, Volume 8, Issue 6, pp 709–720

Using Satellite Remote Sensing to Estimate the Colored Dissolved Organic Matter Absorption Coefficient in Lakes


    • Limnology, Department of Ecology and EvolutionUniversity of Uppsala
  • Donald C. Pierson
    • Limnology, Department of Ecology and EvolutionUniversity of Uppsala
  • Lars Tranvik
    • Limnology, Department of Ecology and EvolutionUniversity of Uppsala
  • Anu Reinart
    • Limnology, Department of Ecology and EvolutionUniversity of Uppsala
  • Sebastian Sobek
    • Limnology, Department of Ecology and EvolutionUniversity of Uppsala
  • Kari Kallio
    • Finnish Environment Institute

DOI: 10.1007/s10021-003-0148-6

Cite this article as:
Kutser, T., Pierson, D.C., Tranvik, L. et al. Ecosystems (2005) 8: 709. doi:10.1007/s10021-003-0148-6


Given the importance of colored dissolved organic matter (CDOM) for the structure and function of lake ecosystems, a method that could estimate the amount of CDOM in lake waters over large geographic areas would be highly desirable. Satellite remote sensing has the potential to resolve this problem. We carried out model simulations to evaluate the suitability of different satellite sensors (Landsat, IKONOS, and the Advanced land Imager [ALI]) to map the amount of CDOM in concentration ranges that occur in boreal lakes of the Nordic countries. The results showed that the 8-bit radiometric resolution of Landsat 7 is not adequate when absorption by CDOM at 420 nm is higher than 3 m−1. On the other hand, the 16-bit radiometric resolution of ALI, a prototype of the next generation of Landsat, is suitable for mapping CDOM in a wider range of concentrations. An ALI image of southern Finland was acquired on 14, July 2002 and in situ measurements were carried out in 15 lakes (18 stations). The results showed that there is a high correlation (R2 = 0.84) between the 565 nm/660 nm ALI band ratio and the CDOM absorption coefficient in lakes. Analysis of 245 lakes in the acquired satellite image showed a normal distribution of CDOM concentration among the lakes. However, the size distribution of lakes was highly skewed toward small lakes, resulting in the CDOM concentration per unit lake area being skewed toward high values. We showed that remote sensing enables synoptic monitoring of the CDOM concentration in a large number of lakes and thus enables scaling up to the level of large ecosystems and biomes.


colored dissolved organic matterremote sensingdissolved organic carbonboreal lakesFinland


Most of the organic matter in the water column of lakes is generally in the dissolved form. The dissolved organic matter (DOM) in lakes consists to a large degree of colored humic substances, which are mainly imported from the terrestrial surroundings to the lakes. Because this complex organic matter is highly colored, the concentration of colored DOM (CDOM) generally correlates closely with the total concentration of dissolved organic carbon (DOC) (see for example, Tranvik 1990; Kallio 1999).

The CDOM imported from the drainage area structures the lake ecosystems in several ways. The light absorption of CDOM affects the acquisition of light by primary producers (Jones 1998). The CDOM efficiently absorbs short-wavelength solar radiation, which results in photochemical reactions as well as the protection of aquatic biota from UV-B induced damage. Moreover, imported terrestrial DOM constitutes a carbon source for heterotrophic bacteria within the lakes, thereby making them partly independent of the primary production of organic substrates within the lakes (Tranvik 1992; Jansson 1998). In addition, the microbial mobilization of energy associated with the imported DOM constitutes a subsidy to the entire aquatic food web and renders the aquatic system net heterotrophic. The net heterotrophic nature of lakes influenced by imported DOM results in supersaturation of the water by carbon dioxide (CO2) and consequently net export of CO2 to the atmosphere. Most lakes are supersaturated (Cole and others 1994), and the degree of supersaturation is to a large extent related to the concentration of DOC and CDOM in the water (Sobek and others 2003).

Because CDOM and associated compounds used by aquatic biota interact with lake ecosystem processes, variation among lakes in the concentration of CDOM has a major effect on ecosystem structure and function. Hence, lakes may range along DOM concentration gradients from being net autotrophic to net heterotrophic (Jansson and others 2000; Prairie and others 2002). Moreover, the concentration of DOM in freshwater environments has implications for the further transport downstream to the marine environment, where it may have a significant influence on coastal ecosystems.

The effect of global warming and climate change on the export of terrestrial DOM to aquatic environments is currently under debate. It has been argued that the concentration of DOM in lakes is regulated by either changing temperature per se (Freeman and others 2001) or climate-induced changes in the landscape–water balance (Tranvik and Jansson 2002). In particular, in the northern boreal zone, lakes have high concentrations of DOM, and there is a wide range in DOM concentration among lakes. To obtain an integrated view of the distribution and availability of DOM in lakes and to investigate the response of lake DOM concentrations to climate change, large-scale monitoring of lakes is needed.

Remote sensing could provide a means of making synoptic measurements of lake DOM over large geographic regions through the mapping of CDOM content in lakes. A successful remote sensing protocol would provide a highly valuable tool for the investigation of the standing stock of DOM at the regional or even the global scale and would make it possible to study seasonal and longer-term changes in DOM.

In this way, the overall impact of DOM on lake ecosystems and their role in biogeochemical processes, such as the export of CO2 to the atmosphere (Sobek and others 2003), could be determined. Information on the export of CO2 by lake systems would contribute to our understanding of the global carbon cycle and help to predict how future climate change will affect the CO2 flux from lakes. In addition, remote sensing would enable us to monitor fluctuations in the DOM concentration of lakes over large geographical areas.

In principle, the measurement of CDOM should be easily done by remote sensing, because in boreal lakes CDOM is the dominant absorbing compound and therefore has pronounced and predictable effects on the reflectance spectra. In practice, such measurements have been hindered by two limitations:
  • Lakes are often small, requiring measurements at high spatial resolution.

  • The high CDOM absorption associated with many boreal lakes results in low reflectance; therefore, satellite measurements of high radiometric sensitivity are required.

The amount of CDOM in lakes has been estimated from remote sensing data collected from boats (Vertucci and Likens. 1989; Arenz and others. 1996; Kutser and others 1998; Hirtle and Rencz 2003) or by using an airborne spectrometer (Kallio and others 2001). Hand-held or airborne spectrometers cannot provide global coverage of CDOM in boreal lakes. Ocean color sensors (for exmaple, SeaWiFS, MERIS) have radiometric sensitivities optimized for measurement in water, but their spatial resolution (1-km) is not adequate for boreal lake measurements. Land-observing satellite sensors (that is, Landsat, Spot) have a spatial resolution that is adequate for boreal lake measurement. However, the radiometric sensitivity of these sensors makes reliable mapping of CDOM in lakes questionable, especially in strongly absorbing lakes. A new generation of land observation sensors, such as the Advanced Land Imager (ALI), has improved spatial spectral and radiometric resolution. In this paper, we evaluate and apply these technical improvements in satellite technology to the synoptic mapping of CDOM in boreal lakes. Using the technique, we show the distribution of CDOM and DOC concentrations among lakes for an area in southern Finland.


Study Site and Image Data

In situ measurements were made in 13 lakes (18 stations) in southern Finland on 14–17, June 2002. A detailed description of the sampling methodology and analytical methods is provided by Strömbeck and Pierson (2001). After filtering through Whatman GF/F filters, water samples were analyzed for the concentration of chlorophyll a (Cchl) using the ISO (1992) standard method, which is based on measuring chlorophyll absorption in ethanol at wavelengths of 665 and 750 nm. For the concentration of suspended particulate inorganic matter (SPIM) (CSPIM), water was filtered through preweighed GF/F filters. To remove the organic particles, the filters were combusted at 550° C for 3 h. The amount of CDOM is expressed by the coefficient aCDOM(420) obtained from spectrophotometric measurements of filtered water (pore size 0.2 μm) in a 10-cm cuvette relative to a reference of distilled water. Samples for dissolved organic carbon (DOC) were passed through 0.2-μm filters (Gelman Supor) using acid-rinsed equipment (Norrman 1993) and measured on a Shimadzu TOC-5000 total carbon analyzer after acidification and purging of inorganic carbon. The samples were stored under dark and cool (4°C) conditions in sterile polypropylene centrifuge vials (Falcon).

The lakes were selected to cover a wide variety of optically different waters. For example, aCDOM(420) varied between 1.28–5.7 m−1; concentration of chlorophyll a (Cchl) varied between 2.0–33.0 μg L−1, and concentration of total suspended solids (CTSS) varied between 0.67–6.60 mg L−1. Concentrations of optically active substances and DOC from the lakes in our study are shown in Table 1. Because most of the studied lakes were relatively small (less than 650 ha), there was one measuring station in each lake. Lake Lohjanjärvi (6,734 ha), with 6 measuring stations, was the exception.
Table 1

Concentrations of Optically Active Substances and Other Characteristics of Finnish Lakes from Water Samples Collected during in situ Measurements









(μg L−1)

(mg L−1)

(mg L−1)

(mg L−1)

(mg L−1)

(m−1 [0.2 μm])































































































































CCHL, concentration of chlorophyll a; CTSS, concentration of total suspended solids; CSPIM, concentration of suspended particulate inorganic matter; CSPOM, concentration of suspended particulate organic matter; DOC, concentration of dissolved organic carbon; aCDOM(420), is absorption coefficient of filtrated water at wavelength 420 nm.

Satellite image acquisition by ALI and Hyperion sensors (both on an EO-1 platform) was attempted at the time of the in situ measurements, but the sky was too cloudy. A less cloudy image, suitable for image analysis, was acquired on July 14, 27 days after the field truth measurements.

The ALI sensor is a prototype of the next-generation Landsat sensor with improved spectral and radiometric resolution and substantial mass, volume, and cost savings. ALI spectral bands in the visible part of the spectrum are practically identical to other multispectral sensors, such as Landsat and IKONOS. ALI has 10 bands: a panchromatic (480–690 nm) with 10-m spatial resolution and nine spectral bands with 30-m spatial resolution (Table 2). The ALI footprint is 37 × 185 km. Hyperion is the first civilian hyperspectral instrument in space. It has 196 useful spectral bands (each approximately 10 nm wide) in the spectral range 430–2,400 nm. Radiometric resolution of these new sensors has also improved. Both ALI and Hyperion have 16-bit resolution versus the 8-bits used by Landsat or the 11-bit resolution of IKONOS, which means that the whole range of detectable signal is divided into 65,536 levels while digitizing the ALI signal, as compared to 256 levels with Landsat. This has important implications when using land-mapping satellites, which have band gains optimized for the high reflectance levels associated with land rather than open water surfaces.
Table 2

Details of Spectral, Spatial, and Radiometric Resolution of Satellite Sensors


ALI Wavelength (nm)

IKONOS Wavelength (nm)

Landsat 7 Wavelength (nm)

Spectral band



































Spectral band


















Spatial resolution




Radiometric resolution




Band gains are given in W m−2 μm−1 sr−1 DN−1, where DN is the digital counts registered by the sensor.

Atmospheric correction of the satellite measurements is critical to the success of aquatic remote sensing. A considerable amount (more than 90% in many cases) of the radiation detected by a satellite sensor is backscattered from the atmosphere without ever penetrating the water. This radiation must be removed before interpretation of the data. It is possible to use images that are not atmospherically corrected to estimate water characteristics, such as the concentration of DOC (Hirtle and Rencz 2003). However, those algorithms are strictly image-specific and cannot be applied to multiple images.

To derive an atmospheric correction that could be applied to the ALI image, we made use of the overlapping Hyperion image, which was acquired simultaneously with the ALI data. The Hyperion swath covers approximately one-fourth of the ALI image. Software package atmospheric correction of the Hyperion image was performed using the ENVI FLAASH (Research Systems Inc. 4090 Pearl East Circle Boulden CO 80301. FLAASH is designed for hyperspectral data and therefore well suited for the Hyperion images. The FLAASH module incorporates the MODTRAN 4 radiation transfer code with all MODTRAN atmosphere and aerosol types so that a unique solution for each image can be calculated. FLAASH also includes a correction for adjacency effects, provides an option to compute a scene-average visibility (aerosol/haze amount), and uses the most advanced techniques for the handling of particularly difficult atmospheric conditions (such as clouds). A midlatitude summer atmospheric model and rural aerosol were used in the FLAASH correction procedure. Remote sensing reflectances obtained from the corrected Hyperion image were then used to atmospherically correct the ALI images using an empirical line approach (see Moran and others 2001 and references within). It was assumed that the atmospheric conditions for the ALI image (37-km swath) were the same as those for the Hyperion image (7.7-km swath).

A relatively large part of the ALI image was covered with cumulus cloud and cloud shadows. Therefore, it would have been complicated to use the image to estimate the total number of lakes in the area. Instead, we used a Landsat 5 image mosaic, available from NASA (, to find the total number of lakes and their surface area in the territory covered by the ALI image. It was difficult to separate very small (few pixels) water bodies from the image processing noise that occurred when using band thresholds to separate water pixels from land pixels. Therefore, we removed all objects with one linear dimension less than 5 pixels (143 m). Visual investigation of the image showed that the number of real water bodies removed during this procedure was very small. Mean CDOM values were calculated for each lake in the ALI image that was not obscured by clouds, and the size of the lakes was estimated from the Landsat image. These data were used for the statistical analysis.

Model Description

A simple bio-optical model (Pierson and Strömbeck 2001), which is similar to others developed for oceanic and coastal waters (for example, Gordon and others 1988; Sathyendranath and others 1989; Kutser and others 2001), was used to simulate the reflectance spectra for the sampled lakes. Irradiance reflectance, or the ratio of upwelling irradiance (Eu) to downwelling irradiance (Ed) below the waters surface (0−), was calculated according to Kirk (1984):
$$ E_u/E_d(\lambda,0)- = (0.975 - 0.629\mu_0)b_b(\lambda )/(a(\lambda)+b_b(\lambda)) $$
where a is the total absorption coefficient, bb is the total backscattering coefficient, λ is the wavelength, and μ0 is the cosine of the solar zenith angle under the water’s surface.
Variations in a and bb were related to the concentrations of optically active substances in the lake water following Eqs. (2) and (3). The total absorption coefficient (m−1) was divided into four additive coefficients, which describe the absorption of water (aw), CDOM (aCDOM), detrital material (ad), and phytoplankton (aph):
$$ a(\lambda) = a_w(\lambda)+a_{CDOM}(\lambda)+a_d(\lambda)+a_{ph}(\lambda) $$
Total backscattering is related to three separate components: water (bbw), phytoplankton (bbph), and SPIM (bbSPIM):
$$ b_b(\lambda) = b_{bw}(\lambda)+b_{bph}(\lambda)+b_{bSPIM}(\lambda) $$

Details of the model equations and parameterization used to derive spectral variations in the substance-specific a and bb coefficients are given in Pierson and Strömbeck (2000, 2001). Radiance reflectance (Lu/Ed0−) is calculated from irradiance reflectance (Eq. [1]) by dividing the latter by the so-called Q factor, which is a wavelength-independent estimate of the ratio of upwelling irradiance (Eu) to upwelling radiance (Lu). Kirk (1994) suggests that the values of Q range between 3.14 and 5. Based on simulations in Swedish lakes (Pierson and Strömbeck 2000), we have estimated Q to have a value of 3.6. Upwelling radiance just below the water surface [Lu(λ,0−)] was estimated by multiplying Lu/Ed(λ, 0-) by a spectra of Ed(λ, 0−) typical for Swedish waters in midsummer. To estimate water-leaving radiance Lw(λ, 0+), Lu(λ, 0−) was divided by a factor of 1.815; this accounts for changes in the angular distribution of the radiance, which results from differences in the refractive index of water versus air (Dekker 1993).

Simulation of Satellite Sensor Response

The response of the different satellite sensors was based on the estimated spectral variations in Lw(λ,0+). For Landsat 7, the Lw values associated with each sensor band were estimated by normalizing the Lw(λ, 0+) spectra to the spectral response for each band, which was obtained from the NASA Landsat 7 Science Data Users Handbook ( Similar spectral response data for the ALI sensor were obtained from the Earth Observing EO-1 Web page ( For IKONOS, a uniform spectral response was assumed for the bandwidths given in Table 2, that is, for band 1, the response at 449 nm was 0, whereas at 450 nm it was 1.

The digital response of the sensor to the simulated radiance in each sensor band was estimated as:
$$ DN(band) = L_w (band)/Gain(band) $$
where DN is the digital counts registered by the sensor and Gain(band) is the unit of radiance measured per digital count, which is in turn a function of the radiometric sensitivity of the sensor and the number of bits in the data format (Table 2). DN values calculated by Eq. (4) were rounded to nearest integer values. Band gains for Landsat 7 were calculated as described in the Landsat 7 Science Data Users Handbook; band gains for IKONOS were obtained from Space Imaging Inc. ( Gain values for ALI were taken from the US Geological Survey description of the sensor (

To simulate the response of each sensor, three separate synthetic data sets were created from three sets of 500 simulations driven by random variations in the optically active substances (chlorophyll, suspended particulate inorganic matter, SPIM, and CDOM). Variations in CDOM and chlorophyll were based on gamma frequency distributions, which were found to fit a series of measurements made in Swedish boreal lakes (Sobek and others 2003). No direct data were available on the SPIM concentration of these lakes. We therefore chose a SPIM distribution that was normally distributed, with a mean and standard deviation of 0.4 mg L−1. This mean is somewhat greater than that measured by us in Lake Vättern and somewhat less than that measured in Stockholm’s archipelago (Strömbeck 2001), Swedish waters that are not highly turbid or influenced by river-borne SPIM or sediment resuspension. Sensor-specific response to variations in CDOM typical of Swedish boreal lakes was calculated based on the simulated variations in CDOM and the DN response of the bands of the sensors under investigation (Landsat, IKONOS, ALI).


Radiometric Sensitivity Effects

Modeled spectral variations in the total absorption coefficient (Eq. [2]) over a range in CDOM concentrations expected for Swedish lakes are shown in Figure 1. Spectral bands of multispectral sensors (Landsat, IKONOS, ALI) are also shown. Given the relatively high absorption coefficient of CDOM in boreal lakes, both bands 1 and 2 (Table 2) will be influenced by variations in CDOM absorption. In band 1, the water-leaving radiance is generally low due to relatively high amounts of CDOM; it is hardly detectable by any remote sensing instrument, and particularly not by satellite-borne instruments designed for land applications. A proper atmospheric correction of band 1 is also the most difficult to achieve. For example, Hirtle and Rencz (2003) found that band 1 reflectance was zero for almost every pixel in an atmospherically corrected Landsat 7 image of Nova Scotia, Canada. Phytoplankton concentrations are typically low in boreal lakes due to low nutrients and a low level of photosynthetically available light due to the absorption of relatively large amounts of CDOM. However, phytoplankton blooms do occur periodically, leading to high chlorophyll concentrations, as measured at some of our study sites (Table 1). The signal in band 1 is influenced more strongly by the absorption of chlorophyll than band 2. For all these reasons, we judged band 2 to be more suitable than band 1 for estimating CDOM concentration. On the third band (630–690 nm), CDOM has only minor effect, and variations in water-leaving radiance will be most influenced by backscattering. Hence, band 3 can be used as a reference. The results of model simulations (Figure 2) clearly show that radiometric resolution is the primary factor limiting our ability to measure CDOM in boreal lakes using multispectral satellites. There is good correlation between the band 2 band 3 ratio and the CDOM absorption coefficient at 420 nm (Figure 2a), which can be described by a power function.
Figure 1

Spectral variations in the total absorption coefficient over an expected range of concentrations of colored dissolved organic matter (CDOM) in Swedish lakes. The shaded intervals represent the range in which a given percent of 500 simulations fell. The approximate positions of multispectral sensor (Landsat, IKONOS, ALI) bands are shown.
Figure 2

Correlation between values of colored dissolved organic matter (CDOM) (measured as absorption coefficient of water samples at 420 nm) and band2/band3 ratios as derived from the modeling results. The same CDOM distribution as used in Figure 1 is used here. a Ratios based on floating point radiance values. b Ratios based on 8-bit radiometric resolution of Landsat. c Ratios based on 11-bit radiometric resolution of IKONOS. d Ratios based on 16-bit radiometric resolution of ALI. The dashed horizontal line shows the mean CDOM absorption coefficient of 3,464 Swedish lakes based on the inventory of Swedish lakes (Swedish Agricultural University 2000).

However, satellites with low radiometric resolution cannot reliably measure CDOM absorption in lakes. Because land remote sensing instruments are adjusted to measure relatively bright targets, the low water-leaving radiance values are spread over too few digital numbers when using the Landsat 8-bit (256 digital level) format. The effect of insufficient digital resolution is clearly seen in Figure 2b when the data points are recalculated for the 8-bit Landsat format. Band ratios for water bodies with high CDOM have only a few discrete values. Even lakes that have CDOM absorption coefficient values substantially lower than the mean value for Swedish and Finnish lakes are within the range where the 8-bit Landsat data fail to reflect variations in CDOM. The 11-bit instrument (IKONOS) is more suitable for estimating the amount of CDOM in boreal lakes (Figure 2C). The IKONOS simulations provide a much more useful algorithm, but there is considerable uncertainty at the highest CDOM levels. The best data are those from ALI (Figure 2d), where the simulated sensor response is nearly identical to the results with a floating point format (Figure 2a). Because the cost of IKONOS images precludes their use for mapping boreal lake CDOM concentrations on a global scale, and because our simulations suggest that ALI is best suited for measuring CDOM, we decided to see how effective ALI imagery would be for the mapping of lake CDOM concentrations.

Atmospheric correction of the ALI image

The empirical line method requires knowledge about reflectance of at least one target within an image (Moran and others 2001). This target reflectance may be measured during image acquisition or estimated from historical data. We had ALI and Hyperion images that had been acquired simultaneously, and the Hyperion swath covered a quarter of the ALI image. The Hyperion image was corrected using the FLAASH atmospheric correction package in ENVI.

After the atmospheric correction we assumed that the remote sensing reflectance spectra obtained from the Hyperion image were real reflectances of different objects. Targets with different reflectances (lakes, sea, a turbid bay, cyanobacterial bloom at sea, fields, road, forest) were selected to cover the entire range of reflectance values available in the image (excluding clouds). Average reflectance spectra were calculated from the Hyperion image for a small area of each target, and then were compared to corresponding average radiance spectra obtained from the atmospherically uncorrected Hyperion image. These data covered the spectral width of the ALI bands, and empirical algorithms were found for each ALI band that predicted atmospherically corrected reflectance from the top of the atmosphere radiance values. These linear relationships (Figure 3) were used to derive remote sensing reflectance values from ALI radiances and thus to perform the atmospheric correction.
Figure 3

Equations for atmospheric correction of ALI images in band 2 (565 nm, left graph) and band 3 (660 nm, right graph). The equations are derived from radiometrically (x-axis) and atmospherically corrected (y-axis) Hyperion images using the empirical line method.

Developing the CDOM Estimation Algorithm

There was almost a month between the ground truth measurements and the acquisition of the ALI and Hyperion images. Although variations in CDOM over monthly time scales during summer months have sometimes been found to be small (for example, see Erm and others 2002; Arst 2003), the time lag may have introduced noise into the correlation between remotely sensed data and ground truth. In a previous study of lakes in northern Michigan (Pace and Cole 2002), the variation in CDOM was found to be synchronous at a regional level. The synchrony was most pronounced for lakes with visible outlets and high CDOM content. These lakes were judged to have shorter water retention time and closer hydrological connections to the drainage system, and are probably most closely related to the relatively high CDOM lakes typical of the area covered in this study. If variation in CDOM was also synchronous in our study area, the possible bias due to the time lag between ground truth sampling and image acquisition may to a large extent be systematic rather than random. Absorption of CDOM (samples filtered with GF/F) was measured in Lake Lohjanjärvi by the Finnish Environment Institute a month after our field trip and just 2 days after the ALI and Hyperion images were acquired. Parallel measurements for June and July were available for five measuring stations in this particular lake. Comparison of these two data sets showed that the concentration of CDOM changed little over time (Figure 4). The correlation between CDOM values from our ground truth measurements in June and by the Finish Environmental Institute in July is high (R2 = 0.90), and the correspondence between sample sets approaches a one-to-one relationship.
Figure 4

Correlation between the aCDOM(420) measured on 14 June 2002 (a month prior to acquisition of the ALI and Hyperion images) and 16 July (2 days after the image acquisition).

For each lake in Table 1, a few pixels (between six and 50) from relatively homogenous areas (near the ground truth sampling site, when possible) were selected from the atmospherically corrected ALI image. An average reflectance spectrum was calculated for the collected pixels of every measuring station, and band 2/band 3 ratios were calculated from the average reflectance spectra. The relationship between the band 2/band 3 ratio calculated from the atmospherically corrected ALI image and CDOM absorption coefficient is similar to that shown in Figure 2d and can be described with a power function:
$$ {\rm a_{CDOM}}(420) = 5.20{\rm x}^{-2.76} $$
where x is the band 2 and 3 ratio (R2 = 0.84). The correlation between aCDOM(420) measured from water samples and estimated from the ALI image is shown in Figure 5. Equation (5) was also used to produce a map of CDOM absorption coefficients at 420 nm shown in Figure 6. The estimated CDOM absorption coefficients in the lakes showed good agreement with our in situ measurement data as was shown in Figure 5 and with typical absorption coefficient values for lakes in the ALI image where historical data about typical CDOM values (Sipelgas and others 2003) or water color (CDOM determined by comparison with standard cobalt chloride disk) (ISO 1994) are available. The CDOM absorption coefficient values in the Gulf of Finland fit well with measurements in the same area 2 days after the image acquisition (L. Sipelgas personal communication).
Figure 5

Correlation between aCDOM(420) measured in situ and estimated from the atmospherically corrected ALI image acquired on 14, July 2002. The dashed line indicates the one-to-one relationship.
Figure 6

Map of southern Finland and the Gulf of Finland derived from the ALI image acquired in 14 July, 2002 showing colored dissolved organic matter (CDOM). The algorithm shown in Figure 4 was used to calculate the CDOM absorption coefficient values from the atmospherically corrected ALI image. The numbers next to the CDOM color bar indicate absorption coefficients by CDOM at 420 nm (m−1). The numbers next to the dissolved organic carbon (DOC) color bar indicate concentrations of DOC in mg L−1 and were calculated from the CDOM values using the equation DOC = 0.9819x + 5.6831 (R2 = 0.78). The Finnish coastline is indicated with a white line.

Variations in the amount of phytoplankton and suspended matter influence the reflectance in bands 2 and 3. Therefore, other optically active substances might also influence the correlation shown in Figure 5. However, the correlation between the chlorophyll concentration and the amount of CDOM in the lakes is low (R2 = 0.47), and the amount of total suspended solids varies independently of CDOM in the lakes (R2 = 0.22). Thus, the band ratio algorithm appears to successfully describe the actual variations in the amounts of CDOM in our study lakes.

Concentrations of CDOM and DOC (Tranvik 1990; Kortelainen 1993; Kallio 1999) are strongly correlated in lake waters; the Finnish lakes in our study showed a similar strong correlation (R2 = 0.78). In a survey of 987 randomly selected lakes throughout Finland, Kortelainen (1993) found a similar correlation between total organic carbon and water color (R2 = 0.86). The correlation between CDOM and DOC allows us to develop an algorithm for estimating also DOC values from the satellite data. However, the correlation between DOC values estimated from the ALI data and from measured water samples was only moderate (R2 = 0.58). This is not surprising, because the specific absorbance of DOC varies among lakes. Different sources of DOC (McKnight and Aiken 1998), as well as different extent of in-lake photochemical and microbial transformations of DOC (Molot and Dillon 1997) result in different aromaticity and absorption properties of the organic matter.

Hirtle and Rencz (2003) have developed algorithms to estimate log(DOC) concentrations in lakes using Landsat data. The algorithms were for both atmospherically uncorrected Landsat data (R2 = 0.72) and atmospherically corrected data (R2 = 0.65). We applied the same algorithms to our ALI data, because ALI and Landsat have almost identical spectral bands. There was no correlation between the estimated and measured log(DOC) values. There are several possible explanations for this findings.

First of all, the relationship between CDOM and DOC concentrations in Nova Scotia may be different from the one that we observed in Finland. Hirtle and Rencz did not measure CDOM absorption in lakes (or at any rate did not present the data in their paper). They simply used statistics to find the correlation between the signal measured by Landsat (band 2) and the DOC values. Radiance measured by a remote sensor (water color) is a direct consequence of the absorption and backscattering of incoming light caused by optically active substances such as phytoplankton pigments, suspended matter, and CDOM. The absorption of light by CDOM is usually the dominant process in the formation of water color in boreal lakes. Part of DOC does not absorb or scatter light and therefore has no effect on the signal detected by a satellite. Thus, the amount DOC can be estimated from satellites if there is a strong correlation between the CDOM and DOC.

The concentration of DOC in the lakes studied by Hirtle and Rencz was relatively low (between 2.8 and 6.4 mg L−1), with just two lakes having high DOC concentrations (9.2 and 13.1 mg L−1 respectively). The low DOC range was suitable for the 8-bit radiometric resolution of Landsat, but the algorithm developed on the basis of these data was not suitable for our more DOC-rich lakes in Finland (6.0 greater than DOC greater than 12.3 mg L−1).

We were able to identify 538 lakes in the area covered by the ALI image using the Landsat mosaic. Lake size varied between 1.8 and 6,734 ha. Their size distribution was similar to that reported by Raatikainen and Kuusisto (1990) for 5,556,012 Finnish lakes with sizes ranging from 0.01 km2 to more than 1,000 km2. Two hundred forty-five of these lakes are shown, at least in part, on the CDOM map in Figure 6. The mean CDOM value was calculated for every lake from the ALI image and the size of each lake was estimated from the Landsat image. The CDOM absorption coefficient of the 245 lakes has a normal distribution, with a mean CDOM absorption coefficient (at 420 nm) of 3.70 m−1 (Figure 7a). However, because most of these lakes are small (Figure 7b), the distribution of absorption coefficient per unit lake area from the image is skewed toward higher absorption coefficient values (mean absorption coefficient per ha lake area 4.20) (Figure 7). This reflects the fact that most of the colored organic matter originates in the watershed. Hence, the ratio of catchment area to lake area (drainage ratio) is positively correlated to the concentration of DOC in lake water (see, for example, Rasmussen and others 1989; Kortelainen 1993). Accordingly, because small lakes tend to have higher drainage ratios, lake size is generally inversely related to absorption coefficient and DOC concentration (Sobek and others 2003). On the other hand, large lakes generally have lower concentrations of CDOM, due to generally lower drainage ratios and longer water retention times, which implies more extensive in-lake degradation of imported CDOM and a higher proportion of less-colored DOM derived from phytoplankton (Curtis 1998). Other parameters of importance for the CDOM concentrations in lakes include land use [for example, the fraction of wetlands in the catchments, Kortelainen (1993)] and annual runoff (Mulholland 2003). By combining the method presented here for the determination of CDOM with other remote sensing and geographic information system (GIS) approaches, we expect to be able to incorporate these and other parameters into studies that will further evaluate the role of terrestrial organic substances in the watershed-lake-atmosphere system.
Figure 7

Histograms of absorption coefficient of colored dissolved organic matter (CDOM) (A) and lake size (B) in southern Finland obtained by combining the CDOM map derived from the ALI image (Figure 6) and the lake map derived from the Landsat mosaic. C Histogram of the CDOM absorption coefficient per lake area.


This work was supported by the Swedish Research Council for Environment, Agricultural Sciences, and Spatial Planning (FORMAS).

Copyright information

© Springer Science+Business Media, Inc. 2005