McKenzie et al. (1998) suggests that the complex relationship between cloud cover and ultraviolet radiation (UV) make it impractical to derive a simple expression that accurately relates the two for all conditions. They report that correlations between cloud estimates and ratios of measured/modelled clear-sky UVB are much weaker when the sun is not obscured by cloud. For cloud estimates, less than 50 % there seems to be little effect of cloud on UVB. One needs to know whether the sun is obscured and the cloud optical depth.

What appears to be missing in the literature to date is a quantification of cloud in the vicinity of the sun. For example, would the proximity to the sun of the 50 % cloud mentioned by McKenzie et al. (1998) above effect the measured UV? Borkowski et al. (1977) quantified all-sky photographs using photometric measurements and applied an empirical orthogonal function analysis to the data. The method allowed representation of the cloud pattern as a composition of basic patterns to which a definite physical meaning could be attached. It was found that the distribution of coefficients of the first eigenvector was governed by the major distribution of clouds in the sky, while those for the second related to the highly variable cloud cover of the area close to zenith, with clouds of different thickness. Although no definitive conclusions could be made, their work indicated that not only the fraction of the sky covered by clouds, but also their distribution, determined the intensity of UV. Long & DeLuisi (1998) suggest that fractional cloud cover alone, except for the extreme of an overcast sky, cannot fully characterise the cloud field. For instance, a cloud cover of 50 % could be due to many small cumulus clouds, or having half the sky covered by a stratus deck. The effects on the downwelling irradiance are often quite different for these two conditions. Additional information provided by a cloud edge to area ratio allows further assessment of the cloud field.

At the present time there has been no reported study of the measurement of cloud properties in the vicinity of the sun using an automated satellite or ground based sky camera system. Cloud images near the sun can be recorded automatically with a sun-tracking camera. In order to process a vast number of images of cloud, it is desirable to automate the image analysis process. Long & DeLuisi (1998) describes the development and Wooldridge (1993) describes the development and evaluation of automated whole-sky camera systems to estimate cloud fraction retrievals. Davis et al. (1992) used colour slides as a source of cloud images. McKenzie et al. (1998), Schafer et al. (1996), and Borkowski et al. (1977) employed ground based whole-sky cameras to study cloud effects on UV at ground level.

Figure 1 The Sky Camera in the laboratory, showing two of the filters used to record cloud images in the vicinity of the sun.

In these studies the degree of solar obstruction and whole-sky cloud amount (not cloud near the sun), were determined visually (not automatically) from the images by subjective methods.

McKenzie et al. (1998) attempted to use a commercially available video camera (Harnett, 1992), however the restrictions of lack of colour, narrow field of view, and the necessity of shading the sensor from direct sunlight limited its suitability.

The sun centred sky camera was situated at Toowoomba, Australia (27.6 ° S, 696 m a.s.l.) at the campus of the University of Southern Queensland (Figure 1). Cloud measurements were recorded automatically by the sky camera and manually by an observer at the site over a period of 12 months (September 1997 to August 1998), at local noon each day. The sky camera system consisted of a standard video camera and a wide-angle lens. A stepper motor was used to point the camera in the direction of the sun in an east-west direction. The altitude angle of the camera was manually adjusted at the end of each week by approximately 2 ° to keep the sun in the centre of the field of view of the camera in a north-south direction. A second stepper motor was used to turn a filter wheel assembly, which was located above the camera lens. The first filter was opaque and ensured that the sun’s rays did not damage the CCD chip of the camera.

Figure 2 A Block diagram of the Sky Camera System showing the major components of camera, PC, controller, stepper motors and UNIX workstation.

A UNIX computer performed digital image processing. Figure 2 shows a block diagram of the sky camera system and the reader is referred to Sabburg & Wong (1997) for a full description.

The images had varying contrasts between the sun, sky, and clouds, and the sun was approximately in the centre of the image (Figure 3a). In the initial version of the image processing software, a non-linear approach was used and this method was evaluated by comparison to visual interpretation of the recorded sky camera images (Sabburg & Wong, 1998). The image processing software positioned the sun at the centre of the image and determined if the sun was obstructed or not. The sun was removed (subtracted), from the image and a measure of cloud cover around the sun was made (Figure 3b). The methodology of the new approach differs by using a linear algorithm that varied from the above method in the areas of filter selection, thresholding, handling bright cloud obstructing the sun and the underestimation of the area of dark cloud (Figure 3c).

Figure 3 a) A typical sky image (in this case with the sun blocked), recorded by the Sky Camera; b) The resulting sky image of a) after image processing with the non-linear algorithm and c) the resulting sky image of a) after image processing with the new linear algorithm. White is cloud and black is sky, except for the sun's disk which is subtracted from the image in case b).

A unique problem which was confronted when analysing the images of the present research, was determining the state of the sun’s disk (whether it was blocked or not). In most cloud imaging to date, the sun has been blocked by a ‘gooseneck’ or shadow band device, thus avoiding the imaging of the bright sun. The manual observer was trained prior to the study by comparison with official meteorological whole-sky cloud cover observations for reported 9 AM and 3 PM data, and was found to be accurate to ± 1 okta. The observer wore protective sunglasses and carefully determined if the sun was blocked by cloud and then estimated the cloud cover around the sun, in a similar field of view as the sky camera.

During the 12-month period, 231 images and 129 corresponding manual observations were collected at local noon, with a solar zenith angle (SZA) ranging from 4.2 to 51 °. Cloud type ranged from Cumulus (36 %), Stratus (9 %), Alto-Cumulus and Alto-Stratus (15 %), sky's containing Cirrus cloud (11 %) and the remainder being other cloud combinations. As in the case of the visual evaluation of the sky images produced by Sabburg & Wong (1998), the manual observer found difficulty in determining the state of the sun's disk and the amount of cloud cover in the presence of Cirrus clouds. Therefore, for the purposes of this validation, the presence of predominant Cirrus cloud cases has not been included. This left 115 cases with corresponding manual observations (Table 1). It should also be noted that for the purposes of this evaluation, partial obstruction of the sun was taken as disk obscured. The cloud amount percentage is that covered by cloud in the 75 ° field of view (FOV) of the sky camera centred on the sun.

Table 1 - Results of comparison between manual sky observation and algorithm derived cloud parameters for 115 images recorded at local noon near the sun.