Cross-validation and robustness of daylight glare metrics

Abstract

This study evaluates the performance and robustness of 22 established and newly proposed glare prediction metrics. Experimental datasets of daylight-dominated workplaces in office-like test rooms were collected from studies by seven research groups in six different locations (Argentina, Denmark, Germany, Israel, Japan and the United States). The variability in experimental setups, locations and research teams allowed reliable evaluation of the performance and robustness of glare metrics for daylight-dominated workplaces. Independent statistical methods were applied to individual datasets and also to one combined dataset to evaluate the performance and robustness of the 22 glare metrics. As performance and robustness are not established in literature, we defined performance as: (1) the ability of the metric value to describe the glare scale (evaluated by Spearman rank correlation), and (2) the ability of the metric to distinguish between disturbing and non-disturbing situations (evaluated by diagnostic receiver operating characteristic curve analysis tests). Furthermore, we defined robustness as the ability of a metric to deliver meaningful results when applied to different datasets and to fail as few as possible statistical tests. Average Spearman rank correlation coefficients in the range of 0.55–0.60 as well as average prediction rates to distinguish between disturbing and non-disturbing glare of 70–75% for several of the metrics indicate their reliability. The results also show that metrics considering the saturation effect as a main input in their equation perform better and are more robust in daylight-dominated workplaces than purely contrast-based metrics or purely empirical metrics. In this study, the daylight glare probability (DGP) delivered the highest performance amongst the tested metrics and was also found to be the most robust. Future research should aim to optimise the terms of glare equations which combine contrast and saturation effects, such as DGP, PGSV or UGRexp, to achieve metrics that also perform reliably in dimmer lighting conditions than the ones explored in this study.

Reconstruction of cumulative daylight illumination fields from high dynamic range imaging: Theory, deployment and in-situ validation

“…16 Perhaps the most active area of applied research in recent years has been for the study of glare. 17,18 Usually this involves the generation of hemispherical fisheye views of potentially glare-inducing scenes, and then relating the physical measures of luminance in the HDR images to subjective assessments of glare. The derivation of incident illuminance using HDRI where the target surface is a uniform diffuse reflector which does not have strong colour saturation is fairly straightforward.…”

Discomfort Glare Cut-Off Values From Field and Laboratory Studies

“…For this study, the best metric of each type according to a recent cross-validation study (Wienold, 2019) was selected. The DGP (Equation 1) developed in (Wienold, 2006) was chosen as the contrast and saturation glare metric; the vertical illuminance (Ev) proposed in (Velds, 2000) as the saturation glare metric; and the CIE Glare Index (CGI) (Equation 2) developed in (Einhorn, 1979) as the contrast glare metric.…”

“…Also based on the ROC curve of the model for a specific borderline, several methods exist to define the best cut-off value, namely the one that offers the best discrimination of the discomfort glare metric values in agreement with the binary subjective assessments. For this study, three of these methods, described in (Wienold, 2019), were applied to derive the best cut-off value for each dataset, and the mean of the three results was used as the best cut-off value for a specific discomfort glare model and borderline. The three methods establish that the best cutoff value is the one that either minimises the distance from the ROC curve to the upper left corner, maximises the Youden index (Equation 3), or maximises a square fitting under the curve.…”

“…As an example, the DGP value representative of 50% of participants being disturbed by discomfort glare in field studies was evaluated as 0.2 (Hirning, 2014) or 0.24 (Mangkuto, 2017). However, from a meta-analysis study combining results from seven laboratory studies (Wienold, 2019), the DGP value derived as the cut-off value between noticeable and disturbing glare was 0.38. Although the cut-off values from the laboratory and field studies were not derived in the same manner, the gap between the values is considerable as it averages around 0.16.…”

Energy Performance And Occupancy-Based Analysis Of Visual And Thermal Comfort For Transmittance Level And Layout Variations Of Semi-Transparent Photovoltaics

“…The value was calculated at the eye level (1.1 m above floor) with a viewing direction towards the computer screen (parallel to the facade). The simplified DGP can be applied for this specific facade since the glazing transmission is rather high, and the saturation effect is dominating (Jan Wienold et al, 2019). As veiling glare threshold, a vertical illuminance on the computer screen of 2000 lux was used (Schierz, Vandahl, and Schmits, 2012).…”