Pixel-Based Image Processing for CIE Standard Sky Classification through ANN

Digital sky images are studied for the definition of sky conditions in accordance with the CIE Standard General Sky Guide. Likewise, adequate image-processing methods are analyzed that highlight key image information, prior to the application of Artificial Neural Network classification algorithms. Twenty-two image-processing methods are reviewed and applied to a broad and unbiased dataset of 1500 sky images recorded in Burgos, Spain, over an extensive experimental campaign. The dataset comprises one hundred images of each CIE standard sky type, previously classified from simultaneous sky scanner data. Color spaces, spectral features, and texture filters image-processing methods are applied. While the use of the traditional RGB color space for image-processing yielded good results (ANN accuracy equal to 86.6%), other color spaces, such as Hue Saturation Value (HSV), which may be more appropriate, increased the accuracy of their global classifications. The use of either the green or the blue monochromatic channels improved sky classification, both for the fifteen CIE standard sky types and for simpler classification into clear, partial, and overcast conditions. The main conclusion was that specific image-processing methods could improve ANN-algorithm accuracy, depending on the image information required for the classification problem.

Forecast accuracy of existing luminance-related spectral sky models and their practical implications for the assessment of the non-image-forming effectiveness of daylight

This paper analyses the forecast accuracy of current state-of-the-art, data-driven, spectral sky models. The aim is threefold: (i) to determine the forecast accuracy of existing spectral sky models based on a large dataset of spatially, spectrally and temporally resolved measurements, (ii) to investigate the practical implications of spectral forecast accuracies for the assessment of spectrally selective responses (here non-image-forming effects are expressed through melanopic irradiance) and (iii) to study if the use of spectral sky models is more appropriate to predict the non-image-forming effectiveness of daylight than the currently assumed CIE standard illuminant D65. The forecast analysis for CIE Standard Overcast Skies (CIE Sky Type 3) showed that the model published by Chain and colleagues in 1999 performed best, whereas the correlated colour temperature distribution can also be represented with the CIE standard illuminant D65. The analysis showed substantial discrepancies in the forecast for clear skies with low luminance turbidity (CIE Sky Type 12) depending on the correlated colour temperature range. Our findings suggest that for CIE 12 skies, even when simulating with the best performing spectral sky model, forecast inaccuracies affect the estimated non-image-forming effectiveness. Nonetheless, the assumption that the spectral distribution of daylight from a CIE 12 sky corresponds with the CIE standard illuminant D65 underestimates the non-image-forming effectiveness to a greater extent. The results advance the understanding of spectral characteristics of daylight and suggest that considering realistic spectral distributions instead of D65 will lead to a difference in the non-image-forming effectiveness assessment.

The New Method To Calibrate Iso/cie General Skies modelled In Artificial Skies

Recently sky luminance distributions under fifteen ISO/CIE sky types were standardised in relative terms, i.e. all elementary sky luminance are normalised by the zenith luminance in nature. However, these standard sky luminance patterns can be characterised also in their physical units, see in CIE215:20014 which can be reduced by suitable intensity scales to simulate the very high natural sky luminance in laboratory under artificial sky domes. For the design of the electric illumination system of the artificial sky the appropriate zenith luminance for each sky type has to be derived respecting the indicatrix and gradation functions of each sky type. Thus, the sky patterns simulated in the artificial sky can be calibrated either by luminance measurements or by analysing fish-eye photos taken from the sky centre showing the smoothness and gradual changes of homogeneous sky type luminance patterns. The sky luminance images can be utilised to proceed with the adjustment or dimming of the LED electrical system to the final calibration check containing zenith luminance and exterior illuminance in the second calibration step.

Photosynthetic Active Radiation, Solar Irradiance and the CIE Standard Sky Classification

Plant growth is directly related to levels of photosynthetic photon flux density, Qp. The improvement of plant-growth models therefore requires accurate estimations of the Qp parameter that is often indirectly calculated on the basis of its relationship with solar irradiation, RS, due to the scarcity of ground measurements of photosynthetic photon flux density. In this experimental campaign in Burgos, Spain, between April 2019 and January 2020, an average value of the Qp/Rs ratio is determined on the basis of measurements at ten-minute intervals. The most influential factor in the Qp/Rs ratio, over and above any daily or seasonal pattern, is the existence of overcast sky conditions. The CIE standard sky classification can be used to establish an unequivocal characterization of the cloudiness conditions of homogeneous skies. In this study, the relation between the CIE standard sky type and Qp/Rs is investigated. Its conclusions were that the Qp/Rs values, the average of which was 1.93±0.15 μmol·J−1, presented statistically significant differences for each CIE standard sky type. The overcast sky types presented the highest values of the ratio, while the clear sky categories presented the lowest and most dispersed values. During the experimental campaign, only two exceptions were noted for covered and partial covered sky-type categories, respectively, sky types 5 and 9. Their values were closer to those of categories classified as clear sky according to the CIE standard. Both categories presented high uniformity in terms of illumination.

Determination of luminance distribution under tropical sky conditions

Introduction. Natural illumination calculations depend on the sky luminance distribution. The most often used diagram of sky luminance in handbooks and guidelines is the luminance distribution in the cloudy sky proposed by Moon and Spencer. This concept actually includes the tropical areas of Vietnam, where the overcast sky and clear sky does not typically occur. To improve the calculation of natural illumination, it is necessary to determine the luminance distribution in the real sky. Materials and methods. In solving the research problem, the real sky types for Hanoi were identified using the 15 international standard sky types with their descriptions by lighting climate, which is provided using the method by R. Kittler. The descriptions are derived from the data on diffuse horizontal illumination Dv, extraterrestrial horizontal illumination Ev and light turbidity coefficient Tv. For a specific sky type, the standard parameters were selected for calculating the luminance distribution of the real sky. Results. The obtained results show that the typical sky type of Hanoi is the partly cloudy sky, no gradation towards zenith, with slight bleaching towards the Sun (type VI). The sky types from October to December are partly cloudy with the obscured Sun (type IX) and partly cloudy with the more luminant circumsolar area (type X). The study shows that the state of cloud coverage has a great influence on the level of diffuse horizontal illumination and luminance distribution under tropical sky conditions. Conclusions. It is revealed that the typical sky type for Hanoi is neither overcast nor clear sky. A typical sky with statistic dominance of cirrus and stratus clouds under effect of high solar radiation of Vietnamese tropical climate gives a high level of diffuse horizontal illuminance. The results show that the difference in luminance distribution between the CIE standard overcast sky and Kittler’s intermediate sky can be resolved at the angles of sky point elevation above horizon γ is higher than 50° with the relative errors below 10 %. In other words, the luminance distribution β of the considered sky type is significant for a system of side natural illumination.

Analysis algorithm for sky type and ice halo recognition in all-sky images

Halo displays, in particular the 22∘ halo, have been captured in long time series of images obtained from total sky imagers (TSIs) at various Atmospheric Radiation Measurement (ARM) sites. Halo displays form if smooth-faced hexagonal ice crystals are present in the optical path. We describe an image analysis algorithm for long time series of TSI images which scores images with respect to the presence of 22∘ halos. Each image is assigned an ice halo score (IHS) for 22∘ halos, as well as a photographic sky type (PST), which differentiates cirrostratus (PST-CS), partially cloudy (PST-PCL), cloudy (PST-CLD), or clear (PST-CLR) within a near-solar image analysis area. The color-resolved radial brightness behavior of the near-solar region is used to define the discriminant properties used to classify photographic sky type and assign an ice halo score. The scoring is based on the tools of multivariate Gaussian analysis applied to a standardized sun-centered image produced from the raw TSI image, following a series of calibrations, rotation, and coordinate transformation. The algorithm is trained based on a training set for each class of images. We present test results on halo observations and photographic sky type for the first 4 months of the year 2018, for TSI images obtained at the Southern Great Plains (SGP) ARM site. A detailed comparison of visual and algorithm scores for the month of March 2018 shows that the algorithm is about 90 % reliable in discriminating the four photographic sky types and identifies 86 % of all visual halos correctly. Numerous instances of halo appearances were identified for the period January through April 2018, with persistence times between 5 and 220 min. Varying by month, we found that between 9 % and 22 % of cirrostratus skies exhibited a full or partial 22∘ halo.

Analysis Algorithm for Sky Type and Ice Halo Recognition in All-Sky Images

Ice halos, in particular the 22° halo, have been captured in long-time series of images obtained from Total Sky Imagers (TSI) at various Atmospheric Radiation Measurement (ARM) sites. Ice halos form if smooth-faced hexagonal ice crystals are present in the optical path. We describe an image analysis algorithm for long-time series of TSI images which identifies images with 22° halos. Each image is assigned an ice halo score (IHS) for 22° halos, as well as a sky type score (STS), which differentiates cirrostratus (CS), partially cloudy (PCL), cloudy (CLD), or clear (CLR) within a near-solar analysis area. The colour-resolved radial brightness behaviour of the near-solar region is used to define the characteristic property spaces used for STS and IHS. The scoring is based on distance from a region in that property space, using tools of multivariate Gaussian analysis. An external expandable master table of characteristic properties allows continued training of the algorithm. Scores are assigned to the standardized sun-centred image produced from the raw TSI image after a series of calibrations, rotation, and coordinate transformation. We present test results on halo observations and sky type for the first four months of the year 2018, for TSI images obtained at the Southern Great Plains (SGP) ARM site. A detailed comparison of visual and algorithm scores for the month of March 2018 shows that the algorithm is about 90 % reliable in discriminating the four sky types, and identifies 86 % of all visual halos correctly. Numerous instances of halo appearances were identified for the period January through April 2018, with persistence times between 5 and 220 minutes. Varying by month, we found that between 9 % and 22 % of cirrostratus skies exhibited a full or partial 22° halo.