We have built a new fisheye camera radiometer to measure the in-water spectral upwelling radiance distribution. This instrument measures the radiance distribution at six wavelengths and obtains a complete suite of measurements (6 spectral data images with associated dark images) in approximately 2 minutes (in clear water). This instrument is much smaller than previous instruments (0.3 m in diameter and 0.3 m long), decreasing the instrument self-shading. It also has improved performance resulting from enhanced sensor sensitivity and a more subtle lens rolloff effect. We describe the instrument, its characterization, and show data examples from both clear and turbid water.
© 2005 Optical Society of America
The radiance distribution is described by the collection of radiance data in all directions coming to a single point. Measurement of the upwelling radiance distribution in the ocean is difficult because of the number of observations required, the variability of the light field due to surface waves, and instrument self-shading, yet the upwelling radiance distribution is vitally important in ocean remote sensing. Ocean remote sensing algorithms have typically treated the ocean as a lambertian surface (radiance independent of viewing geometry). However, because the ocean reflectance varies with view geometry and illumination conditions, knowledge of the upwelling radiance distribution is required for accurate retrieval of oceanic parameters.
At least two previous instruments have used single radiometers, swept over various directions in the zenith and azimuth directions[2,3]. Tyler’s Pend Oreille data set is still one of the fundamental tabulated data sets in the literature. The Aas and Hojerslev data set was used to develop a model of the angular variation of the radiance distribution. Both of these data sets were collected at a single wavelength. Another method to measure the radiance distribution is the use of a fisheye camera lens and a camera. A fisheye lens maps an entire hemispherical field to the image plane using an equidistant projection. This technique was first developed for in-water use by Smith, Austin, and Tyler, using a photographic camera, and a photopic filter. This system allows the radiance distribution to be collected in two images (one upwelling - one downwelling), thus the entire radiance distribution can be obtained quickly, allowing rapid profiling if desired. The difficulty in this system was obtaining radiometric data from the photographic images. In addition, with a fisheye system, the lens radiometric effects must be taken into account, the main feature being lens-rolloff effects.
More recently a series of instruments have been built, for use in the water, based on this fisheye technique, but using electro-optic cameras and remotely controlled spectral filter changers. The first of these  included charge injection device (CID) electro-optic camera systems for both the upwelling and downwelling radiance distribution. This system consisted of three cans of instruments, with upwelling and downwelling systems in separate cans, along with a third can for the control electronics. The filter changer allowed selection of one of 4 spectral filters (25.4 mm interference filters), along with neutral density filters to adjust the overall sensitivity. These cameras were digitized with 8-bit frame grabbers, and did not have an intrinsically high dynamic range. However, by coating the dome/window of the downwelling camera for zenith angles less than 45 degrees, and taking into account the lens system rolloff characteristics, the dynamic range of the downwelling and upwelling radiance distribution could be accommodated. One useful characteristic of the CID architecture was that excess light in one pixel does not spread into neighboring pixels (bloom).
The next system in this series, RADS-II , was developed to take into account advances in camera development (including Charge Coupled Devices, CCD). This system used similar optics and two cooled CCD cameras (512×512 pixels). Both cameras and the control computer were contained in one housing. Along with these components, the system had tilt/roll detectors, a compass, and a multispectral up- and downwelling irradiance system (based on the MER-1032 system [Marine Environmental Radiometer, Biospherical Instruments]). The system housing was a cylinder approximately 0.5 m in diameter and 0.75 meter long. Communication with the system was by an RS-232 link. When the instrument was retrieved, an Ethernet link was used to download data via an ftp transfer to the host computer on deck. While the instrument was in the water, a remote-control program (PCHost) allowed the operator to control the embedded computer functions. The data from the cameras was digitized with a 16 bit digitizer, however system noise reduced overall dynamic range to approximately 12 bits. This instrument was used during several experiments, in particular to gather surface upwelling radiance distribution data [8,9].
Because of the interest in the surface upwelling radiance we found that we were only using the upwelling data from the system. To get accurate radiance distributions, without the effect of ship shadow, required floating the instrument away from the ship, which made it impossible to take upwelling/downwelling profiles or collect useful downwelling information. The size of the instrument, including the buoy to float it, caused a noticeable shadow effect in the images even in clear water . This effect had been seen with the earlier instrument  and was smaller than before, however it was still a noticeable perturbation.
Because of instrument self-shading issues, and the desire to collect only upwelling radiance distribution data, the NURADS instrument was developed. The system also has a filter changer, embedded computer and hard drive, CCD camera, and compass/tilt/roll sensor. However with advances in technology, this whole instrument package is only 0.3 m in diameter and 0.3 m long and also requires less floatation. The reduced self-shading and several other improvements have helped the instrument acquire more accurate radiance distribution data. This paper will describe this instrument, and provide some sample data from the instrument.
2. Fundamental NURADS instrument description
The instrument is based on an Apogee CCD array camera system (AP 260Ep) that uses the KAF-2610E CCD array. The camera housing includes the frame grabber electronics and interfaces directly with the embedded computer (Versalogic Panther, EPM-CPU-6) via a standard parallel interface. The frame grabber digitizes the images at 16 bit resolution, the final system dynamic range will be discussed below. Included in the instrument is a 20 Gbyte 2.5” hard drive on which the camera images and auxiliary data is stored. There is an embedded EZ-Compass-3 (Advanced Orientation Systems, Inc.) that collects the tilt/roll/compass information and transfers it to the computer over a serial interface. This information is logged to the hard drive continuously in the background during camera operation using the Windmill program (Windmill Software, Ltd.). The system uses a fisheye adapter lens, designed for a Nikon Coolpix 950 camera (FC-E8), a custom lens relay system, and a filter changer (MFW, Homeyer) (also controlled over a serial interface) to form the fisheye camera image and provide spectral filtering. The filter changer uses 25.4 mm interference filters, and allows the selection of one of 6 spectral filters for the image. The embedded computer is controlled by a surface laptop computer by use of the Timbuktu (Netopia) remote control software over an Ethernet link. While the data is stored on the embedded computer hard drive, once data collection is completed the data on the hard drive can be extracted via an ftp (file transfer protocol) transfer.
Because of speed advances in the camera frame grabbing technology, enhanced sensitivity of the CCD array, and faster optics, in clear water a complete set of data (6 light images, 6 dark images, one with each spectral filter) can be obtained in 2 minutes. The embedded computer controls the camera with a combination of MaxIM DL (Diffraction Limited), Microsoft Excel, and Microsoft Basic programs. In typical operation, the camera is run continually during the data collection period. By taking multiple data sets, data can be excluded where the camera tilt/roll exceeds some threshold (usually 5 degrees), clouds are determined to be causing a problem with the incident light field, or some other temporary artifact is in the image. Multiple images are also averaged to reduce the effect of bright light rays due to waves at the air-sea interface. In 30 minutes, 15 complete sets of data (approximately 90 Mbytes of data) can be acquired.
3. Characterization and calibration
The method to characterize and calibrate the fish eye systems has been described earlier . We will concentrate on the specific results for this system.
The basic characteristics of the system relevant to the radiance distribution measurements are the lens system rolloff function, camera linearity, camera noise characteristics (dark noise and readout noise), polarization sensitivity, and spectral calibration of filters/system. We will start with camera noise characteristics.
3.1 Camera noise characteristics
To see the background readout noise, and an indication of the dark noise level of the camera system, we averaged dark images (shutter closed) at several different integration times. The images were averaged, pixel-by-pixel to see if the variation was correlated with individual pixels, or was uniform across the array. Figure 3 shows a histogram of this pixel average (left) and the histogram for the standard deviation of the individual pixel averages (right). The Fig. below shows that the spread in integration times is very small for less than 2 sec, at 5 sec there is more variation. In terms of the standard deviation of the individual pixel averages, the average spread does not change significantly even at 5 sec integration. These two graphs imply that at longer integration times, there is some pixel-to-pixel variation in dark counts, but the noise does not increase. Thus subtracting a dark image from the data image is a better strategy then subtracting one overall average number. In our data collection we take a dark image for each data image, and subtract (pixel-by-pixel) this image from the data image. These graphs also show that the intrinsic noise in the system is on the order of 4 counts, thus the system is easily operated as a 14 bit system and pixel averaging can increase the effective dynamic range of the system.
3.2 Camera linearity
The CCD camera is inherently a very linear device, and this system was no exception. This test will not be discussed in detail here, but it is similar to the results seen in the earlier system , with the camera system displaying linearity (within 3%) over 3 orders of magnitude of incident flux.
3.3 Angular calibration
The natural projection of an ideal fisheye lens is a simple linear equation:
where θ (in degrees) is the angle from nadir, r is the radial distance, in pixels, from the center of the image, and K (in degrees/pixel) is a calibration constant found through calibration.
We use a hemispherical dome as the window in the instrument. If the system is constructed, and adjusted properly, the first principal plane of the optical system will be at the center of curvature of the hemispherical dome. At this position, rays which make it through the lens system will enter the dome perpendicular to the local dome surface, thus there will be little refraction, and the angular mapping of pixel location to a given radiance direction is straightforward. To check whether this is done properly we determine the calibration constant K in both air and water. If they are the same, then the instrument is set up properly. Before and after each deployment we do the angular calibration to help determine if any changes have happened in the optics. A typical example is shown below in Fig. 4, for an in-water calibration. The K derived from this graph was 0.469±0.004. The r2 for this regression was 0.9995.
A time series of angular calibrations for a NuRADS instruments is shown in Fig. 5. In this graph the in-water calibrations are shown as open circles, while the in-air calibrations are shown as filled circles. As can be seen, the angular calibration is both very stable, and the differences between the air and water values are small. In some later calibrations, only air or water calibrations were performed to verify instrument stability. The small variations around a constant value are partially real (due to small adjustments of instrument focus or the result of taking apart the lens system and putting it back together) and partially a small error in the calibration results. The small source used in the calibration, when imaged, takes up more than one pixel in the image, so there are errors in determining the true center pixel of the source image.
Note that because of the fisheye projection, the solid angle represented by each pixel varies, depending on nadir angle. Since θ(in radians) is θ=K(π/180) r, then dθ=K(π/180) dr.
The pixel area (dA) is given by dA=rdϕ dr, and solid angle is defined by dΩ=sin θ dθ dϕ. Thus dΩ represented by each pixel (dA) is:
At a nadir angle of 10 degrees, each pixel represents 7×10-5 sr, while at 70 degrees each pixel represents 5×10-5 sr.
3.4 Immersion calibration
The immersion calibration determines the difference in camera system response between measurements made in-air (where most of the calibrations are done) versus those in-water (where the desired measurements are). For radiometers with flat windows this is caused by the difference between air-glass and water-glass transmission and the index of refraction (n2) effect between air (inside the instrument) and water (outside). With the fisheye system, and the hemispherical dome window, there is another effect that offsets this n2 refraction effect. In essence the apparent aperture size of the system varies whether it is in-water or in-air. Hence, we must do a calibration to determine the immersion factor. The calibration is done in the following manner. The instrument is placed, dry, in a barrel and a reflectance plaque is suspended above this barrel at 45 degrees to the vertical. The plaque is illuminated by a 1000W FEL lamp. Images of the plaque are obtained as the water level in the tank is raised above the level of the dome window. Several measurements are made with different water levels, with the window submerged, to determine the water attenuation. The average of a 20×20 pixel area, centered on the plaque, is obtained at each measurement point. The attenuation coefficient of the water is determined from the measurements at the different water levels, and is used to correct for attenuation effects. The apparent radiance that the plaque should have at the front of the camera window is Lwater and Lair, when in water and air respectively. This can be calculated by compensating for the air-water interface effects, and water attenuation as:
Where e-cr is the attenuation from the surface of the water to the front of the dome window, n is the index of refraction of water, and Twater-air is the Fresnel transmission through the air-water interface.) Lwater and Lair, along with the pixel averages when the window was dry (#air) vs. wet (#water) are used to determine the immersion correction, M.
Typical immersion factors are on the order of 1.85.
3.5 Camera lens rolloff
The camera lens rolloff is the variation in system response due to the position in the field of view and includes the geometric effects of the entire lens system. Typically the system response will decrease as one moves from the center of the field of view, towards the edges. The total field of view of the camera is so large that there is not a uniform source to place in front of the system that will fill the entire field of view. Thus the rolloff data is acquired by sequentially imaging a stable source that fills a reasonable amount of the field of view and moving this source around the field of view. This is done in one of two ways. In the first method a reflectance plaque is the source, and data is acquired as this source moves in the field of view in two orthogonal axis. Two orthogonal axis are done to check the azimuthal symmetry of the system. In the second method, a 1 m integrating sphere is used as the source, and the camera is placed as close to the source as possible, without imaging the integrating sphere source ports inside the sphere. This fills a significantly larger portion of the field of view of the system, and with several rotations, allows the entire system to be sampled. In either case a rolloff function is fit to the resulting data vs. angle. Below we show in Fig. 6 both the rolloff function of the NURADS system, along with the rolloff of the previous system, RADS-II. The rolloff of the new system is much less severe than the old system. At 80°, the old rolloff factor was approximately 0.2, while the new system is on the order of 0.9.
The increase in this rolloff factor increases the measurement accuracy at large angles because the inverse of this number is used to correct the radiance images. Thus noise and other errors are only multiplied by a factor of 1.1 at 80° in the new system, rather than 5 as in the old system.
3.6 Spectral calibration
The spectral calibration was performed by measuring the filter transmission of each filter in a spectro-photometer. Since the filters are only nominally 10 nm wide, there are no other sharp features in the system that would significantly effect the calibration. Any well blocked, 2.54 cm interference filter can be used; the characteristics of the filters we chose are shown in Table 1. In addition to the blocking built into these filters, additional blocking was
provided to reduce the infrared response (Schott glass BG18), and Wratten filters for some of the red bands to provide extra blocking for the blue light. Since this instrument is designed to measure upwelling light at the surface, the blocking requirements for the blue light is less severe than a profiling instrument. After constructing the system, with the filter combinations to be used, tests are also performed in the laboratory (imaging an FEL illuminated reflectance plaque) and in full sunlight (imaging a sunlight illuminated reflectance plaque). In these tests, additional filters are placed between the camera and the plaque which either block the passband of the filter, or some other spectral region. Measurements taken in this way can determine the residual out of band response. In these tests, our filter combinations were sufficiently blocked to keep total integrated out of band response to less than 1% for both lamp and solar illuminated targets.
3.7 Polarization Sensitivity
The upwelling radiance distribution in the ocean is partially polarized, thus to make accurate measurements of the total radiance (unpolarized) it is important that the instrument be insensitive to the incident polarization. Measurements of the camera optical system have shown that the polarization sensitivity of the optical system is less than 1%. Our system window is a plastic dome, and while these may show stress bi-refringence  through photoelasticity, as long as the optical system behind the window is polarization insensitive, this will have no effect on the final measurement. We also enter the window at very close to normal incidence, hence Fresnel transmittance has no polarization effect. Thus our polarization sensitivity is less than 1%.
3.8 Absolute calibration
Finally an absolute calibration is performed. We are currently using a radiance source made of a reflecting plaque (99% spectralon plaque) and a 1000W FEL lamp. The history of calibration for one of the instruments is shown in Fig. 7. Note in this Fig., Filter 5 has almost exactly the same numerical values as Filter 4, thus the symbol is hidden. With these calibration coefficients, and an integration time of 1 sec, each count represents on the order of 30 pW cm-2 nm-1 sr-1. Full scale at 1 sec integration time is then 2 µW cm-2 nm-1 sr-1.
4. Sample data
4.1 clear water case
The data shown below was acquired on 10/22/2003 off of Honolulu, Hawaii. The solar zenith angle during these measurements was 38° (refracted in-water angle would be 27°). An image is shown for each wavelength. The water was very clear, the Chlorophyll concentration (Chl) was approximately 0.1 mg/m3.
In these clear water images several things are evident. The first feature is that the minimum in the radiance distribution is actually on the sun side of the nadir, not on the anti-solar side. While it is not as obvious in these false color images as it is in the grey scale images, the anti-solar point is evident as the point where the bright, refracted rays converge. The anti-solar point is also evident in the longer wavelength (red) images, as the place where the instrument self shadow is evident. Since the instrument is a cylinder, with the measurement window in the center of the bottom of the cylinder, the shadow actually extends towards the surface from the anti-solar point. In all the shorter wavelength images, even though the instrument is approximately 30 m or more from the ship, the ship hull is evident on the horizon. It is difficult to quantify the ship shadows magnitude, but (given that the ship was only 15m long) it probably has less than a 1% effect on the upwelling irradiance field. Finally, while the radiance distribution is not isotropic, the range of values from nadir to the horizon is limited to a factor of 3 or less.
These images are individual “snapshots” of the instantaneous upwelling radiance distribution, as such there are various other artifacts in the image. Some of these are evident as the line in the upper right of 412 nm (instrument cable), or some of the other bright spots on the right of the image that are harder to identify. At times we have seen fish in the images. In general we use averages of images when applying the data, in which case the individual artifacts are either masked out or disappear through averaging.
Two convenient simplifications, describing the shape of the upwelling radiance distribution, are the average cosine of the upwelling radiance distribution (µu) and Qu:
Where Eu is the upwelling irradiance, Eou is the upwelling scalar irradiance, and Lu is the nadir upwelling radiance. For an isotropic radiance distribution, Qu would be equal to π and µu would be equal to 0.5. Table 2 shows Qu and µu calculated from the images shown. As can be seen Qu is slightly higher than π, as reflected in the images by the brightening in radiance towards the horizon. µu is slightly less than 0.5. The variation in these factors reflects the variation in the pure water absorption, with Qu increasing towards the longer, red, more absorbed wavelengths. The last column is the Qu predicted by the model of Morel et al.. Our measured Qu is somewhat smaller than the model prediction, except at 616 nm. At 616 nm, instrument self-shadow is obvious in the measurement and may be decreasing Lu more than Eu. In all cases though the two values are within 10% of each other.
Also note that Lu decreases from the blue to the red reflecting the obvious blue color of the water.
4.2 Turbid water case
The next example is taken from a turbid water case 2 data set. This data was obtained in the Chesapeake Bay on 5/19/03. The solar zenith angle (in air) was 33 degrees. The chl was 24 mg/m3.
In these turbid case 2 waters, the radiance distribution has a lot more variation from nadir to the horizon. This is reflected in the Qu and µu factors shown in Table 3. The spectral variation of the parameters, and the underlying radiance distribution, reflects the influence of dissolved organic material in these turbid waters, case 2 waters. In particular, the increased absorption at the lower wavelengths causes Qu to be higher, and µu to be lower than in the clear water case. As the wavelength increases, total absorption decreases, and Qu decreases.
This continues until 616 nm, where the water absorption becomes significant and once again Qu increases. It can be seen that, as expected in this turbid coastal water, the maximum upwelling nadir radiance is in the green, with little light coming out at the blue wavelengths.
Also note the significant amount of nadir radiance at the red wavelength. These Qu’s are within the range shown in . However, the model suggested in  does not extend to these turbid waters, and there is not a good model with which to compare. In fact, this is one of the research areas for which this instrument will be used.
This instrument represents a significant advance in the measurement of the upwelling radiance distribution. The instrument has already been used during several cruises in Hawaii (clear water), the Cheasapeake Bay (turbid coastal water). There are two immediate areas in which data from this instrument will be applied. The first is to improve, or validate, BRDF models for remote sensing. As stated earlier, for remote sensing, the ocean is commonly assumed to be a lambertian surface (radiance independent of viewing geometry). This radiance distribution data can be used to empirically derive the correct variation of the satellite viewed radiance with viewing angle and to test existing and future models. These models are also used to predict Qu, and we are working on a model of Qu in turbid water. In addition, since the radiance distribution is fundamentally dependent on the backscattering phase function of the water, this data can be used to test models of the in-water light field and constrain this phase function.
This work was supported by NASA (NNG04HZ21C). We also thank Dennis Clark (NOAA/NESDIS) for his support in building these instruments. ONR (N000149910008) has also been supported our development of radiance distribution instruments.
References and Links
2. J. E. Tyler, “Radiance distribution as a function of depth in an underwater environment,” Bull. Scripps Inst. Oceanogr. 7, 363–41 (1960).
3. E. Aas and N. K. Hojerslev, “Analysis of underwater radiance distribution observations: apparent optical properties and analytical functions describing the angular radiance distributions,” J. Geophys. Res. 104, 8015–8024 (1999). [CrossRef]
4. K. Miyamoto, “Fish Eye Lens,” J. Opt. Soc. Am. 54, 1060–1061 (1964). [CrossRef]
6. K. J. Voss, “Electro-optic camera system for measurement of the underwater radiance distribution,” Opt. Eng. 28, 241–247 (1989).
7. K. J. Voss and A. L. Chapin , “Next generation in-water radiance distribution camera system,” in Ocean Optics XI, G. D. Gilbert, eds., Proc. SPIE1750, 384–387 (1992).
8. K. J. Voss, C. D. Mobley, L. K. Sundman, J. Ivey, and C. Mazell, “The spectral upwelling radiance distribution in optically shallow waters,” Limnol. Oceanogr. 48, 364–373 (2003). [CrossRef]
9. K. J. Voss and A. Morel, “Bidirectional reflectance function for oceanic waters with varying chlorophyll concentrations: measurements versus predictions,” Limnol. Oceanogr. 50, 698–705 (2005). [CrossRef]
10. J. P. Doyle and K. J. Voss, “3D Instrument Self-Shading effects on in-water multi-directional radiance measurements,” presented at Ocean Optics XV, Monaco, 16–20 Oct. 2000.
11. W. S. Helliwell, G. N. Sullivan, B. Macdonald, and K. J. Voss, “A finite-difference discrete-ordinate solution to the three dimensional radiative transfer equation,” Transport Theory and Statistical Physics 19, 333–356 (1990). [CrossRef]
12. K. J. Voss and G. Zibordi, “Radiometric and geometric calibration of a spectral electro-optic “fisheye’camera radiance distribution system,” J. Atmosph. and Ocean. Techn. 6, 652–662 (1989). [CrossRef]
13. F. L. Pedrotti and L. S. Pedrotti, Introduction to Optics (Prentice Hall, New Jersey, 1993).
14. J. Piskozub, “Effect of ship shadow on in-water irradiance measurements,” Oceanologia 46, 103–112 (2004).
15. A. Morel, D. Antoine, and B. Gentili, “Bidirectional reflectance of oceanic waters: accounting for Raman emission and varying particle scattering phase function,” Appl. Opt. 41, 6289–6306 (2002). [CrossRef] [PubMed]
16. G. Zibordi and J.-F. Berthon, “Relationships between the Q-factor and seawater optical properties in a coastal region,” Limnol. Oceanogr. 46, 1130–1140 (2001). [CrossRef]