Introduction

Organisations collect an ever-increasing amount of data concerning environmental parameters such as temperature, relative humidity, gaseous pollutants, or particulate matter (Vitolo et al. 2015). Persons who are not an expert in data or environmental science but wish to be informed about air quality may be confronted with an information overload, making such data meaningless (Gouws and Tarp 2016). A way to circumvent this problem is by visualising the trends that are present in data of continuously measured environmental parameters. In many cases, the activity of hazards (e.g., human presence, release of gases by furniture, heating systems) varies over time, forcing the environmental parameters to fluctuate. This means that the dynamics shown in graphs provide valuable information about the impact of hazards on their environment. Such changes are expected for outdoor conditions but also occur indoors in museums (Marchetti et al. 2017), chapels (Anaf and Schalm 2019) or churches. Despite the ease with which non-academic experts can understand such graphs, it does not deliver them much information about the impact of these parameters on heritage objects or on human health. To extract such information from the graphs, people need a mental model that derives air quality information from displayed profiles (Slovic 1972). Therefore, a significant fraction of the stakeholders remains unable to read/interpret such data effectively. In addition, the mental model must consider an extra layer of complexity. The strong fluctuations in trends that have been observed in a room will have another impact on for example, a canvas painting, a bronze statue or the health of the visitor in a museum (Anaf and Schalm 2019). For hygroscopic materials such as wood or textile, a change in relative humidity affects their shrinkage and swelling behaviour, which may lead to harm (Anaf et al. 2020; Kalamees et al. 2009; Erhardt and Mecklenburg 1994). Therefore, the relative humidity should remain as constant as possible. For metals, the corrosion rate is reduced when the relative humidity in the room is as low as possible (Schweitzer 2010).

Previous research has already proposed an algorithm that converts environmental parameters to a risk of accelerated degradation for different types of heritage materials or works of art (Anaf et al. 2018). By superimposing the risk information on the trends shown in graphs, moments of worse indoor air quality are easy to identify, even by non-experts (Leyva Pernia 2019; Schalm et al. 2019). For human health, several kinds of air quality indices (AQIs) exist to warn the public when outdoor air pollution is dangerously high (Tan et al. 2021; Plaia and Ruggieri 2011; Kanchan et al. 2015; Zhu and Li 2017; Leyva Pernia 2019). The AQIs could be merged with the environmental measurements and enhance the information in graphs. The applied visualisation methods translate the data obtained into actionable information (Carro et al. 2022). Unfortunately, the algorithms of the AQIs are very different from each other. It turned out to be impractical to combine the knowledge in the existing AQI definitions into a single model as was done in previous research about risk assessment for heritage objects. Therefore, our heritage-based algorithm has been used as a source of inspiration to develop a method that converts measurements of environmental parameters into a level of risk for human health. In that method, we also propose an alternative approach to assess the total risk of a complex hazard by combining individual risks in series or in parallel. The previously developed algorithm used in the heritage context appears to be more general than initially expected.

The new algorithm proposed in this contribution will be applied to a measuring campaign performed in a painting conservation-restoration studio. The studio is a learning environment where students and lecturers work together. A previous investigation has demonstrated that indoor air quality is orientation dependent and thus a vector quantity ('t Hart et al. 2016). This contribution shows that the same environment has different impacts on canvas paintings, panel paintings, students and staff. This means that air quality is also a relative concept that depends on the object/subject that is considered in the study.

Background

The 10 agents of deterioration are a conceptual framework used to categorise the major causes of change, loss or damage to cultural heritage objects (Bacharach 2016). This contribution focuses on the agents of deterioration that are governed by indoor air and lighting: incorrect relative humidity, temperature, light and ultraviolet radiation, and pollutants. Concerning these agents, pollution has the greatest impact on the health of visitors, museum employees or conservators-restorers. In addition, heritage objects and persons might affect each other’s “health”. For example, heritage objects may release a variety of gases (Smedemark et al. 2020) or reagents (Eggert and Fischer 2022) which might affect other heritage materials (Storme et al. 2013). Some of these pollutants contribute to the typical odour of a building or of old books and are sometimes considered olfactory heritage (Bembibre and Strlič 2022; Bembibre Jacobo 2020; Tullett et al. 2022). Concerning human health, some heritage objects have been treated in the past with arsenic, mercury or organochlorine pesticides to protect them from pest or mould infestations (Deering et al. 2020; Rushworth et al. 2014; Wörle et al. 2012). Heating and lighting are needed to ensure proper well-being for humans, but this affects the preservation conditions of heritage objects. In conservation-restoration workshops, chemicals are occasionally used to clean painting surfaces or to remove overpaints. When these products evaporate, they affect human health and heritage objects in the room. In all cases, the overall risk caused by indoor air is determined by many parameters. In addition, the environmental parameters that affect human health are not necessarily the same as those that affect the degradation rate of heritage objects. Unfortunately, monitoring all of them is not feasible for technical or economic reasons. For example, no mid-price sensors are available to detect all the individual volatile organic compounds in air or to analyse the presence of heavy metals in particulate matter.

To perform risk assessments, we need to simplify the complex reality to its essence. Therefore, the overall situation should be described with a limited number of essential properties that can be measured with sensors. These fundamental properties are key risk indicators (Afify et al. 2020; Scandizzo 2005; Shi et al. 2018). In the current work, we based this selection on the opinion of institutions such as the World Health Organization (WHO) (World Health Organization 2006), the European Environmental Agency (EEA 2020) (Guerreiro et al. 2014) and the Environmental Protection Agency (EPA) from the USA (Environmental Protection Agency, 1999). They all use the same key risk indicators: the compounds NO, CO, NO2, O3, SO2, and the particulate matter PM2.5 and PM10. The concentration of these pollutants should be as low as reasonably achievable (ALARA principle) (Hendee and Edwards 1986; Bevelacqua 2010; Prasad et al. 2004).

It is known that the pollutant NO2 has a greater toxic effect on humans than CO2 at the same concentration. For both pollutants, it is essential to consider the different thresholds at which the harm is acceptably low when extracting information on health risk from measured concentrations. Therefore, the concentrations of all pollutants must be converted into a level of risk R so that their aggressiveness can be compared with each other. To assure a similar meaning of Ri for all parameters i, the set of risks {R1, R2, R3, …} must agree with the mathematical conventions given below. They correspond with the conventions that are used for subjective probability (De Finetti et al. 2017, p. 8; Kahneman and Tversky 1972; Machina and Schmeidler 1992).

  • Condition of coherence: All levels of risks must be determined coherently. Two parameters or the same parameter measured at different moments can only be compared quantitatively when they are expressed in the same unit (Marx and Engels 2010, 59–60);

  • Exclusion of inadmissible evaluations: There is a situation so safe that it is almost impossible to find a safer one. This situation corresponds with minimal risk of zero. There is also a maximum risk of 1 because it is practically impossible to find a worse situation;

  • Equivalency: Two situations characterised by risks R1 and R2 are considered as equally dangerous when R1 = R2. This rule shows similarities with the equivalence principle used by Marx about the work and value of commodities (Marx and Engels 2010, 59–60);

  • Negation of risk: The same situation can be described by risk or confidence. The level of confidence is the degree of belief that someone is safe and that he will not be harmed by the hazard. If the level of risk increases, then the level of confidence must decrease. They are complementary concepts. Therefore, the level of confidence is given by 1 – R;

  • Combining risks: The combined risk of R1, R2 and R3 must also always be larger or equal to the lowest possible risk 0 and always be smaller than or equal to the highest possible risk, which is 1.

All environmental parameters are converted to a level of risk through a conversion function. This function consists of piecewise linear functions between reference points. Each reference point combines a specific pollutant concentration (x-axis) and a corresponding level of risk (y-axis). The conversions functions are defined by 5 reference points and are defined in Table 1. The best (i.e., reference point 1) and worst situations (i.e., reference point 5) are associated with a risk of 0 and 1, respectively. The level of risk attributed to points 2, 3 and 4 must be lower than that point 5 (i.e., maximum risk) and higher than that of point 1 (i.e., zero risk). For points 2, 3 and 4, an arbitrary but fixed risk value is attributed while considering their ranking. At the same time, the distance between the points has been maximised to improve the visibility of subtle differences in air quality between the points. This results in incremental steps of 0.25 between points.

Table 1 Different kinds of thresholds associated with a level of risk and the corresponding average concentration. The 5 points are ranked according to the concentration value
Full size table

For all pollutants, zero concentration is the best possible air quality and is associated with zero risk. The literature and legislation about thresholds related to human health suggest 4 kinds of threshold limit values: the occupational (Occ.) and non-occupational (Non-Occ.) context for short-term (ST) and long-term (LT) exposure respectively (EU 1998; 2004; 2008; Flanders 2018; Laamanen et al. 2008). Each category is associated with a reference point. Remarkably, the collected threshold values for each reference point vary significantly from one organisation to the other. For example, the US OSHA thresholds are on average, nearly 40% higher than those of Poland (Schenk et al. 2008). This difference is evidence of a lack of consensus between organisations and countries. Assuming that all these values are opinions that contain some truth, the average value can be considered as the consensus value among all experts. The linear interpolation is used because the trends between the reference points are unknown. Coherence between pollutants is assured because the 5 thresholds for each parameter have the same meaning. The reference points are defined in Table 1.

For some reference points, no threshold limit values could be found in the literature. To estimate the threshold concentration of the missing point, a 2nd-degree polynomial curve fitting of form f(x) = a0 + a1x + a2x2 is performed through the other reference points where x represents the key risk indicator concentration and f(x) the level of risk. Since the level of risk of each reference point is known (see Table 1), the corresponding missing threshold value can be calculated. The fitting is performed by using the function curve_fit from the Python library SciPy (van Rossum 1995; Python Software Foundation 2010; Virtanen et al. 2020). This function uses a nonlinear least squares method to fit the best possible function through the known points. To assure that the fitted function increases monotonously for any possible 2nd degree polynomial, it is considered that a2 and a1 are strictly higher than zero. To ensure that the fitted curve does not retrieve negative concentration values, a0 is also bound to be higher than zero. Table 2 summarises the coefficients of the obtained polynomials, the coefficient of determinations R2 and the estimated values (EV) for the missing points. This polynomial is not used as a conversion function because there is no knowledge available about the trend between the points as is suggested by the polynomial and because we wanted to use a similar way of working as the indoor air quality assessment for heritage objects. In addition, the concept of a piecewise linear function is also used in other air quality indices (Zhu and Li 2017; Plaia and Ruggieri 2011).

Table 2 Coefficients of the 2nd degree polynomial obtained after fitting the known reference points, the coefficient of determination between the points and the polynomial, and the estimated values for the missing points as mentioned in Table 1
Full size table

The conversion function is defined by the 5 reference points defined in Table 1. The relative risk assessment in each region is characterised by linear interpolation between the points. Such assessments make it possible to rank situations as better or worse relative to each other. Such a ranking is quantitative but always relative to the worst possible situation. An additional scale is needed to convert the relative evaluations into an absolute risk. To create the absolute scale, the 5 subsequent regions distinguished by the 5 points must be considered. The absolute risk assessment relies on the verbal descriptor that is attributed to each region (see Table 2). The absolute amount of risk depends on the occupational or non-occupational context. The series of descriptors is inspired by the tolerability of risk concept (Carpio de los Pinos et al. 2021; Bouder et al. 2007; Bouder and Löfstedt 2008), but alternative series do exist. The occupational context consists of 5 consecutive regions, each of which is assigned a dedicated risk level. This results in a commonly used classification involving five levels of risk. This scale is adapted for the non-occupational context, where there are only 3 levels. The most favourable category, “Acceptable”, is achieved when the pollutant concentration remains below the Non-Occ., LT threshold (i.e., point 1). The least favourable category, “Intolerable”, is assigned to all regions with concentrations higher than Non-Occ., ST (i.e., points 3, 4 and 5). The region in between is categorized as “High”, as it falls within the range between the two non-occupational thresholds. This approach aligns with the occupational context, where the “High” category also pertains to the concentration range between the two occupational thresholds (specifically, point 4).

The conversion functions as defined in Tables 1 and 3 transform the environmental parameters into a relative level of risk. This makes it possible to compare different parameters and different situations with each other. The absolute level of risk (i.e., risk acceptability) depends on the occupational or non-occupational context in which a person finds himself. The colour scale visualises the combined information of relative and absolute level of risk. An example of the colour scale used for the O3 concentration in a non-occupational and occupational context is shown in Fig. 1. The risk conversion function combines occupational and non-occupational thresholds in a single model. Therefore, Fig. 1a and b use the same conversion function, and no numerical distinction can be made between them. The main difference between both graphs is the consideration of the absolute risk in the colour scale. For this, the concentration range that the colour scale covers are different. In Fig. 1a, the colour gradually changes between zero concentration and the occupational short-term limit value (i.e., point 5); Fig. 1b uses the same colour scale but varies from zero concentration up to the non-occupational short-term limit which is the highest level of acceptable risk in a non-occupational context (i.e., point 3).

Table 3 The 5 subsequent regions characterised by linear interpolation between 2 points and the corresponding verbal descriptor for an occupational and non-occupational context describing the absolute level of risk. The abbreviations refer to the threshold categories in Table 1
Full size table
Fig. 1
figure 1

Graphical representation of the conversion function for O3 using the jet colour scale to visualize both the relative and absolute level of risk. a Occupational context with a colour scale up to the point corresponding to short-term exposure and b non-occupational context with a colour scale up to the point corresponding to short-term exposure

Full size image

Since humans are simultaneously exposed to several key risk indicators, it is the combination of all these individual risks that must be considered. Each key risk indicator can harm the subject in one way or another. There are multiple ways to determine the overall risk. Examples are the sum of all individual risks, the Euclidean norm of all individual risks (Celebi et al. 2011), the highest value of all the individual key risk indicators or the maximum cumulative ratio (De Brouwere et al. 2014). However, the reality is more complex than that. It is possible that all key risk indicators potentially damage the same part of the human body to some extent, damage a different part of the human body, or that each indicator affects several parts of the human body simultaneously. In addition, one can question when harm occurs: (1) when one of the parts starts to fail or (2) when all the affected parts start to fail. Following the concepts of reliability engineering (Trivedi and Bobbio 2017; Birolini 2017; Straub 2014), a complex system can be considered as a combination of components that work in series or in parallel.

It can be assumed that the key risk indicators vary independently, meaning that the potential effect of key risk indicator i, expressed as level of risk Ri, remains uncertain when the potential impact of all other risks is known. In many cases, antagonist or synergistic effects between indicators occur but are only second order effects. This means that the level of risk Ri cannot be written as a function of the other indicators. In such a situation, the total risk RT of a complex hazard can only be estimated by combining the information of all key risk indicators. In addition, the total risk must obey the previously described mathematical conventions. The serial and parallel combinations (see Fig. 2) adhere to these conventions as the overall risk will always stay within the range of 0 and 1, regardless of the number of individual risks being combined. When the individual risks Ri are combined in series, it means that the total risk RT is low when all individual risks are low. This in series combination bears resemblance to the behaviour of the AND-rule in probability. For risks combined in series, the total level of confidence is the negation of risk and is expressed by 1 − Ri. Therefore, the total level of confidence in the complex hazard is expressed as (1 − R1) AND (1 − R2) AND … AND (1 − Rn). The assessment of the total risk RT is visually shown by the green chain in Fig. 2a, and RT can be calculated by the accompanying formula. In this situation, the combined risk of R1, R2, …, Rn is always larger than the minimum value of the set {R1, R2, …, Rn}. When individual risks are combined in parallel, it means that the total risk RT remains low as long as at least one of the individual risks Ri is low. The parallel combination shares similarities with the behaviour governed by the OR-rule in probability. The total level of confidence 1 − RT is related to all possible situations except R1 AND R2 AND … AND Rn (i.e., red chain in Fig. 2b). Here, the combined risk of R1, R2, …, Rn is always lower and thus safer than the minimum value of the set {R1, R2, …, Rn}. The human body will only remain healthy when none of the key risk indicators of the complex hazard is causing harm. In addition, it seems logical that the risk for the human body cannot be lower than min {R1, R2, …, Rn}. This means that for human health the risks Ri must be combined in series. In this study, the total risk for human health has been calculated with the method shown in Fig. 2a.

Fig. 2
figure 2

Two different ways to evaluate the total level of risk while fulfilling the mathematical conventions. a For the thing to be harmed, it is sufficient that one key risk indicator causes harm; b For the thing to be harmed, it is necessary for all key risk indicators to cause harm at the same time

Full size image

Materials and methods

Measuring campaign

A measuring campaign was performed in a conservation-restoration studio for paintings at a Higher Education Institute. The conservation-restoration studio (see Fig. 3) is in a historical city centre. The room contains an entrance door, several windows shielded by curtains, skylights with UV and IR-blocking foils, and an opening to a second room (not seen in the photos). A ventilation system is present but is only active during working hours. To maintain stable preservation conditions, windows are kept closed to avoid draft and to maintain stable climatic conditions inside the room. However, regulations imposed during the covid-19 period required windows to be opened regularly. Also, the curtains are closed to diminish direct sunlight on the objects of art. Despite this, sunlight can still enter the room through the skylights. The location of the monitoring system in the room can be seen in Fig. 3a, indicated by the arrow. Figure 3b shows the room from a different point of view where the window with curtains and the lighting can be seen. The lighting system consists of TL fluorescent lamps and is usually turned on when people are inside the studio. A mobile humidifier (Defensor PH15) is active in the room to maintain a stable relative humidity. Note that the humidifier is next to the monitoring device. Figure 3c shows the chemicals cabinet and the connection of that cabinet to the ventilation system using a flexible arm. There are also bottles for chemical waste underneath the table on the right side of the chemical cabinet (see below the table on the right of the chemical cabinet in Fig. 3c).

Fig. 3
figure 3

Some images of the conservation-restoration studio at a Higher Education Institute. a View towards the monitoring system indicated by the arrow and entrance door; b view towards the window; c chemicals cabinet

Full size image

Several types of activities are performed in the room such as teaching, restoring paintings and manipulation of solvents. These activities generate fluctuations on the environmental conditions in the room. In some cases, these variations are detrimental to cultural heritage materials (e.g., changes in relative humidity) or harmful to human health (e.g., high CO2 concentration suggesting that persons breathe each other exhaled air which increases the transmission of diseases). Occasionally, students also use solvents to remove for example yellowed varnish layers or to clean their tools. There is also a rectangular table in the centre of the studio where some works of restoration are performed. On that table, some often used utensils can be seen, for example paints and solvents. Both lecturers and students have access to this room. Sometimes, a varnish layer must be applied on a painting using a spray gun in a separate but contiguous room. When a window is opened people inside the room are exposed to external pollution sources (e.g., exhaust gases of traffic). All these activities have an influence on the interpretation of the collected data.

Measuring system

There is a wide range of commercially available pre-calibrated sensors and low-cost sensors used to measure environmental parameters. They can be used to build in-house developed multi-sensor tools that simultaneously measure a broad range of several environmental parameters (Schalm et al. 2019, 2022; Martinez et al. 2022). Such technology is used to build compact and wearable measuring devices to perform exposure-assessment studies of persons (Fanti et al. 2022; 2021) or low-cost systems to evaluate the quality of indoor air (Ródenas García et al. 2022) and outdoor air (Martinez et al. 2022). Some define a low-cost sensor as < $100 and a low-cost monitor consisting of one or more sensors and communication/data components as < $1000 (Morawska et al. 2018). This study focuses on the measurements of many environmental parameters, and some of the sensors that are needed are somewhat more expensive (< $500). Therefore, the monitoring system costs about $5000. In addition, the monitoring system is rather bulky because it has been built inside a metal rack to make it transportable. Due to its size, indoor air quality is analysed at a fixed point in the room throughout the campaign.

The monitoring tool consists of a multi-purpose data logger (DataTaker DT85, Thermo Fischer Scientific, Scoresby Vic, Australia) to which a wide range of off-the-shelf analogue sensors are coupled. Temperature, relative humidity, and CO2 are collected with a GMW90 (Vaisala, Helsinki, Finland). The pressure is measured with a PTB110 (Vaisala, Helsinki, Finland). Particulate matter is recorded with the Shinyei sensors PPD-60 (i.e., it detects all particles larger than 0.5 μm) and PPD-20 (i.e., it detects all particles larger than 1 μm) respectively. Concentrations of CO (CO-B4), NO2 (NO2-B43F), O3 (OX-B431), H2S (H2S-B4), SO2 (SO2-B4) and NO (NO-B4) were collected using the B4-sensors of Alphasense with a 32-mm diameter package (Alphasense, Essex, UK). The concentration of total volatile organic compounds (TVOC) was estimated with a photoionization detector using a 10.6 eV lamp (sensor 000-0022-AH2 of Alphasense, Essex, UK). The Panasonic AMN24112 motion sensor inside the monitoring system is directed toward the door shown in Fig. 3a to obtain some information about the traffic of people in and out of the room and, therefore, to get some indication of the presence of people inside the room. This sensor detects motion within a detection range of 10 m. The monitoring system also includes a rod where 3 additional sensors are installed at the height of about 2 m. The SKL310 and SKU421 sensors (Skye Instruments, UK) monitor the intensity of visible light and UV radiation. They are positioned upward to evaluate the light from lighting and skylights shining on surfaces of objects of art. The third sensor is the HD403TS omni-directional hotwire sensor (Delta Ohm, Soest, The Netherlands) and is used to monitor air speed. All sensors are read out in phase with a frequency of 1 min, while the data logger saves 15 min averaged. In the case of the motion sensor, the data logger counts the number of pulses generated by the sensor in a time window of 5 min. The sampling rate is sufficiently high to consider the discrete time series as a continuous signal.

Calibration information for all sensors is provided by the manufacturers. This information has generally been deemed sufficiently reliable, with the exception of the PM sensors from Shinyei. For these sensors, the calibration curve is only available as a graph in the data sheets. To convert this graph into a usable format, we determined the coordinates of the reference points in that graph using a ruler and recreated the graph in Excel. Regarding the gas sensors from Alphasense, their provided calibration is susceptible to climatic conditions (such as temperature and relative humidity), cross sensitivity and other contextual factors. Calibration experiments of various types have been conducted on the Alphasense gas sensors (González Rivero et al. 2023a, b). The analysed sensors exhibit a strong linear correlation between the sensor signal and the concentration of the target gas. However, the intercept and slope of this relationship may not remain constant over time and space. Interestingly, even the cross-interference of other gases demonstrates a linear response to their concentration. Since the measurement campaign takes place in an indoor environment, it can be assumed that the impact of climatic conditions remains relatively constant. Furthermore, the calibration provided by the supplier has yielded realistic concentration ranges. Although there is an uncertainty about the absolute pollutant concentrations (which affect the conversion into risk), the dynamic in the concentration trends appear to reliable (Schalm et al. 2022).

Data visualisation

The trends in the collected data are visualised with different types of graphs described in the list below. The graphs use colour information to allow a more intuitive analysis of the results and the identification of time-dependent patterns present in the time series by visual inspection of the graphs. The colour scheme jet defined in Matplotlib Python library and its reversed variation are used according to the type of data shown in the graphs (Ari and Ustazhanov 2014; Barrett et al. 2005; Bisong 2019). This colour scale ranges from dark blue through blue, cyan, green, yellow, red, to dark red. For the risk assessment, the worst situation with a risk of 1 is associated with a red colour while the best possible air quality of 1 is associated with a blue colour (Leyva Pernia 2019).

  • Temporal raster plot (carpet plots): In this type of graphic representation, the x-axis shows the days in the year and the y-axis the hours in a day. The values at these coordinates are visualized with a colour from a colour scale. The advantage of this type of chart is that it visualises both short-term (hourly) and long-term (monthly) patterns in a single plot. Carpet plots are created by in-house developed software in Python 3, using the pcolormesh function from Matplotlib library.

  • Strip charts: This type of plot is represented as a horizontal bar where the collected data is organised chronologically, from left to right, and composed of small coloured vertical lines. This type of chart provides a direct interpretation of the time series regarding the impact of air on human health by assigning a colour from a colour map to the value obtained from the risk calculation. Such plots are generated by code developed in Python 3, using the imshow function from the Matplotlib library.

Certain parameters show subtle fluctuations while they have extreme values at rare moments. When using a colour scale that assigns a colour to all measurements between the minimum and maximum value of the parameter [MIN, MAX], the colour contrast in the plots may be insufficient. In such situations, the plot is dominated by the sporadic but tall peaks in the time series (i.e., seen as a few pixels in the plot) while the differences in subtle fluctuations are suppressed. Despite the high dynamic range, it is sometimes necessary to show the subtle fluctuations and the rare high values in a single plot. To optimise the display of the plot, one can adjust the colour contrast using the clipping process by attributing a smaller range of measurements [MIN, TC] with the colour scale and displaying all pixels with values above the clipping threshold TC with the same colour (Santhi and Wahida Banu 2015; Kandhway et al. 2020). For each plot, the clipping threshold has been optimised visually.

Results and discussion

In Fig. 4, four temporal raster plots with the intensity of visible light in lux, UV radiation in mW/m2, the motion of humans in front of the monitoring system, and air speed in m/s are shown. The vertical axis shows the hours of the day (from 0 to 24) and the horizontal axis represents the days the measurements are performed (from November 2019 to February 2021). In Fig. 4a and b, it is possible to see intermittent vertical red stripes that match the same moments. These stripes almost always start at around 8:00 and end between 17:00 and 19:00. They are also correlated by human motion and must be due to switching on and off the TL-lighting. The lighting generates visible light and UV-radiation. The lighting gives indirect information about the times and dates when there are people in the room. For example, stripes are absent during lockdown periods or weekends. This is confirmed by Fig. 4c which gives information about people’s movement inside the studio. In Fig. 4a and b, another smoother pattern can be observed that is related to sunlight. In that pattern, the intensity of visible light and UV is lower during winter months and higher during summer. During the period April–August and during the early hours, the light intensity is substantially higher because sunlight can directly penetrate the room during the early hours. These results are related to seasonal variations and the relative position between the room’s windows/skylights and sun trajectory. It is interesting to note that curtains are closed to avoid direct sunlight on the objects of art, but that TL-lighting emits considerable amounts of UV radiation. As light is explicitly mentioned in the 10 agents of deterioration, this measuring campaign suggests that human activity in the room affects the preservation conditions of the paintings. However, the plots do not show how large the impact is of the lightning on paintings. Figure 4d shows the temporal patterns of the air speed in m/s. As in the previous plots, it is possible to differentiate between regular and lockdown periods. Before the lockdown, higher air speed was observed during working hours, although the temporal pattern is not clearly defined. The draft is probably caused by the ventilation system or by opening/closing doors. During the lockdown period, the situation is reversed, and more draft is noticed in the evenings and nights. After the lockdown period, it was mandatory to keep the windows open, resulting in stronger air drafts (i.e., the red vertical stripes in September and October). In general, one can state that there is a band with elevated air speed between 8:00 and 19:00. This band is caused by switching on the ventilation system just before the working hours and switching off after the working hours.

Fig. 4
figure 4

Temporal raster plots of several parameters: a intensity of visible light (lx), b intensity of UV radiation (mW/m2); c human motion, and d air speed (m/s). Belgium was in lockdown due to the covid-19 pandemic between March 18 and April 25, 2019, but the education institute remained closed until the end of the academic year on 30 June, 2019. Then, the holidays started until half September

Full size image

In Fig. 5, three plots represent the concentration of O3 (ppb), SO2 (ppb) and NO2 (ppb), respectively. The main source of these pollutants is located outdoors. For these plots, temporal patterns during working periods can be observed. These patterns have a rectangular shape related to an increase of concentration during the working periods (from around 8:00 to no more19:00). In the case of the outdoor pollutants, they must enter the building/room by infiltration. However, as we know from Fig. 1c, there were no people during summer times inside the studio to open doors or windows, so indoor fluctuations are most probably caused by natural diffusion. This means that outdoor pollutants easily infiltrate the room. In the case of NO2, a faint opposite behaviour is seen: the concentration decreases instead of increasing during working periods, although the pattern is more subtle. The drop in NO2 concentration in periods where outdoor pollution is expected to be higher is a well-known phenomenon in city air. Its occurrence is based on the chemical transformation of O2 in O3 by NOx compounds (Han et al. 2011). It should be remarked that the SO2 concentrations are close to the detection limit of the sensor and that cross-sensitivity of other pollutants may affect the pattern seen in Fig. 5b. The three pollutants shown in Fig. 5 are harmful to humans and paintings.

Fig. 5
figure 5

Temporal raster plot of outdoor pollutants concentration of a O3, b SO2 and c NO2 in ppb. The grey zones in the raster plots represent moments where data were not obtained during the measurement period

Full size image

In Fig. 6, two plots represent the concentrations of CO2 (ppm) and TVOC (mV), respectively. The primary source of these pollutants is indoors. In the case of CO2 concentration (see Fig. 6a), narrow vertical stripes are observed and are correlated with human presence. This plot shows a pattern similar to the motion plot in Fig. 4c but with the difference that the stripes for CO2 are longer because it needs to diffuse out of the room. The fact that the pollutants generated outdoors show a crisp day-night fluctuations (see Fig. 5) in comparison to the CO2 and that these fluctuations are not only occurring when there is human presence, suggest that they are introduced into the painting studio when the ventilation system is switched on and that they dissipate or decompose quickly when it is off.

Fig. 6
figure 6

Temporal raster plot of indoor pollutants concentration of a CO2 (ppm) and b TVOC (mV)

Full size image

The correlation between the two parameters in Fig. 6 is obvious because they are related to people’s presence. The intensity of the stripes is higher during working hours, and they slowly fade when there is no presence of people in the room. The TVOC pollutants (Fig. 6b) are due to the use of solvents for conservation-restoration actions and/or chemicals used during cleaning activities of the room. In the case of TVOC, the raw signal in voltage after zeroing is shown and not the concentration values because it is impossible to convert the signal caused by a mixture of products into concentration units. The pattern in the TVOC-plot shows most of the time a homogenous blue colour indicating that there are little fluctuations of TVOC inside the room. However, some narrow lines of elevated concentration can be observed. These stripes appear only during working days, and it matches with the moments when restoration activities and activities related to the cleaning of the entire room are performed in the room.

In Fig. 7, the overall risk for human health is presented in both contexts, occupational (i.e., lecturers, staff) and non-occupational conditions (i.e., students). When looking at the trends, the spring and summer period appears to be better than autumn and winter, but it is unclear what is causing that improvement (e.g., no heating, lockdown policy). Figure 7 shows how the overall risk fluctuates in values around 0.5, with some peaks reaching 0.75. This indicates that it is highly probable that the subject exposed to this environment is exceeding the non-occupational thresholds (R = 0.5). In some brief moments, it is exceeding the long-term occupational threshold (R > 0.75). Therefore, it is recommended to manipulate solvents close to the ventilation points.

Fig. 7
figure 7

Overall and individual level risk trends for human health are shown in a graph

Full size image

The information shown in Fig. 7 is also shown in Fig. 8a and b. Figure 8a shows the level of risk for occupational conditions using a strip chart. It visualises the same information as in the graphs of Fig. 7, but it uses colour to show the risk fluctuations. Most of the time, the colours fluctuate from light to darker green. There are short moments where the colour is intense red. Different tones of green are predominant during the full measurements, and these colours correspond to the probability of exceeding a non-occupational short-term threshold. Therefore, for the occupational conditions the level of risk is between low and moderate, except for those short periods where it is possible to see red colours that correspond to a high level of risk.

Fig. 8
figure 8

Same information about the level of risk as shown in Fig. 7 but visualised in a strip chart. a Occupational conditions, and b non-occupational conditions. The grey zones represents moments where data were not collected

Full size image

Figure 8b shows the level of risk for non-occupational conditions. The colours always fluctuate between different tones of red. For non-occupational conditions, this means that very frequently, the level of risk corresponds to a situation between high and intolerable for people exposed to those conditions. By comparing Fig. 8a with b, it is possible to see that the same environmental condition is worse for people exposed in non-occupational contexts. The most relevant key risk indicator with the highest impact on the overall risk is the O3. Ozone in the room can be caused by the infiltration of outdoor pollutants through small openings or generated inside the building, such as (old) copying machines and/or laser printers (Lee et al. 2001). It should be remarked that the last generation of laser printers and copiers are equipped with O3 filters, which break down ozone and limit the leftover emission to a minimum (Dhandapani and Oyama 1997). As the highest O3 concentrations (outdoors) typically occur between 12:00 and 22:00 (see Fig. 5a), it is best to ventilate the room in the morning. Filters containing activated carbon, through which air is passed using fans, can also effectively remove ozone for an extended period.

For human health, the level of risk has been visualised in Figs. 7 and 8, while for heritage objects in the same room, the indoor air quality has been shown in Fig. 9. Although the negative impact of indoor air on humans has been emphasised by using the term “risk” and the positive impact of indoor air on heritage objects by using the term “Indoor air quality”, the colour scale contains in essence the same information where 1 means either high risk or bad indoor air quality. The IAQ for canvas and panel paintings are shown in Fig. 9. For both charts presented in Fig. 9, it is possible to see the general IAQ-index and the individual IAQ-indices for the individual parameters used in the analysis (i.e., relative humidity, temperature, illuminance, ultraviolet light, particulate matter, and ozone). The IAQ shows a similar pattern for both types of heritage objects. However, the general IAQ-index suggests that the same environmental condition in the room is more aggressive for panel paintings than for canvas paintings. By looking at Fig. 9, it is also possible to see which parameter is more relevant or contributes more to the general IAQ-index. For example, a sudden increase in temperature affects canvas paintings more than panel paintings. For that reason, the moments with maximum temperature are more pronounced in the general IAQ-index for canvas paintings. It is clear that relative humidity plays an important role in the general IAQ-index for the paintings. This is because wood and canvas are hygroscopic materials and are affected by humidity.

Fig. 9
figure 9

IAQ-index for objects in painting studio for two different materials. a Canvas and b wood

Full size image

Conclusions

In this case study, several environmental parameters were monitored during a measuring campaign performed in a conservation-restoration studio for paintings. Temporal patterns in the time series, such as seasonal, weekly, and daily trends have been visualised using carpets plots. Some of these patterns can be associated with the presence of people or due to their activity inside the painting studio (e.g., breathing, cleaning, turning on lights, etc.). At the same time, the behaviour of some parameters is associated with external sources (e.g., intensity of sunlight, infiltration of exhaust gases from the outdoor air). We have observed that the environmental parameters can worsen at certain specific moments of the day and that these sudden changes seen as peaks can be attributed to indoor or outdoor hazards. The analysis also shows that human activities such as opening the window or switching on the light affect the air quality of heritage objects. From the dynamics seen in the carpet plots, several recommendations to improve the air quality for both human health (e.g., improve mechanical ventilation, manipulate solvents close to the ventilation points, ventilate the room by opening the windows when the outdoor pollution is lower) and heritage objects (e.g., reduce UV-radiation from illumination system, reduce temperature and relative humidity fluctuations). The existing algorithms to evaluate the impact of indoor air on heritage objects and the new algorithm to evaluate the same air on human health show that the same environment has a different impact on canvas paintings, panel paintings, students, and staff. This means that air quality is not a universal concept but that it depends on the object/subject that is considered in the study.