Towards a framework for light-dosimetry studies: Methodological considerations

For field research of non-visual effects of light, accurate measurement of personal light exposure is required. A consensus framework for light-dosimetry could improve non-visual field research and ensure comparability between studies. Here, we present a review of methodologies used in non-visual light-dosimetry studies published to date, focussing on considerations regarding the measurement and preparation of personal light exposure data. Overall, a large variability in the studies’ methodologies is observed, highlighting the need for a consensus framework. We propose methodological considerations that should be included in such a framework and that can guide future studies. Furthermore, we highlight important points that should be addressed in future research to ensure compatibility between different dosimetry studies. Taken together, this review effort underlines the importance of a systematic approach to light-dosimetry in order to harness all the power of integrative lighting research in real life.


Introduction
Decades of research have shown that light has behavioural and physiological effects unrelated to human vision, mediated by a dedicated neural pathway. Through this pathway, light sets our biological clock, 1 directly affects various aspects of physiology and behaviour 2 and may thereby be related to health and well-being. 3 Much of what is known about these non-visual effects of light has been established by extensive laboratory research, indicating that responses are modulated by different light exposure characteristics. 4 However, real-life light exposure resembles nothing like laboratory stimuli, but consists of complex patterns of light of different quantity and quality, as a result of moving through our indoor and outdoor environments. Therefore, more field research is needed to complement and evaluate findings from controlled laboratory studies and answer pressing questions pertaining to implications in applied contexts, such as architecture and lighting design, therapeutical applications, shiftwork, transcontinental travel and personal lifestyle. 5 Due to the complexity of personal light exposure patterns and the uncontrolled nature of field research, it is crucial that results from different studies are comparable and repeatable, as well as transferable to applied contexts. This can be achieved by agreeing on standardised operating procedures for setting up and reporting lighting research studies. 5  For field research of non-visual effects of light, accurate measurement of personal light exposure is required. A consensus framework for light-dosimetry could improve nonvisual field research and ensure comparability between studies. Here, we present a review of methodologies used in non-visual light-dosimetry studies published to date, focussing on considerations regarding the measurement and preparation of personal light exposure data. Overall, a large variability in the studies' methodologies is observed, highlighting the need for a consensus framework. We propose methodological considerations that should be included in such a framework and that can guide future studies. Furthermore, we highlight important points that should be addressed in future research to ensure compatibility between different dosimetry studies. Taken together, this review effort underlines the importance of a systematic approach to light-dosimetry in order to harness all the power of integrative lighting research in real life. the lighting conditions under investigation. For experimental studies, guidelines for quantifying and reporting lighting conditions have recently been published. [6][7][8] However, these guidelines have only limited applicability for field research, where light exposure cannot be controlled. Thus, to date, it is still unclear how to adequately measure, quantify and analyse personal light exposure data with respect to non-visual responses.
Given the complexity of light exposure and its relationship with non-visual responses in real life, accurate measurement of personal light exposure (i.e. dosimetry) is crucial. The dosimetry process -also called the dosimetry chainconsists of a series of steps (links), each of which has the potential to introduce additional uncertainty in the final research outcome. 9 Broadly, these links can be categorised as (1) selecting optical quantities, (2) calibrating dosimeters, (3) selecting a measurement setup, (4) processing the measured data, (5) calculating light exposure metrics and (6) linking metrics to measured responses. 9 Importantly, different execution of each link may lead to substantial differences in the results, complicating their interpretation and comparison across studies. Therefore, agreement on standardised procedures for each link is urgently needed. 9 In the present article, we set out to laying the groundwork for a consensus framework for nonvisual light-dosimetry studies. To achieve this, we assembled a comprehensive set of dosimetry field studies published to date, which is taken as a basis for discussing methodological considerations for each link in the dosimetry chain. Herein, we review methodologies employed in previous dosimetry studies for measurement and preparation of personal light exposure data. Furthermore, we aim to identify crucial gaps in knowledge that need to be specifically addressed and clarified in future research.
As a brief note on nomenclature, throughout this article, the term light exposure is used to refer to the time series of light a person is exposed to and not to the quantity luminous exposure H V . Furthermore, the term light level is used as a generic term where multiple light quantities are applicable within a given context (e.g. illuminance, alpha-opic irradiance, etc.).

Search strategy
To collect a comprehensive set of light-dosimetry field studies, a forward and backward citation search method was used, allowing for an efficient assembly of relevant studies within the same domain. Two of the first light-dosimetry studies in the context of chronobiology were chosen as a starting point, 10,11 from which eligible dosimetry studies published to date were identified by means of forward and backward citation search using the Web of Science citation database (forward and backward search) and the individual papers' reference list (forward search), within the period of January-March 2021.

Selection criteria
The main objective of this review was to identify light-dosimetry studies that covered the entire dosimetry chain from measurement to quantification and subsequent analysis of the collected light exposure data, the ultimate intent being to highlight methodological considerations in the dosimetry procedure and to identify metrics for the quantification of personal light exposure patterns. Therefore, studies were only eligible if personal light exposure data was measured with wearable light meters over a period of at least 24 hours and if the measured light data was included as a dependent or independent variable in the analysis. This criterion excluded studies that measured 24-hour light exposure with static devices only (i.e. the device was not worn by the subject) or that measured light exposure but did not report any analyses of these measurements. Consequently, many intervention studies where personal light exposure was monitored but not analysed as a major dependent or independent variable were excluded, except for three studies. Specifically, the study by Peeters et al. 12 was included because personal light exposure was a primary dependent variable in the analysis, and Phillips et al. 13 and Zeitzer et al. 14 were included because multiple light exposure metrics were analysed.
In addition, to narrow the scope of this review, we primarily focused on dosimetry studies in the context of the non-visual effects of light, specifically, pertaining to sleep-wake regulation, circadian entrainment and direct physiological and behavioural responses (e.g. alertness, cognitive performance, mood, etc.). As a result, studies that measured personal light exposure in a different context (e.g. myopia, UV-light) were excluded, except for three myopia-related studies where the methodology added novel content to the review. Specifically, the studies by Alvarez and Wildsoet 15 and Ulaganathan et al. 16 were included because effects of sampling frequency on the accuracy of calculated light exposure metrics were investigated. Moreover, the study by Read et al. 17 was included because metrics are described that have not been used in any of the other studies already included in the review.

Final set of studies
In total, 104 studies were deemed eligible and formed the set of dosimetry field studies reviewed here. An overview of the selected studies  and study related characteristics is presented in Table A1.1.

Dosimeter selection and optical quantities
Personal light exposure is usually measured with lightweight wearable devices (dosimeters; see Figure 1). A wide range of different technologies exists on the market, reflected in the variety of dosimeters employed by the reviewed dosimetry studies (see Table 1). The dosimeters can broadly be categorised into dedicated light meters or wristworn actigraphy devices with light sensors (note that most wrist-worn dosimeters can also be adapted to be worn elsewhere). Furthermore, devices differ in their spectral sensitivity and resolution. Most devices listed here measure photopic illuminance (e.g. Actiwatch-L), or spectral irradiance in the visible range across three channels (red, green, blue; e.g. Actiwatch Spectrum). Three devices include a sensor calibrated to approximate the circadian spectral sensitivity (i.e. LuxBlick, Daysimeter and Dimesimeter), whereas only two dosimeters were used that have a higher spectral resolution.
The optical quantities a dosimeter can measure are an important consideration for non-visual dosimetry, since the spectral composition of personal light patterns (i.e. the 'relative spectral diet', see Webler et al. 115 ) can drastically vary over time. Historically, although dosimeters have been developed to measure light for the visual system (i.e. photometric), we know that the description of light in photometric terms is usually not appropriate when studying non-visual effects of light. 116 Consequently, new standards, such as CIE S 026:2018 117 , were developed, defining spectral sensitivity functions, quantities and metrics to describe light for non-visual responses (i.e. alpha-opic quantities).
Currently, the main limitation for the measurement of alpha-opic quantities is the limited availability of dosimetry devices that spectrally match the alpha-opic sensitivities. With the exception of devices used in two studies, 19,26 none of the dosimeters listed here match all alpha-opic sensitivities with sufficient accuracy. 118 Only one of the commercially available devices (i.e. the Actiwatch Spectrum) approximates the melanopic sensitivity function by a linear combination of photosensor outputs. 118 Note that the ActTrust dosimeter can be modified to provide a sufficient spectral match to the melanopic sensitivity curve as described previously 119 ; however, no study in this review used this modified device. Given that the melanopic sensitivity curve matches the sensitivity of non-visual responses for a wide range of conditions, as suggested in a recent comprehensive review of several experimental studies, 120 dosimetry studies could use devices that match this curve (see Lee et al., 66 Van der Maren et al., 69 Price et al. 81 ) until sufficiently spectrally resolved devices become available. Such devices could contain broadband photosensors matching the alpha-opic sensitivities directly (i.e. analogous to the sensor in the Daysimeter); however, these sensors are then constrained by assumptions on the underlying neurophysiology. Therefore, devices from which (parts of) the visible spectrum can be sufficiently recovered in order to calculate alpha-opic quantities offer a significant advantage over the former type of devices, as further elaborated on in Section 4.

Dosimeter assessment and calibration
Light-dosimeters have been found to vary substantially in optical performance between and within models as assessed in a range of studies. 118,[121][122][123] Many of the devices assessed in these specific performance assessment studies were used by the reviewed dosimetry studies. However, only some of the reviewed studies (N = 26) report that dosimeters were validated and calibrated against an industry-standard light sensor (Figure 2(a)). Among the studies that provided a description of the validation and calibration procedure (N = 19), a variety of methods were used; notably, validation was performed for different reference lighting conditions, including artificial light sources (N = 13), simulated daylight (N = 2) and/or under naturalistic indoor or outdoor conditions (N = 6).
The diversity of calibration methods used in dosimetry studies highlights the need for standardised assessment and calibration procedures. Typically, optical performance of photometers is assessed in terms of spectral sensitivity, directional response and response linearity as defined in the standard ISO/CIE 19476:2014. 124 However, current standards are not readily applicable for the characterisation of light-dosimeters for non-visual effects research; therefore, novel methods are needed. 118 Published optical performance metrics can help guiding the selection of an appropriate dosimeter for a given study. For non-visual effects research, devices that have good spectral sensitivity for matching the five α-opic sensitivity functions should be preferred. 117 However, trade-offs between spectral and directional Hubalek et al. 56 Light Eto et al. 38 Light mismatch should be considered, since the arrangement of multiple photosensors may affect directional responses. 118 The impact of directional mismatch may also depend on where the device is worn and how the data are aggregated. That is, high movement at the wrist or aggregation over longer time periods may cancel out directional mismatches. 125 Moreover, the linear range and dynamic resolution that is required should be considered depending on the lighting conditions expected in the study. The standard ISO/CIE 19476:2014 124 also describes calibration procedures for photometry devices; however, this standard may not be applicable to dosimeters that measure non-photometric quantities (e.g. spectral irradiance and alpha-opic irradiance). As a result, several methods have been described by individual studies. 122,[126][127][128] An important aspect to consider during calibration is the selection of a calibration light source that matches the lighting conditions typically encountered during the study. For studies with various lighting conditions, it has been recommended to calibrate devices to an overcast sky at noon, 122 or by averaging across several light sources. 129 Note that ambient conditions (i.e. temperature and humidity) can also substantially impact sensor accuracy and, ideally, should be calibrated for accordingly. However, this calibration is difficult for the many available lightdosimeters that do not include sensors to measure ambient conditions.

Dosimeter position
For research on the non-visual effects of light, the amount of light reaching the eye (corneal light exposure) is of primary interest; therefore, the position of the dosimeter on the body is an important consideration. Among the reviewed studies, only very few measured light exposure at eyelevel (N = 9), whereas most measured at the wrist (N = 67) and some at the chest (N = 22; Figure  2(b)). Measurements at eye-level require a specific setup (e.g. LuxBlick; see Figure 1) and might be perceived as more obtrusive than at the wrist or chest, 130 which may explain the small number of studies measuring at this position. Although both wrist and chest measurements may be less obtrusive, the large number of studies measuring at the wrist can be partly explained by the frequent use of actigraphy devices, allowing concomitant measurement of light exposure and sleep-wake activity within a single device. Only few studies measured actigraphy at the wrist and light at another position with separate devices (N = 12), or with a single device that is transferred to the wrist for actigraphy during sleep (N = 3).

Raw
Yes As most studies did not measure light exposure at eye-level, it is important to study to what extent corneal light exposure can be estimated from measurements at other positions. Yet surprisingly few studies have addressed this question. An early study that is frequently cited reported high correlation between measurements at the wrist and the forehead 10 ; however, the validity of these findings is limited due to dosimeter saturation at higher illuminance levels. Furthermore, correlation does not show the amount and direction of deviations at different light levels. Another study in postsurgical inhospital patients reported little average deviation (<10 lx) between wrist and eye-level measurements up to 5000 lx 131 ; however, the findings may not be generalisable to normal living conditions. To our knowledge, only one study examined different measurement positions under normal living conditions for an extended period, reporting little deviation between illuminance at eye-level and the chest, but large deviation for the wrist, which generally underestimated eye-level exposure, especially at higher illuminance levels. 125 Although this preliminary evidence suggests that measurements at the chest may be more accurate than at the wrist for estimating corneal light exposure, systematic variations in measurements at different positions should be considered. Aarts et al. 123 found that both wrist and chest measurements deviated substantially more from eye-level measurements under indoor conditions compared to outdoor conditions, which was especially pronounced for wrist measurements. Moreover, under indoor conditions, wrist measurements tended to overestimate, and under outdoor conditions underestimate eye-level exposure, corroborating earlier studies. Additionally, under indoor conditions, systematic variations have been observed for different activities, body postures and gaze directions. 132 Interestingly, this study also found that the exposure threshold associated with phase shifts of dim light melatonin onset (DLMO) estimated from eye-level measurements was not associated with phase shifts when using the same threshold for wrist measurements, likely due to overestimation of eye-level exposure.

Recording interval
In light-dosimetry, the recording interval or epoch length defines the rate at which individual light exposure measurements are recorded and therefore determines the temporal resolution and accuracy at which changes in light exposure can be captured. Among the studies under review, epoch lengths ranged from 100 ms to 5 minutes, with 1 minute and 30 seconds being the most used (N = 51 and N = 16, respectively; Figure  2(c)). Note that these epochs indicate the rate at which samples are recorded but do not necessarily reflect the sampling rate of the photosensors, as some dosimeters record an aggregated value of several samples across the given interval.
Epoch lengths are usually selected to optimise battery and memory usage, since dosimeters are often employed continuously for several days in a row. However, while longer epochs may help to conserve battery power and reduce memory load, the accuracy with which light exposure is measured may be affected, resulting in a trade-off between battery/memory usage and accuracy. For example, it has been found that epochs of 3 minutes or longer lead to a loss of accuracy relative to shorter epochs when calculating cumulative light exposure and time spent under bright light conditions. 15,16 On the other hand, while longer epochs may reduce accuracy, it is not yet known how much interpretability is gained by increasing recording rate.

Data preparation 3.2.1 Data cleaning
Due to the uncontrolled nature of lightdosimetry, measured data may contain invalid or artefact data. Two major sources of invalid data are periods where the dosimeter was not worn or when the light sensors were obstructed. Invalid periods were identified by some studies (N = 18) based on concomitant inactivity longer than a given timeframe, which ranged from 5 minutes to 120 minutes across studies. Some studies additionally examined the light data for smooth periods with little fluctuation, indicating invalid periods. 31,69,81 Furthermore, several methods for identifying measurement artefacts due to sensor obstruction have been reported. Some studies (N = 13) identified artefacts as light measurements below a threshold value during day-or wake-time, ranging from 0 lx to10 lx across studies. Other methods involved the identification of temporary drops in the data, 26 unusually high rate of change 74 and outlier detection methods. 105 Moreover, some studies removed remaining noise in the data with smoothing methods, by using simple moving average (SMA) filters with different window sizes between 5 minutes and 20 minutes, 32,79,85,104 or local regression smoothing (LOESS). 108 The latter procedure preserves peaks in the signal better than a SMA filter, but it can be computationally expensive for large datasets. Note that smoothing methods can also be used as an analytical procedure, for example, to quantify light dose in time by mimicking nonvisual response characteristics. 81 Beside smoothing, some studies (N = 15) averaged the data into bins (e.g. hourly averages), which can also be considered a means to remove noise in the data. However, as with smoothing methods, it is important to consider logarithmic transformation when aggregating data, as discussed in the next section.
Although cleaning measured personal light exposure data is an important step in the dosimetry process, nearly no studies have systematically investigated what cleaning methods and parameters are most appropriate for these kinds of data. To our knowledge, only one study has examined identification thresholds for sensor obstruction, showing that dosimetry measurements in a very dim laboratory environment without coverage by clothing did not fall below 1 lx. 88

Logarithmic transformation
Personal light exposure data can cover a large range of light levels over several orders of magnitude and usually follow a log-normal distribution. 133 Therefore, analyses may require logarithmic transformation of the data to ensure a normal distribution and enable interpretability. An important consideration when applying log-transformation is the sequence in which the data are transformed, aggregated and analysed, particularly regarding the question whether the transformation should be applied before or after quantifying the 'raw' light data. Amongst the reviewed studies, less than half applied for a log-transformation, either before (N = 35) or after (N = 12) quantification (Figure 2(d)). A major rationale for transforming the data was to ensure normality for statistical analyses (N = 10). Interestingly, four studies provide an explicit rationale for transforming raw data, 65,92,108,109 referring to log-linearity observed in non-visual dose-response curves. 134,135 Contrastingly, Scheuermaier et al. 88 argue that log-transformation should be applied after aggregating the data, in order to account for brief episodes of very bright light.
Indeed, the sequence of log-transformation may substantially affect the results. Take for example, the hypothetical light exposure pattern of an office worker between 12:00 hours and 13:00 hours, who during work is exposed to 200 lx and who goes outside for a lunch break at 12:30 hours, being exposed to 10 000 lx. The calculated hourly means for this individual would be around 10 3.71 (~5100 lx) with logtransformation after aggregation, and 10 3.15 (~1400 x) with log-transformation before aggregation. Realistically, this difference could be even greater depending on the fluctuation of light levels. This potentially huge impact on aggregated light exposure makes it impossible to compare data, such as mean hourly light exposure between studies that aggregated and transformed the data in a different sequence.
Moreover, analyses of non-visual responses may be affected, such as when using linear models to examine acute effects of hourly light exposure. Note that although it can be argued that mean light exposure may not be the right metric to analyse non-visual effects, it is still the most widely used metric to describe and compare personal light exposure patterns across all studies reviewed here.
Logarithmic stimulus-response relationships are well-established in psychophysics (i.e. Fechner's law), and non-visual responses to light are no exception-logarithmic relationships to light intensity have been described for circadian phase resetting, 136 melatonin suppression 134 and acute alertness. 135 Some findings from cell recordings suggest that logarithmic encoding of light intensity may happen at the level of retinal ganglion cells, whose neuronal firing response is directly proportional to the detected amount of photons on a logarithmic scale. [137][138][139] However, no study has yet specifically addressed what these findings imply for the measurement of time-series light exposure data.

Discussion and recommendations
In this review, we presented an overview of methods employed by previous dosimetry studies to measure and prepare personal light exposure data. Overall, a large variability in methodologies was observed across all studies. Personal light exposure was measured with a variety of dosimeters, at different positions on the body, and with different recording intervals. Very few studies measured spectrally resolved light exposure. Dosimetry devices were often not reported to be calibrated, and studies that calibrated the devices used a variety of different methods. Regarding data preparation, the few studies that report data-driven cleaning procedures used several different methods and parameters to identify invalid light exposure data. Discrepancy was also observed in the application and sequence of log-transformation of the light data. Later, we briefly discuss methodological implications based on these findings and provide recommendations for future dosimetry studies, as well as important points that need to be addressed in future research.

A note on accuracy in dosimetry for non-visual effects of light
As emphasised throughout this review, research of non-visual effects demands an accurate description of lighting conditions to improve the validity of research findings and ensure comparability and transferability of results across studies. 7 This holds especially for dosimetry, to avoid additional uncertainty in the presence of the plethora of confounders in field research. However, since an increase in accuracy often comes at the expense of factors such as practicality and cost, one of the central questions in dosimetry is how much accuracy is required at each dosimetry step, for analysing the relationship between measured non-visual responses and light exposure.
Given that dosimetry inherently refers to the study of the absorption of a physical quantity by the human body, it makes sense that dosimetry methods and metrics are biologically relevant. In theory, many of the methods discussed in this review could be based on neurophysiological mechanisms, for example, using smoothing methods that reflect temporal integration of the light signal. However, one of the central challenges in dosimetry is that our understanding of the mechanisms underlying signal encoding, adaptation and photic integration in the non-visual system is still incomplete. These uncertainties make it difficult to define accuracy limits for spectral and temporal resolution and dynamic range, which are further complicated by large inter-individual differences in light sensitivity. 140 In light of these uncertainties, instead of basing accuracy limits on assumptions of the underlying biology, it might be more appropriate to consider how the physical signal itself can be measured more accurately within the given technological and practical constraints, for instance, by using information theoretic approaches (see Section 4.3).

A proposal for guidelines and recommendations
Based on the findings of this review, we propose guidelines and recommendations for future dosimetry studies, summarised in Table 2. It is important to note that these recommendations are based on current knowledge, whereas several points demand further investigation (see Section 4.3). Therefore, it is crucial that methods and parameters used at each step in the dosimetry process are reported in necessary detail, either in the article itself or provided in Supplemental Material. Insufficient reporting hinders transparency, reproducibility and comparability, which is essential for research of non-visual effects in the field to be successful.
A crucial aspect in dosimetry is the selection of an appropriate dosimeter model, which depends on the context and aims of a given study. For research on the non-visual effects of light, devices that spectrally match the alpha-opic sensitivity curves should be preferred; however, given current limited availability of such devices, at least dosimeters that match the melanopic sensitivity curve should be used. In any case, we strongly recommend to always validate and calibrate dosimeters, and report a (reference to the) description of the calibration setup and procedure, including at least the calibration light sources used and which type(s) of calibration were performed (e.g. spectral, intensity and temperature calibration).
Another important consideration are contextual factors, such as lighting conditions, environments, activities and body positions expected during the study. These factors may determine the amount and direction of deviation from corneal exposure for different measurement positions, which can affect the analysis of non-visual responses. For example, using sensitivity thresholds based on corneal exposure for the analysis of light exposure measured at the wrist under predominantly indoor conditions may prevent detecting an effect, due to overestimation of corneal exposure. Similarly, contextual factors should be considered for selecting dosimeter models, calibration light sources and data cleaning methods.
Furthermore, dosimetry methods should be selected to achieve the highest accuracy within the given study constraints, avoiding assumptions on the underlying biology. The latter is particularly important for data preparation, as it is still unclear how the non-visual system encodes and integrates light information over time. Therefore, we recommend evaluating a range of different methods and parameters during analyses and include a detailed description of the selected cleaning methods and transformation sequence. Moreover, alongside the arithmetic mean, the geometric mean and/or other measures of central tendency should be reported.

Future work
Although the proposed recommendations given above are based on current knowledge and may form the basis of a framework for non-visual light-dosimetry, we have identified several important points that need to be addressed in further research for such a framework to be able to become fully operational (see Table 3). Importantly, more research investigating systematic interactions between factors related to the study context (e.g. lighting conditions, activities, etc.) and dosimetry methods are urgently needed, for example, to evaluate different dosimeter positions or improve data cleaning methods. On a long-term basis, more research studying underlying neurophysiological mechanisms, such as photic integration and sensitivity adaptation, is required to inform dosimetry methods. Such research would be especially insightful regarding the sequence of logarithmic transformation, given the widespread use of this method in dosimetry data analysis.
To facilitate the development of a future consensus framework and standards for light-dosimetry, it is of interest to examine in how far current standards in light metrology can be recycled and adapted. For example, the CIE already proposes standards for the calibration and performance characterisation of photometers (ISO/CIE 19476:2014 124 ). However, these standards are limited to photometric measurements and may not be appropriate for the measurement of alphaopic quantities as recommended in the new standard CIE S 026:2018. 117 Some suggestions for how these standards can be adapted have already been put forward by Price et al. 118 . Similarly, an effort for developing consensus reporting standards for light-intervention studies (ENLIGHT 141 ) is currently under way, which could in future be expanded to light-dosimetry in general. Such reporting standards may also contain guidelines for reporting the intended use of light-dosimetry in intervention studies, since intervention studies often make little use of collected dosimetry data.
In parallel, information theoretic approaches focussing on increasing measurement accuracy within the given technological and practical constraints should be explored. Specifically, novel data-driven methods may be used to increase  spectral and temporal resolution of dosimeters, as described in a complementary paper 142 and briefly introduced in the following. Spectral and temporal resolutions historically have been constrained by technological limits to size, weight, power and cost of dosimeters. However, recent advances in sensor technology and signal processing methods, such as compressed sensing, 143 challenge conventional limitations. Such methods demonstrate that for compressible (i.e. nonrandom) signals, the amount of information encoded can be much smaller than prescribed under classic sampling theorems. Compressed sensing effectively finds the simplest representation of a signal by exploiting regularities in the structure and properties of encoded signals. In practice, sensors based on this method can reduce the amount of encoded information at the source by a factor of 10, 144 enabling high-spectral resolution at a small size and cost. Recent work on a compressed sensing dosimeter named Spectrace, 145 developed by the authors, has already demonstrated the ability to recover 5-nm resolution data over the visible range from 14 narrowband photodiodes, 142 all in an autonomous wearable form no larger than a USB stick. As with compressed spectral sensing, sparse representation and forecasting can also be used to construct adaptive sampling rates that respond to changes in the environment (e.g. sample more when the scene changes and less when it is stable). Such approaches offer low-power and compact hardware forms without information loss. Taken together, information theoretic approaches offer novel ways of increasing accuracy in dosimetry within the given constraints, while avoiding the uncertainty introduced by making assumptions on human biology.

Conclusion
With this article, we discuss the status quo in dosimetry methodology to form a basis for further work towards a consensus framework for light-dosimetry studies. This review is the first to highlight the prevalent variability in methodologies employed by light-dosimetry studies during the past three decades, underscoring the need for standardised operating procedures to improve the comparability and repeatability of dosimetry studies. Moreover, by collecting this diverse set of methodologies, we were able to reveal important gaps in knowledge that need to be addressed in further research. At the same time, we show what methods are available for measuring and preparing personal light exposure, and what to consider and avoid when conducting lightdosimetry studies. With this work, we would like to bring the topic of light-dosimetry to greater attention, with the hope to provide researchers with the required knowledge to perform highquality studies and inspire more research in this direction. Furthermore, we want to emphasise the importance of carefully considering each step in the dosimetry process as each methodological decision may substantially affect the final study results. In the forthcoming second part of this review, methods for quantifying and analysing personal light exposure with respect to the nonvisual system will be discussed.

Acknowledgements
A version of this work was presented at the CIE 2021 Midterm Session.

Declaration of conflicting interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article:    Empty cells with a hyphen indicate that no information was reported.