 0.57 and 0.46 respectively) based on our in situ measurements results. The total absorption coefficient, chlorophyll-a and TSS are all MERIS standard products. Therefore, these products were used in an attempt to estimate TOC, TIC and DIC from MERIS data. We also tested whether any MERIS product is in correlation with lake pCO2 and could be used as a proxy in pCO2 estimation. MERIS atmospheric correction did not work properly in Mälaren station 42 as was seen on Fig. 2B. Therefore, this station was excluded from all further analysis. Fig. 6A shows the best correlation between pCO2 in lake waters and MERIS standard products. The parameter that had the highest correlation was turbidity index. The R2 is not very high (0.56; p = 0.053) but it indicates that there is a relationship between lake water turbidity and its pCO2. Also another water turbidity related product — backscattering by suspended particles retrieved by QAA method (Lee et al., 2002) was in correlation with pCO2 (R2 = 0.53; p = 0.065). It was mentioned above that the sea surface temperature and chlorophyll-a have been used as proxies for pCO2 in ocean waters and DOC could be used as a proxy in inland waters. The potential mechanism between the lake water turbidity (estimated from remote sensing data) and its pCO2 needs further investigation. It must be noted that the pCO2 measurements themselves are associated with some problems. SAMI is measuring pCO2 with 15 minutes interval. Our optical sampling takes usually half an hour to 45 minutes. Therefore, we started measurements in each station with lowering SAMI into water and took it out after other measurements were finished. As a result we had 2–3 pCO2 readings for each station. These 2–3 values were relatively stable, but it would be desirable to have much more readings. Then it will be possible to study whether there are pCO2 fluctuation over the time and are these fluctuations somehow related to any optical water properties that could then be used as proxies for remote sensing estimate of pCO2. As was mentioned above, the laboratory data indicated that chlorophyll-a may be a suitable proxy to estimate the amount of dissolved inorganic carbon in lake water. Indeed, Fig. 6B indicates that for Lake Mälaren there is also reasonable correlation between the MERIS chlorophyll-a product and DIC measured in laboratory (R2 = 0.56, p = 0.086). This suggests that we may be able to predict the DIC from space borne data using the MERIS chlorophyll product as a proxy. Several MERIS products (like Kd(490), TSS) were in reasonable correlation
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
7
Fig. 6. Correlation between lake carbon parameters and selected MERIS products. (A) pCO2 and MERIS turbidity index; (B) dissolved inorganic carbon and MERIS chlorophyll-a product; (C) dissolved inorganic carbon and MERIS total absorption coefficient (excluding the Lake Tämnaren station).
with DIC in Lake Mälaren. Fig. 6C shows the correlation between MERIS total absorption coefficient (at 443 nm) product and DIC in Lake Mälaren (R2 = 0.94, p = 0.0077). There was also correlation between the laboratory measured total suspended matter and the total inorganic carbon concentration. Fig. 7 shows that the correlation also exists between MERIS suspended matter product and TIC (R2 = 0.68, p = 0.34). However, this is the result for Lake Mälaren only (Lake Tämnaren station excluded).
The MERIS CDOM product did not have correlation with laboratory measured CDOM and was not a good proxy for neither DOC nor TOC. However, the MERIS total absorption coefficient product proved to be a reasonable proxy for mapping lake TOC (R2 = 0.71, p = 0.021) and DOC (R2 = 0.74, p = 0.025) as is seen in Fig. 8. The MERIS CDOM product is an estimate of CDOM absorption at 443 nm and the total absorption coefficient product estimates the sum of CDOM and phytoplankton pigments absorption at 443 nm. If the
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
8
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
Fig. 7. Correlation between total inorganic carbon and MERIS suspended matter product in Lake Mälaren.
MERIS total absorption product performs better in estimating lake CDOM than the MERIS CDOM product then there may be problems in MERIS processing chain separating contributions of phytoplankton
and CDOM into the total absorption coefficient. This issue needs further investigation. It must be stressed that the time difference between the in situ sampling and MERIS data collection was up to 2 days. Usually it is not recommended to have more than 3 hours time difference between the sampling and satellite overpass even in optically simple and stable oceanic waters. The studied lakes are very dynamic and weather changed from strong wind and heavy rain showers to calm and clear days. Most probably the changes were the largest in the case of Lake Tämnaren. This lake was sampled first and the time difference with MERIS overpass was the longest. We sampled there in the end of strong wind period. It means that the amount of resuspended material in the lake was huge as the lake is just over 1 m deep. There were relatively large particles in the water column, but most probably large amount of these particles sank before the MERIS image acquisition. Consequently, the optically most extreme conditions and most likely the largest changes between in situ sampling and MERIS image acquisition were the reasons why Lake Tämnaren data differed dramatically from the Lake Mälaren data. There was cyanobacterial bloom present in all sampling stations and this had serious impact on water reflectance as can be seen in Fig. 2. Thus, the conditions were not favourable for remote sensing of lake carbon. Despite the suboptimal conditions we were able to find reasonable correlations between several MERIS products and different in situ measured carbon fractions.
Fig. 8. Correlation between the MERIS total absorption coefficient product and the total organic carbon (A) and dissolved organic carbon (B) in lakes Mälaren and Tämnaren.
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
It must be noted that the pilot study had a limited number of measuring stations. Therefore, it is not possible to say with high level of confidence which lake carbon parameters can be mapped with MERIS/OLCI type sensors and which parameters cannot be mapped with remote sensing. However, our limited study pointed out that most of the studied parameters could potentially be estimated by means of band ratio type algorithms and using different neural network based MERIS products as proxies. The study also revealed relationships between some optical parameters (total absorption coefficient, turbidity index) and some lake carbon characteristics. It would be reasonable to investigate whether these relationships were just statistics valid in limited number of water bodies or are there causal relationships between the optical water properties and lake carbon parameters. 4. Conclusions The results of this study suggest that the MERIS CDOM product was not suitable for estimating CDOM in lakes Mälaren and Tämnaren and was not a good proxy for mapping lake DOC and TOC from space. On the other hand the MERIS total absorption coefficient product and a simple green to red band ratio calculated from MERIS data were suitable proxies to estimate lake CDOM, DOC and TOC from space. Statistical analysis performed with in situ data suggested that some carbon parameters are correlated with concentrations of optically active substances and/or optical characteristics of the studied lakes. For example DIC was correlated with chlorophyll-a, and there was strong correlation between the total absorption coefficient and TOC as well as with DOC. Analysing the relationships between MERIS CoastColour products and in situ measured water carbon characteristics suggested that some MERIS products, like the turbidity index, suspended matter, and chlorophyll-a concentrations should allow to estimate lake carbon parameters like DIC, TIC and pCO2 from space in Lake Mälaren and/or in both the studied lakes. Although the concentrations of optically active substances and the lake carbon fractions from CDOM to pCO2 varied in relatively wide range in the studied lakes there is still strong need to test the validity of these results in other types of lakes, for example in oligotrophic, hypertrophic and sediment laden lakes, before it can be confirmed that the results are valid more globally than for certain boreal lakes. This applies specially to the parameters that have no obvious causal relationship with the optical water properties. The relatively small number of sampling stations does not allow us to make strong conclusions about the suitability of OLCI for mapping different carbon parameters in lakes globally. Nevertheless, the results of our pilot study suggest that estimating of several lake carbon fractions from space is feasible. Acknowledgements The study was funded by FORMAS strong research environment project “The Colour of Water — interplay with climate, and effects on drinking water supply.” Part of the data processing costs were covered from Estonian Basic Research Grant SF0180009As11. References Alikas, K., & Reinart, A. (2008). Validation of the MERIS products on large European lakes: Peipsi, Vänern and Vättern. Hydrobiologia, 599, 161–168. Campbell, G., Phinn, S. R., Dekker, A. G., & Brando, V. E. (2011). Remote sensing of water quality in an Australian tropical freshwater impoundment using matrix inversion and MERIS images. Remote Sensing of Environment, 115, 2402–2414. Chang, N. -B., & Vannah, B. (2012). Monitoring the total organic carbon concentrations in a lake with the integrated data fusion and machine learning (IDFM) technique. Proceedings of SPIE, 8513, http://dx.doi.org/10.1117/12.927632. Hadjimitsis, D. G., & Clayton, C. (2011). Field spectroscopy for assisting water quality monitoring and assessment in water treatment reservoirs using atmospheric corrected satellite remotely sensed imagery. Remote Sensing, 3, 362–377.
9
Kallio, K. (1999). Absorption properties of dissolved organic matter in Finnish lakes. Proceedings of Estonian Academy of Sciences. Biology and Ecology, 48, 75–83. Kallio, K., Attila, J., Härmä, P., Koponen, S., Pulliainen, J., Hyytiäinen, U. -M., et al. (2008). Landsat ETM + images in the estimation of seasonal lake water quality in boreal river basins. Environmental Management, 42, 511–522. Kallio, K., Kutser, T., Hannonen, T., Koponen, S., Pulliainen, J., Vepsäläinen, J., et al. (2001). Retrieval of water quality from airborne imaging spectrometry of various lake types in different seasons. The Science of the Total Environment, 268, 59–77. Köhler, S. J., Kothawala, D., Futter, M. N., Liungman, O., & Tranvik, L. (2013). In-lake processes offset increased terrestrial inputs of dissolved organic carbon and color to lakes. Plos One, 8, e70598. Kutser, T. (2004). Quantitative detection of chlorophyll in cyanobacterial blooms by satellite remote sensing. Limnology and Oceanography, 49, 2179–2189. Kutser, T. (2012). The possibility of using the Landsat image archive for monitoring long time trends in coloured dissolved organic matter concentration in lake waters. Remote Sensing of Environment, 123, 334–338. Kutser, T., Arst, H., Mäekivi, S., Leppänen, J. -M., & Blanco, A. (1997). Monitoring of algae blooms by optical remote sensing. In A. Spiteri (Ed.), Remote Sensing '96 (pp. 161–166). Rotterdam: Balkema. Kutser, T., Herlevi, A., Kallio, K., & Arst, H. (2001). A hyperspectral model for interpretation of passive optical remote sensing data from turbid lakes. The Science of the Total Environment, 268, 47–58. Kutser, T., Pierson, D. C., Kallio, K., Reinart, A., & Sobek, S. (2005). Mapping lake CDOM by satellite remote sensing. Remote Sensing of Environment, 94, 535–540. Kutser, T., Pierson, D. C., Tranvik, L. J., Reinart, A., Sobek, S., & Kallio, K. (2005). Using satellite remote sensing to estimate the colored dissolved organic matter absorption coefficient in lakes. Ecosystems, 8, 709–720. Kutser, T., Tranvik, L. J., & Pierson, D. C. (2009). Variations in colored dissolved organic matter between boreal lakes studied by satellite remote sensing. Journal of Applied Remote Sensing, 3, 033538. Kutser, T., Vahtmäe, E., Paavel, B., & Kauer, T. (2013). Removing glint effects from field radiometry data measured in optically complex coastal and inland waters. Remote Sensing of Environment, 133, 85–89. Lee, Z., Carder, K. L., & Arnone, R. A. (2002). Deriving inherent optical properties from water color: A multiband quasi-analytical algorithm for optically deep waters. Applied Optics, 41, 5755–5772. Matilainen, A., Vepsalainen, M., & Sillanpää, M. (2010). Natural organic matter removal by coagulation during drinking water treatment: A review. Advances in Colloid and Interface Science, 159, 189–197. McDonald, T. A., & Komulainen, H. (2005). Carcinogenicity of the chlorination disinfection by-product MX. Journal of Environmental Science and Health, Part C: Environmental Carcinogenesis and Ecotoxicology Reviews, 23, 163–214. Mitchell, B. G., Kahru, M., Wieland, J., & Stramska, M. (2003). Determination of spectral absorption coefficients of particles, dissolved material and phytoplankton for discrete water samples. Ocean optics protocols for satellite ocean color sensor validation, revision 4. Inherent optical properties: Instruments, characterizations, field measurements and data analysis protocols, vol. IV. (pp. 39–64). Mittenzwey, K. H., Gitelson, A. A., & Kondratiev, K. Y. (1992). Determination of chlorophyll-a of inland waters on the basis of spectral reflectance. Limnology and Oceanography, 37, 147–149. Moses, W. J., Gitelson, A. A., Berdnikov, S., Saprygin, V., & Povazhnyi, V. (2012). Operational MERIS-based NIR-red algorithms for estimating chlorophyll-a concentrations in coastal waters — the Azov Sea case study. Remote Sensing of Environment, 121, 118–124. Shuchman, R. A., Leshkevich, P., Sayers, M. J., Johengen, T. H., Brooks, C. N., & Pozdnyakov, D. (2013). An algorithm to retrieve chlorophyll, dissolved organic carbon, and suspended minerals from Great Lakes satellite data. Journal of Great Lakes Research, 39, 14–33. Sobek, S., Algesten, G., Bergström, A. -K., Jansson, M., & Tranvik, L. J. (2003). The catchment and climate regulation of pCO2 in boreal lakes. Global Change Biology, 9, 630–641. Song, K., Zhang, B., Wang, Z., Li, F., Duan, H., & Guo, Y. (2006). Water TOC and TP concentration estimation using Landsat TM data with empirical algorithms in Chagan Lake, China. Proceedings of IGARSS 2006 (pp. 3438–3442). Stramski, D., Reynolds, R. A., Kahru, M., & Mitchell, B. G. (1999). Estimation of particulate organic carbon in the ocean from satellite remote sensing. Science, 285, 239–242. Tranvik, L. J. (1990). Bacterioplankton growth on fractions of dissolved organic carbon of different molecular weights from humic and clear waters. Applied and Environmental Microbiology, 56, 1672–1677. Tranvik, L. J., Downing, J. A., Cotner, J. B., Loiselle, S., Striegl, R. G., Ballatore, T. J., et al. (2009). Lakes and impoundments as regulators of carbon cycling and climate. Limnology and Oceanography, 54, 2298–2314. Vertucci, F. A., & Likens, G. E. (1989). Spectral reflectance and water quality of Adirondack mountain region lakes. Limnology and Oceanography, 34, 1656–1672. Wetzel, R. G. (2001). Limnology: Lake and river ecosystems (3rd ed.). Elsevier (1006 pp.). Zhang, Y., Qin, B., Zhu, G., Zhang, L., & Yang, L. (2007). Chromophoric dissolved organic matter (CDOM) absorption characteristics in relation to fluorescence in Lake Taihu, China, a large shallow subtropical lake. Hydrobiologia, 581, 43–52. Zhu, W., Yu, Q., Tian, Y. Q., Becker, B.L., Zheng, T., & Carrick, H. J. (2014). An assessment of remote sensing algorithms for colored dissolved organic matter in complex freshwater environments. Remote Sensing of Environment, 140, 766–778.
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
Contents lists available at ScienceDirect
Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse
Estimating lake carbon fractions from remote sensing data Tiit Kutser a,b,⁎, Charles Verpoorter c,b, Birgot Paavel a, Lars J. Tranvik b a b c
Estonian Marine Institute, University of Tartu, Mäealuse 14, Tallinn, 12618, Estonia Evolutionary Biology Centre, University of Uppsala, Norbyvägen 18D, 75236, Uppsala, Sweden INSU-CNRS, UMR 8187, LOG, Laboratoire d'Océanologie et des Géosciences, Université de Lille Nord de France, ULCO, 32 avenue Foch, 62930 Wimereux, France
a r t i c l e
i n f o
Article history: Received 30 December 2013 Received in revised form 14 May 2014 Accepted 14 May 2014 Available online xxxx Keywords: Remote sensing Lakes CDOM DOC TOC MERIS
a b s t r a c t Issues like monitoring lake water quality, studying the role of lakes in the global carbon cycle or the response of lakes to global change require data more frequently and/or over much larger areas than the in situ water quality monitoring networks can provide. The aim of our study was to investigate whether it is feasible to estimate different lake carbon fractions (CDOM, DOC, TOC, DIC, TIC and pCO2) from space using sensors like OLCI on future Sentinel 3. In situ measurements were carried out in eight measuring stations in two Swedish lakes within 2 days of MERIS overpass. The results suggest that the MERIS CDOM product was not suitable for estimating CDOM in lakes Mälaren and Tämnaren and was not a good proxy for mapping lake DOC and TOC from space. However, a simple green to red band ratio and some other MERIS products like the total absorption coefficient, turbidity index, suspended matter and chlorophyll-a were correlated with different carbon fractions and could potentially be used as proxies to map these lake carbon fractions (CDOM, DOC, TOC, DIC, TIC and pCO2) from space. © 2014 Elsevier Inc. All rights reserved.
1. Introduction Knowing the lake carbon content is important from several aspects. For example, the increase in the amount of dissolved organic matter in lakes used as drinking water resources increases the cost of water treatment (Matilainen, Vepsalainen, & Sillanpää, 2010) and the risk of cancer (McDonald & Komulainen, 2005). Moreover, recent studies (Tranvik et al., 2009) show that lakes play an important role in the global carbon cycle. It is obvious that remote sensing has to be used in order to get spatial and temporal coverage needed for water management, and there is no other option to clarify the role of lakes in the global carbon cycle than by means of remote sensing. First attempts to map lake carbon content with remote sensing were made already in the 1980s. For example Vertucci & Likens (1989) estimated dissolved organic carbon (DOC) and absorbance at 360 nm (CDOM) in Adirondack lakes from reflectance measured with a hand held spectrometer from a boat. Different airborne and satellite sensors have been used in mapping lake dissolved organic matter or its coloured component CDOM (coloured dissolved organic matter) since then (Kallio et al., 2008, 2001; Kutser, Herlevi, Kallio, & Arst, 2001; Kutser, Pierson, Kallio, Reinart, & Sobek, 2005; Kutser, Pierson, Tranvik, et al., 2005; Kutser, Tranvik, & Pierson, 2009; Shuchman et al., 2013; Zhu et al., 2014). The sensors used in these studies (Landsat, ALI, and Hyperion) were not suitable for operative monitoring and/or global studies ⁎ Corresponding author at: Estonian Marine Institute, University of Tartu, Mäealuse 14, 12618 Tallinn, Estonia. Tel.: +372 6718947; fax: +372 6718900. E-mail address: [email protected] (T. Kutser).
due to either low spatial coverage, low temporal coverage or low radiometric sensitivity. MERIS provided global coverage and high temporal coverage, had sufficient radiometric sensitivity and was shown to provide reliable CDOM estimates in boreal lakes (Zhu et al., 2014). The only disadvantage of this sensor was 300 m spatial resolution which was suitable only for relatively large lakes while the majority of the lakes on Earth are too small to be studied with sensors with such spatial resolution. The situation with sensors potentially suitable for lake carbon studies is changing now. Landsat 8 and planned Sentinel-2 missions will provide data for smaller lakes and Sentinel-3 will provide data for larger lakes. Therefore, it is now timely to evaluate what satellite sensors can actually provide for lake carbon studies. Most of the carbon in lakes is usually in dissolved form (Wetzel, 2001). Therefore, there is often good correlation between the dissolved organic carbon and total organic carbon (TOC). Some studies have shown that the lake TOC can be mapped with remote sensing (Chang & Vannah, 2012; Song et al., 2006). Both of these studies used Landsat TM imagery and data from one lake. Although Landsat imagery has been used in some other studies for lake carbon mapping (Kallio et al., 2008; Kutser, 2012) it is generally known that the low signal to noise ratio and 8-bit radiometric resolution (256 gray levels) are suboptimal for aquatic studies because just a few of the gray levels in each band have to describe the whole possible variations in optical properties of water bodies. Stramski, Reynolds, Kahru, and Mitchell (1999) have shown that particulate organic carbon (POC) can be estimated with remote sensing in Southern Ocean waters. The algorithm was based on a good correlation between the particulate backscattering coefficient at 510 nm and
http://dx.doi.org/10.1016/j.rse.2014.05.020 0034-4257/© 2014 Elsevier Inc. All rights reserved.
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
2
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
POC. However, this relationship varied in different parts of the ocean and the correlation was good for log-transformed variables. Hadjimitsis & Clayton (2011) used hand-held spectrometer data, recalculated it to Landsat bands and developed a statistical POC retrieval algorithm for water treatment reservoirs. Backscattering was not measured in these water bodies. Therefore, it is not known whether the backscattering from organic particles made their quantitative mapping with remote sensing feasible. It must be noted that both of these studies use the term particulate organic carbon (POC), but neither of them measured actually carbon. The actual measured parameter was suspended particulate organic matter (SPOM) which was determined as the difference between the dry weight of total suspended solids (TSS) and suspended particulate inorganic matter (SPIM) that is left after combustion of the TSS filters. Sobek, Algesten, Bergström, Jansson, and Tranvik (2003) have shown that pCO2 and DOC are correlated (R2 = 0.51) during ice free season in many boreal lakes. This should allow us to estimate roughly pCO2 from space if we can estimate DOC with sufficient accuracy. We measured pCO2 in both Mälaren and Tämnaren in order to test this hypothesis. The aim of our study was to carry out a small scale experiment and test which lake carbon fractions could potentially be mapped with remote sensing. In this pilot study we tried to estimate from remote sensing data both the parameters that have direct impact on water colour (like CDOM), but we tested also the possibility of estimation of characteristics which do not have known direct impact on water colour (like pCO2). The only satellite sensor suitable for lake carbon studies that was available during our pilot study was MERIS. However, the results of this study allow us to test the potential of future OLCI sensor onboard Sentinel-3 as this sensor will be MERIS follow up. 2. Study sites and data Most of the measurements were carried out in Lake Mälaren (Fig. 1) on September 14–18, 2011 while one sampling station was in Lake Tämnaren. Lake Mälaren is the third largest lake in Sweden. Its surface area is 1140 km2. It is a gemorphologically very complicated lake and its basins behave quite differently from optical point of view (Köhler et al., 2013; Kutser, 2012). Lake Tämnaren is a much smaller (32.6 km2) and a very shallow (max 1.7 m) lake. Lake Mälaren was chosen as a study object because of its optical variability while Lake
Tämnaren was the main study site during 2011 in the Colour of Water project in the frame of which our pilot study was carried out. The lake is located about 50 km north of Uppsala and was equipped with a flux tower and underwater instrumentation for continuous carbon flux studies during 2011 ice free period. Weather conditions were quite variable throughout the fieldwork week. During the first day we experienced strong wind and rain showers while during the last days the conditions were nearly perfect from remote sensing point of view — calm and cloud free. The conditions were most extreme in Lake Tämnaren as strong wind and very shallow water depth caused resuspension of considerable amount of sediments. This, however, guaranteed that the water was optically deep as the Secchi depth was just 0.3 m in the sampling station while the water depth was 1.5 m. The in situ reflectance measurements were performed with Trios Ramses spectrometers (spectral range of 350–900 nm). The measurements were carried out in two different ways. At first we measured reflectance as a ratio of upwelling radiance to downwelling irradiance while both sensors were above the water surface. Second set of measurements was carried out having upwelling radiance sensor (which is equipped with a 5 cm black plastic tube) just below the water surface. Ten spectra were measured and averaged in both cases. The first measurements gave us reflectance with sky and sun glint while the second series gave true reflectance without glint. Absorption, attenuation and scattering coefficient measurements were carried out in situ with WetLabs AC-S, volume scattering measurements with WetLabs ECO-VSF3, and backscattering measurements with Wetlabs ECO-BB3 and ECO-VSF3 for which the wavelengths were chosen complementary to cover the visible part of spectrum (BB3 412 nm, 595 nm and 715 nm; VSF3 460 nm, 532 nm, and 660 nm). In situ pCO2 measurements were performed with SAMI by Sunburst Sensors. The carbon fractions measured in the laboratory were dissolved organic carbon (DOC), dissolved inorganic carbon (DIC), total organic carbon (TOC) and total inorganic carbon (TIC). The carbon samples were collected in acid-rinsed (10% HCl) plastic vials by filling the bottle almost to the top. The sample set was not preserved with acid at the time of collection. The samples were packed in ice and frozen gel packs and delivered to laboratory within hours where most of the analysis where performed immediately. Water chemistry samples were divided into subsamples in the laboratory and thus for analysing dissolved
Fig. 1. Landsat image of Lake Mälaren showing the sophisticated geomorphology of the lake. Sampling stations are indicated with red stars. Lake Tämnaren is located 50 km north from Uppsala and is not shown in this map.
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
organic carbon (DOC), dissolved inorganic carbon (DIC), total organic carbon (TOC), total absorbance for wavelength range from 200 nm to 1100 nm, coloured dissolved organic matter (CDOM), chlorophyll-a and total suspended solid (TSS). Chlorophyll-a was measured with ISO standard methods. TSS was measured gravimetrically and its organic and inorganic components were measured separately. The amount of suspended particulate inorganic matter (SPIM) was determined by combusting TSS filters at 550 °C for 30 minutes. The amount of suspended particulate organic matter (SPOM) was determined by subtracting SPIM from dry weight of TSS. Water samples for analysis of DOC and measuring CDOM absorbance spectra were filtered with a gentle vacuum (N0.17 atm). Precombusted Whatman glass-fiber (GF/F) filters (0.7-μm pore size) were used. DOC concentrations were measured in precombusted borosilicate glass vials (50 ml) using a Siever 900 Total Organic Carbon analyser (General Electric Analytical Instruments, Boulder, CO, USA) equipped with an autosampler. DOC was measured in triplicate maintaining an analytical error to within ±0.5%. Blanks were negligible and the coefficient of variation between injections in a given sample was ±0.5%. Samples for determination of CDOM spectral absorbance were stored in refrigerator (4 to 8 °C). Before measurements the samples were warmed to room temperature. Absorbance spectra of both total and CDOM (filtered) absorption were measured using a spectrophotometer (Lambda 40, Perkin Elmer, Waltham, MA, USA) and Suprasil quartz 1 cm path length cells with ultraviolet (UV) oxidized Milli-Q water as the blank and reference (Mitchell, Kahru, Wieland, & Stramska, 2003). The absorbance of the filtrate was then recorded at wavelengths ranging from 200 to 1100 nm with an interval of 1 nm, but the value at 420 nm was used as the measure of CDOM quantity. MERIS full resolution data were downloaded from CoastColour project archive. Both Level 2R and Level 2W data were used. MERIS imagery was available on September 16th. Previous days were cloudy. Most of Lake Mälaren was between two MERIS swaths on 17th and most of the lake was covered with clouds during MERIS overpass on September 18th. Thus, we had one good image for the 5 days of our measurements campaign. Fortunately the image was collected in the middle of our fieldwork and the time difference between water sampling and MERIS data collection was less than 2 days. MERIS standard absorption and backscattering products are retrieved by means of neural network. For example the MERIS CDOM product is estimated as CDOM absorption coefficient at 443 nm (in m−1). On the other hand the MERIS chlorophyll-a product is a conversion of pigment absorption coefficient at 443 nm and MERIS total suspended matter product is a conversion from total scattering coefficient derived with neural network. MERIS CoastColour product contains also IOP values derived by QAA method (Lee, Carder, & Arnone, 2002) besides the standard neural network approach. These products were also used in the analysis.
3
3. Results and discussion Summary of the laboratory measurements is given in Table 1. It can be seen from variability of all measured parameters that different parts of Lake Mälaren can be considered as different water bodies for the developing and validating remote sensing products. Several parameters in Lake Tämnaren were significantly different from all stations of Lake Mälaren. For example phytoplankton biomass (chlorophyll-a) was 2–15 times higher in Lake Tämnaren and the total absorption coefficient was significantly higher than in any station in Lake Mälaren. The Lake Tämnaren total absorption coefficient was mainly high due to the high amount of resuspended particles in the water column. Most of the particles were of organic origin but also the concentration of inorganic particles was twice higher in Tämnaren than in Mälaren. Usually DOC is more than 90% from TOC in lake environments. This was the case in Mälaren where the percentage was between 94% and 99%. However, it was just 78% in Tämnaren where the strong resuspension of sediments increased the amount of particulate organic material. In situ reflectance spectra were measured in two different ways as was described above. Glint free spectra are shown in Fig. 2. These were measured directly although they can be retrieved from above water measurements using the glint removal method proposed by Kutser, Vahtmäe, Paavel, & Kauer (2013). The highest reflectance was measured in sampling station 3 where the chlorophyll-a concentration was 50.8 μg/l. Chlorophyll-a concentration was twice as high in Lake Tämnaren, but the reflectance values were much lower there due to strong absorption of light by particulate material in water. On the other hand, the higher concentration of biomass is clearly visible in Lake Tämnaren reflectance spectrum as the peak at 704–706 nm is relatively much higher than in any of the Lake Mälaren stations. The height of the peak near 700 nm is often used as an indicator of phytoplankton biomass (Kutser, Arst, Mäekivi, Leppänen, & Blanco, 1997; Mittenzwey, Gitelson, & Kondratiev, 1992; Moses, Gitelson, Berdnikov, Saprygin, & Povazhnyi, 2012). It is clearly observable from the reflectance spectra that cyanobacteria were present in relatively large quantities in all sampled waters as the phycocyanin absorption feature at 624 nm is found in the reflectance spectra of all stations. Surprisingly, the cyanobacterial presence was the least obvious in reflectance spectra of station 51 where we carried out two reflectance measurements — one series of measurement was carried out in the situation where the bloom was mixed in the upper water column and one series of measurements was carried in the situation where where cyanobacterial colonies were floating on the water surface. In both cases the phycocyanin absorption was hardly detectable, chlorophyll-a absorption at 676 nm was relatively low and the peak near 700 nm was also small despite chlorophyll-a concentration was 13.5 μg/l. The absence of the peak and high reflectance values at NIR wavelengths in the case of surface floating cyanobacteria was surprising as the high NIR reflectance of cyanobacterial blooms can be observed even from
Table 1 Results of the laboratory measurements of water samples collected in lakes Tämnaren and Mälaren.
Tämnaren Mälaren 42 Mälaren 11 Mälaren 9 Mälaren3 Mälaren 34 Mälaren 51 Mälaren 17
Chl
CDOM(420)
absTOT(420)
TSS
SPOM
SPIM
DOC
DIC
TIC
TOC
CO2
114.1 10.8 39.4 26 50.8 7.1 13.5 22.4
6.31 5.71 4.82 4.21 8.06 2.16 2.43 3.31
55 7.23 11.14 9.01 19.52 2.97 4.11 9.22
64.95 21.1 39.71 29.93 40.05 21.59 18.89
40.62 10.12 25.66 19.88 31.78 10.65 7.37
24.33 10.98 14.05 10.05 11.28 10.94 11.52
18,713 12,362 10,516 9526 11,466 8202 8867 8986
17,700 25,737 7732 6842 5329 9138 13,875 7560
18,400 27,100 7796 6955 5443 9310 14,100 7779
24,116 12,950 10,837 9816 12,108 8322 9119 9546
109 145 196 111 259 106 177 105
Chl — chlorophyll-a in μg/l; CDOM(420) — absorption of filtered water samples at 420 nm in m−1; absTOT(420) — absorption of unfiltered water samples at 420 nm in m−1; TSS — total suspended solids in mg/l; SPOM — suspended particulate organic matter in mg/l; SPIM — suspended particulate inorganic matter in mg/l; DOC — dissolved organic carbon in ppb; DIC — dissolved inorganic carbon in ppb; TIC — total inorganic carbon in ppb; TOC — total organic carbon in ppb; CO2 — carbon dioxide partial pressure in water in ppm. The number in the station name indicates the number of each station in lake Mälaren monitoring program.
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
4
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
Fig. 2. Glint-free reflectance (Lu/Ed) spectra measured with Ramses spectrometers (A) and MERIS reflectance (Lw/Ed) for the same stations extracted from CoastColour L2R imagery (B). Numbers in the legend indicate the number of each sampling station in lake Mälaren monitoring program (for interpretation of the references to colour in this figure legend, the reader is referred to the web version of the article).
space (Kutser, 2004). In station 51 the cyanobacteria were present as aggregates of a few millimetres in diameter. This explains why the bloom was clearly visible and why the reflectance signal was low although the bloom was floating on the water surface — the isolated aggregates covered just a small fraction of the area sampled by Ramses spectrometers (usually a few cm2) and the signal was dominated by relatively clear water. In other stations the cyanobacteria were present in individual colonies. Although the biomass was higher in the other stations the bloom just made water greenish and turbid which is visually not so spectacular like large aggregations of colonies of cyanobacteria in relatively clear water. However, the high amount of individual colonies well mixed in the water gave much stronger remote sensing signal than the biomass that was present in larger aggregations. MERIS imagery was available for September 16 as was mentioned above. Reflectance from CoastColour L2R product was compared with the in situ reflectance data (Fig. 2). MERIS reflectance in station 42 differed significantly from the in situ measured reflectance. The reason may be small clouds and their effects in the vicinity of this station. MERIS results from this station were excluded from further analysis due to the incorrect reflectance. Comparing the spectra in Fig. 2 panels A and B gives an impression that there is something is wrong with the shape of MERIS reflectance. For example MERIS reflectance has often a peak at 619 nm where the in situ spectra have a minimum due to phycocyanin absorption. The actual reason for the unusual reflectance shape is not bad atmospheric
correction, but the absence of spectral bands around 590 nm and 650 nm. At these wavelengths are the true peaks in Mälaren and Tämnaren reflectance spectra. It can also be observed in the MERIS spectra that the relative depth of the chlorophyll-a absorption feature in red part of spectrum is smaller than in field spectra. This is partly caused by the fact that the location
Fig. 3. Correlation between CDOM measured from water samples and MERIS CoastColour CDOM product. Note that the laboratory and satellite product use different wavelengths.
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
of the peak near 700 nm is shifting towards red with increasing phytoplankton concentration. The actual peak is usually at shorter wavelengths than the MERIS 709 nm band and MERIS cannot capture the actual height of this peak. Also the maximum in chlorophyll-a absorption occurs between MERIS 664 nm and 680 nm bands. It means that the MERIS spectra are smoothed in the red part of spectrum compared to the field reflectance data. Remote sensing can provide information about water properties if these properties affect the reflectance of light or are correlated with a parameter that affects light reflectance from a water body. There is
5
good correlation between DOC and its coloured component CDOM in many lakes (Kallio, 1999; Tranvik, 1990; Zhang, Qin, Zhu, Zhang, & Yang, 2007). This was the case also in our study where the correlation between DOC and CDOM measured from water samples was strong (R2 = 0.92; p = 0.0022) in Lake Mälaren but the Lake Tämnaren station did not fit with the DOC-CDOM relationship. The first logical option was to try to estimate lake DOC by using MERIS CDOM product. However, we found no correlation between the in situ measured and MERIS CDOM estimates (Fig. 3). The correlation between the MERIS CDOM product and measured DOC was even worse. The results of other authors (Alikas &
Fig. 4. Correlation between CDOM measured from water samples and estimated from Ramses spectra using the band ratio algorithm by Kutser, Pierson, Kallio, et al., 2005 (A); between the green to red band ratio and measured DOC (B); the band ratio and TOC (C).
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
6
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
Reinart, 2008; Campbell, Phinn, Dekker, & Brando, 2011) are similar to ours — MERIS CDOM product does not perform well in inland water bodies. MERIS CoastColour products contain also CDOM produced be QAA algorithm (Lee et al., 2002). However, the QAA algorithm failed in estimating CDOM absorption for half of the stations. There are many empirical CDOM retrieval algorithms (Kallio et al., 2008, 2001; Kutser, Pierson, Kallio, et al., 2005; Kutser, Pierson, Tranvik, et al., 2005; Kutser et al., 2009; Shuchman et al., 2013) and analytical methods for retrieving lake CDOM (Kutser et al., 2001) besides the MERIS neural network approach. For example Zhu et al. (2014) have tested several of the band ratio type algorithms in optically complex inland waters. One of the band ratio algorithms that worked in boreal lakes (Kutser, Pierson, Kallio, et al., 2005; Kutser, Pierson, Tranvik, et al., 2005) and performed well over wider variety of lakes according to Zhu et al. (2014) is green to red band ratio. Before applying this ratio to MERIS data we tested it with field reflectance. We recalculated Ramses spectra into broad bands of ALI/Landsat 8 and tested how the green to red band ratio is performing in the studied lakes. There was good correlation between measured CDOM and CDOM estimated from Ramses data using the band ratio algorithm by Kutser, Pierson, Kallio, et al. (2005) as is seen in Fig. 4A (R2 = 0.78; p = 0.0037). We used Ramses data and the same band ratio for estimating DOC and TOC concentration in lakes Mälaren and Tämnaren. The results were not very good as can be seen in Fig. 4B and C. The correlation between the band ratio and measured DOC concentration was poor (R2 = 0.35, p = 0.22) and the correlation with TOC was even lower (R2 = 0.30, p = 0.32). It is clearly seen in both graphs that the one sampling point from Lake Tämnaren differs significantly from all the results from Lake Mälaren. Removing the sampling station of Lake Tämnaren improves the correlation significantly for both DOC (R2 = 0.59; p = 0.075). and TOC (R2 = 0.62; p = 0.062). Approximate equivalents of ALI/Landsat green (525–605 nm) and red (630–690 nm) bands are MERIS bands 6 (559 nm) and 8 (664 nm). The green to red ratio was calculated from MERIS reflectance data. It is seen in Fig. 5 that the b6/b8 ratio had a very good correlation with measured CDOM (R2 = 0.81, p = 0.002). Thus, this ratio works better in CDOM retrieval than the MERIS neural network based CDOM product. We tested whether or not this band ratio is suitable for estimating DOC and TOC from MERIS data, but the correlation coefficients were low (R2 = 0.43, p = 0.50 and R2 = 0.40, p = 0.45 respectively). It can be argued that using a power function is not justified and the high correlation coefficient seen in Fig. 5 is misleadingly high due to the distribution of the band ratio values. However, there are modelling studies carried out for different satellites (Landsat 7, ALI, IKONOS) and
Fig. 5. Correlation between MERIS band 6 and band 8 ratio and in situ measured CDOM absorbance at 420 nm.
for wide range of concentrations of optically active water substances (Kutser, Pierson, Tranvik, et al., 2005) as well as studies based on real satellite and in situ data (Kutser, Pierson, Kallio, et al., 2005; Kutser, Pierson, Tranvik, et al., 2005; Zhu et al., 2014) showing that the green to red band ratio and power functions, similar to that presented in Fig. 5, perform well in estimating lake CDOM. MERIS processors produce many different water quality products from total-, pigments-, and CDOM absorption coefficients to estimates of chlorophyll-a, and TSS concentrations as well as parameters like turbidity index. Before testing suitability of different MERIS products as proxies for some lake carbon parameters we analysed relationships between different parameters in our in situ data. Analysis of the in situ results showed that the total absorption coefficient at 420 nm measured from water samples is well correlated with DOC (R2 = 0.99; p = 0.000004) and TOC (R2 = 0.99; p = 0.000004). Most of the suspended matter in the studied lakes was of organic origin and there was very strong correlation between TSS and SPOM (R2 = 0.92; p = 0.0022). The MERIS TSS product is derived from suspended sediment backscattering coefficient (at 443 nm) by neural network. We do not have backscattering sensors that measure at 443 nm, but one of the six wavelengths, where we measured backscattering coefficient, is 460 nm. We can confirm that there is also very strong correlation between the backscattering coefficient at 460 nm and the TSS concentration (R2 = 0.96). From the carbon parameters under investigation TIC was slightly correlated with the TSS concentration, and DIC was correlated with chlorophyll-a (correlation coefficients 0.57 and 0.46 respectively) based on our in situ measurements results. The total absorption coefficient, chlorophyll-a and TSS are all MERIS standard products. Therefore, these products were used in an attempt to estimate TOC, TIC and DIC from MERIS data. We also tested whether any MERIS product is in correlation with lake pCO2 and could be used as a proxy in pCO2 estimation. MERIS atmospheric correction did not work properly in Mälaren station 42 as was seen on Fig. 2B. Therefore, this station was excluded from all further analysis. Fig. 6A shows the best correlation between pCO2 in lake waters and MERIS standard products. The parameter that had the highest correlation was turbidity index. The R2 is not very high (0.56; p = 0.053) but it indicates that there is a relationship between lake water turbidity and its pCO2. Also another water turbidity related product — backscattering by suspended particles retrieved by QAA method (Lee et al., 2002) was in correlation with pCO2 (R2 = 0.53; p = 0.065). It was mentioned above that the sea surface temperature and chlorophyll-a have been used as proxies for pCO2 in ocean waters and DOC could be used as a proxy in inland waters. The potential mechanism between the lake water turbidity (estimated from remote sensing data) and its pCO2 needs further investigation. It must be noted that the pCO2 measurements themselves are associated with some problems. SAMI is measuring pCO2 with 15 minutes interval. Our optical sampling takes usually half an hour to 45 minutes. Therefore, we started measurements in each station with lowering SAMI into water and took it out after other measurements were finished. As a result we had 2–3 pCO2 readings for each station. These 2–3 values were relatively stable, but it would be desirable to have much more readings. Then it will be possible to study whether there are pCO2 fluctuation over the time and are these fluctuations somehow related to any optical water properties that could then be used as proxies for remote sensing estimate of pCO2. As was mentioned above, the laboratory data indicated that chlorophyll-a may be a suitable proxy to estimate the amount of dissolved inorganic carbon in lake water. Indeed, Fig. 6B indicates that for Lake Mälaren there is also reasonable correlation between the MERIS chlorophyll-a product and DIC measured in laboratory (R2 = 0.56, p = 0.086). This suggests that we may be able to predict the DIC from space borne data using the MERIS chlorophyll product as a proxy. Several MERIS products (like Kd(490), TSS) were in reasonable correlation
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
7
Fig. 6. Correlation between lake carbon parameters and selected MERIS products. (A) pCO2 and MERIS turbidity index; (B) dissolved inorganic carbon and MERIS chlorophyll-a product; (C) dissolved inorganic carbon and MERIS total absorption coefficient (excluding the Lake Tämnaren station).
with DIC in Lake Mälaren. Fig. 6C shows the correlation between MERIS total absorption coefficient (at 443 nm) product and DIC in Lake Mälaren (R2 = 0.94, p = 0.0077). There was also correlation between the laboratory measured total suspended matter and the total inorganic carbon concentration. Fig. 7 shows that the correlation also exists between MERIS suspended matter product and TIC (R2 = 0.68, p = 0.34). However, this is the result for Lake Mälaren only (Lake Tämnaren station excluded).
The MERIS CDOM product did not have correlation with laboratory measured CDOM and was not a good proxy for neither DOC nor TOC. However, the MERIS total absorption coefficient product proved to be a reasonable proxy for mapping lake TOC (R2 = 0.71, p = 0.021) and DOC (R2 = 0.74, p = 0.025) as is seen in Fig. 8. The MERIS CDOM product is an estimate of CDOM absorption at 443 nm and the total absorption coefficient product estimates the sum of CDOM and phytoplankton pigments absorption at 443 nm. If the
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
8
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
Fig. 7. Correlation between total inorganic carbon and MERIS suspended matter product in Lake Mälaren.
MERIS total absorption product performs better in estimating lake CDOM than the MERIS CDOM product then there may be problems in MERIS processing chain separating contributions of phytoplankton
and CDOM into the total absorption coefficient. This issue needs further investigation. It must be stressed that the time difference between the in situ sampling and MERIS data collection was up to 2 days. Usually it is not recommended to have more than 3 hours time difference between the sampling and satellite overpass even in optically simple and stable oceanic waters. The studied lakes are very dynamic and weather changed from strong wind and heavy rain showers to calm and clear days. Most probably the changes were the largest in the case of Lake Tämnaren. This lake was sampled first and the time difference with MERIS overpass was the longest. We sampled there in the end of strong wind period. It means that the amount of resuspended material in the lake was huge as the lake is just over 1 m deep. There were relatively large particles in the water column, but most probably large amount of these particles sank before the MERIS image acquisition. Consequently, the optically most extreme conditions and most likely the largest changes between in situ sampling and MERIS image acquisition were the reasons why Lake Tämnaren data differed dramatically from the Lake Mälaren data. There was cyanobacterial bloom present in all sampling stations and this had serious impact on water reflectance as can be seen in Fig. 2. Thus, the conditions were not favourable for remote sensing of lake carbon. Despite the suboptimal conditions we were able to find reasonable correlations between several MERIS products and different in situ measured carbon fractions.
Fig. 8. Correlation between the MERIS total absorption coefficient product and the total organic carbon (A) and dissolved organic carbon (B) in lakes Mälaren and Tämnaren.
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020
T. Kutser et al. / Remote Sensing of Environment xxx (2014) xxx–xxx
It must be noted that the pilot study had a limited number of measuring stations. Therefore, it is not possible to say with high level of confidence which lake carbon parameters can be mapped with MERIS/OLCI type sensors and which parameters cannot be mapped with remote sensing. However, our limited study pointed out that most of the studied parameters could potentially be estimated by means of band ratio type algorithms and using different neural network based MERIS products as proxies. The study also revealed relationships between some optical parameters (total absorption coefficient, turbidity index) and some lake carbon characteristics. It would be reasonable to investigate whether these relationships were just statistics valid in limited number of water bodies or are there causal relationships between the optical water properties and lake carbon parameters. 4. Conclusions The results of this study suggest that the MERIS CDOM product was not suitable for estimating CDOM in lakes Mälaren and Tämnaren and was not a good proxy for mapping lake DOC and TOC from space. On the other hand the MERIS total absorption coefficient product and a simple green to red band ratio calculated from MERIS data were suitable proxies to estimate lake CDOM, DOC and TOC from space. Statistical analysis performed with in situ data suggested that some carbon parameters are correlated with concentrations of optically active substances and/or optical characteristics of the studied lakes. For example DIC was correlated with chlorophyll-a, and there was strong correlation between the total absorption coefficient and TOC as well as with DOC. Analysing the relationships between MERIS CoastColour products and in situ measured water carbon characteristics suggested that some MERIS products, like the turbidity index, suspended matter, and chlorophyll-a concentrations should allow to estimate lake carbon parameters like DIC, TIC and pCO2 from space in Lake Mälaren and/or in both the studied lakes. Although the concentrations of optically active substances and the lake carbon fractions from CDOM to pCO2 varied in relatively wide range in the studied lakes there is still strong need to test the validity of these results in other types of lakes, for example in oligotrophic, hypertrophic and sediment laden lakes, before it can be confirmed that the results are valid more globally than for certain boreal lakes. This applies specially to the parameters that have no obvious causal relationship with the optical water properties. The relatively small number of sampling stations does not allow us to make strong conclusions about the suitability of OLCI for mapping different carbon parameters in lakes globally. Nevertheless, the results of our pilot study suggest that estimating of several lake carbon fractions from space is feasible. Acknowledgements The study was funded by FORMAS strong research environment project “The Colour of Water — interplay with climate, and effects on drinking water supply.” Part of the data processing costs were covered from Estonian Basic Research Grant SF0180009As11. References Alikas, K., & Reinart, A. (2008). Validation of the MERIS products on large European lakes: Peipsi, Vänern and Vättern. Hydrobiologia, 599, 161–168. Campbell, G., Phinn, S. R., Dekker, A. G., & Brando, V. E. (2011). Remote sensing of water quality in an Australian tropical freshwater impoundment using matrix inversion and MERIS images. Remote Sensing of Environment, 115, 2402–2414. Chang, N. -B., & Vannah, B. (2012). Monitoring the total organic carbon concentrations in a lake with the integrated data fusion and machine learning (IDFM) technique. Proceedings of SPIE, 8513, http://dx.doi.org/10.1117/12.927632. Hadjimitsis, D. G., & Clayton, C. (2011). Field spectroscopy for assisting water quality monitoring and assessment in water treatment reservoirs using atmospheric corrected satellite remotely sensed imagery. Remote Sensing, 3, 362–377.
9
Kallio, K. (1999). Absorption properties of dissolved organic matter in Finnish lakes. Proceedings of Estonian Academy of Sciences. Biology and Ecology, 48, 75–83. Kallio, K., Attila, J., Härmä, P., Koponen, S., Pulliainen, J., Hyytiäinen, U. -M., et al. (2008). Landsat ETM + images in the estimation of seasonal lake water quality in boreal river basins. Environmental Management, 42, 511–522. Kallio, K., Kutser, T., Hannonen, T., Koponen, S., Pulliainen, J., Vepsäläinen, J., et al. (2001). Retrieval of water quality from airborne imaging spectrometry of various lake types in different seasons. The Science of the Total Environment, 268, 59–77. Köhler, S. J., Kothawala, D., Futter, M. N., Liungman, O., & Tranvik, L. (2013). In-lake processes offset increased terrestrial inputs of dissolved organic carbon and color to lakes. Plos One, 8, e70598. Kutser, T. (2004). Quantitative detection of chlorophyll in cyanobacterial blooms by satellite remote sensing. Limnology and Oceanography, 49, 2179–2189. Kutser, T. (2012). The possibility of using the Landsat image archive for monitoring long time trends in coloured dissolved organic matter concentration in lake waters. Remote Sensing of Environment, 123, 334–338. Kutser, T., Arst, H., Mäekivi, S., Leppänen, J. -M., & Blanco, A. (1997). Monitoring of algae blooms by optical remote sensing. In A. Spiteri (Ed.), Remote Sensing '96 (pp. 161–166). Rotterdam: Balkema. Kutser, T., Herlevi, A., Kallio, K., & Arst, H. (2001). A hyperspectral model for interpretation of passive optical remote sensing data from turbid lakes. The Science of the Total Environment, 268, 47–58. Kutser, T., Pierson, D. C., Kallio, K., Reinart, A., & Sobek, S. (2005). Mapping lake CDOM by satellite remote sensing. Remote Sensing of Environment, 94, 535–540. Kutser, T., Pierson, D. C., Tranvik, L. J., Reinart, A., Sobek, S., & Kallio, K. (2005). Using satellite remote sensing to estimate the colored dissolved organic matter absorption coefficient in lakes. Ecosystems, 8, 709–720. Kutser, T., Tranvik, L. J., & Pierson, D. C. (2009). Variations in colored dissolved organic matter between boreal lakes studied by satellite remote sensing. Journal of Applied Remote Sensing, 3, 033538. Kutser, T., Vahtmäe, E., Paavel, B., & Kauer, T. (2013). Removing glint effects from field radiometry data measured in optically complex coastal and inland waters. Remote Sensing of Environment, 133, 85–89. Lee, Z., Carder, K. L., & Arnone, R. A. (2002). Deriving inherent optical properties from water color: A multiband quasi-analytical algorithm for optically deep waters. Applied Optics, 41, 5755–5772. Matilainen, A., Vepsalainen, M., & Sillanpää, M. (2010). Natural organic matter removal by coagulation during drinking water treatment: A review. Advances in Colloid and Interface Science, 159, 189–197. McDonald, T. A., & Komulainen, H. (2005). Carcinogenicity of the chlorination disinfection by-product MX. Journal of Environmental Science and Health, Part C: Environmental Carcinogenesis and Ecotoxicology Reviews, 23, 163–214. Mitchell, B. G., Kahru, M., Wieland, J., & Stramska, M. (2003). Determination of spectral absorption coefficients of particles, dissolved material and phytoplankton for discrete water samples. Ocean optics protocols for satellite ocean color sensor validation, revision 4. Inherent optical properties: Instruments, characterizations, field measurements and data analysis protocols, vol. IV. (pp. 39–64). Mittenzwey, K. H., Gitelson, A. A., & Kondratiev, K. Y. (1992). Determination of chlorophyll-a of inland waters on the basis of spectral reflectance. Limnology and Oceanography, 37, 147–149. Moses, W. J., Gitelson, A. A., Berdnikov, S., Saprygin, V., & Povazhnyi, V. (2012). Operational MERIS-based NIR-red algorithms for estimating chlorophyll-a concentrations in coastal waters — the Azov Sea case study. Remote Sensing of Environment, 121, 118–124. Shuchman, R. A., Leshkevich, P., Sayers, M. J., Johengen, T. H., Brooks, C. N., & Pozdnyakov, D. (2013). An algorithm to retrieve chlorophyll, dissolved organic carbon, and suspended minerals from Great Lakes satellite data. Journal of Great Lakes Research, 39, 14–33. Sobek, S., Algesten, G., Bergström, A. -K., Jansson, M., & Tranvik, L. J. (2003). The catchment and climate regulation of pCO2 in boreal lakes. Global Change Biology, 9, 630–641. Song, K., Zhang, B., Wang, Z., Li, F., Duan, H., & Guo, Y. (2006). Water TOC and TP concentration estimation using Landsat TM data with empirical algorithms in Chagan Lake, China. Proceedings of IGARSS 2006 (pp. 3438–3442). Stramski, D., Reynolds, R. A., Kahru, M., & Mitchell, B. G. (1999). Estimation of particulate organic carbon in the ocean from satellite remote sensing. Science, 285, 239–242. Tranvik, L. J. (1990). Bacterioplankton growth on fractions of dissolved organic carbon of different molecular weights from humic and clear waters. Applied and Environmental Microbiology, 56, 1672–1677. Tranvik, L. J., Downing, J. A., Cotner, J. B., Loiselle, S., Striegl, R. G., Ballatore, T. J., et al. (2009). Lakes and impoundments as regulators of carbon cycling and climate. Limnology and Oceanography, 54, 2298–2314. Vertucci, F. A., & Likens, G. E. (1989). Spectral reflectance and water quality of Adirondack mountain region lakes. Limnology and Oceanography, 34, 1656–1672. Wetzel, R. G. (2001). Limnology: Lake and river ecosystems (3rd ed.). Elsevier (1006 pp.). Zhang, Y., Qin, B., Zhu, G., Zhang, L., & Yang, L. (2007). Chromophoric dissolved organic matter (CDOM) absorption characteristics in relation to fluorescence in Lake Taihu, China, a large shallow subtropical lake. Hydrobiologia, 581, 43–52. Zhu, W., Yu, Q., Tian, Y. Q., Becker, B.L., Zheng, T., & Carrick, H. J. (2014). An assessment of remote sensing algorithms for colored dissolved organic matter in complex freshwater environments. Remote Sensing of Environment, 140, 766–778.
Please cite this article as: Kutser, T., et al., Estimating lake carbon fractions from remote sensing data, Remote Sensing of Environment (2014), http://dx.doi.org/10.1016/j.rse.2014.05.020