Use of IoT sensing and occupant surveys for determining the resilience of buildings to forest fire generated PM2.5

Wildfires and associated emissions of particulate matter pose significant environmental and health concerns. In this study we propose tools to evaluate building resilience to extreme episodes of outdoor particulate matter using a combination of indoor and outdoor IoT measurements, coupled with survey-based information of occupants’ perception and behaviour. We demonstrated the application of the tools on two buildings with different modes of ventilation during the Chico Camp fire event. We characterized the resilience of the buildings on different temporal and spatial scales using the well-established I/O ratio and a newly proposed E-index that evaluates indoor concentration in the context of adopted 24-hour exposure thresholds. Indoor PM2.5 concentration during the entire Chico Camp Fire event was 21 μg/m3 for 4th Street (Mechanically Ventilated) and 36 μg/m3 for Wurster Hall (Naturally Ventilated). The cumulative median I/O ratio during the fire event was 0.27 for 4th Street and 0.67 for Wurster Hall. Overall E-index for 4th Street was 0.82, suggesting that the whole building was resilient to outdoor air pollution while overall E-index was 1.69 for Wurster Hall suggesting that interventions are necessary. The survey revealed that occupant perception of workplace air quality aligns with measured PM2.5 in the two buildings. The results also highlight that a large portion of occupants wore face masks, even though the PM2.5 concentration was below WHO threshold level. The results of our study demonstrate the utility of the proposed IoT-enabled and survey tools to assess the degree of protection from air pollution of outdoor origin for a single building or across a portfolio of buildings. The proposed survey tool also provides direct links between the PM2.5 levels and occupants’ perception and behavior.


Introduction
In recent years, we have observed increased wildfire frequency and intensity in response to global warming [1]. Many regions prone to fire are forecasted to have increased frequency of wildfires and associated air pollution episodes [2]. Wildfire smoke includes carbon dioxide, PLOS  building or room scale) and temporal scales (instantaneous and long term) to demonstrate the application of building assessment tools. We used two buildings located in Berkeley (California) during the two-week Chico Camp fire in November 2018; one building was equipped with mechanical ventilation and the other with natural ventilation. By reliance on combination of I/O ratio metric, survey questionnaires and a newly proposed E-index, the study sought to quantify the resilience of buildings against particle penetration indoors from outdoor origin.

Materials and methods
We used two types of indoor PM 2.5 sensors (Clarity Inc, and Senseware), one type of outdoor PM 2.5 sensor (Clarity Inc), and one type of CO 2 sensor (Senseware) placed on and in two commercial buildings in Berkeley, CA. One of the buildings was naturally ventilated, while the other was mechanically ventilated. We also developed and distributed a survey to gather feedback on occupant experience in their workplace during the Chico Camp fire compared to their typical experience in the space. With this data, we developed and evaluated tools to quantify and assess each building's resilience to the extreme air pollution episode.

Experimental apparatus
Both Clarity and Senseware PM 2.5 nodes (Fig 1) count the particle number using the principle of light scattering, which are characteristic for optical particle counters. The accuracy of the all the sensors was the same-within ±10 μg/m 3 in the range of 0 to 100 μg/m 3 and ±10% in the range of 100 to 1000 μg/m 3 . Data was collected on 15-minute intervals for Clarity nodes on 1-minute intervals for Senseware nodes. As typical for optical particle counters, Clarity and Senseware nodes are factory-calibrated with Arizona Test Dust (ATD). Calibration with ATD does not robustly account for many aerosol particle properties, such as density, shape, refractive index, and absorption. Therefore, if aerosol properties of the particles measured in the field do not have properties similar to ATD (as is likely the case for wildfires), the optical particle counters can produce inaccurate readings. There is a lack of understanding of the wildfire source profiles and resulting aerosol composition [22]; thus making the field calibration challenging. To arrive at accurate measurements, the Clarity nodes apply co-location of nodes and post-processing correction factors to the raw outdoor measurements using a local government measurement stations that indirectly include information on particle optical properties. For this dataset, the Clarity node corrections for particle optical properties were extracted from the California Air Resources Board Oakland Laney College measurement station. Clarity Inc. typically only applies these corrections to outdoor nodes; however, we applied the same correction to indoor Clarity nodes because under the Chico Camp fire scenario indoor PM 2.5 dominantly originated from the outdoors. Senseware nodes do not make corrections to the raw measurements and are calibrated with the ATD. Senseware CO 2 sensors measured CO 2 levels with an accuracy of ±50 ppm. In Wurster Hall, the CO 2 sensors were paired with PM 2.5 measurements, while in 4 th Street they were placed in two representative locations. Senseware contact monitors-(Model COZIR-LP) were mounted on the windows in Wurster Hall to detect window position. Detection was binary (i.e., open/closed) and did not provide information on opening area or window angle.

Quality assurance
Prior to deployment, Clarity Inc. and Senseware PM 2.5 nodes were compared side-by-side and with Grimm aerosol spectrometer (model 11-A, GRIMM Aerosol Technik GmbH, Ainring, Germany) that provides time-and size-resolved data for aerosol particles. Clarity and Senseware PM 2.5 nodes were within ±8 μg/m 3 compared to Grimm measurements at ambient PM 2.5 levels ranging from 10 μg/m 3 to 45 μg/m 3 , which is within the given accuracy for each node.

Experimental design
We selected two buildings, 1608 4th Street Building (mechanically ventilated) and Wurster Hall (mixed-mode ventilation operating as fully naturally ventilated during the study period) to deploy sensors. Both buildings were located in Berkeley (California). In March 2018, we deployed indoor and outdoor PM 2.5 sensors, indoor CO 2 sensors in both buildings and window contact sensors in Wurster Hall. Berkeley was affected by air pollution from the Chico Camp fire from November 8 th until November 21 st , 2018. The current study provides analysis of data collected during the Chico Camp fire.
1608 4th Street Building (abbreviated as 4 th Street) is a mechanically ventilated building with centralized heating, ventilation and air-conditioning (HVAC) system. The HVAC system utilizes three stages of air filtration: the first stage of filtration is by a MERV 8 pleated filter, the second stage is the Gas Phase filter, and the final filter stage is a high-efficiency MERV 13 filter. The measure CO 2 decay profile during non-occupied hours were used to calculate infiltration rates (details in the Supporting Information). Infiltration was below 0.3 air exchanges per hour suggesting that the building is airtight without operable windows. The building has approximately 270 full-time occupants.
Wurster Hall-Mixed-mode building with seasonal changeover, meaning that the building is fully naturally ventilated when the building is operating in cooling, as was the case during the Chico Camp fire. The building relies on operable windows for ventilation and cooling. The building has a high level of infiltration, as shown by typical CO 2 levels below 550 ppm during normal operation. The building has approximately 300 full-time occupants.

Placement of outdoor and indoor sensors. 1608 4th Street
Building-is occupied on the 2 nd , 3 rd and 4 th floor. Floors had the same layouts; therefore, considering the available number of sensors, we placed all sensors on the 3 rd floor. We placed one outdoor PM 2.5 sensor (Clarity node) on the roof of the building, and 15 PM 2.5 sensors (14 Senseware nodes and one Clarity node) at different indoor locations. S1 Fig in the Supporting Information provides a floor plan with sensor locations in the 4 th Street. Each of the floors consists of large open floor office area with high partitions housing 71 stations, six private offices, two team meeting rooms, one large team meeting room, and two phone booths. We placed one PM 2.5 and one CO 2 sensor in the private office, and the remaining sensors in the open floor office. We wanted to measure if the building envelope was leaky, so we placed PM 2.5 sensors near outside walls with windows so that we can conservatively estimate building operation.
Wurster Hall-consisted of two building wings with separate HVAC systems. The south wing is four stories and consisted of classrooms, open plan offices, small and large meeting rooms, and several private offices. The north wing is 10 stories and consisted almost entirely of large open plan offices. Our study focused on the south wing of the building. We placed one PM 2.5 sensor (Clarity node) on the roof above the 4 th floor, and 11 PM 2.5 sensors (10 Senseware nodes and one Clarity node). Private offices were the most common space type in Wurster Hall's south wing; therefore, we placed the majority of PM 2.5 and CO 2 sensors in that environment. Additionally, we placed sensors in one meeting room with operable windows and in one small open plan office with four occupants. This particular open plan office was chosen also because it was adjacent to the hallway with an entrance door to the building. S2 Fig in the Supporting Information provides a floor plan with sensor locations in the Wurster Hall. We placed contact sensors on operable windows; each room had at least one manually operated awning-type window, and some rooms had two or more depending on the size of the space.

Data analysis
An assumption for data analysis was that the dominant source of indoor PM 2.5 originated from the outdoors. Although this assumption is not generally valid at typical outdoor PM 2.5 concentrations, in this case, the PM 2.5 data collected before and after the Chico Camp Fire showed typical indoor PM 2.5 concentrations below 3 μg/m 3 , implying that indoor sources were not significant relative to the outdoor PM 2.5 during forest fires. The average indoor raw PM 2.5 concentrations (i.e., not corrected for particle properties) measured in each building during the Chico Camp fire followed trends of outdoor measurements and were approximately 80 times greater than the average concentrations measured the week before and after the Chico Camp fire. This validates the assumption that all indoor particles can be treated as having outdoor origin during the Chico Camp fire.
As mentioned, Clarity Inc. corrects measured data to include information about PM 2.5 composition, while Senseware measurements rely on ATD calibration. Due to this disparity, the analysis uses raw measurements when combining or comparing data from the Clarity and Senseware nodes, and the Clarity corrected measurements when discussing observed indoor and outdoor PM 2.5 concentrations. For the analysis, we define building operating hours as 07:00 to 19:00. When presenting the results, we identify the median value with the interquartile range in parentheses (i.e., the range between the 25 th and 75 th percentiles), for example median PM 2.5 concentration of 21 μg/m 3 (IQR = 13 μg/m 3 ).

Outdoor PM 2.5 comparison.
We compared local (i.e., building) outdoor PM 2.5 concentrations to nearby regional weather station measurements to evaluate the current assumption that regional measurements are representative of individual exposure.

Building assessment tools.
We assessed the following tools for characterizing building resiliency during extreme pollution events based on IoT environmental sensing: (i). Comparison of indoor PM 2.5 concentrations to the WHO exposure threshold value for PM 2.5 . We chose the WHO guidelines because it is the strictest, with the maximum 24-hour mean exposure of 25 μg/m 3 . Alternative guidelines such as EPA may be used, depending on the region of the world. We compared median hourly indoor PM 2.5 concentration (i.e., median value from all indoor sensors) to WHO 24-hour mean exposure guideline, and calculated the Exceedance index (E-index), as shown in Eq 1, on an hourly basis. The E-index, is a unitless value that informs by how much hourly PM 2.5 concentration exceeds the recommended level. We can then calculate the number or percent of hours that indoor PM 2.5 concentration exceeded specified levels during the air pollution episode, or the whole building average E-index. It is a tool to evaluate occupant exposure that can be compared across buildings during extreme air pollution events. This can also be done on a space-by-space basis within the same building.
(ii). Comparison of indoor and outdoor PM 2.5 concentrations. Outdoor PM 2.5 can enter the indoor environment mainly by penetration (uncontrolled via building cracks and doors) or ingress (controlled via mechanical ventilation system or windows). From the building operation perspective, quantification of pathways and the quantity of outdoor particles that enter the indoor environment is one of the most important aspects of building resilience to extreme air pollution events. Considering negligible indoor PM 2.5 sources, we quantified PM 2.5 penetration using indoor to outdoor particle (I/O) ratio (Eq 2) using outdoor and indoor PM 2.5 measurements. Eq 2 is appropriate when the indoor pollution sources can be neglected, as we determined is the case for this event. We can calculate I/ O ratio at different temporal (i.e., cumulative episode or hourly) and spatial (i.e., whole building or by room) scales, which informs different scenarios of interest. The differences between I/O ratios among nearby buildings is typically attributed to differences in indoor particle sources, geometry of building cracks, wind direction and intensity, ventilation strategy and rate, and air filtration (16). We supplemented the I/O ratio analysis with information on infiltration rates calculated using CO 2 decay and knowledge of the HVAC system operation.
Instantaneous I/O ratio was calculated for each indoor sensor location using hourly mean indoor and outdoor PM 2.5 concentration (Eq 2, where C in (t) and C out (t) are the hourly means). We defined instantaneous as hourly in this study, but the calculation could be done at smaller time steps if necessary for a building or space. We used raw measurements from all sensors so data was comparable between sensor manufacturer. To calculate the whole building Instantaneous I/O ratio, we used the median hourly mean PM 2.5 from the indoor sensors compared to the hourly mean outdoor PM 2.5 concentration; we used the median value instead of the mean because it is robust to outlier instances, such as a window being left open which occurred in one space. Instantaneous I/O ratio is a tool that enables quick-response comparison between buildings and between spaces within a building on a short time scale.
Cumulative I/O ratio was calculated for each building to represent the overall whole building I/O ratio over a given period of time post-event, which was the Chico Camp fire period for this study. It is the median value of all the mean hourly I/O ratios at each sensor location in a building during an event. While the Instantaneous I/O ratios can vary depending on environmental conditions and driving forces (e.g. air temperatures and outdoor wind speeds), the Cumulative I/O ratio provides an overall value to compare buildings' abilities to prevent infiltration and penetration of outdoor pollutants during extreme pollution events. The Cumulative I/O ratio can also be calculated at each sensor location to compare between spaces within a building.

Survey of occupants.
It may not always be feasible to measure indoor and outdoor PM 2.5 for every building due to sensor availability and cost. A survey is an economical and simple method to gather data; however, it relies on self-reported information from occupants which may not always reflect the environmental conditions. This study compares the indoor PM 2.5 measurements with occupant feedback on perceived air quality, impacts to productivity from air quality, and behavior changes during the Chico Camp fire compared to their typical experience in the space. We then evaluated the potential effectiveness of surveys as a building assessment tool.
The voluntary web-based survey was distributed to full-time occupants in each building between March 13 and March 22, 2019. University of California at Berkeley Institutional Review Board specifically approved this study (obtained Protocol ID:2019-02-11802). Although there is a four-month gap between the fire and occupants' survey responses, the intensity of air pollution during the Chico Camp fire was atypical enough for the Berkeley area that we think the responses are valid representations of occupants' perception and behavior during that time. The survey used a five-point scale from -2 to +2 to rate occupant satisfaction from "dissatisfied" to "satisfied" or"interfered" to"enhanced", with a neutral vote as the middle value. The survey had two parts: first, the occupants were asked about their typical experience in the space, and then secondly, about their experiences during the Chico Camp fire. The survey main topics and rating scale included items listed in Table 1, and survey questions can be found in Supplemental information (Wildfire air quality survey).
Campus closures occurred on Nov 16,19, and 20 in response to outdoor pollutant levels, with the exception of some services that included staff in 4 th Street; therefore, many 4 th Street occupants worked in their building during the closure while Wurster Hall was closed. The survey controlled for and separated out these days to understand whether occupants in either building worked in their workplace on those days.

Statistical tools used
We analyzed the IoT sensing and survey data using R version 3.5.0 software [23]. We performed statistical analysis on the measured sensor data to compare between sensor locations (i.e., between buildings, between buildings and a regional weather station, and within buildings) and to compare survey responses between buildings and between typical conditions (i.e., non-forest fire) and forest fire responses. The data under consideration was not normally distributed, so we used non-parametric tests. To assess statistical significance between measured PM 2.5 concentrations at different locations, we used a two-sided Wilcoxon Rank-Sum test, also known as Mann-Whitney test. To compare IAQ satisfaction survey responses between buildings, we also used a two-sided Wilcoxon Rank-Sum test. We could have used a one-sided test because the hypothesis is that occupants in Wurster Hall had lower satisfaction than those in 4th Street, but we used a more conservative test. We tested all scenarios to a 95% confidence level, meaning statistical significance was considered when the p value was below 0.05. There is no guidance for effect size specific to occupant survey responses; we used Cohen's d value and effect size recommended thresholds from [24] and [25].

Outdoor PM 2.5 results: Spatial variability
Results in Fig 2 shows

Indoor PM 2.5 results
Results in this section include (i) indoor PM 2.5 concentrations and (ii) I/O ratio. In order to appropriately inform conclusions about the buildings from I/O ratios, we referenced information about the building infiltration and HVAC system operating hours to understand the underlying mechanisms that contributed to particle penetration and ingress.

Comparison of PM 2.5 concentration to WHO guidance.
We evaluated measured indoor PM 2.5 concentrations to assess building resilience from the occupant exposure perspective because it affects occupant health, productivity, and comfort. Results in Fig 3 show that the median hourly indoor PM 2.5 concentration over the entire Chico Camp Fire event was Theses values represent a general overview of the whole building indoor levels, which indicates that there was a difference in exposure levels between the two buildings. The overall whole building indoor PM 2.5 levels provide a coarse indicator of IAQ when they are benchmarked against recommended thresholds, such as WHO 24 hr exposure limit of 25 μg/m 3 or EPA limit of 35 μg/m 3 . Based on this simple comparison, we conclude that 4 th Street is more resilient than Wurster Hall to the extreme air pollution event. This cumulative comparison is possible only after the air pollution event has occurred and is useful for ranking buildings across a large portfolio (e.g. campus, district, town) and identifying which buildings should have resource priority.
From the occupants' perspective, PM 2.5 concentration and amount of time they were exposed play a critical role in health implications. In order to evaluate this in the context of building resilience, we must simultaneously evaluate ambient concentration and exposure time, which is done with the developed E-index tool. In Fig 4, the x-axis represents days during Results in Fig 4 show that 4 th Street whole building E-index was below WHO suggested threshold for 66% of the hours (indicated as E-index of 0) and was one or greater for 34% of the hours, with only 3% being two or greater, meaning that only 3% of the pollution episode hours had median indoor PM 2.5 concentration more than or equal to twice the WHO threshold level. Overall average E-index, calculated by averaging hourly values for the entire pollution episode, for 4 th Street was 0.82, suggesting that the building as a whole was resilient to outdoor air pollution during this episode, and that the tight building envelope paired with two-staged particle filtration of MERV 8 and MERV 13 at the air handler was effective at blocking PM 2.5 penetration and providing acceptable IAQ conditions.
Results for the Wurster Hall show that PM 2.5 concentration exceeded the recommended level for 77% of the hours. E-index was in the range from one to less than two for 48% of the hours, from two to less than four for 25% of the hours, and four or above for 3% of the hours. From an hourly analysis, we can see that Wurster Hall had periods that could potentially cause acute effects on occupants, especially for those with pulmonary health issues. One potential way to think about the role of interventions would be to eliminate periods where E-index was above a specified threshold (e.g. level that causes acute effects). For the entire pollution episode, overall average E-index was 1.69, suggesting that Wurster Hall requires intervention in order to be operational during an air pollution episode. Although overall average E-index enables us to compare and rank building resilience performance, the simplification to a single number representation omits important results: that for 29% of the hours Wurster Hall indoor concentration levels exceeded the WHO 24h-mean PM 2.5 exposure thresholds by more than two and was at a level that might cause acute health effects during occupied hours.

Cumulative and instantaneous I/O ratio.
As described in [17], I/O ratio provides information about the quantity of outdoor particles that enter indoors when indoor sources can be neglected. Considering cumulative I/O ratio at the whole building scale, the median I/ O ratio over the entire pollution episode was 0.27 for 4 th Street and 0.67 for Wurster Hall. Cumulative whole building I/O ratio is a general indicator of building resilience and can aid in comparing amongst buildings, but it does not contain information on more granular spatial or temporal variations. In Fig 5, we show the cumulative hourly I/O ratio across all sensors and IoT sensing tool for determining building resilience across the entire forest fire event. This can aid in identifying time dependent operational aspects at the whole building scale.
In some instances, the cumulative I/O ratio can provide insights about PM 2.5 penetration and ingress. As seen in Fig 5, there were trends in I/O ratio by hour of day during the entire pollution episode in each building. The 4 th Street had a higher probability of a larger I/O ratio during the HVAC system operating hours compared to the non-operating hours. During this period, median I/O ratio was 0.33 during operating hours compared to 0.21 during non-operating hours. This indicated that PM 2.5 ingress is the dominant pathway of outdoor particle entry. The higher I/O ratios observed in Wurster Hall compared to 4 th Street is due to the building being less airtight. Wurster Hall has operable windows designed for natural ventilation, and although the windows were closed (based on contact sensor readings) during the pollution event, the CO 2 decay rate shows infiltration of 0.4 h -1 , as opposed to 0.1 h -1 for 4 th Street.
Cumulative I/O ratio can also be viewed on smaller spatial scales, such as individual rooms or zones, which allows for comparison and assessment within a building, similar to the comparison between buildings. We can use cumulative I/O ratio as a tool to assess (postevent) the effectiveness of an intervention, such as changes in building control strategy, economizer regime, air handler filter grade, or changes in recirculated-to-outdoor air ratio. Local level of intervention might improve conditions in a particular room or zone without necessity for larger scale intervention or any significant effect on a building scale.
Cumulative I/O ratio describes overall resilience to an air pollution episode but is limited in that it is only a post-event evaluation tool. IoT sensing can also be used to show instantaneous I/O ratio that can aid in real-time evaluation, such as quickly assessing whether a portable filter is appropriately sized for a room. For example, at the room scale, the instantaneous (e.g., hourly) I/O ratio results in Figs 6A and 7B for Wurster Hall show that different indoor locations can vary from the median building I/O ratio (i.e., the median of all indoor sensors). Fig  6A depicts the effectiveness of a portable filter intervention in room 348 on Nov. 15 th . When in use, the portable filter quickly reduced the I/O ratio by more than 50%. The portable filter was manually turned on when the occupant entered the room then off when they left and that was repeated one more time during the day. This results in the sharp decreases and increases in I/O ratio in the middle of the day. Fig 6B shows that room 232, which is adjacent to a hallway with an entrance door to the building, can have~40% higher instantaneous I/O ratio compared to the whole building median. This is likely due to elevated infiltration of particles caused by opening the entrance doors and movement of people [26]. This space can be considered as a pollution hot zone in the building and interventions should be considered, especially considering that the space had five occupants. As described in the introduction, current guidelines only have qualitative suggestions about portable filters, and this tool enables quantification of effectiveness of that or any intervention.
Instantaneous I/O ratio can also be viewed at the whole building scale, as shown in Fig 7  which depicts the median I/O ratio from all indoor sensors at each hour for the entire period of the event. Results show that in each hour 4th Street building was more resilient to high outdoor PM 2.5 levels and allowed less PM 2.5 to enter the building compared to Wurster Hall. In this way, Instantaneous I/O ratio can be used to evaluate the effectiveness of an intervention at the whole building scale in real time, such as increasing the filter grade in air handling units, or adjusting the outdoor air supply rate.

Spatial and temporal variability within the buildings.
Placement of indoor PM 2.5 sensors is important for the evaluation of the building resilience; however, there is limited guidance on sensor density and appropriate placement. RESET and WELL v2 [27] suggest a sensing densities of one sensor per 500 m 2 and one sensor per 325 m 2 , respectively. Sensing

Survey of occupants
From the occupant perspective, resilience can be assessed based on how disruptive an air pollution event was to "business as usual" operation. We assessed this using a survey to  understand occupants' air quality perception, self-reported productivity, and behaviour. Survey questions can be found in Supporting Information.
We received 95 complete responses in 4 th Street (MV) and 101 in Wurster Hall (NV) from full-time occupants, correlating to response rates of 37% and 34%, respectively. In 4 th Street, 78% have worked in the space for more than 1 year, and 89% of respondents work more than 30 hours per week. The demographics of Wurster Hall respondents were similar: 60% have worked in the space for more than 1 year, with the next largest group being those that have worked between 7 to 12 months. 66% of respondents work more than 20 hours per week in the building (45% of respondents work more than 30 hours per week). There was a fairly even distribution across age groups between 21 and over 50 in both buildings. It should be noted that there was minimal campus-wide or building-specific guidance on indoor pollution levels, strategies occupants could take in their workplace, or alternative places to work. Fig 9 shows the results of occupant satisfaction with indoor air quality during typical conditions and during the Chico Camp fire in the two building types. Under typical conditions, occupant satisfaction with IAQ was relatively low in both buildings, with 25% and 41% voting 'slightly satisfied' or 'satisfied' in 4 th Street and Wurster Hall, respectively. There was a statistically significant difference in mean satisfaction scores (p<0.05), with Wurster Hall having higher satisfaction; however, the effect size was small to negligible (d = 0.3). There were 54% and 45% of respondents voting 'slightly dissatisfied' or 'dissatisfied' with IAQ in 4 th Street and Wurster Hall, respectively. The lower satisfaction in 4 th Street can be explained by the building's proximity to a cement facility, and, according to the building manager, occupants often complain of poor IAQ as a result even though there is a mechanical ventilation system.

Satisfaction with indoor air quality (IAQ).
Previous studies have found that humans are not accurate sensors of small changes in air pollutant concentrations when exposed to typical levels (i.e., typical to their experience), and that humans associate air quality with aspects of thermal comfort [28][29][30]. However, the survey responses indicate that during the extreme air pollution event, occupant perception aligned with measured PM 2.5 concentrations in the two study buildings. In Wurster Hall, dissatisfaction with IAQ during the forest fire (i.e., votes for 'slightly dissatisfied' and 'dissatisfied') increased from 45% to 89% of respondents, with a statistically significant difference in mean satisfaction vote (p<0.05) and a moderate to large effect size (d = 1.3) compared to typical conditions. Remarkably, there was only a slight increase in IAQ dissatisfaction from 54% to 57% in 4 th Street, and the difference in mean satisfaction vote was not statistically significant (p = 0.48) compared to typical conditions. The larger increase in dissatisfaction in Wurster Hall than 4 th Street aligns with the buildings' low resilience, as seen in Fig 3 indoor  A confounding factor that we could not control is the air temperature in the buildings, which can impact the IAQ perception. Along with mechanical ventilation, 4 th Street has air conditioning while Wurster Hall relies on operable windows for cooling. Outdoor temperatures were moderate at the time, with daily temperature highs between 15 to 22˚C. From the survey responses, air temperatures were only an issue in Wurster Hall classrooms with high occupancies. Three respondents mentioned that they opened windows or saw windows opened because classrooms became stuffy and hot, and they were forced to make a tradeoff with thermal comfort. Other locations such as private offices where the majority of full-time occupants reside were not likely affected by the air temperatures.

Perceived impact to work productivity.
Poor air quality during the Chico Camp fire may have affected productivity by disrupting work schedules or routines, and by causing mental and physical stress on occupants. There is limited research linking wildfire pollutants, including PM 2.5 and carbon monoxide, to decreased cognitive abilities. We asked a series of questions to assess how occupants felt IAQ in their workplace affected productivity, including cognitive abilities and psychological state. Responses can help to assess productivity loss during the extreme pollution event and provide information for an intervention cost-benefit analysis [31,32]. Fig 10 shows the self-reported impacts of workplace IAQ on occupants' ability to get their work done. Similar to occupants' satisfaction with IAQ in Fig 9, occupants in both buildings reported similar levels of impact under typical conditions, with slightly higher reports of interference in 4 th Street. Responses for the Chico Camp fire period are in line with previously reported trends of Wurster Hall occupants reporting more severe impact. Both buildings have statistically significant (p<0.05) difference in mean response between typical conditions and during the Chico Camp fire, but the percent of respondents that reported that workplace IAQ interferes (i.e., considering 'somewhat interferes' and 'interferes') with productivity increased more in Wurster Hall (38% to 81%) than in 4 th Street (45% to 52%) when comparing typical conditions to the Chico Camp fire.
In addition to asking directly about productivity, we wanted to understand how occupants felt (i.e., their psychological state) in regards to IAQ once they entered their building from the outside where there was high pollutant concentration, as seen in Figs 3 and 4. Although personal exposure varied by mode of transportation (e.g., by bike, by bus, walking, or by car) and use of N95 face masks, all occupants had some level of exposure and were likely aware of the highly publicized outdoor air quality. The results, shown in Fig 11, have a significant difference (p<0.05) in mean relief vote between the buildings with a small effect size (d = 0.42). More respondents felt relieved in 4 th Street and more felt not relieved in Wurster Hall, which aligns with the overall trend of measured performance for those buildings. Although the PM 2.5 concentrations in 4th Street were below WHO recommended level the majority of the time, 72% of respondents reported being 'not relieved' or only 'slightly relieved'. This could be due to indoor PM 2.5 concentrations being higher than occupants are typically exposed to, unsatisfactory IAQ during typical conditions, or lack of knowledge that indoor pollution was below the risk levels for healthy adults.
When survey results from Figs 10, 11 and 12 are combined with IoT PM 2.5 measurements in Figs 4 and 5, we see a trend that shows occupants perceived lower PM 2.5 in 4th Street compared to Wurster Hall. This provides evidence that surveys can be a useful tool in intervention assessment and aid in intervention decision-making during the severe outdoor air pollution events.

Occupant behavior and strategies to reduce exposure.
Another approach to understand occupants' perception of IAQ is by reviewing their actions to reduce exposure to pollutants during the Chico Camp fire. Occupants identified which building features they controlled and what strategies or devices, if any, they used to reduce exposure to pollutants while in their workplace during the Chico Camp fire. The results in Fig 12 shows that 24% in 4 th Street and 29% in Wurster Hall wore a face mask in their workplace at some point during the fire. A respondent from Wurster Hall stated, "During the forest fires, smoke filled the room to the point that I was wearing my mask inside." N95 face masks can reduce individual exposure, but are still a controversial issue especially related to effectiveness of non-fitted masks [13] and recommendations for some groups (e.g., pregnant women, small children, people with respiratory challenges) not to wear them. Additionally, we can reasonably expect that wearing face masks affects productivity by obstructing normal operation within the building, reducing interaction with colleagues, degrading general comfort, and increasing CO 2 inhalation which has been linked to cognitive function [33].
While occupants' direct responses about satisfaction and perceived impacts in each building align with IoT PM 2.5 sensing measurements, their reported behavior for wearing face masks did not. Despite indoor PM 2.5 concentration being typically below the WHO threshold level in 4th Street and occupants generally reporting higher satisfaction with and lower impact from IAQ during the Chico Camp fire, occupants still wore masks indoors at about the same rate as those in Wurster Hall. The use of face masks in both buildings may have been driven by concern over visual appearance of outdoor air conditions (e.g., grey fog), publicized air quality index (AQI) for outside air, lack of knowledge about air quality in the workplace, and lack of alternative solutions. The question assumes that respondents were negatively impacted by outdoor conditions; we included a response for "I was not affected by the outdoor conditions", indicated as "not affected". Other strategies to avoid pollutant exposure include operating interior doors and windows to the outside in Wurster Hall. As shown in Fig 12, 4 th Street occupants did not report high use of any controllable features, mostly because the space is largely open office (i.e., few interior doors) with a limited number of operable windows in select locations. The use of portable air filters was low in both building, with 11% using this device in 4 th Street and 16% using in Wurster Hall, less than the number of occupants wearing face masks. We did not collect information on the air filter type or size, but as seen in Fig 6A, a portable air filter can improve IAQ in these events. Some of the possible reasons that portable filters were not more widely used are upfront costs and lack of knowledge or means to measure effectiveness of air filters, which must be properly sized for room volume.
3.3.4 Survey as a building performance assessment tool. Comparing the IoT PM 2.5 measurements with the occupant IAQ satisfaction results shows that a survey has the potential to be an effective indicator of general IAQ during the high outdoor air pollution events; however, the survey only provides qualitative information and can assess whether a building is able to maintain "business-as-usual" for occupants. As in the case of Wurster Hall (NV), the survey indicated low IAQ satisfaction, but sensor measurements were necessary to quantify the indoor pollutant concentrations and mechanisms for air pollutant entry into the building. As opposed to the IoT devices, a survey is more often used as a retrospective tool (i.e., can only be applied after the air pollution event occurs), and it could be used to assess the resiliency of buildings or effectiveness of interventions to maintain "business-as-usual".
As indicated by respondents wearing face masks indoors, maintaining PM 2.5 concentration levels within the prescribed limits, either through building level features such as air handler's filter grade or alternatives solutions like portable air filters, might not be enough to prevent occupants from feeling at risk. A key missing piece was ensuring that occupants are aware of the conditions in the space. Results from this survey only point to this possibility, and need to be validated. Without sensor equipment from this study, neither 4th Street nor Wurster Hall had mechanisms to monitor indoor PM 2.5 levels and thus, there was a limited ability to provide building-specific information. This identifies a gap and an additional aspect to consider for a survey to probe into the effectiveness of communicating air quality data.

Recommendations for building resilience evaluation
The general trending direction of IoT environmental sensing application and research is to create urban air pollution maps and increase knowledge about the pollution distribution on a city scale; however, this maintains the assumption that outdoor measurement is a good proxy for personal exposure of the entire population within the vicinity of the measurement location. This study shows new and additional value of dense urban environmental IoT sensing networks and addresses the missing gap in integration between urban and building scale IoT sensing, especially considering that humans in developed parts of the world spend approximately 90% of their time indoors [34].
The use of these tools is widespread and can serve several purposes including benchmarking building performance, assessing the effectiveness of interventions, and understanding occupant perception and behavior. Benchmarking building resilience is useful for an entity that manages a large building portfolio (e.g. university campus consisting of~100 buildings). E-index and I/O ratio are benchmarking tools that can rank and prioritize buildings for improvements and inform occupants of safety. The use and potential applications of these tools at various temporal and spatial scales are listed in Table 2. E-index provides information on exposure level compared to a specified threshold (e.g., WHO) and can identify "safe" buildings or spaces within buildings; similarly, it will identify buildings with high air pollutant exposure or hotspots within buildings. I/O ratio can quickly assess the effectiveness of potential interventions on a whole building or space-by-space scale, such as seen in Fig 6 with  Creating a database with E-index and/or I/O ratio for a large and diverse set of buildings (i.e., with various typologies, ventilation air delivery strategies, cooling and heating systems, control sequences, facade air tightness, or pressure differentials between indoor and outdoors) will aid in identifying safe buildings and effective practices that can be implemented when extreme air pollution events occur. Together with low-cost IoT sensing, these tools can map out the indoor air pollution risk in a city or campus to inform organizational policies during an extreme event. This could lead to incorporating requirements for building resilience against wildfire-associated air pollution into the design stage of buildings. It can also lead to the development of clean air safe shelters for those at risk, such as homeless people, those who were evacuated from the wildfires affected areas, those with respiratory problems, or those who cannot afford portable filters.
Using the suggested criteria for a resilient building during extreme events, which is to maintain "business as usual" occupant behavior and perception, buildings can be considered successful if occupants feel safe enough indoors not to wear face masks. It is necessary to understand how to achieve this, and one component in this process is collecting IAQ information which can be achieved with IoT sensing. In future work, the survey can be used as an assessment tool and can be reduced to the questions necessary to understand perception and behavior. Additionally, it should include aspects about the effectiveness of methods to increase occupant awareness of indoor air quality during pollution episodes. Further, if IoT sensing measurements are not available or feasible for a building, the study results indicate that occupant perception could be used as a proxy for IAQ; although additional studies in various buildings should be conducted to confirm this.
In interpreting the study results, a few limitations should be recognized. Firstly, sample of only two buildings and a relatively limited number of responses were used for the analysis. As such, the I/O ratio, E-index and occupancy response results reported here cannot be indicative of all buildings and all people. Secondly, the assumption that one local measurement location (placing the sensor on the roof of each of the buildings) is sufficient to represent the outdoor pollution level for the entire building and conclude how the building operates in respect to the penetration of outdoor pollution indoors. Another limitation is a restricted utility of I/O ratio for recommending the specific building interventions, unless used for building resiliency evaluation against the elevated outdoor air pollution. However, the tools can be effective in dry areas prone to wildfires and in heavily polluted cities around the world. Notwithstanding the limitations, our study has not only provided useful preliminary data on which further research, but also offered a methodological contribution towards the assessment of building protection from air pollution of outdoor origin elsewhere.

Conclusions
The tools we used in this study are methods to evaluate building resilience to extreme outdoor air pollution episodes using IoT sensing indoors and outdoors, coupled with survey-based information of occupant perception and behaviour. We demonstrated the application of the tools on two buildings with different modes of ventilation. We characterized the resilience of the buildings on different temporal and spatial scales using the well-established I/O ratio and a newly proposed E-index. These tools can be used for a single building or across a portfolio of buildings for resilience ranking. Meanwhile, the survey tool provides qualitative insights on the connection between air pollution levels and occupants' perception and behavior, which allows assessment of a building's ability to be resilient, meaning that occupants can maintain "business as usual" view. Results indicate that even under the extreme outdoor air pollution events, outdoor PM 2.5 concentrations can be significantly different across a region, and therefore, outdoor sensors need to be localized to properly evaluate building resilience. Using the tools developed in this study comparing the resilience of mechanically ventilated 4 th Street building and naturally ventilated Wurster Hall building indicates: • Indoor PM 2.5 concentration over the entire Chico Camp Fire event was 21 μg/m 3 for 4 th Street and 36 μg/m 3 for Wurster Hall. The cumulative median I/O ratio over the entire pollution episode was 0.27 for 4 th Street and 0.67 for Wurster Hall. Overall E-index for 4 th Street was 0.82, suggesting that the whole building was resilient to air pollution while overall Eindex was 1.69 for Wurster Hall suggesting that interventions are necessary in order to be operational during an air pollution episode.
• The survey revealed that occupant perception of workplace IAQ aligns with measured PM 2.5 in the two buildings, with greater self-reported impacts in Wurster Hall which had significantly worse IAQ than 4th Street. After entering indoor environment coming from outside, occupants in Wurster Hall felt significantly less relieved (p<0.05) than those in 4th Street As many as 31% of occupants that did not feel relieved in Wurster Hall and only 13% in 4th Street Building. The results also revealed that a large portion of occupants wore face masks in both buildings, even though the PM 2.5 concentration was below the WHO threshold level. In 4th Street, which was likely driven by lack of alternative solutions and knowledge of PM2.5 levels. This suggests that occupants did not feel or did not know that the indoor air was at the acceptable level. We use this behaviour to suggest criteria for building resiliency.