1. Introduction
Remote sensing offers a particularly useful way to monitor water resources over different space and time scales. Satellite spatial resolution, on the order of 1 km, while useful in shelf seas, cannot be used in coastal areas and inland waters where shallow water and land borders contaminate the images. In such regions,
in situ or airborne mounted systems must be employed. At present, narrow-band, multi-spectral radiometers have been used to measure optical properties of natural waters to yield information on many water quality parameters, such as chlorophyll [
1–
7], mineral suspended sediments (MSS) [
8–
10] and yellow substance [
11–
13]. The use of relatively complex multi-spectral instruments may be costly, however. This current study explores the possibilities of employing a conventional digital camera, as an alternative low-cost technique, to estimate water composition from optical properties of the water surface.
The concept of using a conventional imaging camera as a remote sensing instrument is not unknown. The Kodak DCS460c
® digital camera, a camera designed for photojournalism, has been used for area mapping of vegetation [
14], and urban areas [
15]. Directional reflectance of grass and of pine trees was studied using a standard CCD camera, an Electrim EDC-1000C
® [
16]. The benefits of aerial imaging photography have also been applied to hydrographical research. A setup of four synchronous analogue K-17 cameras and filters mounted on an airplane was applied to measure the roughness of the sea surface from photographs of the sun’s glitter [
17]. Frontal systems in a coastal zone were studied using an airborne standard 35 mm camera [
18]. Ocean surface currents were accurately estimated using a digital camera system on an aircraft, by deriving the Doppler shift of gravity waves from the images [
19]. An airborne Kodak DCS460c was applied in studies of the dynamics of a river plume, discharging into coastal water [
20]. Cameras have been used
in situ as well, an example is the oceanographic camera system designed to measure underwater radiance distribution in natural waters [
21]. The system contained two analogue Nikon cameras placed back-to-back, each with a fisheye lens and could operate at depths up to 100 m. In more recent times, an
in situ Sony Mavica FD 83
® digital camera, with a pipe breaking the water surface attached to its lens, was used to estimate MSS levels in coastal waters [
20]. Operating a Nikon Coolpix 885 digital camera in a similar fashion, yellow substance and chlorophyll concentrations were successfully monitored in Galway Bay, Ireland [
22].
Employing an off the shelf digital camera as an
in situ optical instrument to assess water composition is a relatively novel technique [
20,
22]. The method described in those papers is the focus of this paper. We explored remote sensing algorithms known to calculate yellow substance and chlorophyll concentrations, as these are the optically active components that control the water colour in Galway Bay [
22]. The remote sensing algorithms we applied were based on observations of subsurface spectral reflectance
R(λi), defined by the ratio of up welling (ir)radiance and down welling (ir)radiance. Yellow substance absorption decreases exponentially with increasing wavelength from ultraviolet to near infrared wavelengths, giving water a yellow colour [
23]. Other names for yellow substance are: coloured dissolved organic matter or CDOM, aquatic or marine humus, gilvin and gelbstoff. To quantify the concentration of yellow substance, the absorption coefficient at 440 nm, ah(440), is commonly used [
23]. When yellow substance and water are the main light absorbers, the red/blue reflectance ratio will linearly increase with yellow substance concentration [
11–
13]:
Known remote sensing algorithms for assessing chlorophyll concentration, [Chl], are described by a log-log relationship with the blue/green reflectance ratio [
1,
2,
4,
5,
24]:
Chlorophyll absorption is low in the green, but high in the blue part of the spectrum, and hence the blue/green colour ratio of reflectance decreases with increasing chlorophyll concentration, if chlorophyll dominates the water colour.
The outline of this paper is as follows. First it is demonstrated how images of water leaving light were acquired using different set-ups, and how their RGB values were derived. In the following sections is described how the response spectra of the ECOshot and CP885 digital cameras were measured, and how the Ocean Colour Sensor, an in situ irradiance meter designed for ocean colour observations, was used. Next, the required lighting conditions for using the in situ camera are explained. In the final Methods section, details about the surveys in Galway Bay, Ireland, and in the Southern Rockall Trough in the North Atlantic, can be found. The response spectra of the two cameras are presented in Results. These spectra show that the digital cameras are fundamentally three-band radiometers. The observed relationships between RGB ratios and water quality parameters, obtained using the in situ CP885 during numerous surveys, are described in the following section. In the last Results section, the findings of the in situ CP885 and ECOshot are compared. The digital camera methodology and its results are evaluated in the Discussion, which includes a concise conclusion.
4. Discussion
It has been shown how ‘off the shelf’ digital cameras can be used as an optical instrument to quantify water quality parameters of marine waters. Measured response spectra of the CP885 and the ECOshot illustrate that these digital cameras can be regarded as three-band radiometers. The red, green and blue bands are broad and overlapping, similar to human vision [
32], but unlike more traditional narrow-band radiometers that are generally used in remote sensing of natural waters [
4,
38]. Despite this, known remote sensing algorithms, designed for more traditional radiometers, were successfully applied to
RGB values of digital pictures of water leaving light to quantify water quality parameters. In addition, observed
RGB values of the CP885 closely followed the responses in the corresponding red, green and cyan channels of the OCS spectral irradiance meter, while capturing water leaving light (
Figure 5). It was concluded that the simplification of the digital camera as a three-band radiometer was a valid first approximation, when explaining the observed relationships between
RGB values, obtained
in situ, and water quality.
The results of an initial survey in Galway Bay, when the CP885 was used with a tube breaking the water surface attached to its lens [
22], were successfully repeated in numerous subsequent surveys, under a wide range of lighting conditions. This included extremely low light levels, during the surveys in November and February, due to both low solar elevation angles and completely overcast skies. Surface yellow substance concentrations could be estimated from
R/B values using a general calibration (
Equation 6). Further research is needed to test its validity in water bodies other than Galway Bay. For the approximation of chlorophyll concentrations from
B/G values, survey specific calibrations would be recommended as it would normally improve the precision of the estimation (
Table 3b). The CP885 was deployed in a floating housing (
Figure 1b) in the Southern Rockall Trough in the North Atlantic. The strong relation between obtained
B/G values and SeaWIFS chlorophyll
a (
Figure 7) suggested that the camera could be used in clearer waters than in Galway Bay. The MSS concentrations in the sampled waters were too low, around 1 mg L
−1, to affect the optical signal in a detectable manner. A correspondence between
R/G and MSS, as in coastal waters off Arklow where MSS varied between 2 and 10 mg L
−1 [
20], was not retrieved. An effect of varying MSS levels on the remote sensing algorithms for yellow substance [
11] and chlorophyll detection [
24,
39] wasn’t observed either. It would be interesting to further investigate the effect of higher MSS concentrations on digital camera observations.
The variation in the relationships between
RGB ratios and water quality parameters (
Table 3) was not proven to be a consequence of the digital camera methodology, such as changing lighting conditions or white balancing effects. If anything, the variation in the relationships between
B/G and chlorophyll
, and between
R/B and yellow substance, demonstrated how sensitive the
in situ camera was to properties of the water body, i.e. to level of covariance between chlorophyll and yellow substance [
36], and plume depth (
Figure 6) respectively. The latter results highlighted the possibility of the
in situ camera’s use as a depth indicator.
The uncertainty in the estimation of yellow substance concentration using
R/B values was about ±0.2 m
−1. The natural log precision of the estimation of chlorophyll concentration from
B/G values varied between ±0.3 and ±0.4, implying an error of −25% to 35% and of −33% to 50% respectively. The magnitude of this error was big, but agreed with previous estimations of chlorophyll concentration using two band algorithms [
3,
24,
39,
40]. The statistical errors in the ah(440) and chlorophyll estimations could be justified through consideration of a number of factors: variation in the colour of natural sunlight (10% for any colour ratio), inhomogeneous vertical distributions of optically active components in the water column (e.g. varying plume depth), variation in the presence of the other water quality parameters, errors in measurements of ah(440) and chlorophyll concentration, and patchiness of the water. Tube depth variation during measurements was not found to contribute to significant errors in waters as clear as in Galway Bay. It was concluded that, except for colour changes of sunlight, the uncertainties were unrelated to the actual camera deployment technique.
The
RGB values of the
in situ CP885 and ECOshot cameras, obtained in the surface waters of Galway Bay, were mostly similar (
Figure 9). Subtle differences between the two cameras were partly explained by their different response spectra (
Figure 3), in combination with the spectral distribution of the captured water leaving light. The ratio of
G values obtained using the ECOshot, and using the CP885, was 1.04±0.04 on average. This ratio agreed with the estimated sensitivities of their green bands,
sg,ECOshot/sg,CP885, being 1.05 (
Equation 5,
Table 1). Using
Equation 5 to calculate similar relations for
in situ R and
B values was not successful. Despite the higher peak transmission of the ECOshot’s blue band (
Figure 3), its
B values didn’t exceed those obtained using the CP885. A possible explanation could be that the CP885’s centre wavelength of the blue band was 30 nm longer, closer to cyan, than the centre wavelength of the ECOshot’s blue band (
Table 1). OCS reflectance of the water surface in Galway Bay was two to three times higher in its cyan channel (490 nm), than in its blue channel (439 nm) [
41]. The higher
in situ ECOshot’s
R values, compared to corresponding
R values of the CP885, could not be simply explained by the different response spectra of the two cameras, indicating that more complex colour encoding was at work. The ECOshot’s red channel was narrower and its peak transmission lower, while the centre wavelengths of the red channels of the CP885 and ECOshot were equal (
Table 1). In addition, the response of the red channel to blue light, for the CP885, was not as strong in the ECOshot (
Figure 3). The differences between the two cameras were minor.
The proportionality between yellow substance and
R/B appeared to be similar for the ECOshot (
Equation 7) and the CP885 (
Equation 8), while the CP885 was slightly more sensitive to changes in chlorophyll concentration (
Equations 9&
10). It would appear that the CP885 is more effective in monitoring chlorophyll, because its blue band is closer to cyan (
Figure 3;
Table 1).