Comparison of Simulation Methods for Glare Risk Assessment with Roller Shades

Abstract

Daylight discomfort glare evaluation is important when selecting shading properties. New standards recommend allowable glare frequency limits but do not specify the modeling accuracy required for annual glare risk assessment. Fast simulation tools allow users to perform hourly glare evaluations within minutes. However, reliable evaluation of glare through roller shades requires accurate modeling of their specular and diffuse transmission characteristics, affected by color, materials, and weaving technology. This study presents a systematic comparison between commonly used glare simulation methods against the “ground truth” Radiance ray-tracing tool rpict in terms of hourly daylight glare probability (DGP), hourly vertical illuminance (Ev), and annual visual discomfort frequency. The results are presented for two shade fabrics using light transmission models with and without a peak extraction algorithm (Radiance–aBSDF and Radiance–BSDF) for the specular component. The impact of sky/sun discretization on glare prediction is also discussed. The results show that the Radiance 5–Phase Method (5PM) is superior when modeling direct sunlight and DGP through shades, while other investigated methods (3–Phase Method, imageless DGP, ClimateStudio Annual Glare) are not as robust for that purpose. Users are encouraged to understand the underlying assumptions in the imageless methods to avoid errors when simulating glare, especially due to the contrast effects.

1. Introduction

1.1. Glare Risk Assessment

In the process of selecting roller shades, designers have good intuitions about subjective design criteria like color and appearance of fabrics, but often lack an understanding of how to properly quantify and mitigate direct sunlight and glare, especially for occupants seated near windows [1]. Simulating interior lighting and luminance distributions with fabric shades helps designers and practitioners quantify their performance and select shade properties that admit sufficient daylight with a reduced risk of glare. Discomfort glare is caused either by excessive light or contrast caused by bright sources in the field of view (such as a bright window surface). Although there is no universal way to predict glare, daylight glare probability (DGP) has been recognized as the most robust metric for evaluating glare [1,2]. DGP considers factors that may cause adaptation- and/or contrast-based discomfort glare; the adaptation (saturation) term in the DGP equation is based on vertical illuminance, and the contrast term depends on the intensity and size of glare sources with respect to their position and the background luminance. Typically, a luminance map that captures the occupant’s field of view is required to detect and evaluate information of glare sources including their intensity, size, and position using Evalglare [3,4] in order to fully calculate DGP. The Radiance ray-tracing tool rpict is often used as a reference “ground truth” simulation for validating other simulation methods [5,6,7,8,9]; rpict takes minutes to hours to simulate a point-in-time image depending on the level of fidelity.

The complexity and lengthy time of simulating luminance maps have hindered a wide adoption of image-based DGP annual analysis for evaluating visual discomfort. In addition, glare risk assessment not only depends on space configuration and sky conditions, but also relies on the relative position and view direction of the occupant(s). Estimating DGP from simulated luminance maps at numerous locations and views in a space may thoroughly quantify the amount of glare perceived by occupants [10], but that is practically infeasible using “conventional” ray-tracing tools such as rpict or even matrix-based daylight modeling methods. Several faster methods for predicting image-based glare have been proposed, including Accelerad [11] and Raytraverse [9]. Accelerad takes advantage of GPU to enable faster rpict and matrix-based daylight simulation, but it does not support the Radiance material types Radiance BSDF and aBSDF yet. Raytraverse allows efficient daylight modeling by employing adaptive sampling in the light field and simulating only the important features; it has been validated against rpict across several complex fenestration systems but requires users to be familiar with the Linux operating system. EN 17037:2018 “Daylight of Buildings” Standard (EN 17037) [12] recommends that annual visual discomfort frequency assessed at the most critical views (time percentage during which DGP exceeds a predefined threshold) should not exceed 5% of the working time in a year. Several studies have applied frequency limits to select maximum shade properties for glare protection using vertical illuminance (Ev) or enhanced simplified DGP (sDGP) [13,14]. When multiple view directions are considered, spatial glare autonomy (sGA) and spatial disturbing glare (sDG) can be used to quantify spatial–temporal variations in DGP. sGA is the fraction of views in a space that are visually comfortable (visual discomfort frequency < 5% where the DGP threshold is 0.4), while sDG is the fraction of views that fail to provide visual comfort (visual discomfort frequency > 5%, DGP threshold is 0.38). Ideally, occupants anywhere in a space should not experience discomfort glare for a significant amount of time, so sDG and sGA should be targeted at 0 and 100, respectively. It is tricky to use spatial–temporal metrics to make design decisions; for example, spaces with persistent intolerant glare for some critical views may still result in low sDG.

1.2. Imageless Methods

To facilitate fast glare analysis and spatial mapping, new methods including imageless DGP (iDGP) [15], available in Honeybee (HB) [16] and Annual Glare tool in ClimateStudio (CS) [17], have been developed to conveniently allow users to perform annual DGP simulations within a few seconds. These tools support a growing interest in daylighting and building design that provide visual comfort, defined by glare, daylight availability, daylight uniformity, and view out. Both imageless DGP methods are presented as plugins within the Rhino v.7 3D modeling software [18], allowing a seamless exchange in building information and simulation feedback, for users of any level of expertise in daylighting. To compare and select design options, sDG evaluated by CS Annual Glare has been used as an objective function or a design criterion together with other visual comfort metrics [19,20,21] and energy metrics [22,23,24,25,26]. Investigators who have used iDGP for evaluating sGA have also considered other design criteria including visual comfort [27,28], energy savings [29], and thermal comfort [30].

Table 1. Summary of studies that use ClimateStudio (CS) Annual Glare and Honeybee imageless DGP.

References	Investigated Systems	Imageless Method
[26]	Glazing	CS Annual Glare
[19]	Responsive façade modules	CS Annual Glare
[24]	Translucent–photoluminescent coating for glazing	CS Annual Glare
[31]	Building massing	CS Annual Glare
[32]	Building massing and façades (including exterior shades)	CS Annual Glare
[20]	Building massing, skylights, and exterior shades	CS Annual Glare
[21]	Top openings	CS Annual Glare
[25]	Tabular daylight devices	CS Annual Glare
[23]	Skylights	CS Annual Glare
[28]	Exterior shades	iDGP
[27]	Skylights	HB–iDGP
[29]	Exterior shades and glazing	HB–iDGP
[30]	Static overhangs or kinetic blinds shades and electro-chromic glazing	HB–iDGP

summarizes the investigated systems considered in the studies that have applied imageless methods for glare assessment, and none of those studies have investigated interior roller shades.

Unlike rpict and matrix-based methods, iDGP and CS Annual Glare predict DGP without simulating images. iDGP calculates vertical illuminance using the 2–Phase Method (2PM) (with eight ambient bounces) for the brightness term, and deduces the luminance, size, and position of all potential glare sources without generating an image for the contrast term. Provided that the glare is due to the direct view to the sun and sky, iDGP determines the luminance contributions of sky patches by examining whether each observed sky patch reaches a luminance level to be considered as a glare source, based on results of 2PM (with 0 ambient bounce). Like iDGP, CS Annual Glare calculates the vertical illuminance and estimates contrast due to the solar disc. CS integrates progressive path-tracing version of Radiance raytracer with hardware acceleration to expediate simulations.

There is a lack of general understanding about how different Radiance-based methods predict DGP and the inherent assumptions in those methods. When investigators apply CS Annual Glare or iDGP, they may assume that (i) the imageless DGP methods exactly followed the original DGP equations [33] without simplified calculations [22,25,26,28,29], or (ii) CS or iDGP was a field-validated simulation tool for DGP calculations [20,21,24,27,30,31], because they were based on Radiance. While the original DGP equation is calculated based on information extracted from luminance maps, CS Annual Glare and iDGP calculate vertical illuminance but estimate contrast due to the solar orb or sky patches without generating images. Although both CS Annual Glare and iDGP are based on Radiance and CS has been validated for calculating illuminance [21,25,34], solar irradiation [24], and image-based DGP [35], no studies have shown that those imageless methods have comparable accuracy with field measurements or “ground-truth” models such as rpict across different fenestration systems. Wasilewski et al. [36] compared several fast daylight simulation tools against rpict for annual glare simulations and found that CS Annual Glare systematically underpredicted DGP for most studied fenestration systems including roller shades. iDGP was first validated by the 2–Phase Method (2PM) image-based method for a single view [15], but 2PM could not accurately model the direct solar contribution as the sun is misrepresented by several sky patches. Wasilewski et al. [36] further compared iDGP against Radiance tool rpict and concluded that iDGP offers sufficient accuracy in predicting DGP for systems that involve only clear glazing and rough specular transmissions where DGP is mainly driven by vertical illuminance and is not applicable to glazing of low visible transmissions, roller shades, or specular reflections.

While imageless methods implemented in Rhino plugins deliver simulation results to users almost immediately, users may still need to delve a bit deeper into the definitions and applications of DGP to avoid a misuse of DGP. DGP is a useful metric for quantifying discomfort glare for a space with some glare protection; evaluating DGP for a space without any shading device will inevitably result in intolerant glare at some point. Demonstrating that the proposed shading or façade designs will result in less sDG or DGP than a baseline without any shading [19,24,37] does not necessarily mean that the proposed designs would significantly reduce the risk of glare. Comparing DGP or sDG across designs without shading [26,31] is rather meaningless for practical glare protection recommendations. In addition, some studies have calculated annual average DGP which does not provide critical information about visual comfort in the space [19,26]; or presented luminance maps without their corresponding luminance scales or DGP values [19,31]; or even evaluated DGP at work plane levels [27,32].

Besides the imageless methods, CS and HB also provide other tools for glare evaluations. Within HB, users can perform rpict and 2PM image-based calculations, but the 3–Phase Method (3PM) and 5–Phase Method (5PM) image-based calculations are currently not available. CS offers another tool called Radiance Render which can render luminance maps evaluated from a selected view. Similar to rpict, CS Radiance Render generates point-in-time luminance maps and can be time-consuming for generating high-quality luminance maps; thus, investigators use CS Radiance Render to perform static analysis at selected dates and times for glare analysis [19,24,31,34,35,38]. However, it is difficult to select all the times with increased glare risk throughout the year for an overall evaluation. Takhmasib et al. [35] found that DGP simulated from CS Radiance Render was aligned with sDGP and DGP evaluated from field measurements; however, they only investigated low-transmittance modular facades during times when the sun was not in the field of view, so the validation of CS Radiance Render in this study was limited.

1.3. Matrix-Based Methods

To overcome the computation and accuracy challenges presented in DGP evaluation, considerable research effort has been made to improve the accuracy and speed of glare simulation methods. A major breakthrough that enables practical and efficient daylight simulations has been brought by ray-tracing-based matrix methods with parametric capabilities to recycle calculations. All matrix-based simulation methods are either direct implementations or advanced derivatives of the 2–Phase or daylight coefficient method (2PM or DC) [39]. The use of 2PM considers discretized skies and space configurations as two independent factors and is expressed by a multiplication of a daylight coefficient matrix (depending on space configurations) and a sky vector (size and luminance of sky patches).

As a variant on 2PM, 3PM separates the light transport between the sky and the interior space into three phases [39]: exterior transport, fenestration transmission represented by bidirectional scattering distribution function (BSDF), and interior transport. BSDF describes the transmission and reflection between each pair of incoming and outgoing angles with the capability of characterizing optical properties of any complex fenestration systems (CFSs). The use of 3PM enables parametric annual simulations with Klems’ BSDF data (with 145 pairs of incident and existing angles), allowing efficient daylight and glare assessment for various CFSs, operable shading, locations, and orientations. To evaluate the glare performance of roller shades for a specific view direction and space configuration, one only needs to replace the fenestration transmission (the BSDF) of the shade material and reuse the other two matrices. The 3PM provides valid results for illuminance calculations [40], but introduces some errors in direct sun contributions (which are important for glare analysis) due to discretized data in a matrix form [6]. Persistence errors in 3PM are two-fold, as follows: (1) the Klems basis has an average resolution of 13.5° apex angle, but the sun orb has a 0.5° apex angle; (2) when a Tregenza sky matrix is mapped to a Klems BSFD, the sky and Klems subdivisions do not align seamlessly; as a result, the sun orb is spread over several sky patches, leading to misrepresentations of the size and intensity of the sun orb [6]. Although the resolution of sky subdivisions and BSDF can be higher, computation of 3PM becomes increasingly time-intensive and practically impossible using a 1.5° resolution (36,866 patches) on a Windows operating system (used by most designers and engineers) [6].

The spatial distribution of solar intensity is important for glare analysis because contrast-based glare is due to intense direct light from small glare sources, like the sun orb. To improve accuracy, the 5PM replaces the direct solar component in 3PM with the direct sun calculation (cds, a derivative of 2PM) using the intensity of the sun and its actual solid angle rather than several sky patches as in 3PM [39]. That makes the 5PM less efficient by requiring a full ray-tracing calculation (one ambient bounce) for a fixed CFS for each potential sun position. In cds, high-resolution BSDFs (with a tensor tree angular basis) or actual geometry can be used to model CFSs; tensor tree BSDFs (2^2k × 2^2k matrix, where k could be specified for generating high-resolution BSDFs) are more accurate than Klems’ BSDFs (145 × 145 matrix) and describe how light is scattered from 2^2k directions by a CFS. The number of potential sun positions is determined by the MF parameter (using the Reinhart subdivision scheme), where the MF can range from one to six, corresponding to 145 to 5185 solar positions. A recent study proposed a modified 5PM (M–5PM) by using continuous sun positions rather than placing suns at the center of Reinhart patches [41]. Despite the ability of M–5PM to model exact sun positions and retain parametric capability, it is less computationally efficient than 5PM with an MF less than 3. The 5PM was refined following a validation study completed by Geisler-Moroder et al. [5] through inclusion of the luminance of the shading system to improve the accuracy of the luminance maps and evaluation of visual discomfort. Illuminance and image-based glare metrics predicted using 5PM with high-resolution BSDFs for CFSs were validated against field measurements. The 5PM clearly outperforms 3PM and is reliable to perform precise calculation of the direct sun component [5,6]. DGP values predicted using 5PM with clear glazing systems [5,6] and roller shades [7] were also validated against “ground truth” rpict static results.

1.4. Modeling Roller Shades as Complex Fenestration Systems

Modeling light transmission through complex fenestration systems (CFSs), for each pair of incoming and outgoing angles, (represented by BSDF datasets) is difficult and requires hours of expensive and tedious measurements. The specular transmission properties are important for evaluating contrast-based glare discomfort, especially for “see-through” systems like roller shades. Fabric properties are angle-dependent and anisotropic and depend on color, material types, and weaving patterns.

Even with high-resolution BSDFs (applied in the Radiance BSDF material primitive), the simulated view through shades is blurred and distorted, and the solar disc is enlarged and dimmed leading to misrepresentation of its actual size and intensity [42]. To resolve this limitation, a peak extraction algorithm (PE) (implemented in the Radiance aBSDF material primitive) was developed for CFSs with a “see-through” component (such as roller shades) to model the solar disc with its actual size and intensity as seen in the BSDF dataset and preserve views through CFSs, by isolating specular or near-specular transmission from any tabulated BSDFs [42,43,44]. Radiance aBSDF should be used when the system has a view component such as roller shades (even with a small openness factor). When PE is used, a human retinal blurring function needs to be applied on simulated images to avoid overestimating the DGP perceived by human eyes [42]. Both Radiance BSDF and aBSDF materials require a tabulated BSDF in xml format for describing optical properties.

Fabric shades have complex optical properties inherent in their three-dimensional structures, affected by color, materials, and weaving technology. This makes them one of the most challenging CFS to study in terms of glare prediction. To accurately model light scattering by CFS like roller shades, a generalized method to derive data-driven BSDFs was proposed [42,45]; however, measuring the properties of fabric is expensive, time-consuming, and hardly practical for most designers and practitioners. The ray-tracing tool genBSDF [46] can theoretically be used to compute BSDFs based on actual geometry and materials of CFSs but describing detailed geometric descriptions of threads woven in various colors, patterns, and materials is difficult. The last option is to generate isotropic BSDF datasets based on analytical equations.

In this study, analytical isotropic models are used to describe the direct and diffuse transmission of fabrics. There are three analytical models available for describing the properties of fabric shades: (1) Kotey [47]; (2) Modified-Kotey [48]; and (3) the model by Wienold et al. [14], each of which was derived based on measurements of sample fabrics. Modified-Kotey is a refinement of the original Kotey based on measurements of a larger dataset, and both models estimate angular properties of fabric shades using only openness factor (Tv,n-n, or normal–normal direct visible transmittance) and total visible transmittance (Tv,n-h, or total hemispherical visible illuminance), which are typically provided by manufacturers. Original-Kotey estimates the direct-direct visible transmittance (Tv,dir-dir) with a cut-off angle (beyond which Tv,dir-dir diminishes to 0) around 65° which slightly depends on Tv,n-n, while Modified-Kotey does not apply a cut-off angle potentially leading to an overestimation of visual discomfort [45]. To estimate Tv,dir-dir, the model developed by Wienold et al. [14] requires a cut-off angle as an additional input; it was previously implemented in the Radiance BRTDfunc material primitive and used to be the standard for modeling roller shades prior to the development of the aBSDF material [7,14].

1.5. Scope of Study and Objectives

To protect occupants from glare, EN 17037 [12] recommends that DGP should not exceed disturbing limits for more than 5% of occupied hours but does not specify the modeling accuracy required for annual glare risk assessment. Simulation methods for calculating DGP vary in accuracy, complexity, and computational time. Despite the development of fast daylight and glare prediction simulation tools, there is a lack of understanding of the modeling assumptions, limitations, and applications of these tools, especially for modeling interior roller shades. The necessity of comparing glare simulation methods is underlined by the fact that some methods may introduce considerable errors when modeling direct light transmission through CFSs, although the investigated methods are based on Radiance. There is only one available study that compares hourly DGP values across faster methods for efficient DGP calculations [36], but there are no investigations available on the accuracy of annual discomfort visual frequency among different simulation methods. This study compares the accuracy of common glare simulation methods (3PM, 5PM, CS–Annual Glare, HB–iDGP) against “ground truth” Radiance ray-tracing tool rpict, and quantifies errors in predicting DGP, Ev, and annual visual discomfort frequency in spaces with roller shades. Methods both with and without PE (Radiance–aBSDF and Radiance–BSDF/Matrix–Only) are compared, using two commonly used fabric shades as representative examples.

2. Materials and Methods

The overall workflow is shown in Figure 1. Commonly used methods are compared against the “ground-truth” Radiance ray-tracing tool rpict in terms of glare assessment for fabric shades. This study focuses on the risk of glare when the sun is in the field of view and annual hourly simulations are performed for a typical private office space and two view directions (window-facing, and window-side/parallel). The metrics related to glare assessment, including hourly DGP, hourly Ev, and visual discomfort frequency, were computed with different simulation methods, and several error metrics were calculated against the ground truth. Evalglare v2.0 [3] was used to evaluate DGP and Ev for image-based methods including rpict, 5PM, and 3PM, while methods that deduce the information of glare sources without rendering images (CS–Annual Glare and HB–iDGP) directly output DGP (and Ev) results at the end of the simulation.

2.1. Case Study and Model Inputs

For all glare simulation methods being investigated in this study, the Perez all-weather sky model was used to model sky luminance distributions, using TMY3 hourly weather data for Chicago (41.9° N, 87.6° W). For rpict, Radiance gendaylit was used to generate a Radiance scene description of a continuous luminance distribution and of the sun and the sky using “perezlum.cal”. For matrix-based methods, Radiance gendaymtx was used to generate a matrix of sky patches, where the sun contribution was represented by the four nearest sky patches. A direct sky matrix without any diffuse contribution was used to calculate the direct solar contribution in 3PM–D, while a discretized sun matrix was used in cds.

A typical private office (4 m × 5 m × 3 m) was used for glare assessment in all simulations (Figure 2), considering two view directions (window-facing and window-side/parallel). The office has a south-facing façade with a window-to-wall-ratio of 60%. The occupant was assumed to sit 1.1 m away from the window at an equal distance from the side walls (eye height = 1.2 m). Opaque surfaces were modeled as Radiance plastic material primitives with commonly used reflectance (wall: 50%, floor: 20%, ceiling: 80%). The glazing had a normal visible transmittance of 75%; it was modeled as a Radiance glass material in rpict, cds of 5PM, iDGP, and CS, and it was represented by a Klems-based BSDF dataset (generated through Radiance genBSDF) in 3PM. Two commonly used roller shade products were included in the simulation: a dark-colored fabric (Tv,n-n, Tv,n-h) = (3%, 4%) and a light-colored fabric (Tv,n-n, Tv,n-h) = (5%, 10%). The selected fabrics had different openness factors and visible transmittance which determined their daylight performance, glare protection ability, and view clarity. For example, the dark fabric offered a clearer outside view and had pronounced contrast effects, while the light fabric admitted more daylight and would result in higher saturation-based glare. The effect of the openness factor and color on glare performance can be captured by comparing the results between different cases using the selected fabrics.

2.2. Daylight Simulation Modeling and Parameters

The quality (accuracy and efficiency) of Radiance simulations depends on the selection of Radiance parameters and the complexity of the scene. Interpreting useful ranges of Radiance parameters and their corresponding quality [49,50] was straightforward, but investigation and in-depth analysis of the collective effects of those parameters may be required, for specific modeling needs. For example, a higher ambient division (-ad) helps to reduce errors in indirect calculations by sampling more rays to calculate the ambient value of a point, but lower values may achieve smooth ambient results through oversampling. Ambient bounces (-ab) represent the maximum number of bounces for the diffuse (indirect) calculation, and a single ambient bounce (-ab 1) was used to calculate the direct solar contribution in 3PM–D and cds of 5PM (Table 2). The number (=$1+144\times MF$) of discretized skies or potential solar locations on a hemispherical basis was specified by a particular value of MF (ranging from 1 to 6). For cds of 5PM, a higher MF increased the solar spatial accuracy but resulted in time-intensive computation. M-5PM which uses exact sun positions rather than placing suns at the center of Reinhart Patches was not applied in this study for two reasons: (i) M-5PM is less efficient than 5PM with an MF less than 3 and (ii) exact sun positions are important for point-in-time glare evaluations but may not be necessary for annual glare assessment. MF has a significant impact on iDGP which estimates glare sources based on luminance contribution of each sky patch, and finer discretized sky patches can better represent the intensity and size of the potential glare sources. The impact of MF on 5PM and iDGP is discussed in Section 4.1 and Section 4.2, respectively.

For image-based simulation methods (rpict, 3PM, 5PM), luminance maps were simulated with an image size of 800 × 800 pixels, which was the minimum resolution to sufficiently represent the sun orb that was not partially occluded (with 2 pixels to model a 0.5° solar disc, considering a 180° fisheye image). Radiance parameters for rpict were adopted from a similar study which used rpict as a reference simulation method (Wasilewski et al. [36]), except that the maximum error of the indirect calculation or ambient accuracy (-aa) was changed from 0.075 to 0.2 (Table 2). Our investigation showed that the change in -aa did not produce any significant difference in DGP results (Appendix A) but using a small -aa approximately quadrupled the rendering time. Radiance parameters recommended by several tutorials [51,52] and previous investigations [53,54] were used in 3PM and 5PM (Table 2). One of the objectives of this work was to compare the performance of commonly used fast glare simulation tools, so the default simulation settings in CS–Annual Glare and HB–iDGP were kept unchanged (Table 2). For CS–Annual Glare, users could conveniently adjust the -ab, weight limit (-lw), and the total number of samples to control simulation accuracy; customizing Radiance parameters in HB–iDGP is not straightforward, because users need to change the source code. Evalglare was used to evaluate Ev and DGP from HDR images simulated by rpict, 3PM, and 5PM. HB–iDGP and CS–Annual Glare directly outputted DGP without generating images, while CS–Annual Glare also outputted Ev.

Table 2. Radiance parameters.

Radiance Parameters
rpict [36]		-dp 4096 -dt 0.01 -dt 1 -ds 0.02 -dr 3 -ms 0.025 -ss 16 -st 0.01 -lr 12 -lw 1e-5 -av 0 0 0 -aa 0.2 -ar 600 -ab 6 -ad 1500 -as 750 -ps 3 -pt 0.04
5PM [52,53,54,55]	3PM	Sky matrix: -m1 (MF = 1) Daylight matrix: -c 1500 -ab 4 -ad 1024 -lw 9.76e-4 View matrix: -c 10 -ab 10 -ad 65536 -lw 1.53e-5 -x 800 -y 800
	3PM–D	Sky matrix: -m1 -d (MF = 1) Daylight matrix: -c 1500 -ab 4 -ad 1024 -lw 9.76e-4 View matrix: -c 10 -ab 1 -ad 65536 -lw 1.53e-5 -x 800 -y 800
	cds	Sky matrix: -m3 (MF = 3) Daylight coefficient matrix: -ab 1 -ad 1024 -dc 1 -dt 0 -dj0 -x 800 -y 800 MF = 3
CS–Annual Glare		(Default) -ab 6 -lw 0.01 with 64 samples per pass and 100 passes
HB–iDGP		(Default) Sky matrix: -m1 (MF = 1) Daylight coefficient matrix: -ab 10 -ad 25000 -as 4096 -c 1 -dc 0.75 -dp 512 -dr 3 -ds 0.05 -dt 0.15 -lr 8 -lw 4e-07 -ss 1 -dt 0.15

Table 2. Radiance parameters.

Radiance Parameters
rpict [36]		-dp 4096 -dt 0.01 -dt 1 -ds 0.02 -dr 3 -ms 0.025 -ss 16 -st 0.01 -lr 12 -lw 1e-5 -av 0 0 0 -aa 0.2 -ar 600 -ab 6 -ad 1500 -as 750 -ps 3 -pt 0.04
5PM [52,53,54,55]	3PM	Sky matrix: -m1 (MF = 1) Daylight matrix: -c 1500 -ab 4 -ad 1024 -lw 9.76e-4 View matrix: -c 10 -ab 10 -ad 65536 -lw 1.53e-5 -x 800 -y 800
	3PM–D	Sky matrix: -m1 -d (MF = 1) Daylight matrix: -c 1500 -ab 4 -ad 1024 -lw 9.76e-4 View matrix: -c 10 -ab 1 -ad 65536 -lw 1.53e-5 -x 800 -y 800
	cds	Sky matrix: -m3 (MF = 3) Daylight coefficient matrix: -ab 1 -ad 1024 -dc 1 -dt 0 -dj0 -x 800 -y 800 MF = 3
CS–Annual Glare		(Default) -ab 6 -lw 0.01 with 64 samples per pass and 100 passes
HB–iDGP		(Default) Sky matrix: -m1 (MF = 1) Daylight coefficient matrix: -ab 10 -ad 25000 -as 4096 -c 1 -dc 0.75 -dp 512 -dr 3 -ds 0.05 -dt 0.15 -lr 8 -lw 4e-07 -ss 1 -dt 0.15

2.3. Light Transmission through Roller Shades

Isotropic BSDF datasets were used to describe the optical properties of the roller shade and were generated based on analytical models, parametrized by two shade properties: openness factor (Tv,n-n) and total hemispherical visible transmittance (Tv,n-h). A “gaussKoteyBSDF.cal” file [55] was used to define the analytical equations that characterized angular optical properties and converted them to bidirectional-scattering direction functions. Here, we presented a “mixed” model (Figure 3) to improve the accuracy of calculations. The original *cal file was based on Modified-Kotey model [48]; however, the specular transmission equations were replaced with the Original-Kotey angular direct-direct equations [47] to account for a cut-off angle of around 65° in this study (Figure 3). When the cut-off angle was around 65°, the model developed by Wienold et al. [14] and the Original-Kotey model had similar angular direct-direct transmission (Tv,dir-dir) and should result in similar contrast-based glare predictions (Figure 4). In other cases, Wang et al. [45,49] showed that the Wienold et al. model with a custom-selected cut-off angle performed better in high contrast conditions. In reality, cut-off angles vary with shade openness factor, visible transmittance, fabric color, materials, and weave technology. Lu et al. [56] showed that properly modeling cut-off angles and angular transmission has a significant impact on glare prediction (see Section 4.5). Based on the analytical equations, Radiance bsdf2ttree was used to generate high-resolution BSDF datasets (-t 3 -g 6), which described the light scattering between 2¹² ingoing directions and 2¹² outgoing directions. To represent the flux transport through shades in 3PM and 3PM–D (of 5PM), a low-resolution Klems-based BSDF was generated using Radiance bsdf2klems.

For daylight modeling methods (rpict, cds in 5PM, CS–Annual Glare, iDGP) that treat CFSs as part of direct-direct light transmission room configurations, fabric shades were characterized by a Radiance aBSDF (with the peak extraction algorithm) or BSDF material type, to model direct and diffuse light transmission based on tensor-tree BSDF datasets. To mimic how blurring happens in human eyes, a blurring filter (Lorentzian function simulated by Gaussian function with FWHM of 11) was applied to simulated images if Radiance aBSDF was used [42]. For 3PM, the shades were modeled as a fenestration matrix represented by a Klems-based BSDF dataset, rather than a material primitive.

2.4. Variables and Compatibility between Different Glare Simulation Methods

Table 3. Outputs and compatibility with PE between different glare simulation methods (marked with “✓”).

Modeling Method		rpict	5PM	3PM	CS–Annual Glare	HB–iDGP
Output	DGP	✓	✓	✓	✓	✓
	Ev	✓	✓	✓	✓
Material type for roller shades	With PE (Radiance–aBSDF)	✓	✓		✓
	Without PE (Radiance–BSDF)	✓	✓	✓	✓	✓

lists the compatibility of glare metrics and light transmission models (with and without PE) between the different simulation methods. Hourly DGP, hourly Ev, and annual visual discomfort frequency are compared against rpict. The annual visual discomfort frequency is defined as the percentage of working hours (8:30 am to 6:30 pm) over a year during which predicted DGP is greater than a predefined threshold (5% according to EN 17037 [12]. However, it was impractical to simulate all hourly images of investigated scenarios using rpict due to the extremely long computation time. Previous investigations have shown that if the sun (or another very bright source) was not within the FOV, glare mostly happened due to the saturation effects driven by vertical illuminance. In that case, the DGP results predicted by rpict were not significantly different from 5PM [7]. For these reasons, and to make rpict calculations feasible on a Windows operating system, hourly simulations were performed (with shades on) whenever the sun was in the field of view (FOV) of the person. This was checked by detecting whether there was a glare source that had a luminance greater than 50,000 cd/m², as seen through the window without shades. Following this logic, the occupant would perceive the sun for 1296 h from the window-facing view and for 631 h from the side view, respectively (out of 3650 total occupied hours) in Chicago.

Table 3 outlines the simulation outputs and indicates whether PE is compatible with the investigated methods. Note that PE cannot be applied to 2PM, iDGP, or 3PM, because those methods fail to satisfy the required conditions to activate PE. PE happens when all the following conditions are true [42]: (1) Radiance aBSDF is used; (2) BSDF datasets used in Radiance aBSDF have a sufficiently pronounced peak in the “see-through” direction from the point of concern; and (3) the point of concern is the direct beam of one or more light sources or is a view ray looking at the surface. Light transmission through fenestration in 3PM was represented by a BSDF dataset in xml format directly rather than a Radiance material primitive, while 2PM and iDGP failed to meet the third condition because they used sky patches represented as part of distant glow surfaces, which are neither a direct light source nor a direct view. Therefore, the results predicted using rpict with Radiance aBSDF were compared only with 5PM and CS simulation results, as shown in Table 3. After simulations were performed, annual visual discomfort frequency was calculated based on predicted hourly DGP for each investigated method.

3. Results

Annual simulation results from the investigated methods (5PM, 3PM, ClimateStudio Annual Glare (CS–Annual Glare), Honeybee imageless DGP (HB–iDGP)) are compared against the results of “ground-truth” Radiance raytracing tool rpict, considering a dark-colored fabric (Tv,n-n, Tv,n-h) = (3%, 4%) and a light-colored fabric (Tv,n-n, Tv,n-h) = (5%, 10%) covering the entire window and two occupant view directions (window-facing and window-side). Glare results predicted by simulation methods combined with PE (Radiance–aBSDF) and without PE (Radiance–BSDF/Matrix–Only) are compared against rpict with and without PE, respectively, for a fair comparison. In addition, the errors of using Radiance–BSDF to represent roller shades are discussed separately in Section 4.3.

3.1. DGP Comparison and Errors in Different Methods

Figure 5 presents hourly DGP scatterplots to compare results from investigated methods against rpict, with and without PE, for the two fabrics and the two occupant views. Figure 6 shows root mean squared error (RMSE), mean absolute error (MAE), and mean biased error (MBE) in DGP for each case (Equations (1)–(3)). The relative error (Equation (4)) distribution in hourly DGP using each investigated method is presented in Figure 7. These results help to identify the errors and limitations in each method, discussed below.

$$RMSE=\sqrt{{\sum }_{i=1}^n\frac{{\left({\widehat{y}}_i-y_i\right)}^2}{n}}$$

(1)

$$MAE=\frac{1}{n}{\sum }_{i=1}^n|{\widehat{y}}_i-y_i|$$

(2)

$$MBE=\frac{1}{n}{\sum }_{i=1}^n({\widehat{y}}_i-y_i)$$

(3)

$$Relative Error={\widehat{y}}_i-y_i$$

(4)

where $y_i$ is glare metric (DGP or Ev) predicted with rpict; ${\widehat{y}}_i$ is glare metric (DGP or Ev) predicted with each investigated method; and n is the number of timesteps for comparison.

The DGP results predicted using 5PM without PE show consistent alignment with the results of rpict. Although applying PE in 5PM introduces a level of uncertainty due to the sampling mechanism and discrete spatial resolution of the solar disc, the overall trend in DGP using Radiance–aBSDF in 5PM and rpict remains similar (Figure 5, bottom). Across all scenarios (considering the two fabrics, two views with and without PE), 5PM results in the least errors, indicating that it is a robust and accurate method for predicting DGP. The MBE of 5PM with and without PE is around 0; the MAE without PE is below 0.01, but with PE it increases to 0.015–0.025, with the largest deviation observed for the side view direction with the lighter fabric (Figure 6). The outliers could be explained by the use of discretized sun discs and by the sampling variations within the PE algorithm, especially when the sun is near the façade or the cut-off angle of the fabric. Relative errors of DGP predicted using 5PM, with and without PE, are centered around 0 (Figure 7). The outliers are due to discretized solar positions in 5PM and occur only when the sun is near the edge of the façade. The sun orb may consistently appear near the edge of the façade in 5PM at some timesteps (due to a low MF value), while it may be partially or fully occluded in rpict which is based on a continuous sun path, and vice versa.

The 3PM and CS systematically underpredict DGP across all scenarios with negative relative errors in a considerable amount (Figure 7). The MBE of 3PM and CS–Annual Glare for predicting DGP is also negative (Figure 6), confirming an underestimation of DGP. Most importantly, the MAE of 3PM and CS without PE is around 0.04, and the MAE of CS with PE ranges from 0.04 to 0.09, leading to a categorical change in glare classification. This error is comparable or higher to the size of each DGP scale, indicating that both methods are inadequate for categorical or even binary DGP prediction. Previous studies also show that 5PM is superior to 3PM for modeling the direct light component [5,6] which is crucial for evaluating the contrast term of DGP, and CS–Annual Glare underestimates DGP in case studies of various CFSs even with high-quality simulation settings [36].

Finally, HB–iDGP has inconsistent performance for the different fabrics and view directions. The DGP results of HB–iDGP match the predictions of rpict and 5PM only for a window-facing view with a large openness factor, indicating that it is not a robust method for DGP prediction for other cases. For the darker fabric, HB–iDGP underpredicts DGP, with a MAE as high as 0.06 and negative MBE below 0.05.

3.2. Vertical Illuminance (Ev) Comparison and Errors in Different Methods

Overall, in terms of hourly Ev, the results of the different methods are similar compared to rpict (Figure 8). Note that Ev cannot explicitly be exported from the HB–iDGP model and therefore is not shown in the results. The relative errors in Ev are centered around 0, while the 5PM with PE and the CS–Annual Glare models have more outliers with significant errors in Ev (Figure 9). The 5PM outliers can be explained by its discretized spatial resolution of the sun orb and can be reduced if a higher MF (which defines the number of possible solar positions) is used, as discussed later (in Section 4.1). The fact that different methods result in similar Ev values shows that the uncertainty in predicting DGP mainly comes from estimation of the contrast term, and that all the models are able to classify glare in low-contrast, high-adaptation scenarios when the sun and specular reflections are not in the field of view.

3.3. Visual Discomfort Frequency

The overall glare risk assessment with roller shades is evaluated by computing the annual visual discomfort frequency for a specific space configuration and climate, which is defined as the percentage of occupied hours when DGP (evaluated at a view direction) is greater than a typical DGP threshold. The thresholds selected in this study are 0.38 (Wienold et al. [14]), 0.4 (EN 17037 [12]), and 0.45 (Wienold et al. [14] and EN 17037 [12]), which correspond to the proposed (and the originally published) limits of noticeable, disturbing, and intolerable glare, respectively. When the sun is predicted not to be in the FOV through the shades, DGP is assumed to be imperceptible (DGP < 0.35). This assumption was made to focus on the accuracy of investigated methods when modeling the direct sun component.

Table 4. Annual visual discomfort frequency computed with 5PM, 3PM, ClimateStudio Annual Glare (CS), Honeybee imageless DGP (HB–iDGP), and rpict, with PE (“highlighted”) and without PE.

View			With PE Radiance–aBSDF			Without PE Radiance–BSDF/Matrix–Only
	Shade Property (Tv,n-n, Tv,n-h)	DGP Threshold	rpict	5PM	CS	rpict	5PM	3PM	CS	HB–iDGP
Window- Facing	(5%, 10%)	0.38	15.7%	15.8%	9.1%	7.2%	6.9%	3.1%	1.9%	6.7%
		0.4	13.9%	13.9%	7.5%	5.7%	5.4%	2.0%	0.6%	5.3%
		0.45	9.7%	9.4%	4.2%	3.3%	3.1%	0.3%	0.0%	2.6%
	(3%, 4%)	0.38	13.5%	13.2%	1.2%	1.0%	0.9%	0.0%	0.0%	0.0%
		0.4	10.6%	10.3%	0.2%	0.2%	0.4%	0.0%	0.0%	0.0%
		0.45	4.8%	5.0%	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%
Window- Side	(5%, 10%)	0.38	3.9%	4.4%	2.8%	0.0%	0.0%	0.0%	0.0%	0.0%
		0.4	2.7%	2.9%	1.7%	0.0%	0.0%	0.0%	0.0%	0.0%
		0.45	0.1%	0.1%	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%
	(3%, 4%)	0.38	2.4%	2.7%	1.7%	0.0%	0.0%	0.0%	0.0%	0.0%
		0.4	1.6%	1.8%	0.9%	0.0%	0.0%	0.0%	0.0%	0.0%
		0.45	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%	0.0%

lists the discomfort frequency results with all modeling methods. Despite slight deviations in hourly DGP and Ev predicted by 5PM compared to rpict, the differences in annual visual discomfort frequency between 5PM and rpict with and without PE are within 0.5% (corresponding to only 18 h in the year), indicating that 5PM is a robust method for evaluating annual glare risk assessment with roller shades. The discomfort frequency is much higher with Radiance–aBSDF due to the accurate evaluation of specular transmission and high contrast effects— the same observations have been noted in the literature [43,57].

The other simulation methods consistently and significantly underpredict the risk of glare. Even with Radiance–aBSDF, CS–Annual Glare predicts much lower values than rpict. For example, for the dark-colored fabric and window-facing view, rpict results in annual visual discomfort of 13.5%, 10.6%, and 4.8% for the three DGP limits, whereas CS–Annual Glare results in almost zero risk of glare in all cases. With Radiance–BSDF, CS–Annual Glare also results in significantly lower visual discomfort frequency than rpict. HB–iDGP results in similar annual visual discomfort frequency as rpict only when the fabric has a larger openness factor, and the view is towards the façade.

Overall, all simulation methods (except 5PM) will result in significant errors in annual visual discomfort frequency compared to the ground truth (rpict). These errors will misclassify fabrics with respect to their ability to mitigate glare according to EN 14501:2021 “Blinds and shutter” standard [58] or EN 17037 [12]. Therefore, these tools should not be used for design and recommendations for glare protection with roller shades without extra care, and users need to understand the risks, potential errors, and limitations with these methods.

4. Discussion

4.1. Effect of MF on Results of 5PM

The MF parameter defines the number of potential solar positions in cds of 5PM (based on Reinhart subdivision scheme). The cds requires a full raytracing for a fixed CFS and each potential solar position, so a higher MF leads to a higher spatial accuracy of solar positions at the expense of computation time. The impact of MF on DGP prediction with 5PM was studied to investigate whether a lower MF could still result in similar glare prediction compared to rpict. Figure 10 plots hourly DGP predicted using 5PM with different MFs against rpict. The results from the two methods are aligned even when the smallest MF = 1 is used, despite the larger spread.

More importantly, the differences in annual visual discomfort frequency estimated using 5PM with a different MF compared to rpict are less than 1% (Figure 11). This result indicates that the lowest MF could be auto-selected to perform annual simulations efficiently with sufficient accuracy for a specific space and view direction. The level of sun/sky subdivisions has an impact on how the sky/sun and circumsolar regions are perceived by occupants, but this perceived difference does not necessarily translate into major deviations in predicted DGP values and annual visual discomfort frequency for the case or roller shades. Our observations on the effect of MF on DGP predictions agree with Subramaniam et al. [59], who found that the degree of sky subdivisions had little impact on DGP values evaluated using image-based 2PM for a pure glazing system.

4.2. Effect of MF on Results of iDGP

The MF parameter determines the resolution of sky subdivisions in 2PM and iDGP. In HB–iDGP, the default MF is 1, so that the sky is discretized into 145 patches (144 skies + ground). To study whether iDGP with a higher MF can reach consistent performance in glare evaluation, we compare hourly DGP predicted using iDGP with several MFs against rpict using the Radiance parameters presented in

Table 5. Radiance parameters in iDGP.

Radiance Parameters
iDGP	Sky matrix: -m1, 3, 6 (MF = 1, 3, 6) Daylight coefficient matrix: -lw 1e 8 -ab 7 -lr -12 -ad 80000

. Note that iDGP is performed in the Command window rather than Honeybee for this analysis; to change MF in Honeybee, users can follow the instructions on the Ladybug Discourse [60].

The results of Figure 12 show that iDGP with the maximum MF (6, corresponding to 5184 skies + ground) and rpict with PE exhibit pronounced similarity in DGP predictions for larger openness factors and when the view is towards the façade, while DGP values are underestimated in other scenarios. With a higher MF, iDGP results in higher DGP values because the direct sky patches become finer with higher luminance values, increasing the chance to deduce a glare source.

The effects of MF on glare prediction by iDGP are not consistent with rpict with Radiance–BSDF. An MF of 1 and 3 in iDGP generally agrees with rpict without PE for different fabrics and view directions, while the highest MF always overestimates DGP when Radiance BSDF is used. The selection of MF is scenario-based, requiring users to tune and select MFs for different cases (whenever the CFS, view direction, or space configuration is changed), so iDGP could provide comparable results to “ground truth” rpict.

4.3. The Impact of Peak Extraction (Radiance aBSDF) on Correct Glare Classification

Both Radiance BSDF and aBSDF material types take a BSDF dataset in the xml format to model the light transmission of complex fenestration systems. However, even with high-resolution BSDFs, small glare sources like the solar disc can be misrepresented by a larger size and lower intensity [42]. In the previous results, the modeling approach without PE (Radiance–BSDF) for all methods was made for the purpose of fair and relative comparison. However, the Radiance–BSDF material (without PE) tends to smear the specular transmission through CFSs with a “see-through” component (as with fabric shades), which results in an enlarged sun orb with a lower intensity than its true value (always less than 1 million cd/m² in this study). Because of this, high-contrast discomfort glare is underestimated with this approach.

For that reason, as mentioned earlier, Radiance–aBSDF (with PE) should be used when the roller shades have a view component (such the common products selected in this study), to accurately model the apparent solar disc and specular light transmission [42,43,45]. Radiance–aBSDF adopts the PE algorithm to isolate specular and near specular transmission from any tabulated BSDFs to obtain better representation of small glare sources, despite some discrepancies with field measurements [45].

Figure 13 illustrates the difference in hourly DGP predicted with PE (Radiance–aBSDF) and without PE (Radiance–BSDF) for the studied fabrics and view directions. DGP values are much higher when PE is activated (with most values above the diagonal line), indicating a significant difference in glare prediction between the two material types for all studied cases.

Table 6. Annual visual discomfort frequency predicted with PE (Radiance–aBSDF) and without PE (Radiance–BSDF) using 5PM (MF = 3) for the two fabrics and view directions.

View	Shade Property (Tv,n-n, Tv,n-h)	DGP Threshold	aBSDF (with PE)	BSDF (without PE)
Window-Facing	(5%, 10%)	0.38	15.8%	6.9%
		0.4	13.9%	5.4%
		0.45	9.4%	3.1%
	(3%, 4%)	0.38	13.2%	0.9%
		0.4	10.3%	0.4%
		0.45	5.0%	0.0%
Window-Side	(5%, 10%)	0.38	4.4%	0.0%
		0.4	2.9%	0.0%
		0.45	0.1%	0.0%
	(3%, 4%)	0.38	2.7%	0.0%
		0.4	1.8%	0.0%
		0.45	0.0%	0.0%

lists the differences in annual visual discomfort frequency estimated with and without PE for both fabrics and view directions, as well as with different DGP thresholds (from perceptible to intolerable glare). More specifically, the dark-colored fabric (3%, 4%) for the window-facing view will cause perceptible/just disturbing glare (DGP > 0.4) for 10.3% of occupied hours throughout the year with Radiance–aBSDF, way above the recommended 5% limit (EN 17037 [12]); but the discomfort frequency with Radiance–BSDF is reduced to only 0.4% for the same case, indicating that the fabric is glare-safe according to EN 17037 [12]. Similarly, the discomfort frequency for the light-colored fabric (5%, 10%) is only 5.4% using Radiance–BSDF but equal to 13.9% using Radiance–aBSDF. Previous studies in office buildings with fabric shades [61,62,63] have confirmed that fabrics with an openness factor equal or higher than 3–4% will cause significant discomfort and glare. In summary, the results show that using Radiance–BDSF will underestimate DGP predictions for roller shades with a view component and may only be appropriate when the fabric shade is highly diffuse, where glare is mainly due to high adaptation rather than contrast.

4.4. Computation Time

When selecting a daylight simulation method, one needs to consider both the efficiency and accuracy of the available methods. The “ground-truth” simulation method rpict takes 25 min (based on the settings in Table 2) to generate a point-in-time luminance map, making it unfeasible to produce annual hourly results for glare risk assessment, although low-quality Radiance settings may still result in similar DGP predictions. CS–Annual Glare and HB–iDGP predict annual hourly DGP within a few seconds; however, the results of this study show that CS–Annual Glare cannot predict DGP with sufficient accuracy for the case of roller shades, and the selection of MF in iDGP needs fine-tuning.

DGP predictions by 5PM with and without PE consistently match the results of rpict across different scenarios. With Linux-based parallel cloud computing and sufficient memory, 5PM could evaluate the annual risk assessment for several fabric shades within a few hours. Assisted by cloud computing, cds in 5PM can be complete within a few minutes (based on the settings in Table 2) or several hours depending on the selection of MFs, room configurations, and available cloud processing units. The view matrix calculations in 3PM and 3PM–D can be time-consuming for accurate simulations, but they can be reused if the room geometry and view are kept the same. Multiplication calculations and Evalglare evaluations are the most time-intensive part in 5PM where parallel computing cannot make calculations faster. In addition, the new tool Raytraverse [36] has been shown to produce similar DGP results with rpict across several fenestration systems including roller shades [36], and could be a strong alternative to 5PM due to its computation efficiency, especially if a user-friendly version is developed in the future.

4.5. Errors in Analytical Light Transmission Models through Roller Shades

Fabric shades have anisotropic behaviors with varying light scattering at different incident angles. Cut-off angles in direct-direct light transmission and angular transmission of fabrics depend on their color, materials, and weaving technology. Wang et al. [45] showed that cut-off angles are lower for low-openness fabrics. Lu et al. [56] showed that the cut-off angle in analytical models can be properly selected when compared with data-driven BSDFs (which, however, are very limited). In this study, errors in isotropic models were not investigated (this was out of the scope of comparing available methods), and all fabrics were assumed to have a cut-off angle of 65°. However, the cut-off angle and angular transmission have a significant impact on glare prediction and need to be modeled properly for the reliable assessment of glare risk with fabrics. For example, when the cut-off angle is changed from 65° to 55°, the annual visual discomfort frequency (estimated with 5PM and Radiance–aBSDF) drops by more than 20% (

Table 7. Change in annual visual discomfort frequency estimated with 5PM using aBSDF (PE) when the cut-off angle of a fabric (3%, 4%) is changed from 65° to 55° (with equations for direct-direct visible transmittance derived by Kotey et al. [47] and Wienold et al. [14]).

DGP Threshold	Annual Visual Discomfort Frequency		Change (%)
	Cut-Off Angle: 65°	Cut-Off Angle: 55°
0.38	13.2%	10.1%	−23.5%
0.4	10.3%	7.9%	−23.3%
0.45	5.0%	3.7%	−26.0%

). This limitation does not affect the relative comparisons of the investigated methods against rpict, because all the methods have the same BSDF datasets as inputs. The scope of this work was a consistent comparison between existing glare simulation methods and not to investigate potential errors introduced by uncertainty in BSDFs using analytical models or other simplifications such as space configuration, etc.

5. Conclusions

Fast daylight and glare prediction simulation tools contain several modeling assumptions which are not always obvious, especially when modeling CFSs such as roller shades. This study has a specific and narrow scope: to compare common glare simulation methods (3PM, 5PM HB–iDGP, CS–Annual Glare) against the “ground truth” Radiance ray-tracing tool rpict using a robust modeling framework, and quantify errors in predicting DGP, Ev, and annual visual discomfort frequency in spaces with roller shades. In parallel, the results with PE (Radiance–aBSDF) and without PE (Radiance–BSDF, or Matrix–Only as in 3PM) models are compared, using two commonly used fabric shades as representative examples. Glare evaluation results using 5PM, with and without PE, are in good agreement with rpict across different shading properties and view directions, despite some uncertainties introduced by sampling variations and discretized solar positions in the sun coefficient calculations. Overall, the annual visual discomfort frequency predicted by 5PM shows no obvious deviations from rpict. With parallel cloud computing and sufficient disc memory, 5PM could reliably evaluate annual glare in spaces with roller shades, provided that the anisotropic properties can be properly modeled (using data-driven BSDFs or measurement-aided analytical models). On the other hand, the 3PM consistently underpredicts DGP.

Fast simulation tools such as CS–Annual Glare and HB–iDGP have enabled users across all levels of expertise to conveniently perform hourly DGP evaluations within a few seconds. We encourage users to understand the underlying assumptions and inherent uncertainty of the imageless methods, because they may introduce significant errors when simulating glare with roller shades compared to “ground-truth” simulations, especially due to the contrast effects. CS–Annual Glare with and without PE consistently underestimates DGP. iDGP may be applicable for scenarios when the view is directed towards the façade and the shade has relatively high openness; for other scenarios and CFSs, users need to carefully select the number of sky subdivisions to achieve comparable accuracy with rpict. For low-contrast, high-adaptation scenarios (such as when the sun is not within the field of view), the fast simulation models can be used to calculate vertical illuminance and adaptation-based glare. Finally, the results of this study are limited to roller shades and cannot be extrapolated to other complex fenestration systems.