New metrics for noise exposure related to mental health

D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 1 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 New metrics for noise exposure related to mental health Project title: Early Environmental quality and life-course mental health effect Project Number: 874724 Project Acronym: Equal-Life Work package 3 – Deliverable 3.6 HISTORY OF CHANGES Version Publication date Change 1.0 01-12-2023 Suggested Table of Content 2.0 15-11-2024 Draft send out for review 3.0 31-12-2024 Final version H2020 program Grant agreement numbers 874724 (Equal-Life), Project start date: Jan 1st 2020 Duration: 60 months Project Coördinator: Irene van Kamp (RIVM) WP (number and title) Equal-Life WP3 Deliverable Number Equal-Life D3.6 Deliverable Title A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health Due date 28-02-2024 Actual date 31-12-2024 Dissemination Level Public Lead beneficiary 12 – UGent Responsible authors D. Botteldooren E-mail / phone responsible authors +32 9 2649968 Co-authors A. Can T. Renterghem L. Dekoninck Reviewers K. Persson Waye Contributor: in-depth data sleep Gothenborg P. Lercher Contributor: in-depth data sleep Alpine R. Bogers Contributor: ABCD cohort tests L. Sulc Contributor: all cohort random forest data, ADHD J. Gulliver I. Van Kamp D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 3 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Contents Executive Summary .................................................................................................................................... 4 1. Introduction ........................................................................................................................................ 6 1.1. Context ....................................................................................................................................... 6 1.2. Organisation of the report ......................................................................................................... 7 2. New noise indicators for mental health assessment ......................................................................... 7 2.1. The need for new indicators ...................................................................................................... 7 2.2. Criteria towards the definition of noise indicators .................................................................... 8 2.3. Sleep disturbance index ........................................................................................................... 10 2.4. Absence of restorative periods ................................................................................................ 12 2.5. Overview of diurnal indicators ................................................................................................. 15 3. Indicators derived from measurements ........................................................................................... 21 3.1. Towards new indicators via clustering of outdoor measurements .......................................... 21 3.2. Towards new indicators via clustering of indoor measurements ............................................ 29 3.3. Outdoor to indoor .................................................................................................................... 35 3.4. Comparing insulation between Gothenburg and Ghent .......................................................... 42 4. Modelled road traffic noise indicators tested on cohorts ............................................................... 43 4.1. Road traffic noise indicator modelling ..................................................................................... 43 4.2. ABCD ......................................................................................................................................... 44 4.3. Alpine ........................................................................................................................................ 48 5. Combining all available knowledge .................................................................................................. 52 5.1. Methodology for selecting indicators for exposome assessment ........................................... 52 5.1. Correlation and overlap between indicators ........................................................................... 53 5.2. Relevance for mental health, wellbeing and cognitive development ...................................... 58 5.3. Analysis of the graph ................................................................................................................ 62 6. Software for calculating new metrics ............................................................................................... 65 7. Overall conclusion and outlook ........................................................................................................ 66 8. References ........................................................................................................................................ 68 D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 4 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Executive Summary In epidemiological studies focusing on mental health, well-being, and cognitive development, noise exposure is often addressed in a rather imprecise manner. Historically, the lack of affordable and flexible noise monitoring devices necessitated reliance on computational noise models. However, the choice of noise model and its underlying assumptions can lead to significant discrepancies in calculated exposure levels. Three key limitations in current practices can be identified: 1. Façade Noise Levels and Sleep Disturbance: Noise levels are typically calculated for the most exposed façade, neglecting the possibility that a dwelling may have a quieter side where the bedroom—crucial for sleep, an identified mediator for many studied effects—may be located. Furthermore, for assessing sleep disturbance, indoor noise levels are more relevant. Given modern building codes emphasizing energy efficiency, open windows during sleep are no longer standard. Unfortunately, conventional noise models perform poorly in accounting for shielded façades and courtyard acoustics, which can lead to underestimations of exposure in such areas. 2. Simplistic Noise Indicators: The most commonly used indicators, such as façade Lden and Lnight, rely on A-weighted equivalent sound levels, which are straightforward to calculate. While these indicators show high spatial correlation with more sophisticated metrics, they lack the validity needed for nuanced analyses. As a result, researchers often forego more complex calculation models despite their higher accuracy in certain contexts. 3. Limited Source Representation: Noise exposure calculations frequently focus on specific sources (e.g., road traffic) or subsets thereof (e.g., traffic on major roads), excluding other contributors. However, this limitation is partially offset by the fact that specific sources produce characteristic spectro-temporal sound patterns, which can aid in identifying associations with health outcomes. In the Equal-Life project, the aforementioned shortcomings were addressed by integrating direct measurements and introducing conceptually robust noise indicators, alongside methods for their efficient calculation. To analyse the complete sound environment—both outside the bedroom window and inside the children's rooms—data from measurements conducted in Gothenburg and the Ghent area were utilized. A wide range of noise indicators proposed in the literature were calculated for each location at 15minute intervals. Clustering this extensive dataset revealed that specific combinations of noise indicators are particularly effective in identifying distinct "disturbances." These clusters highlighted either unique disturbed sound environments specific to individual measurement sites or common patterns across multiple locations, such as the morning rush hour. Notably, the largest cluster, representing an undisturbed urban living sound environment, frequently occurred at night but persisted for extended periods during the day in many areas. Indoor sound climate analysis based on the Gothenburg measurements revealed that indoor noise levels were less influenced by outdoor sound events than anticipated. This was determined by analysing the number of indoor sound peaks attributable to outdoor sources, as well as clustering indoor sound environments and correlating each cluster with outdoor measurements. Interestingly, only a limited number of indoor sound environment classes showed strong associations with outdoor noise indicators. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 5 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Currently, it is not feasible to efficiently calculate the contribution of all sound sources to detailed spectro-temporal exposure patterns at any location across Europe. As a result, the calculation models in this study focused on road traffic noise. Indicators for road traffic noise were selected based on three criteria: validity, applicability, and transparency, as well as their relevance to expected pathways affecting mental health and cognitive development—namely, sleep, stress, and restoration. For sleep, the focus was on accurately predicting indoor sleep-disturbing sound events. A novel indicator was introduced, synthesizing insights from laboratory and home sleep studies. Regarding stress, it was hypothesized that the home environment—particularly during the evening hours—serves as a critical space for restoration after a stressful day for both parents and children. Consequently, evening median noise levels were identified as a relevant metric. Calculating noise events and median levels, which exclude single transient events, requires precise estimates of local traffic dynamics, including both direct exposure and background noise. To address this, experience from the QSIDE project, which focused on quiet side noise estimation, was integrated into the model. However, despite simplifying assumptions, these models remain too resource-intensive in terms of labour and computational power to be efficiently deployed at scale across Europe. To overcome this limitation, a machine learning-based surrogate model was developed. This approach, combined with simplified traffic intensity estimates and OpenStreetMap data, enables the resulting software to estimate traffic noise indicators at any location in Europe with significantly improved efficiency. The relevance of the proposed indicators and the associated calculation model was assessed through direct correlations with selected outcomes from the ABCD and Alpine cohorts, as well as the preschool children in-depth study. These analyses revealed a small but statistically significant Spearman correlation, suggesting the potential utility of these indicators. However, the proposed indicators also correlated with Lden and Lnight values previously calculated for these cohorts and with proxies such as the distance to the nearest main road. These relationships were further explored using a knowledge graph that integrated findings from previous studies and other components of the Equal-Life project. Additionally, insights from a prior Shapley analysis of the surrogate model (D3.5) supported these observations. The analysis highlighted that the proposed sleep disturbance indicator for different age groups, calculated at the least exposed façade, and the evening median exposure indicators were strongly associated with the presence of major roads within a few hundred meters. Sleep disturbance at the most exposed façade, in contrast, was linked to the length of nearby roads and traditionally calculated Lden values. In summary, there is sufficient direct and indirect evidence to support the relevance of the sound environment—when characterized using innovative indicators—for mental health, well-being, and cognitive development. This warrants continued use of the sleep disturbance index at both the least and most exposed façades of dwellings, as well as median or 90th percentile noise levels at these façades. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 6 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 1. Introduction 1.1. Context It has long been debated whether indicators for noise that are based on equivalent level and standard diurnal patterns suggested by the environmental noise directive (END, EC 2002/49/EC) and widely used in WHO reviews, namely Lnight and Lden, are sufficient to assess all possible effects of environmental sound on health and well-being. For Equal-Life in particular, making indicators more specific for children and young people related to their different diurnal activity pattern and specific sensitivities might be useful. In relation to mental health and well-being, Equal-Life focuses on several potential pathways: sleep, stress and restoration, and coping. In this report we focus on sleep and restoration. The importance of sleep for mental health, wellbeing, and cognitive development is well established (see D1.1 for theoretical considerations). For stress and restoration, a slightly unconventional 1 point of view is taken: stress (both parental and child) has many different causes that are out of control of those involved in creating livable neighborhoods. Yet guaranteeing a place and time for restoration could be a target of urban planning. Providing green and blue space is studied in several other deliverables of Equal-Life, so here we focus in particular on guaranteeing restorative time during time spend at home in the evening. Sound level measurements at home provide a wealth of information not only on the sound as disturbance but also about the presence of nature, humans, recreation, industry, etc. In 2024 sound recognition embedded in the measurement devices (the so-called edge) could reveal all these details. However the current report will use as the finest spectro-temporal resolution one-third octave bands and 125 msec time sampling as more refined recognition was not available at the time of the measurement and privacy regulation has not been adapted to accommodate automatic sound recognition in monitoring. The main innovation in processing sound measurements for Equal-Life consists in clustering sound environments based on the multitude of indicators that can be calculated from these measurements. The software for calculating indicators as well as for clustering them is made available (see Section 6). Modelling of the contribution of specific sources – here road traffic – to the overall sound environment in Europe is mostly done based on CNOSSOS. This model is developed as a tool for strategic noise mapping and targets the standard indicators Lden and Lnight. The propagation part of this model allows to obtain more refined spectro-temporal characteristics, yet for the strategic mapping purposes of the END, its application is restricted by the input data and sound power calculation. This model was first extended to include contributions of individual vehicles and therefore allow calculation of a wealth of new indicators. It was also extended with the QSIDE model for scattering into shielded areas. Such a modelling effort however requires huge computational power, hence a hybrid surrogate model based 1 Traditionally, environmental noise at home is considered an environmental stressor rather than a factor that prohibits restoration at home. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 7 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 on open source data was created that can be used anywhere in Europe to get an estimate of the exposures relevant for mental health and wellbeing. Validation of the new indicators is limited in this report. Further validation could be found in WP1 and WP7 deliverables. 1.2. Organisation of the report The report starts by identifying new indicators based on theoretical considerations. This is related to the work already presented in D3.5 and conference publications [97][98][99][99][101]. Some parts are repeated here for the convenience of the reader (Section 2). The second part of the report introduces techniques for extracting indicators from measurements and clustering them to typical sound environments. The latter is in line with the exposome concept that combines multiple aspects of the living environment of the child. Both outdoor and indoor measurements are considered (Section 3). The third part of the report extends the use of new indicators to the broader cohorts. Here the surrogate model plays a crucial role and the focus is on traffic noise. Proof of concept for the model and the proposed indicators is given on the ABCD and the Alpine cohort (Section 4). While combining the multitude of indicators for mental health and wellbeing into the exposome concept, evidence from Equal-Life’s own results from in-depth studies and cohort studies have to be combined with prior knowledge considering the overlap between indicators and their relevance for the studied outcomes. To this end, a graph-based methodology is proposed in Section 5 and applied to environmental sound indicators. A final Section briefly discusses the open source software. 2. New noise indicators for mental health assessment 2.1. The need for new indicators In general, studies that have investigated the impact of noise on mental health in the past thirty years were based on a methodology that consists of cross-referencing average noise levels obtained from modelling, with the results of standardised tests such as the Strength and Difficulties Questionnaire (SDQ). A recent review [1] 2022) showed that noise exposure, assessed by energetic indicators, has significant associations with non-auditory health effects: psychophysiological, cognitive development, mental health and sleep effects. Percentile and event-based indicators provided significant associations to cognitive performance tasks and well-being dimension aspects. If an overall effect is observed [7], there is no real consensus between studies: some studies tend to show that the effects are not established [1][2][83][84], while others highlight an effect [85][86][87][88][89][90][91][92][5][6] 2 . This 2 More elaborate investigation of the state-of-the-art can be found in WP1 deliverables and publications. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 8 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 lack of consensus may be due to the variability of exposures, as some studies focus on road or air traffic noise exposure situations, some at home and others at school. Further, the variability in the age groups studied may also explain differences in the outcomes. But it is also likely that the acoustic indicators selected itself, which consist of an average of noise levels over long periods, mask some of the effects. Indeed, unlike other pollutants that are active in the definition of the exposome, noise is a non-cumulative quantity: it is not necessarily only the dose of noise received that counts, the temporal distribution of noise levels possibly plays a role. An overview of possible improvements can be found in [93][94]. Some studies have shown for example the advantage of introducing the temporal noise dimension via and additional noise metric: the Intermittency Ratio when assessing annoyance [9], a more transparent indicator reflecting the importance of noticing a sound [95]. Amongst the temporal dimensions, the access to restorative periods of calm, the number and magnitude of noise peaks, can be of interest. Sleep disturbance is also a dimension in noise exposure that should be targeted, as [7] states that “further research on residential childhood (nighttime) traffic noise exposure is needed to determine if the risk of conduct disorders is indeed increased by transportation noise”. In addition, there is much to be gained by understanding which dimensions in noise are responsible for the effects to better target noise mitigation measures: the dose, the absence of quiet periods, noise peaks, certain sound frequencies? There is therefore a need to refine the proposed acoustic indicators to better characterise the links with children's mental health. However, acoustic impact studies are often limited to analyses based on modelled Lden levels, as this is the indicator used in the noise modelling required by the European Noise Directive for the production of strategic noise maps, and standard methodologies do not provide access to other indicators. In addition, it is very difficult – not feasible until recently – to use measurement data for cohorts of several hundred individuals. Fortunately, recent developments offer new insights: • Numerous acoustic indicators have been proposed to characterize noise environments, some of which have been tested in sound pleasantness studies. In addition, the impact of noise dynamics on annoyance and night-time awakenings has been demonstrated [18][19]. • Recently, noise prediction models have made it possible to estimate so-called dynamic indicators, either through costly modelling based on dynamic road traffic modelling [20][21][22][23][96], or statistically through the development of machine learning models [26]. The question of which indicators should be used to study the links between noise and mental health therefore needs to be re-examined. 2.2. Criteria towards the definition of noise indicators Noise indicators can be used for a variety of purpose: to characterize the sound environment, to describe health effects, to support decision-making, or to communicate with the community. Depending on the D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 9 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 objectives sought, the selection criteria to be met may vary. In addition, the different temporal dynamics that characterize sound environments need to be taken into account. The computation of long-term indicators In noise, several time scales are intertwined. On a fifteen minutes timescale, the acoustic environment is relatively stable [10]; in other words, the indicators that describe it vary little from one quarter of an hour to the next. However, even on this time scale, the acoustic environment has its own dynamics, in that it varies from one second to the next (or from one 125 ms time frame to the next). The role of acoustic indicators calculated on the short term, for instance every 15 minutes, is to capture this dynamic, which encompass mean sound levels, the temporal distribution of noise levels, background noise, the number and intensity of sound events, the spectral content. Consequently, over the course of a day, ideally 96 different 15-min sound environments (or 24 different one-hour sound environments) should be characterized by these indicators. Then at the 24-hour time scale, calculating statistics on these 15 min-calculated indicators (maximum, minimum, median, arithmetic mean, percentiles, etc.) serves to highlight, within a day, the arithmetic mean of noise peaks, a percentile on background noise, and so on. The way in which short-term information is aggregated must make sense in terms of the sequences of daily mobility/activity (time at home for instance), and the processes (cognitive or otherwise) leading to a health effect, if the objective is to characterize this effect. If, for example, the aim is to highlight periods of restoration, one can assume that percentiles showing the absence of noisy periods will be of interest. Sub-periods of the day may also be considered, if one wishes to target certain effects of noise, or to highlight certain characteristics of the sound environment (hours relating to sleep, for example). It is these aggregated indicators that can be related to effects. It should be noted that certain indicators, by their design, are already defined for 24-hour periods or for some given subperiods. This is the case, for instance, with the Sleep Disturbance Index (see 2.3), the Absence of Restorative periods ARP (2.4), or estimates on the number of awakenings. Finally, it should be noted that sound environments exhibit a high degree of repeatability from one day to the next. It is therefore usual to show indicators for a typical day only. However, it is also possible to consider statistics at an annual scale by conducting annual statistics on the indicators calculated over a 24-hour period. This can, for example, help emphasize calm periods during weekends or holidays, which are likely to be beneficial from a mental health perspective. However, if the input data for the calculations are modelled, this would mean having access to noise modelling every day, which is costly. Criteria for indicators selection. The qualification of acoustic indicators can be based on the following three criteria: “validity”, “practical applicability”, “transparency” (see Figure 1). These three dimensions have been discussed in the past [13]. The selection of indicators should be considered in light of these three dimensions. If the primary objective is the characterization of effects, then the "validity" criterion should take precedence. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 16 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 ARP Sum of LA50,15mn values (minus 50, with a floor value at 0), over the [16:00-21:00] period at the least exposed facade. See Section 2.4 𝐴𝑅𝑃 = 1 20 ∑ 𝑅𝑒𝐿𝑢(min ℎ𝑜𝑚𝑒𝐿𝐴50,15𝑚𝑖𝑛 −50) 21:00 16:00 . … . . . E . . LA90[16:00-21:00] Evening restorativeness potential for children and young adults: Arithmetical average of the LA90 values in the [16:00-21:00] period at home High continuous noise in the evening will prevent restorativeness . .. .. .. E … LA50[16:00-21:00] Evening restorativeness potential for children and young adults : Arithmetical average of the LA50 values in the [16:00-21:00] period at home High median noise levels in the evening will prevent restorativeness . .. .. .. E … LAeq[16:00-21:00] Energetic average of the 5 values of LAeq,1h (resp. 20 values of LAeq,15mn) in the [16:00-21:00] period High noise levels in the evening will prevent restorativeness .. .. .. … E … EN70[16:00-21:00] Arithmetical average of EN70 (number of events above 70 dB) values in the [16:00-21:00] period A high number of noise events in the evening will prevent restorativeness . .. . . .. … ET70[16:00-21:00] Arithmetical average of ET70 (time above 70 dB) values in the [16:00-21:00] period This indicator possibly underlines intermittent noise (vehicles pass-byes for .. .. . .. … D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 17 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 instance) which will prevent restorativeness NPE Noise Peaks Evening: Weighted average of all the thresholds indicators, calculated over the [16:00-21:00] period. A specific weight is given to each time period and each threshold value. 𝑁𝑃𝐸 =∑ ∑ 𝛼(𝑡)𝛽(𝑡ℎ𝑟𝑒𝑠ℎ) 𝐸𝑇𝑡ℎ𝑒𝑠ℎ,𝑡 𝑡ℎ𝑟𝑒𝑠ℎ=80 𝑡ℎ𝑟𝑒𝑠ℎ=65 𝑡=21 𝑡=16 With 𝛼(𝑡) a weight for the time period, 𝛽(𝑡ℎ𝑟𝑒𝑠ℎ) a weight for the threshold value. At first, one proposes 𝛼(t) = (t-11)/5 between 16:00 and 21:00. Other weighting can be discussed. At first, one proposes 𝛽 =1 for each threshold (this is equivalent to considering a weight of 1 to events above 65dB, 2 to events above 70dB, 3 to events above 75 dB, 4 to events above 80dB, since an event above 80 dB is counted in each of the for cited ET values. This indicators describes the presence of noise events in the evening period. .. .. . .. … EPE Noise Peaks Emergence Evening: Weighted average of all emergence indicators, calculated over the [16:00-21:00] period. A specific This indicators describes the presence of noise events in the evening period .. .. . .. E .. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 18 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 weight is given to each time period and each threshold value. 𝐸𝑃𝐸 = ∑ ∑ 𝛼(𝑡)𝛽(𝑡ℎ𝑟𝑒𝑠ℎ) 𝐸𝑇𝐿50+𝑡ℎ𝑟𝑒𝑠ℎ,𝑡 𝑡ℎ𝑟𝑒𝑠ℎ=10 𝑡ℎ𝑟𝑒𝑠ℎ=5 𝑡=21 𝑡=16 With 𝛼(𝑡) a weight for the time period, 𝛽(𝑡ℎ𝑟𝑒𝑠ℎ) a weight for the threshold value. The weights are chosen as 𝛼(t) = (t-11)/5 between 16:00 and 21:00 which makes the weights vary between 1 and 2. The weight 𝛽(3)=1 and 𝛽(10)=1 while all other thresholds are not considered. measured as emergence above the 15minute LA50. LA05[20:00-22:00] Maximal value of the LA05 values in the [20:0022:00] period A significant proportion of noisy levels during peripheral periods of the children night is likely to shorten sleep duration . .. . . . .. LA05[05:00-07:00] Maximal value of the LA05 values in the [05:0007:00] period A significant proportion of noisy levels during peripheral periods of the children night is likely to shorten sleep duration . .. . . . .. SDI Indicator proposed during Equal-Life, based on the probability of awakening for each 15-minute epoch of the night, summed over the children sleep duration. See Section 2.3 … … . E . D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 19 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 ET[21:00-07:00] Arithmetical average of the ET70 in the [21:0007:00] period Highlights the number of noise peaks during the sleep period .. .. . .. ET[01:00-05:00] Arithmetical average of the ET70 in the [01:0005:00] period Highlights the number of noise peaks during the deep sleep period .. .. . .. NPN Noise Peaks at Night: Weighted average of all the thresholds indicators, calculated over the [21:00-07:00] period. A specific weight is given to each time period and each threshold value. 𝑁𝑃𝑁 =∑ ∑ 𝛼(𝑡)𝛽(𝑡ℎ𝑟𝑒𝑠ℎ) 𝐸𝑇𝑡ℎ𝑒𝑠ℎ,𝑡 𝑡ℎ𝑟𝑒𝑠ℎ=80 𝑡ℎ𝑟𝑒𝑠ℎ=65 𝑡=6 𝑡=21 With 𝛼(𝑡) a weight for the time period, 𝛽(𝑡ℎ𝑟𝑒𝑠ℎ) a weight for the threshold value. At first, one proposes 𝛼(t) = (t-19)/2 between 21:00 and 23:00, 𝛼(t) = 2 between 23:00 and 05:00, 𝛼(t) = (9-t)/2 between 05:00 and 07:00. Other weighting can be discussed. At first, one proposes 𝛽 =1 for each threshold (this is equivalent to considering a weight of 1 to events above 65dB, 2 to events above 70dB, 3 to events above 75 dB, 4 to events above 80dB, since an event More complete than the ET indicators, since it considers all the thresholds and different time steps. In that sense it is not far from relying on probability of awakenings .. .. .. . .. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 20 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 above 80 dB is counted in each of the for cited ET values. Lnight,children Energetic average of the 10 values of LAeq,1h (resp. 40 values of LAeq,15mn) in the [21:00-07:00] period Lnight is often seen as a good descriptor of sleep disturbance .. . .. … E … Lden,age The evening (restorative period) and night start earlier for young children hence age-dependent D,E,N periods are defined Lden is a well known indicator used all over Europe. A slight increase in validity might be observed by accounting for the time of the day when children of different ages are usually at home. .. … E .. Nden Same as above but for loudness (accounting for spectral weighting more accurately than Aweighting) .. .. E . 3. Indicators derived from measurements 3.1. Towards new indicators via clustering of outdoor measurements When there is an opportunity to assess the living environment via (noise) measurements, the question naturally arises how to aggregate the measurements to relevant long-term indicators while preserving as much as the information as possible. 3.1.1. Dataset In Equal-Life two new measurement datasets were collected in the context of the in-depth studies in The Ghent area and in Gothenburg. Both measurement campaigns have in common that the measurements were conducted just outside the bedroom window of the child and that the raw data consisted of 1/3 octave bands acquired with a sampling rate of 8 times per second (which is becoming a de facto standard). This strong similarity allowed us to pool the data from both in-depth studies in the analysis below. Linking the raw data to long-term effects is best done in multiple steps. Here we advocate that a 15minute time interval would be a useful intermittent step as discussed above, and in line with the literature [10]. Today, powerful artificial intelligence tools are available that would allow us to identify the sounds that contribute to the 15-minute outdoor sound environments. However, at the conceptualisation of Equal-Life this possibility was not foreseen and hence we rely on the classical 15minute indicators described above. These indicators derived from the literature in environmental acoustics accurately describe sound environments, with the aim of capturing their effects on human health. The pooled dataset is thus translated to 113050 15-minute intervals collected at 145 locations (which represents a total of more than 1177 days of data). 3.1.2. Methodology The 15-minute intervals are clustered using a powerful combination of UMAP for creating a reduced 8dimensional embedding of the 150-dimensional data and hdbscan for labelling the clusters that emerge in the reduced space based on data density [61][62]. Hdbscan has the advantage that it can ignore data point that are difficult to cluster because they are e.g. very far from other datapoints or e.g. on the border between two clusters. Prior to mapping, the dataset is rescaled to map all indicators to a range around 0 using a sklearn StandardScaler. Parameters for UMAP were n_neighbors=30, min_dist=0.0, n_components=8, random_state=42; parameters for hdbscan were min_samples=20, min_cluster_size=500. Note that these clusters implicitly include exposome components beyond noise as they for example can identify the presence of animal vocalisations and hence the presence of biodiverse green. Once the clusters of 15-minute sound environments are obtained, diurnal patterns can be further analysed. Once more, clustering of the measurement locations based on the diurnal pattern of cluster membership can shed some light on how urban environments develop over the day. Once more, UMAP and hdbscan are used. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 22 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 3.1.3. Results Clustering of 15-minute sound environments based on the above mentioned 150 indicators resulted in 31 clusters. Figure 3 shows the embeddings in 8 dimensions obtained with UMAP, highlighting the second-most populated cluster. Figure 3 clustering of 15-minute sound environments based on UMAP, shown in 8 dimensions plotted two by two (upper left=dimensions 1 &2, upper right= dimensions 3&4, lower left= dimensions 5&6, lower right = dimensions 7&8). The second most populated cluster obtained with hdbscan is highlighted. The number of 15-minute intervals in each cluster is very different depending on the cluster. E.g. cluster 14 contains 59000 15-minute observations, cluster 10 has over 5800 and all other clusters contain less than 2500 elements. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 23 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 To explore the meaning of each of the sound environment clusters, a heatmap is constructed relating each cluster to all the 15-minute indicator. In view of the skewness of the distributions, we opted for plotting the median value of each indicator calculated over all members of the cluster. The stars indicate whether the inter-quartile interval on the indicator values over the cluster is below 0.2 and gives an indication on whether the indicator takes a value that is consistent over the cluster (Figure 4). This analysis of the clusters reveals that the most populated cluster (14) has low noise levels with few events. Low frequencies are moderately present and sharpness indicators are medium. This cluster that is found often and at many locations, seems to indicate a quiet environment. The second most populated cluster (10) has slightly elevated statistical levels with high number and has more tonal components in the 1250 Hz third octave band. As the most populated cluster it has systematically below average event counts and event duration when events are identified against a fixed threshold. Emergence is low but not systematically because of the elevated continuous sound. Sharpness is below average in general as well as statistical. These observations suggest the presence of a ventilation, air conditioning, or other mechanical system close by. The probability of sleep disturbance remains low. A traffic noise model will not be able to distinguish this from the most populated cluster as it is unaware of such sources. The third most populated cluster (29) has a clearly higher sharpness S01 statistic indication short, high pitch events. This shortness of events is also recognized in the event count above LA10 and the absence of longer, loud events. In combination with an investigation of the time of occurrence, we conclude that this corresponds to a typical urban environment with added bird sound, e.g. a morning chorus. A slightly higher probability of sleep disturbance is observed, especially for the slightly more sensitive 6 to 12 year olds. Yet one should keep in mind that this analysis does not take into account the hour of the day yet and thus PSD may be high when most children have already woken up or left the home. Also here, traffic noise models will not be able to distinguish this cluster. Now let us turn to some of the louder clusters. Cluster 13 is mainly observed at one location. The high peaks exceeding various thresholds in combination with a broad spectrum indicate the presence of important traffic infrastructure, rail in combination with road traffic. Cluster 6 is very similar but has lower event counts and less variations in them and combines this with surprisingly high levels at high frequencies. Cluster 2 is characterised by higher emergence counts and a higher spectral canter or gravity. Secondly, the importance of indicators in distinguishing the different sound environments is considered. Median values of levels and statistical levels differ between clusters, but they do seem to take a consistent value as interquartile intervals are large. This might be due to the variation in absolute levels caused by shielding and reflection near the bedroom window which the clustering approach seems to be able to ignore. The parameters related to sudden increase in level (DLAxx) also vary between clusters, but the interquartile interval between 15-minute intervals remains large. Note that the average value D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 24 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 has no effect whatsoever as the average increase and decrease compensate. Spectral information summarised in the centre of gravity of the spectrum, seems to be consistently low in some of the clusters The most surprising observation nevertheless relates to tonality. Low frequency tonality has no effect on the clustering, yet many smaller clusters seem to be formed based on a higher-than-average tonality in specific frequency bands above 500 Hz. This might be due to the presence of cooling and ventilation units near the bedroom windows, which are quite often at the side of the house shielded from traffic and street sounds. Loudness and sharpness show similar trends as other level indicators. From the composite indicators, the probability of sleep disturbance has the smallest spread within some of the (smaller) clusters. Figure 4 Heat map of the median value of each indicator in the clusters obtained purely based on measurement data; stars indicate where the interquartile interval over all observations in this class is below 0.2 giving an indication of the consistency of values over the 15-minute intervals belonging to this cluster (second is a new clustering, ordered by number of measurements in the cluster. Figure 5 illustrates in which measurement locations the clusters are found. The most popular sound environment clusters, 14 and 10, are found at many locations. They also occur at all hours of the day (Figure 6). But also, some of the less populated ones such as 8, 29, and 30 are quite generally encountered and typically occur at certain hours of the day (e.g. early morning). But there are also clusters that are very specific for a single location or a couple of locations, e.g. 11, 13, 15, 16, 24. Except for the last one, these are all very loud situations, with slightly different characteristics. Node 24 seems to be unique in its extremely low background. All in all, the clustering based on 15-minute noise indicators does not seem to differentiate between different flavours of quietness but creates more variations while characterising highly exposed areas. This identification of unique situations may be due to the abundance of indicators focussing on noise D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 25 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 and noise events. These unique situations may be relevant for mental health but will most likely be very difficult to predict. Figure 5 distribution of occurrence of the clusters at each measurement location. The size of the circle indicates the probability of a 15 minute epoch at that location to belong to this cluster . Dwelling numbers have been replaced by 0 for the Ghent dataset and 1 for the Gothenburg dataset. Figure 6 distribution of occurrence of the clusters over the hours of the day. The size of the circle indicates the number of occurrences. Time is in UTC, local time is obtained by adding 1 or 2 hours depending on daylight saving. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 32 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Figure 11 distribution of occurrence of the clusters at each measurement location. The size of the circle indicates the probability of occurrence normalised by Dwelling. Figure 12 distribution of occurrence of the clusters over the hours of the night. The size of the circle indicates the number of occurrences. Time is in local time at Gothenburg. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 33 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Indoor-outdoor comparison Comparison of the cluster of the 15-minute outdoor measurement to the cluster of the indoor measurement for the corresponding 15-minute interval (Figure 13) in general does not reveal a strong one-to-one mapping of indoor and outdoor sound environments. Considering the indoor cluster representing a tranquil environment (21), it can be seen that it is found often in combination with outdoor cluster 14, which is by far the most populated cluster identified as a quiet outdoor environment before. But it can also be seen that the indoor tranquil cluster occurs in combination with different outdoor clusters such as cluster 4 which is also tranquil but has less sharp events or cluster 10 which has even less sharp events. In the same line, the loud cluster 24 that was expected to contain evening indoor sound sources such as voices, music, etc., also occurs together with most outdoor sound environments. Similarly, indoor cluster 7 which was assumed to contain cooling or air conditioning noise co-occurs with several outdoor clusters. Of particular interest is indoor noise cluster 22 which is characterised by continuous rather high noise levels with a relatively low frequency spectrum and a high PSD. It is found in combination with outdoor clusters 5, 6, 15, 9, which are clusters corresponding to loud environments with high PSD but with somewhat different temporal structure, probably all related to traffic. Figure 13 distribution of co-occurrence of clusters based on indoor and outdoor measurements. Size of the circles is proportional to the number of co-occurrent 15-minute intervals normalised by the square root of the total number of occurrences of the two classes that are combined. Clusters are ordered by number of elements with the largest clusters on the right and the top. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 34 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Although clustering of outdoor noise environment seems to be poorly related to indoor nighttime noise clustering, outdoor noise indicators might still be useful to estimate indoor noise clusters. Hence a heatmap of median values of outdoor noise indicators for each indoor noise cluster has been calculated (normalisation is done with respect to all coupled measurements using StandardScaler) and displayed in Figure 14. The difference in outdoor indicators between the most [populated clusters 21 and 24 vanished, which is consistent with this difference being mainly caused by indoor sound. The small difference tends towards less quiet environments where the indoor sound occurs, which could be related to lifestyle differences between neighbourhoods. The indoor sound cluster 7 that had high sharpness probably due to ventilation also has some sharpness outside, indicating that this sound source might also affect the outdoor measurement. The indoor cluster 22, which has already been identified as possibly governed by outdoor (traffic) sound, shows clearly different outdoor indicators such as equivalent levels, statistical levels, and PSD. Cluster 13 like cluster 7 has high indoor sharpness, but this difference is less pronounced in the outdoor indicators which is consistent with an indoor source being responsible for the difference. Figure 14 Heat map of the median value of each outdoor indicator in the indoor noise clusters obtained purely based on measurement data; stars indicate where the interquartile interval over all observations in this class is below 0.2 giving an indication of the consistency of values over the 15-minute intervals belonging to this cluster, ordered by number of measurements in the cluster. All in all, it can be concluded that both clustering of outdoor sound environments and considering single indicators measured outside are insufficient for describing indoor sound environments for most bedrooms (of 18 year olds). From the 11285 combined indoor-outdoor 15-minute measurements, 4002 (35%) cannot clearly be clustered because they either fall between clusters or are very unique, 3682 (32%, clusters 21 and 24) cannot be explained by outdoor sound, 540 (4.7%, clusters 7 and 13) are probably related to ventilation and air conditioning, mainly inside the room, but also noticeable outside D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 35 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 the bedroom window. Contrary, 216 (1.9%, cluster 22) 15-minute measurement epochs could be linked to distinct outdoor sources, probably traffic. For the remaining 26% of 15-minute interval indoor noise clusters, differences in outdoor indicators seem to hint that outdoor sound environments have an influence while outdoor sound clusters are less convincing. However, these clusters often occur at only a few locations and hence conclusions may be influenced by accidental correlations of outdoor environment and indoor sources. 3.3. Outdoor to indoor 3.3.1. From outdoor to indoor sound levels Deliverable 3.7 focuses specifically on building sound insulation. In view of the indoor and outdoor clustering of sound environments with exposome in mind performed above, it is worth investigating in a more classical way how outdoor sound affects indoor sound levels on the same dataset. Indoor levels of noise (and other environmental pollutants) are partly caused by outdoor sources. For some indicators related to sleep disturbance / sleep quality, disturbance of learning in a school environment, and partly even for restoration, the indoor micro-environment is more relevant than the outdoor micro-environment. Without having access to the dwelling or a description of the dwelling by the inhabitant, estimating the state of the indoor environment is rather tedious. What's more, measurements inside buildings, which are fairly rare, do not a priori give any indication of whether the noise levels measured are linked to activity inside the building, or to noise coming from outside. Finally, the acoustic performance of buildings can vary considerably depending on the year of construction, the materials used, etc., and it is therefore very imprecise to rely on standard attenuation values. Successfully establishing building attenuation values from joint outdoor/indoor measurements is therefore of crucial interest. In this section, the Equal-Life measurement campaign in Gothenburg, which contains simultaneous measurements indoor and outdoor, is analysed, in order to propose a methodological framework to answer these concerns. 3.3.2. Analyse of the Gothenburg dataset In cases where indoor micro-environments are impacted by numerous noises originating from housing units, determining the contribution from the outdoors to indoor sound levels can be tedious. However, these outdoor noises are responsible for a significant proportion of health effects and impacts on mental health, as they correspond to sources of endured noise (aircraft noise, road traffic noise, human activities in the street, etc.). Moreover, the relationships established between noise exposure and annoyance are based on road traffic, railway or aircraft noise at the facades of residences. It is therefore essential to know how to distinguish between outdoor noise and indoor noise in noise measured inside homes. This section proposes an in-situ method for determining housing attenuation values in 1/3 octave bands, aiming to determine acoustic indicators beyond simple LAeq, both for indoor and outdoor contributions. The idea is therefore ultimately to decompose the indoor Lf,125ms time series into an outdoor contribution and an indoor contribution. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 36 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Description of the dataset 108 locations were chosen in the Gothenburg region, with simultaneous measurements carried out inside homes and on building facades over a period of 4 days. For practical reasons, 10 noise sensors were moved from one home to another, so that the measurements were not taken simultaneously in all 108 locations. The measurements were taken in 1/3 octave bands from 6.3 Hz to 20 kHz, with a temporal resolution of 125 ms. They were limited to measurements between 9 p.m. and 8 a.m., in order to focus on night time exposure, that are associated with periods of restorative activity and potential sleep disturbance. The sites are relatively quiet, partly because of the night-time periods selected. In detail, the LAeq calculated over the whole measurement campaign duration at the 108 sites ranged from 37.1 dB(A) for the quietest location to 65.8 dB(A) for the loudest one, with an average of 46.6 dB(A). This makes it essential but tedious to be able to distinguish between indoor and outdoor contributions to the indoor sound micro-environment. The proposed method relies on a detailed analysis of noise level attenuations by juxtaposing the time series of outdoor and indoor LAeq,1s. Indeed, in the case of relatively low outdoor noise levels, this method allows for the extraction of outdoor noise contribution while eliminating noise generated inside buildings. However, this requires ensuring that there is no temporal misalignment between the time series. A preliminary analysis thus involved temporally aligning the time series to ensure that outdoor noise peaks corresponded to indoor noise peaks on a second-by-second scale. An autocorrelation function was applied to the 108 data points with the ccf function of R, per hour data packet, to point and correct for any potential temporal drifts. 36% of hourly frames exhibit a temporal misalignment of less than 5 seconds, 27% of less than 1 second. It is noteworthy that for hourly frames with significant temporal misalignment, this is often associated with time series exhibiting low correlation and very low levels of both outdoor and indoor noise. As a result, they most likely correspond to time slots and locations where outdoor noise is very low and has a limited impact on indoor noise. For the subsequent stages of the study, the temporally aligned data is utilized. Furthermore, to prevent very small misalignments, such as 125 ms, from disrupting the association, sliding averages are used, as explained below. Description of the method Figure 15 illustrates the time series of LAeq,125ms for outdoor and indoor measurements for nodes 1036 and node 1001. The figure for node 1036 clearly demonstrates that each outdoor noise peak is associated with an indoor peak. The indoor time series exhibits additional peaks, which are necessarily related to events occurring indoor. For node 1001, the outdoor noise contains significantly fewer peaks, and indoor noise is higher; so it would be necessary here to focus on the loudest peak to estimate the attenuation. The objective will therefore be to automate the search for the attenuation value by selecting the relevant elements in the time series. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 37 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Figure 15 Illustration of LAeq,1s time series indoor and outdoor. Top: node 1036. Bottom: node 1001 The proposed method involves, once the time series are aligned, first selecting the data points where the difference in sound level between the outdoor and indoor is large enough to indicate that the indoor noise originates from the outdoor. This process is conducted for each one-third octave band, as noise peaks outdoors do not necessarily occur at the same frequencies. For example, brake noise, horn noise, or aircraft noise each have distinct frequency content. Therefore, since the goal is to determine attenuation by frequency band, it is necessary to select the emerging time periods by frequency band. Thus, for each 1/3 octave band f between 20 Hz and 20 kHz, the 125 ms data points are selected where the difference in sound levels between outdoor and indoor Lf,outdoor,125ms - Lf,indoor,125ms exceeds 20 dB(A). This ensures that the Lf,indoor,125ms sound level most likely corresponds to a contribution originating from the outdoor environment. Then, only the noisiest 28880 125ms time-frames (which makes 30 minutes) are kept, which correspond approximately to 1% of the dataset. That way, the method retains for each 1/3 octave band f the most prominent noise peaks indoors that originate from the outside. Tests on the variability of the attenuation calculated at this value of 1% have shown that it is a good compromise, and that the determined attenuation values vary little for thresholds in this order of magnitude. The assumption made is that the attenuation of noise by the building is not dependent on the outdoor noise level, and that therefore identifying emerging outdoor noise peaks is sufficient to determine the attenuation. In practice, it is possible that distant, weaker noises correspond to a different attenuation because the directivity of the sound source is not the same, which may involve different physical phenomena, for example if the sound wave passes through both the window and walls with a different directivity. In practice, verifying this would require an extremely low noise level inside the dwellings, which is outside the scope of this study. However, the hypothesis adopted here, consisting of focusing on high outdoor levels in order to determine in situ attenuation, is in line with the regulatory method, D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 38 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 which determines attenuation on the basis of controlled laboratory studies, using a powerful standard noise source. Finally, once the subset of selected data is constituted at the 1/3 octave band f for each of the 108 locations, the attenuation A(f) is determined as the acoustic mean of the a Lf,outdoor,125ms - Lf,indoor,125ms values on the subset of selected data. Note that for each 125 ms period, the sliding value of LAeq,10s(t) replaces LAeq,125ms(t) to avoid issues caused by small temporal misalignments. Considering measurement points separated by 3 meters, the sound propagation delay between the two points—approximately 10 ms—can be neglected. This substitution has a very minor impact on the calculated attenuation values when alignment is perfect, as noise peaks typically have slower dynamics and LAeq gives greater weight to higher levels. However, it prevents aberrant results in cases where the time series are slightly misaligned. Figure 16 Boxplot of the attenuation values determined per 1/3 octave band, over the 108 nodes Figure 16 shows the boxplots of attenuation values per one-third octave band between 20 Hz and 20 kHz, determined for the 108 points. The attenuation curve shows a typical pattern up to 4 kHz, increasing with frequency, after which it becomes less realistic: the curve would logically continue to increase (mass law), yet a decrease is observed. This is due to outdoor levels not being high enough to provide a reliable estimate of attenuation at higher frequencies (such low levels are impossible to measure accurately and are lost in the indoor background noise). We can hypothesize attenuation values above 40 dB for frequencies above 4 kHz; however, this has no impact on the indoor values determined, precisely because of the low outdoor levels at these higher frequencies. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 39 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Figure 17 shows the histogram of attenuation values determined in LAeq. These values are obtained by applying frequency-specific attenuation, assuming pink noise outdoors. The median of the attenuation values is 36.0 dB, with interquartile ranges of 33.1 dB and 40.0 dB. Figure 17 Distribution of the Attenuation values calculated. Once the attenuation value A(f) is determined for a given location, it is possible to calculate the contribution to Lf,indoor,125ms from the corresponding Lf,outdoor,125ms simply by subtracting the attenuation value A(f). This results, for each 1/3 octave band f in two indoor time series, Lf,indoor,125ms and Lf,indoor_from_outdoor,125ms, the latter describing the outdoor contribution to the indoor level. A third time series, Lf,indoor_from_indoor,125ms, represents the acoustic difference between these two latter ones and describes the indoor contribution to indoor sound levels. From these three time-series, typical acoustic indicators, but discriminating between indoor and outdoor contributions, can be calculated. D3.7 gives a few alternatives for handling brut sound insulation. Results To illustrate the possibilities offered by the discrimination of indoor and outdoor contributions to residential noise, various common acoustic indicators have been calculated every 1h in addition to LAeq,1h: statistical indicators, representing each hour sound levels from LAmin to LAmax, passing through various fractile indices: LA95 for the level exceeded 95% of the time, LA90 for the level exceeded 90% of the time, LA50 for the median sound level, LA10, LA5 and LA1 for sound levels exceeded respectively 10%, 5% and 1% of the time. Figure 18, Figure 19, and Figure 20 depict the boxplots, for each hour of the day, calculated over the 108 locations, for LAeq, LA10 and LAmax, respectively. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 40 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Figure 18 LAeq,indoor values, for both indoor and outdoor contributions, calculated over the 108 locations. Figure 19 LA10,indoor values, for both indoor and outdoor contributions, calculated over the 108 locations. Figure 20 LAmax,indoor values, for both indoor and outdoor contributions, calculated over the 108 locations. Discussion Determining the contribution of outdoor noise to the noise measured indoors is challenging, especially in a nighttime noise context where outdoor noise levels are relatively low. In this section, an ad-hoc method is proposed, which has the advantage of determining various indoor acoustic indicators, aiming to improve associations with health effects and mental health. However, several limitations are identified: • It would be interesting to confront the method with diverse datasets, involving daytime measurements. Indeed, the contrasts between outdoor sound levels and indoor sound levels are more significant during the day, which would simplify the determination of the sought attenuation value. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 41 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 • In the analyzed dataset, the indoor contribution to indoor sound levels seems quite high, suggesting that the attenuation values might be underestimated. This could be verified with daytime data containing more pronounced noise peaks, to see if the attenuation is higher or not. • Finally, the assumption is made here that attenuation is independent of outdoor sound level. It is possible that maximum levels are more or less attenuated than the rest. The physics of sound propagation tends to suggest that maximum levels are more attenuated than others. In the example of road traffic, maximum levels are attenuated more strongly than low levels as one moves away from a sound source, and this compression of noise dynamics is well-known in the literature (a vehicle is considered a point source whereas traffic flow is considered a linear source). However, other factors may contradict this for certain outdoor/indoor exposure cases, depending on the directivity of the source, the geometric configuration of the housing and its windows, etc. To test this, one would need to rely on a dataset with highly varied outdoor sound levels and exceptionally low indoor levels, which is beyond the scope of this study and the usual real exposure situations encountered. 3.3.3. Analyse of Ghent dataset The first dataset consists of 72 unique living rooms along an inner ring road 2x2 lanes with traffic lights in the city of Ghent. The non-normalised sound insulation (Lp,out-Lp,in) is obtained from a measurements in front of the front door facing the road (façade level) and measurement inside near the (closed) window of the living room facing the road. The noise source is the traffic sound naturally occurring during the day. All dwellings are exposed to approximately the same noise level, LAeq=65-70 dBA. During pre-processing only data are retained where the traffic noise is well above any background hum that may exist in the house. More details regarding the context of these measurements can be found in [54]. This dataset illustrates that significant differences can be found in brut sound insulations even for dwellings in a similar environment in the same city. Figure 21 shows the median and interquartile interval of the measurements. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 48 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Attention-related indicators BS and ROO may be influenced by noise events during the evening and night and restorative periods in the evening respectively. One could hypothesise that sleep disturbance is a mediator for BS, essentially a reaction time. On the other hand, sleep related indicators do not influence ROO as much as the median level in the evening, hence a wake restorative period. The cohort analysis in Table 7 of D7.2 shows that the average street betweenness scores of every street segment within 300m from each geo-coordinate (avg_betw_300) is the most important physical exposome indicator together with the slope within 300m (slope) in the random forest analysis. Avg_betw_300 is an important input to the noise indicator model as it reflects the expected traffic intensity on the street. Amongst the top 10 most important indicators, some differences can be observed between BS and ROO. For BS, the top 10 contains streets_lenght_500 (total length (m) of street sections within 500 m buffer from residence) and int_close_300 (The mean street intersection closeness scores of every street segment within 300m), while for ROO, int_close_100 (The mean street intersection closeness scores of every street segment within 100m), mj_road_distance (Euclidean distance (m) between residence and the closest major road) and road_distance (Euclidean distance to the closest road) can be found. The latter may indicate the presence of streets with more and more continuous traffic, which may indeed lead to higher L50. Hence there is some consistency between the random forest analysis and the correlations with the advanced indicators that are found here. Finally, we cannot exclude methodological issues. All indicators are calculated based on distance to streets, edge betweenness, and building intensity. This might not be accurate enough to predict e.g. the event-related indicators at the most exposed façade. Some data such as new houses may be missing in the building database, some streets may be wrongly categorised. All these limitations of calculated indicators may have affected the outcome. E.g. inaccurate traffic intensity and composition will have a stronger effect on the prediction of indicators at the most exposed façade that include event indicators. Similarly, the absence of important sound sources like rail and air traffic in the model may result in eventrelated indicators not popping up. 4.3. Alpine The Alpine cohort was mainly used to compare calculation methods for noise indicators. Details on the cohort and the in-depth study on sleep can be found in D7.2 and D1.2. Here we focus on the kindl_mental, kindl_prosocial, and kindl_hedonic subscales because these data were gathered with approximately 1500 children while the in-depth variables were only available for 156 children. As the equivalent levels caused by rail and road traffic were calculated with multiple models, these have been compared first. Three models are investigated: Model 1: sources: traffic data obtained from local authorities; propagation parameters: dwellings, noise barrier, terrain obtained from local authorities; propagation model: Harmonoise 2.5D [4]; approximate setup time: months; approximate run time: weeks. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 49 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Model 2: sources: open street map (OSM) with traffic estimate based on betweenness; propagation parameters: dwellings from OSM; propagation model: Equal-Life surrogate model (D5.3); approximate setup time: hours; approximate run time: one day. Model 3: sources: open street map (OSM) with traffic estimate based on betweenness; propagation parameters: dwellings from OSM; propagation model: Equal-Life surrogate model (D5.3, improved model 6); approximate setup time: hours; approximate run time: one day. Table 6 shows the Spearman’s correlation coefficient between the selected outcome variables and the Lnight calculated using the above models. Although correlations are extremely small, the size of the dataset makes some of them significant. Firstly, we explore the interrelationship between the outcome variables kindl and reported sleep quality, which is considered as a pathway in Equal-Life. The correlation between sleep and kindl-mental is high and positive while it is negative for kind-prosocial and kindlhedonic. In general, Lnight correlates weakly with sleep problems but this relationship is statistically significant except for Lnight highway. The latter might indicate that noise events are important for sleep disturbance. Comparing both models it can be seen that the Equal-Life surrogate model gives roughly the same results as the classical model which requires orders of magnitude more setup time and calculation time. Spearman’s correlation between the results of both models are in the order of magnitude of 0.48 for the most and 0.46 for the least exposed façade, which reveals that the models do not calculate exactly the same quantity but that both capture the essential features for predicting sleep problems. Table 6 Spearman’s correlation between elements of the KINDL scale, sleep quality and Lnight calculated using the different models for different sources. All correlations are very weak but because of the size of the Alpine dataset (n~1500) some are statistically significant at p<0.05 (bold). Spearman’s correlation above 0.05 are highlighted. mental prosocial hedonic sleep problems mental - -0.450 -0.406 0.381 prosocial -0.450 - 0.724 -0.279 hedonic -0.406 0.724 - -0.275 sleep problems 0.381 -0.279 -0.275 - Model 1 Lnight highway 0.043 -0.028 -0.038 0.028 Lnight main road 0.031 0.003 -0.010 0.066 Lnight rail 0.036 -0.015 -0.011 0.064 Lnight total 0.046 -0.026 -0.030 0.081 Model 2 Lnight road least 0.018 0.017 0.020 0.082 Lnight road most 0.051 0.009 0.009 0.056 Model 3 Lnight road least 0.026 0.024 -0.006 0.075 Lnight road most 0.051 0.023 0.014 0.079 D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 50 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 To further investigate the potential relevance of different indicators, Spearman’s correlation between elements of the kindl well-being scale and sleep problems at the one hand and various proposed indicators at the other is calculated and presented in Table 7. The elements of the kindl scale correlate significantly with sleep problems. Noise indicators focussing on the night: SDI, Lnight, Len (an Lden with time intervals according to the child’s age), and LAeq (24hours) correlate significantly with sleep problems. Indicators for noise during the evening also correlate but this is probably due to the fact that they correlate strongly with LAeq and Lnight: for the equivalent levels correlation is close to perfect, for statistical levels it is between 0.5 and 0.6. The only indicator that correlates significantly with any of the kindl subscale values is ARP, an indicator based on the value of L50 above a given threshold, designed as an indicator for the absence of restorative periods during the night. Table 7 Spearman’s correlation between elements of the KINDL scale, sleep quality and several noise indicators calculated using model 3. All correlations are very weak but because of the size of the Alpine dataset (n~1500) some are statistically significant at p<0.05 (bold). Spearman’s correlation above 0.05 are highlighted. mental prosocial hedonic sleep problems mental - -0.444 -0.408 0.390 prosocial -0.444 - 0.725 -0.270 hedonic -0.408 0.725 - -0.278 sleep problems 0.390 -0.270 -0.278 - SDI_0_3l 0.050 0.038 0.006 0.064 SDI_3_6l 0.042 0.029 -0.002 0.084 SDI_6_12l 0.035 0.026 -0.004 0.091 SDI_12_18l 0.039 0.030 0.000 0.091 Lnightl 0.026 0.024 -0.006 0.075 SDI_0_3m 0.040 0.028 0.012 0.077 SDI_3_6m 0.032 0.030 0.016 0.073 SDI_6_12m 0.028 0.030 0.012 0.072 SDI_12_18m 0.028 0.031 0.012 0.068 Lnightm 0.051 0.023 0.014 0.080 ARP 0.093 0.001 -0.027 0.013 LAeqm 0.054 0.015 0.010 0.083 LA90_evm 0.020 -0.006 -0.006 0.048 N90_evm 0.017 -0.002 -0.004 0.046 LA50_evm 0.012 0.015 0.001 0.048 N50_evm 0.008 0.020 0.004 0.047 LAeq_evm 0.056 0.010 0.006 0.084 Len_3_12m 0.037 0.020 0.012 0.081 Len_12_18m 0.036 0.022 0.013 0.079 EPEm 0.025 0.017 0.017 -0.017 LAeql 0.030 0.023 -0.006 0.079 D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 51 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 LA90_evl 0.001 0.022 -0.007 0.044 N90_evl -0.002 0.026 -0.001 0.050 LA50_evl 0.004 0.034 -0.003 0.059 N50_evl 0.002 0.035 -0.003 0.058 LAeq_evl 0.034 0.020 -0.009 0.082 Len_3_12l 0.020 0.030 0.001 0.069 Len_12_18l 0.017 0.033 0.002 0.067 EPEl 0.033 0.002 0.002 -0.015 Discussion Lnight does not show any significant Spearman’s correlation with any of the components of the kindl wellbeing scale. This holds for traffic noise as a whole (calculation model 1) and road traffic noise (calculation models 1, 2, and 3). Reported sleep problems are not specifically related to noise in the questionnaire. Thus, in contrast to the results of questionnaires specifically asking for sleep and noise, the variance explained by Lnight is expected to be lower: Spearman’s correlation for road traffic noise is between 0.06 and 0.08 for all models considered. A few interesting observations can be made: (1) highway Lnight shows much less correlation than the same indicators calculated for the main road and rail traffic. This could indicate that noise peaks are more relevant than the steady noise of the highway. However, one should keep in mind the specific situation of this Alpine area where the highway impact is limited to the main valley. In this study area, night time railway activity cannot be neglected compared to road traffic contributions to sleep disturbance. Comparing models shows that for road traffic noise the Lnight calculated with the extensive model and with the surrogate models gives roughly the same correlations. This indicates that the main contributions to the dynamic noise are indeed captured by the surrogate model. Model 3 which is based on the 6th variant of the trained surrogate model (see D3.5) gives more similar correlations for the most and least exposed façade than model 2. These observations are in line with the results reported in [65]. There a linear regression coefficient of the order of 0.1 is found for the relationship between Lnight and sleep problems. Moreover, using structural equation modelling it is shown that 0.03 of the effect occurs through the perceived neighbourhood quality indicator. Most of the effect also persists when the availability of green and a garden is taken into account. From the new indicators, all nighttime indicators correlate significantly with sleep problems. Even with the more balanced model 3 indicators calculated at the least exposed façade slightly outperform those calculated at the most exposed façade. This trend is generally observed, but we should caution the reader that this may also be due to a inaccuracies in the modelling. LAeq during the evening or even over the whole day show a very similar correlation with sleep problems as the nighttime indicators. This is probably due to the very strong correlation between both sets of indicators caused by the underlying traffic situation in this area. On the contrary, statistical levels correlate less. These are determined by continuous traffic such as the traffic on the highway and it was already established above that highway noise contributes less to sleep problems. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 52 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Surprisingly, the only indicator that correlates significantly with any of the kindl subscales is ARP. In addition equivalent levels show some effect on the mental health subscale. 5. Combining all available knowledge 5.1. Methodology for selecting indicators for exposome assessment The Equal-Life project aims at relating mental health, cognitive development, and well-being of children and young people to the exposome, including both physical and social components. In this deliverable, the focus is on a specific group of indicators for the sound environment within the broader definition of exposome. The selection of suitable indicators and concepts can nevertheless follow the same methodology as the general definition of (physical) exposome. Ideally it is based on both prior knowledge from the systematic literature review (WP1) and relationships observed in the cohort studies (D7.3) and in-depth research (D1.3). As discussed in Section 2.2, indicators should fulfil multiple requirements. To assess the validity of indicators, proof of a significant relationship with specific effects such as ADHD, internalizing/externalizing problems, selective attention should be found. Each indicator is also representative for a concept or dimension that is believed to affect the outcomes. These various relationships are shown in Figure 22. Figure 22. Well-being, Mental health, and Cognitive development of children is affected by the exposome over early life. This exposome can be defined by a number of concepts. These concepts are measured via precisely defined indicators that have a degree of representativeness for the concepts. WM&C are also made operational via a set of defined and measurable effects which contribute to WM&C with a certain severity. The weights in the graph are extracted from evidence-based expert judgement. Relationships represented by dashed lines are deducted from these weights whenever possible needed. The methodology starts by condensing knowledge on indicators and indicator-effect relationships in a database (excel sheets) containing (1) the indicators for physical and social exposure that have been used, either pre-existing in some of the cohort studies, calculated within the Equal-Life project, or simply drawn from literature; (2) the relationships between these indicators and between these indicators and outcomes. The indicator sheet has additional information on: • the concept that it is an indicator for (e.g. air pollution); D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 53 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 • how representative the indicator is for this concept, rated between 0=not at all and 1=extremely based on expert opinions; this is Wi,c • how the indicator is obtained (calculated/measured); • and whether the indicator relies solely on open data available in the EU. The latter aspect is important to allow to produce tools that are directly applicable in the whole EU avoiding the huge effort to collect local data. Additional information entered for each relationship in the relationships sheet includes: • the source of evidence for proposing this relationship; • the strength rated on a [0,1] scale, 1 indicating that the relationship is extremely important; • the confidence of belief in this relationship rated on a [0,1] scale, 1 indicating absolute certainty; this confidence can come from cohort studies, in-depth studies or literature when the relationship is between an exposome indicator and an outcome and from the calculation process or from literature when it is between exposome indicators. Relationships come in two forms, relationships between indicators and effects, with strength measures as Wi,e and relationships between indicators with strength measures as Wi,i. The quantification of relationships can be guided by the data (e.g. correlations over a single cohort) but will require human expert knowledge. Once all relationships have been quantified, this knowledge can be represented in a knowledge graph. Centrality of a node in this graph is then used to measure the importance of the indicator. 5.1. Correlation and overlap between indicators To estimate the overlap between indicators and associated weights, as Wi,I one can partly rely on the correlation (Pearson or Spearman) between indicators over a certain dataset. Yet in addition theoretical considerations as well as in depth knowledge on the construction of the indicators needs to be used. Moreover, some indicators are only available in some datasets (cohorts). SDI_0_3l SDI_3_6l SDI_6_12l SDI_12_18l ARP Lnightl SDI_0_3m SDI_3_6m SDI_6_12m SDI_12_18m Lnightm LAeqm LA90_evm N90_evm LA50_evm N50_evm LAeq_evm Len_3_12m Len_12_18 m EPEm LAeql LA90_evl N90_evl LA50_evl N50_evl LAeq_evl Len_3_12l Len_12_18l EPEl SDI_0_3l 1.00 0.89 0.74 0.75 0.46 0.64 0.32 0.33 0.32 0.32 0.32 0.34 0.27 0.36 0.29 0.37 0.35 0.37 0.37 0.02 0.62 0.24 0.32 0.34 0.47 0.61 0.59 0.58 0.13 SDI_3_6l 0.89 1.00 0.93 0.93 0.35 0.81 0.39 0.43 0.44 0.44 0.42 0.44 0.40 0.48 0.43 0.49 0.44 0.48 0.48 -0.02 0.78 0.37 0.44 0.48 0.61 0.77 0.76 0.75 0.09 SDI_6_12l 0.74 0.93 1.00 1.00 0.27 0.91 0.48 0.56 0.59 0.59 0.54 0.54 0.50 0.52 0.58 0.60 0.54 0.58 0.59 -0.03 0.87 0.46 0.49 0.61 0.69 0.86 0.86 0.86 0.04 SDI_12_18l 0.75 0.93 1.00 1.00 0.27 0.90 0.49 0.57 0.60 0.60 0.55 0.54 0.49 0.51 0.57 0.59 0.54 0.58 0.59 -0.02 0.86 0.45 0.48 0.60 0.68 0.84 0.85 0.84 0.04 ARP 0.46 0.35 0.27 0.27 1.00 0.20 0.15 0.15 0.13 0.14 0.15 0.18 0.23 0.38 0.20 0.31 0.19 0.22 0.22 -0.13 0.21 0.25 0.45 0.25 0.44 0.21 0.23 0.24 -0.08 Lnightl 0.64 0.81 0.91 0.90 0.20 1.00 0.39 0.48 0.53 0.52 0.48 0.47 0.51 0.48 0.57 0.53 0.47 0.52 0.53 -0.06 0.97 0.49 0.48 0.63 0.64 0.96 0.96 0.95 0.01 SDI_0_3m 0.32 0.39 0.48 0.49 0.15 0.39 1.00 0.95 0.86 0.87 0.88 0.86 0.42 0.44 0.63 0.73 0.85 0.86 0.86 0.23 0.36 0.25 0.23 0.37 0.39 0.35 0.37 0.38 0.00 SDI_3_6m 0.33 0.43 0.56 0.57 0.15 0.48 0.95 1.00 0.97 0.97 0.94 0.91 0.48 0.47 0.71 0.76 0.90 0.91 0.91 0.23 0.44 0.32 0.29 0.46 0.47 0.44 0.46 0.47 -0.03 SDI_6_12m 0.32 0.44 0.59 0.60 0.13 0.53 0.86 0.97 1.00 1.00 0.93 0.89 0.52 0.49 0.75 0.76 0.88 0.91 0.91 0.20 0.50 0.37 0.33 0.53 0.52 0.49 0.52 0.53 -0.06 SDI_12_18m 0.32 0.44 0.59 0.60 0.14 0.52 0.87 0.97 1.00 1.00 0.92 0.88 0.52 0.48 0.75 0.76 0.87 0.90 0.90 0.20 0.49 0.37 0.33 0.53 0.52 0.48 0.51 0.52 -0.06 Lnightm 0.32 0.42 0.54 0.55 0.15 0.48 0.88 0.94 0.93 0.92 1.00 0.98 0.44 0.44 0.64 0.69 0.96 0.96 0.95 0.31 0.46 0.27 0.25 0.42 0.42 0.45 0.47 0.47 0.02 LAeqm 0.34 0.44 0.54 0.54 0.18 0.47 0.86 0.91 0.89 0.88 0.98 1.00 0.46 0.50 0.64 0.71 1.00 0.99 0.98 0.29 0.45 0.28 0.28 0.41 0.44 0.45 0.47 0.47 0.01 LA90_evm 0.27 0.40 0.50 0.49 0.23 0.51 0.42 0.48 0.52 0.52 0.44 0.46 1.00 0.87 0.90 0.82 0.47 0.57 0.60 -0.55 0.54 0.89 0.80 0.85 0.79 0.54 0.63 0.65 -0.69 N90_evm 0.36 0.48 0.52 0.51 0.38 0.48 0.44 0.47 0.49 0.48 0.44 0.50 0.87 1.00 0.79 0.88 0.51 0.61 0.63 -0.44 0.50 0.72 0.80 0.71 0.77 0.50 0.57 0.58 -0.51 LA50_evm 0.29 0.43 0.58 0.57 0.20 0.57 0.63 0.71 0.75 0.75 0.64 0.64 0.90 0.79 1.00 0.92 0.64 0.73 0.76 -0.36 0.57 0.77 0.70 0.86 0.80 0.57 0.65 0.68 -0.50 N50_evm 0.37 0.49 0.60 0.59 0.31 0.53 0.73 0.76 0.76 0.76 0.69 0.71 0.82 0.88 0.92 1.00 0.71 0.79 0.82 -0.27 0.52 0.65 0.68 0.73 0.76 0.52 0.58 0.60 -0.38 LAeq_evm 0.35 0.44 0.54 0.54 0.19 0.47 0.85 0.90 0.88 0.87 0.96 1.00 0.47 0.51 0.64 0.71 1.00 0.99 0.98 0.28 0.45 0.29 0.28 0.41 0.44 0.45 0.47 0.47 0.00 Len_3_12m 0.37 0.48 0.58 0.58 0.22 0.52 0.86 0.91 0.91 0.90 0.96 0.99 0.57 0.61 0.73 0.79 0.99 1.00 1.00 0.18 0.50 0.39 0.39 0.51 0.53 0.50 0.53 0.53 -0.07 Len_12_18m 0.37 0.48 0.59 0.59 0.22 0.53 0.86 0.91 0.91 0.90 0.95 0.98 0.60 0.63 0.76 0.82 0.98 1.00 1.00 0.15 0.52 0.42 0.42 0.54 0.56 0.51 0.54 0.55 -0.10 EPEm 0.02 -0.02 -0.03 -0.02 -0.13 -0.06 0.23 0.23 0.20 0.20 0.31 0.29 -0.55 -0.44 -0.36 -0.27 0.28 0.18 0.15 1.00 -0.09 -0.67 -0.62 -0.51 -0.47 -0.09 -0.17 -0.20 0.72 LAeql 0.62 0.78 0.87 0.86 0.21 0.97 0.36 0.44 0.50 0.49 0.46 0.45 0.54 0.50 0.57 0.52 0.45 0.50 0.52 -0.09 1.00 0.53 0.51 0.65 0.66 1.00 0.99 0.98 -0.03 LA90_evl 0.24 0.37 0.46 0.45 0.25 0.49 0.25 0.32 0.37 0.37 0.27 0.28 0.89 0.72 0.77 0.65 0.29 0.39 0.42 -0.67 0.53 1.00 0.88 0.92 0.84 0.54 0.63 0.66 -0.72 N90_evl 0.32 0.44 0.49 0.48 0.45 0.48 0.23 0.29 0.33 0.33 0.25 0.28 0.80 0.80 0.70 0.68 0.28 0.39 0.42 -0.62 0.51 0.88 1.00 0.82 0.90 0.52 0.60 0.62 -0.57 LA50_evl 0.34 0.48 0.61 0.60 0.25 0.63 0.37 0.46 0.53 0.53 0.42 0.41 0.85 0.71 0.86 0.73 0.41 0.51 0.54 -0.51 0.65 0.92 0.82 1.00 0.93 0.65 0.75 0.77 -0.57 N50_evl 0.47 0.61 0.69 0.68 0.44 0.64 0.39 0.47 0.52 0.52 0.42 0.44 0.79 0.77 0.80 0.76 0.44 0.53 0.56 -0.47 0.66 0.84 0.90 0.93 1.00 0.66 0.74 0.76 -0.47 LAeq_evl 0.61 0.77 0.86 0.84 0.21 0.96 0.35 0.44 0.49 0.48 0.45 0.45 0.54 0.50 0.57 0.52 0.45 0.50 0.51 -0.09 1.00 0.54 0.52 0.65 0.66 1.00 0.99 0.98 -0.03 Len_3_12l 0.59 0.76 0.86 0.85 0.23 0.96 0.37 0.46 0.52 0.51 0.47 0.47 0.63 0.57 0.65 0.58 0.47 0.53 0.54 -0.17 0.99 0.63 0.60 0.75 0.74 0.99 1.00 1.00 -0.14 Len_12_18l 0.58 0.75 0.86 0.84 0.24 0.95 0.38 0.47 0.53 0.52 0.47 0.47 0.65 0.58 0.68 0.60 0.47 0.53 0.55 -0.20 0.98 0.66 0.62 0.77 0.76 0.98 1.00 1.00 -0.18 EPEl 0.13 0.09 0.04 0.04 -0.08 0.01 0.00 -0.03 -0.06 -0.06 0.02 0.01 -0.69 -0.51 -0.50 -0.38 0.00 -0.07 -0.10 0.72 -0.03 -0.72 -0.57 -0.57 -0.47 -0.03 -0.14 -0.18 1.00 Table 8 Pearson’s cross correlation between the road traffic noise indicators on the ABCD cohort. Orange highlights strong positive correlations, green highlights strong negative correlations. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 55 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 SDI_0_3l SDI_3_6l SDI_6_12l SDI_12_18l ARP Lnightl SDI_0_3m SDI_3_6m SDI_6_12m SDI_12_18m Lnightm LAeqm LA90_evm N90_evm LA50_evm N50_evm LAeq_evm Len_3_12m Len_12_18m EPEm LAeql LA90_evl N90_evl LA50_evl N50_evl LAeq_evl Len_3_12l Len_12_18l EPEl dist_streets_2018 -0.04 -0.03 0.00 0.00 0.03 0.02 0.04 0.03 0.02 0.03 -0.03 -0.06 0.10 0.07 0.12 0.09 -0.06 -0.02 0.00 -0.07 0.02 0.12 0.11 0.12 0.09 0.02 0.04 0.04 -0.08 length_streets_100_2018 0.01 0.00 -0.04 -0.05 -0.08 -0.05 -0.16 -0.15 -0.15 -0.16 -0.09 -0.07 -0.20 -0.15 -0.24 -0.21 -0.07 -0.11 -0.12 0.06 -0.05 -0.17 -0.17 -0.19 -0.16 -0.05 -0.08 -0.09 0.13 length_streets_300_2018 -0.04 -0.07 -0.10 -0.10 -0.08 -0.15 -0.02 -0.05 -0.07 -0.07 -0.03 -0.01 -0.25 -0.21 -0.24 -0.18 -0.01 -0.06 -0.08 0.10 -0.17 -0.21 -0.23 -0.22 -0.20 -0.17 -0.19 -0.20 0.13 length_streets_500_2018 -0.06 -0.09 -0.11 -0.10 -0.08 -0.17 0.02 0.01 -0.01 -0.01 0.02 0.02 -0.23 -0.22 -0.18 -0.14 0.02 -0.02 -0.04 0.11 -0.20 -0.20 -0.23 -0.19 -0.18 -0.20 -0.21 -0.21 0.09 dist_roads_2018 -0.04 -0.05 -0.11 -0.12 0.05 -0.06 -0.22 -0.25 -0.27 -0.27 -0.24 -0.20 -0.04 0.01 -0.15 -0.12 -0.19 -0.19 -0.19 -0.13 -0.05 -0.04 0.02 -0.12 -0.07 -0.04 -0.05 -0.06 -0.02 length_roads_100_2018 0.08 0.12 0.19 0.20 -0.02 0.13 0.38 0.40 0.40 0.41 0.35 0.31 0.09 0.05 0.22 0.22 0.30 0.30 0.30 0.18 0.10 0.06 0.02 0.13 0.11 0.10 0.11 0.11 0.04 length_roads_300_2018 -0.01 0.01 0.08 0.09 -0.04 0.04 0.15 0.19 0.21 0.22 0.17 0.14 0.06 -0.01 0.15 0.10 0.13 0.13 0.13 0.05 0.04 0.08 0.01 0.14 0.09 0.03 0.05 0.06 -0.06 length_roads_500_2018 -0.02 -0.02 0.01 0.02 -0.03 -0.01 0.10 0.12 0.13 0.13 0.11 0.09 0.10 0.02 0.12 0.07 0.08 0.09 0.09 -0.03 0.00 0.14 0.05 0.15 0.08 0.00 0.03 0.04 -0.13 dist_majorroads_2018 -0.13 -0.19 -0.20 -0.19 -0.08 -0.20 -0.12 -0.15 -0.15 -0.15 -0.13 -0.16 -0.46 -0.36 -0.36 -0.33 -0.16 -0.20 -0.21 0.39 -0.22 -0.46 -0.37 -0.39 -0.37 -0.23 -0.27 -0.28 0.46 length_majorroads_100 0.22 0.23 0.21 0.21 0.25 0.17 0.27 0.27 0.26 0.27 0.25 0.28 0.23 0.36 0.28 0.39 0.29 0.30 0.31 -0.04 0.17 0.14 0.21 0.18 0.26 0.17 0.18 0.18 -0.07 length_majorroads_300 0.21 0.30 0.29 0.27 0.16 0.27 0.14 0.19 0.21 0.21 0.18 0.22 0.36 0.45 0.36 0.40 0.23 0.27 0.28 -0.21 0.29 0.30 0.39 0.33 0.41 0.29 0.31 0.32 -0.22 length_majorroads_500 0.19 0.27 0.26 0.24 0.14 0.25 0.09 0.13 0.15 0.14 0.11 0.16 0.41 0.44 0.36 0.37 0.17 0.21 0.23 -0.32 0.28 0.38 0.44 0.38 0.43 0.29 0.32 0.33 -0.31 built_ESM_300_2015 -0.14 -0.19 -0.20 -0.19 -0.08 -0.28 0.02 0.00 -0.02 -0.02 -0.01 -0.01 -0.23 -0.20 -0.16 -0.12 -0.02 -0.05 -0.07 0.06 -0.29 -0.18 -0.20 -0.17 -0.16 -0.30 -0.29 -0.29 0.03 built_ISA_300_2015 -0.12 -0.18 -0.22 -0.20 -0.05 -0.30 0.04 0.01 -0.03 -0.02 -0.01 0.00 -0.25 -0.19 -0.17 -0.10 0.00 -0.05 -0.06 0.07 -0.32 -0.23 -0.21 -0.21 -0.18 -0.33 -0.32 -0.32 0.02 infrast_100_2018 0.08 0.10 0.11 0.12 0.04 0.07 0.18 0.19 0.19 0.19 0.19 0.19 -0.05 0.02 0.07 0.11 0.18 0.17 0.16 0.17 0.05 -0.11 -0.06 -0.03 0.01 0.04 0.03 0.02 0.14 urbanhigh_500_2018 -0.10 -0.14 -0.16 -0.14 -0.07 -0.23 0.03 0.02 0.00 0.00 0.01 0.01 -0.14 -0.14 -0.11 -0.08 0.01 -0.02 -0.03 0.06 -0.23 -0.10 -0.15 -0.12 -0.13 -0.24 -0.22 -0.22 0.00 urbanlow_500_2018 0.07 0.11 0.13 0.13 -0.03 0.19 -0.02 0.01 0.03 0.02 0.01 0.00 0.09 0.06 0.07 0.03 -0.01 0.02 0.02 0.02 0.20 0.08 0.06 0.11 0.07 0.20 0.19 0.19 0.07 green_urb_500_2018 0.07 0.11 0.14 0.13 0.04 0.17 -0.02 0.01 0.04 0.04 0.03 0.02 0.23 0.17 0.16 0.09 0.02 0.06 0.07 -0.10 0.19 0.24 0.21 0.22 0.17 0.20 0.21 0.21 -0.10 water_500_2018 -0.02 -0.06 -0.09 -0.09 -0.01 -0.05 -0.04 -0.09 -0.12 -0.11 -0.09 -0.08 -0.17 -0.12 -0.14 -0.09 -0.08 -0.10 -0.11 0.05 -0.06 -0.22 -0.15 -0.22 -0.15 -0.07 -0.09 -0.10 0.09 lum_500_2018 0.11 0.15 0.19 0.18 0.01 0.25 -0.01 0.02 0.06 0.05 0.04 0.03 0.19 0.14 0.17 0.10 0.03 0.06 0.07 -0.07 0.26 0.18 0.16 0.22 0.17 0.26 0.26 0.26 -0.02 pop_WP_300_2020 -0.01 -0.02 -0.01 0.00 -0.03 -0.09 0.14 0.15 0.15 0.15 0.14 0.14 0.02 -0.01 0.06 0.07 0.14 0.12 0.12 0.01 -0.10 0.05 -0.01 0.06 0.04 -0.10 -0.08 -0.07 -0.08 ndvi_5yrs_all_300_2011 0.09 0.15 0.19 0.18 0.04 0.25 -0.03 0.01 0.05 0.05 0.03 0.01 0.24 0.17 0.16 0.08 0.02 0.06 0.07 -0.05 0.27 0.25 0.20 0.24 0.17 0.28 0.28 0.28 -0.03 ndvi_5yrs_all_std_300 0.07 0.13 0.18 0.18 0.04 0.23 0.04 0.08 0.11 0.11 0.09 0.08 0.37 0.26 0.28 0.19 0.08 0.12 0.14 -0.23 0.25 0.38 0.31 0.33 0.26 0.25 0.28 0.29 -0.24 ndvi_5yrs_greenest_300 0.09 0.14 0.19 0.18 0.04 0.24 -0.01 0.03 0.07 0.07 0.05 0.03 0.27 0.18 0.19 0.10 0.03 0.08 0.09 -0.07 0.26 0.28 0.22 0.26 0.19 0.27 0.28 0.28 -0.06 ndvi_5yrs_greenest_std_300 0.05 0.11 0.15 0.14 0.02 0.21 0.00 0.03 0.05 0.05 0.04 0.03 0.23 0.18 0.17 0.11 0.03 0.06 0.07 -0.11 0.22 0.23 0.21 0.21 0.18 0.22 0.23 0.23 -0.11 msavi_5yrs_all_300_2011 0.08 0.14 0.18 0.17 0.05 0.24 -0.05 -0.01 0.02 0.02 0.01 -0.01 0.20 0.16 0.13 0.06 -0.01 0.03 0.04 -0.03 0.26 0.21 0.18 0.20 0.15 0.27 0.27 0.27 0.01 D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 56 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 msavi_5yrs_all_std_300 0.10 0.17 0.22 0.21 0.06 0.28 0.00 0.04 0.08 0.07 0.06 0.05 0.32 0.24 0.23 0.15 0.05 0.10 0.11 -0.13 0.29 0.31 0.28 0.29 0.23 0.30 0.31 0.32 -0.12 msavi_5yrs_greenest_300 0.09 0.15 0.19 0.18 0.05 0.25 -0.04 0.00 0.04 0.03 0.02 0.01 0.23 0.18 0.16 0.08 0.01 0.05 0.06 -0.05 0.27 0.24 0.21 0.22 0.17 0.28 0.28 0.28 -0.02 msavi_5yrs_greenest_std_300 0.06 0.11 0.15 0.14 0.04 0.21 0.00 0.04 0.06 0.06 0.05 0.04 0.24 0.19 0.17 0.11 0.04 0.08 0.09 -0.07 0.22 0.23 0.21 0.21 0.17 0.23 0.24 0.24 -0.08 stops_100_2018 0.02 0.05 0.08 0.09 -0.01 0.05 0.16 0.17 0.17 0.17 0.14 0.13 0.00 0.00 0.06 0.07 0.12 0.12 0.12 0.12 0.03 -0.01 -0.02 0.02 0.02 0.03 0.03 0.03 0.06 lines_100_2018 0.17 0.19 0.20 0.21 0.17 0.14 0.29 0.31 0.30 0.31 0.27 0.27 0.16 0.22 0.23 0.29 0.26 0.28 0.28 0.06 0.12 0.11 0.15 0.15 0.19 0.11 0.13 0.13 -0.02 lines_500_2018 0.10 0.12 0.11 0.11 0.11 0.08 0.08 0.10 0.10 0.10 0.08 0.11 0.24 0.24 0.19 0.20 0.12 0.14 0.15 -0.18 0.08 0.24 0.25 0.22 0.24 0.08 0.11 0.12 -0.22 ne_DEM_500 -0.05 -0.09 -0.11 -0.10 0.00 -0.18 0.09 0.06 0.03 0.04 0.04 0.05 0.01 -0.02 0.00 0.04 0.05 0.03 0.03 -0.12 -0.20 0.02 -0.02 -0.03 -0.01 -0.20 -0.18 -0.17 -0.18 treecover_2015_300 0.07 0.11 0.16 0.16 0.03 0.19 0.05 0.10 0.13 0.12 0.12 0.09 0.29 0.18 0.23 0.14 0.09 0.13 0.14 -0.12 0.21 0.31 0.21 0.29 0.20 0.21 0.24 0.24 -0.15 treecover_pxls_2015_300 -0.05 -0.04 -0.04 -0.05 0.00 -0.02 -0.13 -0.14 -0.13 -0.13 -0.15 -0.15 0.05 0.02 -0.01 -0.05 -0.15 -0.13 -0.12 -0.10 0.00 0.07 0.05 0.04 0.01 0.01 0.01 0.01 -0.07 BEV_DICHTH -0.07 -0.11 -0.13 -0.12 -0.02 -0.20 0.08 0.05 0.03 0.03 0.03 0.04 -0.06 -0.07 -0.04 0.01 0.05 0.02 0.01 -0.05 -0.21 -0.06 -0.08 -0.09 -0.07 -0.21 -0.20 -0.19 -0.07 OAD -0.07 -0.11 -0.14 -0.13 -0.01 -0.22 0.10 0.08 0.06 0.06 0.06 0.07 -0.04 -0.05 -0.01 0.03 0.07 0.05 0.04 -0.09 -0.24 -0.02 -0.06 -0.05 -0.04 -0.25 -0.22 -0.21 -0.14 STED -0.05 -0.07 -0.08 -0.08 0.01 -0.07 -0.11 -0.13 -0.13 -0.13 -0.14 -0.14 -0.09 -0.04 -0.08 -0.07 -0.14 -0.13 -0.13 0.02 -0.06 -0.11 -0.04 -0.10 -0.07 -0.06 -0.07 -0.08 0.06 P_HOOG_INK -0.07 -0.10 -0.11 -0.10 -0.01 -0.11 0.01 -0.01 -0.03 -0.02 -0.03 -0.04 -0.12 -0.10 -0.10 -0.06 -0.04 -0.05 -0.06 0.05 -0.14 -0.12 -0.10 -0.14 -0.11 -0.14 -0.14 -0.14 -0.01 P_NIET_ACT 0.10 0.14 0.15 0.14 0.02 0.13 0.06 0.08 0.10 0.09 0.09 0.10 0.15 0.13 0.12 0.11 0.11 0.12 0.12 -0.03 0.15 0.15 0.13 0.16 0.15 0.15 0.16 0.16 -0.01 P_PENS_ONT 0.12 0.17 0.22 0.22 -0.02 0.24 0.07 0.12 0.14 0.13 0.14 0.13 0.12 0.06 0.12 0.07 0.12 0.13 0.14 0.04 0.24 0.13 0.06 0.16 0.11 0.24 0.24 0.23 0.02 P_UIT_ONTV 0.08 0.11 0.11 0.11 0.01 0.11 0.01 0.03 0.05 0.04 0.05 0.06 0.11 0.08 0.07 0.05 0.07 0.07 0.08 -0.01 0.13 0.10 0.08 0.11 0.09 0.14 0.13 0.13 0.02 Table 9 Pearson’s cross correlation between the road traffic noise indicators and a selection of other relevant indicators on the ABCD cohort. Orange highlights strong positive correlations, green highlights strong negative correlations. Table 8 shows the correlation between the road traffic noise indicators calculated using the surrogate model proposed in D3.5 over all locations in the ABCD cohort. The proposed sleep disturbance index (SDI) correlates across all age categories and with Lnight and it does so both for the most exposed as for the least exposed façade. This is not unexpected as all of these indicators or variants on the same underlying idea: road traffic noise events disturb sleep. Similarly, the indicators for restoration in the evening: LA90, LA50 and N90, N50, correlate strongly amongst each other by construction, but they correlate slightly less with the sleep disturbance indicators. The same indicators calculated at the least and most exposed façade show some, but lower correlation reflecting the different contributions from close by and distant traffic. The ARP (absence of restorative periods) mostly correlates with the N50 (and N90) at the least exposed façade where its definition is based on. The negative correlation between EPE and LA90 and others are explained by the fact that this indicator is based on emergence above background. Also all other correlations that can be observed in Table 8 match expectations. Table 9 illustrates another interesting overlap between the noise indicators and a selection from the other indicators calculated at the dwelling of the ABCD cohort. For the selection, for each group (e.g. by buffer distance) the most strongly correlating index is retrained. The first relationship that is clearly illustrated is the driver-pressure relationship: the presence of road traffic causes noise. While interpreting the indicators for the presence of roads, one should keep in mind the difference between streets, roads, and major roads that is explained in deliverable 5.1. The total length of major roads (highways and primary roads) within a certain radius correlates most with background levels, LA90, LA50, etc. and the circle of 500m is still relevant. The length of roads within a short distance from the dwelling correlates most with sleep disturbance indices calculated at the most exposed façade. Noise indicator values at the least exposed façade are not only influenced by the closest road but contain contributions from other roads and thus the correlation is less strong. The negative correlation with the presence of streets at short distances with LA90, LA50, etc. at the one hand and levels at the least exposed façade at the other needs further reflection. We suspect that this reflects typical housing situations: persons living in living neighbourhoods with lots of small streets probably live further away from highways and primary streets that cause the high background levels. These results confirm what was reported in D3.5 whare Shapley analysis was used to understand what drives the traffic noise model. The correlations between noise indicators and the other indicators in Table 9 reflect the urban morphology and where people live. These might be specific for the region covered by the ABCD cohort. Most correlations can clearly be explained but might not seem trivial for some reasons. For example, high urbanisation correlates negatively with road traffic noise indicators at the least exposed façade due to the increased screening by buildings resulting in quiet backyards. Several indicators for the presence of green within 300m correlate positively with the road traffic noise indicators, especially at the least exposed façade and the background levels. This might be because also green buffers around main roads are counted as green and thus their presence may indicated the presence of traffic noise unscreened by buildings. Such correlations are not to be included in the Wi,i. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 64 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Figure 24 graph representation of the indicators and their connections highlighting noise indicators in blue text and nodes corresponding to effect indicators as red dots. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 65 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 Figure 25 graph representation of the indicators connected to ADHD in second order, that is by direct evidence or by evidence relating the indicator and ADHD to at least one common indicator. 6. Software for calculating new metrics Two software codes are made available: (1) the surrogate model for calculating advanced road traffic noise indicators in Table 3; (2) the calculation of indicators and clustering based on 1/3 octave band measurements with 125ms resolution. The code is available on Github: waves-acoustics/Equal-Life_noise: Horizon 2020 Equal-Life project datadriven and hybrid models for new metrics for noise exposure related to mental health (D3.6) D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 66 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 The surrogate model (1) is purely based on Open Streetmap (OSM) data and can thus be run everywhere in Europe and beyond. The outcome will nevertheless be influenced by the local accuracy of OSM. It is most efficient to run the code for a group of receiver points in a region or large city. Using it for producing maps is sub-optimal as it is assumed that all receivers are connected to a dwelling. CPU-time needed depends largely on the density of buildings and streets around the receiver points. This code has been tested on all cohorts in Equal-Life. Instructions for use can be found in the README file. The clustering model (2) contains two parts: (a) a python script for calculating a multitude of noise indicators based on the spectral measurements. Some code adaptation is needed to accommodate for the formatting of the measurement file (CSV file, excel file) produced by your measurement equipment. (b) an assignment of your measurement to one of the clusters obtained in this work. This code should be seen as a step in future research as the relevance of the clusters on large datasets has not been established due to the lack of broad noise measurement campaigns combined with wellbeing, mental health, and cognitive development assessment. 7. Overall conclusion and outlook This deliverable addressed the challenge of characterizing environmental sound at home in a manner relevant to the mental health, well-being, and cognitive development of children—the primary focus of Equal-Life. Evaluating the relevance of novel sound environment indicators remains complex; traditional metrics such as Lden and Lnight have been widely used due to their transparency and practical applicability, and much of the existing evidence, where available, pertains to these indicators. However, our findings demonstrate that more specifically tailored indicators, designed with pathways and a priori hypotheses regarding noise effects, may exhibit stronger predictive value for certain outcomes, including SDQ, KINDL, cognitive development, and attention, thus leading to indicators with higher validity. In this report, simple correlation analyses were conducted to explore these relationships, with more sophisticated analyses deferred to relevant work packages. Notably, the findings suggest that specific noise indicators may hold greater relevance for particular subcomponents of scales like SDQ and KINDL. Cluster analysis of sound measurements taken at the bedroom windows of children during in-depth studies in WP1—5-7-year-olds in the preschool study in Ghent and 18-year-olds in the sleep study in Gothenburg—revealed that the majority of 15-minute epochs could be categorized as representing a tranquil urban sound climate. The remaining clusters were either common across locations, often tied to specific times of the day, or identified as distinctly disturbing location. A similar cluster analysis of indoor measurements (bedrooms of 18-year-olds) indicated that the indoor sound climate is minimally influenced by the outdoor sound environment, with only a small proportion of indoor clusters showing a relationship to outdoor noise indicators. Further analyses, which linked indoor and outdoor sound events by estimating brute sound insulation, confirmed that indoor sound events rarely have a direct external origin. These findings suggest that a commonly used linear regression approach between outdoor noise levels and sleep quality may not be suitable. However, it is important to note that these measurements were conducted in a Nordic country, where high thermal insulation—and consequently D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 67 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 high acoustic insulation—could significantly influence the results. Thus, these conclusions regarding the indoor sound climate may not be generalizable across other EU regions. The analysis of measurements highlighted the significance of diurnal patterns, prompting the exploration of more specific indicators tailored to children's daily activities. The Equal-Life machinelearning-based surrogate model for advanced road traffic noise indicators provides hourly values for metrics such as the number of events exceeding a threshold and LA50 (refer to Section 4.1 for a comprehensive list). These hourly indicators can be aggregated in various ways over a 24-hour period, depending on the child's age and activity schedule. Based on a priori considerations, several potentially relevant configurations were proposed to complement the Equal-Life indicators, SDI and ARP. All these indicators are calculated using the surrogate software, which has been made available to the research community. The models can be applied in various contexts depending on the data at hand: using local traffic data when available, the Equal-Life traffic model from D3.2, or the default configuration based on OpenStreetMap (OSM). For the development of the latter, a hybrid approach was employed. Measurements taken at the most exposed façade were used to infer the relationship between OSM road type, connectivity, and traffic intensities. The resulting "pseudo-traffic intensities" represent plausible traffic estimates derived from the observed measurements. From an exposome perspective, the variety of noise indicators must be contextualized within their broader environment. Traffic noise indicators, for example, are linked to metrics characterizing neighbourhood activities, such as proximity to major roads, road network connectivity, and the degree of urbanization. They are also associated with indicators of co-exposures driven by similar underlying factors, such as NO₂ levels or perceived traffic safety. However, accidental correlations may arise within specific areas or populations, complicating interpretations. To address this complexity, a knowledge graph was introduced as a tool for systematically mapping and exploring relationships. By integrating all known associations into the graph, researchers can examine multiple interactions and dependencies from various perspectives. In this deliverable, knowledge from D7.2 regarding one outcome—ADHD— was combined with relationships identified from a specific group of exposures: home sound environments. The introduction of the knowledge graph partly compensates for the possible critique that only simple Spearman correlation is used to show the potential validity of some of the new indicators. From an exposome perspective, it is not crucial to identify the specific exposure paths that influence outcomes. This approach can be extended in the coming months to incorporate additional exposome components and outcomes, broadening the scope of the analysis. By integrating our insights into the effects of noise exposure and proposing novel metrics, alongside providing software for their efficient calculation, we anticipate that researchers beyond the current Equal-Life consortium will engage in broader investigations of noise effects. This collaborative effort has the potential to overcome the limitations outlined in the initial discussions and advance the understanding of noise-related impacts on health and development. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 68 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 8. References [1] Terzakis, M. E., Dohmen, M., van Kamp, I., & Hornikx, M. (2022). Noise Indicators Relating to Non-Auditory Health Effects in Children—A Systematic Literature Review. International Journal of Environmental Research and Public Health, 19(23), 15633.. [2] S.A. Stansfeld, C. Clark, R.M. Cameron, T. Alfred, J. Head, M.M. Haines, I. van Kamp, E. van Kempen, I. Lopez-Barrio, Aircraft and road traffic noise exposure and children's mental health, Journal of Environmental Psychology, Volume 29, Issue 2, 2009, Pages 203-207, [3] Zare Sakhvidi, F., Zare Sakhvidi, M.J., Mehrparvar, A.H. et al. Environmental Noise Exposure and Neurodevelopmental and Mental Health Problems in Children: a Systematic Review. Curr Envir Health Rpt 5, 365–374 (2018). [4] Dreger, S., Meyer, N., Fromme, H., Bolte, G. Environmental noise and incident mental health problems: A prospective cohort study among school children in Germany, Environmental Research, Volume 143, Part A, 2015, Pages 49-54, [5] Lercher, P., Evans, G.W., Meis, M., et al. Ambient neighbourhood noise and children's mental health. Occupational and Environmental Medicine 2002;59:380-386. [6] Crombie, R., Clark, C. & Stansfeld, S.A. Environmental noise exposure, early biological risk and mental health in nine to ten year old children: a cross-sectional field study. Environ Health 10, 39 (2011). https://doi.org/10.1186/1476-069X-10-39. [7] Schubert, M., Hegewald, J., Freiberg, A., Starke, K.R., Augustin, F., Riedel-Heller, S.G., Zeeb, H., Seidler, A. Behavioral and Emotional Disorders and Transportation Noise among Children and Adolescents: A Systematic Review and Meta-Analysis. Int J Environ Res Public Health. 2019 Sep 10;16(18):3336. doi: 10.3390/ijerph16183336. PMID: 31510007; PMCID: PMC6765874. [8] Huhdanpää, H.; Morales-Muñoz, I.; Aronen, E.T.; Pölkki, P.; Saarenpää-Heikkilä, O.; Paunio, T.; Kylliäinen, A.; Paavonen, E.J. Sleep Difficulties in Infancy Are Associated with Symptoms of Inattention and Hyperactivity at the Age of 5 Years: A Longitudinal Study. J. Dev. Behav. Pediatrics 2019, 40, 432–440 [9] J. M. Wunderli et al., “Intermittency ratio: A metric reflecting short-term temporal variations of transportation noise exposure,” J Expo Sci Environ Epidemiol, vol. 26, no. 6, 2016, doi: 10.1038/jes.2015.56. [10] Brocolini L, Lavandier C, Quoy M, Ribeiro C. Measurements of acoustic environments for urban soundscapes: choice of homogeneous periods, optimization of durations, and selection of indicators. J Acoust Soc Am. 2013 Jul;134(1):813-21. doi: 10.1121/1.4807809. PMID: 23862887. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 69 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 [11] van Kamp, I., Simon, S., Notley, H., Baliatsas, C., van Kempen, E. Evidence Relating to Environmental Noise Exposure and Annoyance, Sleep Disturbance, Cardio-Vascular and Metabolic Health Outcomes in the Context of IGCB (N): A Scoping Review of New Evidence. Int J Environ Res Public Health. 2020 Apr 26;17(9):3016. doi: 10.3390/ijerph17093016. PMID: 32357581; PMCID: PMC7246943. [12] Babisch W, Houthuijs D, Pershagen G, Cadum E, Katsouyanni K, Velonakis M, Dudley ML, Marohn HD, Swart W, Breugelmans O, Bluhm G, Selander J, Vigna-Taglianti F, Pisani S, Haralabidis A, Dimakopoulou K, Zachos I, Järup L; HYENA Consortium. Annoyance due to aircraft noise has increased over the years--results of the HYENA study. Environ Int. 2009 Nov;35(8):1169-76. doi: 10.1016/j.envint.2009.07.012. Epub 2009 Aug 21. PMID: 19699524. [13] Can, A., Aumond, P., Michel, S., De Coensel, B., Ribeiro, C., Botteldooren, D., Lavandier, C. (2016b). Comparison of noise indicators in an urban context, Proceedings of Internoise 2016, Hambourg (Allemagne), 21-24 August, 2016. [14] Floud S, Blangiardo M, Clark C, de Hoogh K, Babisch W, Houthuijs D, Swart W, Pershagen G, Katsouyanni K, Velonakis M, Vigna-Taglianti F, Cadum E, Hansell AL. Exposure to aircraft and road traffic noise and associations with heart disease and stroke in six European countries: a cross-sectional study. Environ Health. 2013 Oct 16;12:89. doi: 10.1186/1476-069X-12-89. PMID: 24131577; PMCID: PMC4015897. [15] Héritier H, Vienneau D, Frei P, Eze IC, Brink M, Probst-Hensch N, Röösli M. The association between road traffic noise exposure, annoyance and health-related quality of life (HRQOL). Int J Environ Res Public Health. 2014 Dec 5;11(12):12652-67. doi: 10.3390/ijerph111212652. PMID: 25489999; PMCID: PMC4276638. [16] Hume KI, Brink M, Basner M. Effects of environmental noise on sleep. Noise Health 2012;14:297-302 [17] McGuire, S., & Basner, M. (2020). Predictions of sleep disturbance for different nighttime airport operation strategies using a new Markov state transition sleep model. Euronoise 2015. [18] Schomer, P.D. (2005). Criteria for assessment of noise annoyance. Noise Control Engineering, 53(4), 132–144. [19] Fujii, K., Atagi, J., & Ando, Y. (2002). Temporal and spatial factors of traffic noise and its annoyance. Journal of Temporal Design in Architecture and the Environment, 2(1), 33–41. [20] Can, A., Leclercq, L., & Lelong, J. (2008). Dynamic estimation of urban traffic noise: influence of traffic and noise source representations. Applied Acoustics, 69(10), 858-867 [21] De Coensel, B., De Muer, T., Yperman, I., & Botteldoren, D. (2005). The influence of traffic flow dynamics on urban soundscape. Applied Acoustics, 66, 175–194. [22] Can, A., Leclercq, L., Lelong, J., & Botteldooren, D. (2010b). Traffic noise spectrum analysis: Dynamic modeling vs. Experimental observations. Applied Acoustics. 2010; 71(8): 764-770. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 70 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 [23] Luo, W.L., Cai, M., Li, F., & Liu, J.K. (2012). Dynamic modeling of road traffic noise around buildings in an urban area. Noise Control Engineering Journal, 60(4). [24] Can, A., & Gauvreau, B. (2015). Describing and classifying urban sound environments with a relevant set of physical indicators. Journal of the Acoustical Society of America, 137(1), 208218. [25] Torija, A.J., Ruiz, D.P., & Ramos-Ridao A.F. (2013). Application of a methodology for categorizing and differentiating urban soundscapes using acoustical descriptors and sementic-differential attributes. The Journal of the Acoustical Society of America, 134(1), 791-802. [26] Van Renterghem, T.; Le Bescond, V.; Dekoninck, L.; Botteldooren, D. Advanced Noise Indicator Mapping Relying on a City Microphone Network. Sensors 2023, 23, 5865. https://doi.org/10.3390/s23135865 [27] Wass, S. v., Smith, C. G., Daubney, K. R., Suata, Z. M., Clackson, K., Begum, A., & Mirza, F. U. (2019). Influences of environmental stressors on autonomic function in 12-month-old infants: understanding early common pathways to atypical emotion regulation and cognitive. [28] S. Myllyntausta, J. Virkkala, P. Salo, J. Varjo, L. Rekola, and V. Hongisto, “Effect of the frequency spectrum of road traffic noise on sleep: A polysomnographic study,” J Acoust Soc Am, vol. 147, no. 4, 2020, doi: 10.1121/10.0000985. [29] M. G. Smith, M. Ögren, J. Ageborg Morsing, and K. Persson Waye, “Effects of ground-borne noise from railway tunnels on sleep: A polysomnographic study,” Build Environ, vol. 149, 2019, doi: 10.1016/j.buildenv.2018.12.009. [30] J. M. Wunderli et al., “Intermittency ratio: A metric reflecting short-term temporal variations of transportation noise exposure,” J Expo Sci Environ Epidemiol, vol. 26, no. 6, 2016, doi: 10.1038/jes.2015.56. [31] B. Griefahn, P. Bröde, A. Marks, and M. Basner, “Autonomic arousals related to traffic noise during sleep,” Sleep, vol. 31, no. 4, 2008, doi: 10.1093/sleep/31.4.569. [32] M. Basner and S. McGuire, “WHO environmental noise guidelines for the european region: A systematic review on environmental noise and effects on sleep,” International Journal of Environmental Research and Public Health, vol. 15, no. 3. 2018. doi: 10.3390/ijerph15030519. [33] M. Basner, U. Müller, and E. M. Elmenhorst, “Single and combined effects of air, road, and rail traffic noise on sleep and recuperation,” Sleep, vol. 34, no. 1, 2011, doi: 10.1093/sleep/34.1.11. [34] M. G. Smith, M. Younes, D. Aeschbach, E. M. Elmenhorst, U. Müller, and M. Basner, “Traffic noise-induced changes in wake-propensity measured with the Odds-Ratio Product (ORP),” Science of the Total Environment, vol. 805, 2022, doi: 10.1016/j.scitotenv.2021.150191. [35] S. Myllyntausta, J. Virkkala, P. Salo, J. Varjo, L. Rekola, and V. Hongisto, “Effect of the frequency spectrum of road traffic noise on sleep: A polysomnographic study,” J Acoust Soc Am, vol. 147, no. 4, 2020, doi: 10.1121/10.0000985. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 71 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 [36] M. G. Smith, M. Ögren, J. Ageborg Morsing, and K. Persson Waye, “Effects of ground-borne noise from railway tunnels on sleep: A polysomnographic study,” Build Environ, vol. 149, 2019, doi: 10.1016/j.buildenv.2018.12.009. [37] M. Basner and U. Siebert, “Markov processes for the prediction of aircraft noise effects on sleep,” Medical Decision Making, vol. 30, no. 2, 2010, doi: 10.1177/0272989X09342751. [38] B. de Coensel, T. de Muer, I. Yperman, and D. Botteldooren, “The influence of traffic flow dynamics on urban soundscapes,” Applied Acoustics, vol. 66, no. 2, 2005, doi: 10.1016/j.apacoust.2004.07.012. [39] R. D. Chervin, J. L. Fetterolf, D. L. Ruzicka, B. J. Thelen, and J. W. Burns, “Sleep stage dynamics differ between children with and without obstructive sleep apnea,” Sleep, vol. 32, no. 10, 2009, doi: 10.1093/sleep/32.10.1325. [40] E. A. Lopez-Poveda, “Development of Fundamental Aspects of Human Auditory Perception,” in Development of Auditory and Vestibular Systems: Fourth Edition, 2014. doi: 10.1016/B978-012-408088-1.00010-5. [41] E. Öhrström, E. Hadzibajramovic, M. Holmes, and H. Svensson, “Effects of road traffic noise on sleep: Studies on children and adults,” J Environ Psychol, vol. 26, no. 2, 2006, doi: 10.1016/j.jenvp.2006.06.004. [42] A. Muzet, “Environmental noise, sleep and health,” Sleep Medicine Reviews, vol. 11, no. 2. 2007. doi: 10.1016/j.smrv.2006.09.001. [43] C. Clark et al., “Implications of the RANCH study for exploring mechanisms of noise effects on cognition,” in International Congress on Noise Control Engineering 2005, INTERNOISE 2005, 2005. [44] M. Atzori et al., “Locus ceruleus norepinephrine release: a central regulator of CNS spatio-temporal activation?,” Front Synaptic Neurosci, vol. 8, p. 25, 2016. [45] B. F. Sadacca, A. M. Wikenheiser, and G. Schoenbaum, “Toward a theoretical role for tonic norepinephrine in the orbitofrontal cortex in facilitating flexible learning,” Neuroscience, vol. 345, pp. 124–129, 2017. [46] R. S. Ulrich, “View through a window may influence recovery from surgery,” Science (1979), vol. 224, no. 4647, 1984, doi: 10.1126/science.6143402. [47] [18] R. S. Ulrich, R. F. Simons, B. D. Losito, E. Fiorito, M. A. Miles, and M. Zelson, “Stress recovery during exposure to natural and urban environments,” J Environ Psychol, vol. 11, no. 3, 1991, doi: 10.1016/S0272-4944(05)80184-7. [48] J. J. Alvarsson, S. Wiens, and M. E. Nilsson, “Stress recovery during exposure to nature sound and environmental noise,” Int J Environ Res Public Health, vol. 7, no. 3, 2010, doi: 10.3390/ijerph7031036. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 72 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 [49] T. van Renterghem, “Towards explaining the positive effect of vegetation on the perception of environmental noise,” Urban Forestry and Urban Greening, vol. 40. 2019. doi: 10.1016/j.ufug.2018.03.007. [50] C. J. W. McClure, H. E. Ware, J. Carlisle, G. Kaltenecker, and J. R. Barber, “An experimental investigation into the effects of traffic noise on distributions of birds: Avoiding the phantom road,” Proceedings of the Royal Society B: Biological Sciences, vol. 280, no. 1773, 2013, doi: 10.1098/rspb.2013.2290. [51] M. N. Verzijden, E. A. P. Ripmeester, V. R. Ohms, P. Snelderwaard, and H. Slabbekoorn, “Immediate spectral flexibility in singing chiffchaffs during experimental exposure to highway noise,” Journal of Experimental Biology, vol. 213, no. 15, 2010, doi: 10.1242/jeb.038299. [52] H. Brumm, “The impact of environmental noise on song amplitude in a territorial bird,” Journal of Animal Ecology, vol. 73, no. 3, 2004, doi: 10.1111/j.0021-8790.2004.00814.x. [53] H. N. Li, C. K. Chau, and S. K. Tang, “Can surrounding greenery reduce noise annoyance at home?,” Science of the Total Environment, vol. 408, no. 20, 2010, doi: 10.1016/j.scitotenv.2010.06.025. [54] T. van Renterghem and D. Botteldooren, “View on outdoor vegetation reduces noise annoyance for dwellers near busy roads,” Landsc Urban Plan, vol. 148, 2016, doi: 10.1016/j.landurbplan.2015.12.018. [55] E. Öhrström, A. Skånberg, H. Svensson, and A. Gidlöf-Gunnarsson, “Effects of road traffic noise and the benefit of access to quietness,” J Sound Vib, vol. 295, no. 1–2, 2006, doi: 10.1016/j.jsv.2005.11.034. [56] Y. de Kluizenaar, S. A. Janssen, H. Vos, E. M. Salomons, H. Zhou, and F. van den Berg, “Road traffic noise and annoyance: A quantification of the effect of quiet side exposure at dwellings,” Int J Environ Res Public Health, vol. 10, no. 6, 2013, doi: 10.3390/ijerph10062258. [57] T. van Renterghem and D. Botteldooren, “Focused study on the quiet side effect in dwellings highly exposed to road traffic noise,” Int J Environ Res Public Health, vol. 9, no. 12, 2012, doi: 10.3390/ijerph9124292. [58] A. Dzhambov and P. Lercher, “Noise in restorative environments-Perceptions, positive and negative environmental components in Tyrolean children,” in Proceedings on Internoise, Glasgow, 2022. [59] B. Locher et al., “Differences between outdoor and indoor sound levels for open, tilted, and closed windows,” Int J Environ Res Public Health, vol. 15, no. 1, 2018, doi: 10.3390/ijerph15010149. [60] J. Defrance et al., “Outdoor sound propagation reference model developed in the European harmonoise project,” Acta Acustica united with Acustica, vol. 93, no. 2, 2007. D3.6 – A report and open source code on data-driven and hybrid models for new metrics for outdoor and indoor noise exposure related to mental health 73 This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 874724 [61] McInnes, L., Healy, J., & Melville, J. (2018). Umap: Uniform manifold approximation and projection for dimension reduction. arXiv preprint arXiv:1802.03426. [62] McInnes, L., Healy, J., & Astels, S. (2017). hdbscan: Hierarchical density based clustering. J. Open Source Softw., 2(11), 205. [63] Goodman, A., & Goodman, R. (2009). Strengths and difficulties questionnaire as a dimensional measure of child mental health. Journal of the American Academy of Child & Adolescent Psychiatry, 48(4), 400-403. [64] Ravens-Sieberer, U., Herdman, M., Devine, J., Otto, C., Bullinger, M., Rose, M., & Klasen, F. (2014). The European KIDSCREEN approach to measure quality of life and well-being in children: development, current application, and future advances. Quality of life research, 23, 791-803. [65] Dzhambov, A. M., Lercher, P., & Botteldooren, D. (2024). Childhood sound disturbance and sleep problems in Alpine valleys with high levels of traffic exposures and greenspace. Environmental Research, 242, 117642. [66] Stansfeld, S. A., Haines, M. M., Burr, M., Berry, B., & Lercher, P. (2000). A review of environmental noise and mental health. Noise and Health, 2(8), 1-8. [67] Clark, C., & Paunovic, K. (2018). WHO environmental noise guidelines for the European region: a systematic review on environmental noise and quality of life, wellbeing and mental health. International journal of environmental research and public health, 15(11), 2400. [68] Hahad, O., Kuntic, M., Al-Kindi, S., Kuntic, I., Gilan, D., Petrowski, K., ... & Münzel, T. (2024). Noise and mental health: evidence, mechanisms, and consequences. Journal of Exposure Science & Environmental Epidemiology, 1-8. [69] Raess, M., Brentani, A. V. M., Flückiger, B., de Campos, B. L. D. A., Fink, G., & Röösli, M. (2022). Association between community noise and children’s cognitive and behavioral development: A prospective cohort study. Environment international, 158, 106961. [70] Tangermann, L., Vienneau, D., Hattendorf, J., Saucy, A., Künzli, N., Schäffer, B., ... & Röösli, M. (2022). The association of road traffic noise with problem behaviour in adolescents: A cohort study. Environmental research, 207, 112645. [71] Li, A., Martino, E., Mansour, A., & Bentley, R. (2022). Environmental noise exposure and mental health: evidence from a population-based longitudinal study. American journal of preventive medicine, 63(2), e39-e48. [72] Lim, J., Kweon, K., Kim, H. W., Cho, S. W., Park, J., & Sim, C. S. (2018). Negative impact of noise and noise sensitivity on mental health in childhood. Noise and Health, 20(96), 199-211. [73] Sygna, K., Aasvang, G. M., Aamodt, G., Oftedal, B., & Krog, N. H. (2014). Road traffic noise, sleep and mental health. Environmental research, 131, 17-24. [74] Dzhambov, A. M., Lercher, P., Vincens, N., Waye, K. P., Klatte, M., Leist, L., ... & van Kamp, I. (2023). Protective effect of restorative possibilities on cognitive function and mental health in