Fast prediction of indoor airflow distribution inspired by synthetic image generation artificial intelligence.

Prediction of indoor airflow distribution often relies on high-fidelity, computationally intensive computational fluid dynamics (CFD) simulations. Artificial intelligence (AI) models trained by CFD data can be used for fast and accurate prediction of indoor airflow, but current methods have limitations, such as only predicting limited outputs rather than the entire flow field. Furthermore, conventional AI models are not always designed to predict different outputs based on a continuous input range, and instead make predictions for one or a few discrete inputs. This work addresses these gaps using a conditional generative adversarial network (CGAN) model approach, which is inspired by current state-of-the-art AI for synthetic image generation. We create a new Boundary Condition CGAN (BC-CGAN) model by extending the original CGAN model to generate 2D airflow distribution images based on a continuous input parameter, such as a boundary condition. Additionally, we design a novel feature-driven algorithm to strategically generate training data, with the goal of minimizing the amount of computationally expensive data while ensuring training quality of the AI model. The BC-CGAN model is evaluated for two benchmark airflow cases: an isothermal lid-driven cavity flow and a non-isothermal mixed convection flow with a heated box. We also investigate the performance of the BC-CGAN models when training is stopped based on different levels of validation error criteria. The results show that the trained BC-CGAN model can predict the 2D distribution of velocity and temperature with less than 5% relative error and up to about 75,000 times faster when compared to reference CFD simulations. The proposed feature-driven algorithm shows potential for reducing the amount of data and epochs required to train the AI models while maintaining prediction accuracy, particularly when the flow changes non-linearly with respect to an input.

Quantifying spatiotemporal variability in occupant exposure to an indoor airborne contaminant with an uncertain source location.

Well-mixed zone models are often employed to compute indoor air quality and occupant exposures. While effective, a potential downside to assuming instantaneous, perfect mixing is underpredicting exposures to high intermittent concentrations within a room. When such cases are of concern, more spatially resolved models, like computational-fluid dynamics methods, are used for some or all of the zones. But, these models have higher computational costs and require more input information. A preferred compromise would be to continue with a multi-zone modeling approach for all rooms, but with a better assessment of the spatial variability within a room. To do so, we present a quantitative method for estimating a room's spatiotemporal variability, based on influential room parameters. Our proposed method disaggregates variability into the variability in a room's average concentration, and the spatial variability within the room relative to that average. This enables a detailed assessment of how variability in particular room parameters impacts the uncertain occupant exposures. To demonstrate the utility of this method, we simulate contaminant dispersion for a variety of possible source locations. We compute breathing-zone exposure during the releasing (source is active) and decaying (source is removed) periods. Using CFD methods, we found after a 30 minutes release the average standard deviation in the spatial distribution of exposure was approximately 28% of the source average exposure, whereas variability in the different average exposures was lower, only 10% of the total average. We also find that although uncertainty in the source location leads to variability in the average magnitude of transient exposure, it does not have a particularly large influence on the spatial distribution during the decaying period, or on the average contaminant removal rate. By systematically characterizing a room's average concentration, its variability, and the spatial variability within the room important insights can be gained as to how much uncertainty is introduced into occupant exposure predictions by assuming a uniform in-room contaminant concentration. We discuss how the results of these characterizations can improve our understanding of the uncertainty in occupant exposures relative to well-mixed models.

Tradeoffs among indoor air quality, financial costs, and CO2 emissions for HVAC operation strategies to mitigate indoor virus in U.S. office buildings.

Adapting building operation during the COVID-19 pandemic to improve indoor air quality (IAQ) while ensuring sustainable solutions in terms of costs and CO2 emissions is challenging and limited in literature. Our previous study investigated different HVAC operation strategies, including increased filtration using MERV 10, MERV 13, or HEPA filters, as well as supplying 100% outdoor air into buildings for a system initially sized for MERV 10 filtration. This paper significantly extends that research by systematically analyzing the potential financial and environmental impact for different locations in the U.S. The previous medium office building system model is improved to account for operation in different climates. New evaluation metrics are created to consider the comprehensive impact of improving IAQ on costs and CO2 emissions, using dynamic emission factors for electricity generation depending on the location. HVAC operation strategies are studied in five different locations across the United States, with distinct climates and electricity sources. In four of the five locations, MERV 13 filtration offers the best improvement in IAQ per increase in costs and emissions relative to MERV 10. The exception is the mildest climate of San Diego, where use of 100% outdoor air provides the best IAQ with a limited increase in costs and emissions. A system not sized for HEPA filtration can lead to increased costs and emissions without much improvement in IAQ.

A Time-Varying Model for Predicting Formaldehyde Emission Rates in Homes.

Recent studies have succeeded in relating emissions of various volatile organic compounds to material mass diffusion transfer using detailed empirical characteristics of each of the individual emitting materials. While significant, the resulting models are often scenario specific and/or require a host of individual component parameters to estimate emission rates. This study developed an approach to estimate aggregated emissions rates based on a wide number of field measurements. We used a multi-parameter regression model based on previous mass transfer models to predict formaldehyde emission rate for a whole dwelling using field-measured, time-resolved formaldehyde concentrations, air exchange rates, and indoor environmental parameters in 63 California single-family houses built between 2011 and 2017. The resulting model provides time-varying formaldehyde emission rates, normalized by floor area, for each study home, assuming a well-mixed mass balance transport model of the home, and a well-mixed layer transport model of indoor surfaces. The surface layer model asserts an equilibrium concentration within the surface layer of the emitted materials that is a function of temperature and RH; the dwelling ventilation rate serves as a surrogate for indoor concentration. We also developed a more generic emission model that is suitable for broad prediction of emission for a population of buildings. This model is also based on measurements aggregated from 27 homes from the same study. We showed that errors in predicting household formaldehyde concentrations using this approach were substantially less than those using a traditional constant emission rate model, despite requiring less unique building information.

Assessing the use of portable air cleaners for reducing exposure to airborne diseases in a conference room with thermal stratification.

The COVID-19 pandemic has highlighted the need for strategies that mitigate the risk of aerosol disease transmission in indoor environments with different ventilation strategies. It is necessary for building operators to be able to estimate and compare the relative impacts of different mitigation strategies to determine suitable strategies for a particular situation. Using a validated CFD model, this study simulates the dispersion of exhaled contaminants in a thermally stratified conference room with overhead heating. The impacts of portable air-cleaners (PACs) on the room airflow and contaminant distribution were evaluated for different PAC locations and flow rates, as well as for different room setups (socially distanced or fully occupied). To obtain a holistic view of a strategy's impacts under different release scenarios, we simultaneously model the steady-state distribution of aerosolized virus contaminants from eight distinct sources in 18 cases for a total of 144 release scenarios. The simulations show that the location of the source, the PAC settings, and the room set-up can impact the average exposure and PAC effectiveness. For this studied case, the PACs reduced the room average exposure by 31%-66% relative to the baseline case. Some occupant locations were shown to have a higher-than-average exposure, particularly those seated near the airflow outlet, and occupants closest to sources tended to see the highest exposure from said source. We found that these PACs were effective at reducing the stratification caused by overhead heating, and also identified at least one sub-optimal location for placing a PAC in this space.

Measured influence of overhead HVAC on exposure to airborne contaminants from simulated speaking in a meeting and a classroom.

Tracer gas experiments were conducted in a 158 m3 room with overhead supply diffusers to study dispersion of contaminants from simulated speaking in physically distanced meeting and classroom configurations. The room was contained within a 237 m3 cell with open plenum return to the HVAC system. Heated manikins at desks and a researcher operating the tracer release apparatus presented 8-9 thermal plumes. Experiments were conducted under conditions of no forced air and neutral, cooled, or heated air supplied at 980-1100 cmh, and with/out 20% outdoor air. CO2 was released at the head of one manikin in each experiment to simulate small (<5 µm diameter) respiratory aerosols. The metric of exposure relative to perfectly mixed (ERM) is introduced to quantify impacts, based on measurements at manikin heads and at three heights in the center and corners of the room. Chilled or neutral supply air provided good mixing with ERMs close to one. Thermal stratification during heating produced higher ERMs at most manikins: 25% were ≥2.5 and the highest were >5× perfectly mixed conditions. Operation of two within-zone air cleaners together moving ≥400 cmh vertically in the room provided enough mixing to mitigate elevated exposure variations.

Investigation of HVAC operation strategies for office buildings during COVID-19 pandemic.

To minimize the indoor transmission of contaminants, such as the virus that can lead to COVID-19, buildings must provide the best indoor air quality possible. Improving indoor air quality can be achieved through the building's HVAC system to decrease any concentration of indoor contaminants by dilution and/or by source removal. However, doing so has practical downsides on the HVAC operation that are not always quantified in the literature. This paper develops a temporal simulation capability that is used to investigate the indoor virus concentration and operational cost of an HVAC system for two mitigation strategies: (1) supplying 100% outdoor air into the building and (2) using different HVAC filters, including MERV 10, MERV 13, and HEPA filters. These strategies are applied to a hypothetical medium office building consisting of five occupied zones and located in a cold and dry climate. We modeled the building using the Modelica Buildings library and developed new models for HVAC filtration and virus transmission to evaluate COVID-19 scenarios. We show that the ASHRAE-recommended MERV 13 filtration reduces the average virus concentration by about 10% when compared to MERV 10 filtration, with an increase in site energy consumption of about 3%. In contrast, the use of 100% outdoor air reduces the average indoor concentration by about an additional 1% compared to MERV 13 filtration, but significantly increases heating energy consumption. Use of HEPA filtration increases the average indoor concentration and energy consumption compared to MERV 13 filtration due to the high resistance of the HEPA filter.

Automating the interpretation of PM2.5 time-resolved measurements using a data-driven approach.

The rapid development of automated measurement equipment enables researchers to collect greater quantities of time-resolved data from indoor and outdoor environments. While significant, the interpretation of the resulting data can be a time-consuming effort. This paper introduces an automated process of interpreting PM2.5 time-resolved data and differentiating PM2.5 emissions resulting from indoor and outdoor sources. We use Random Forest (RF), a machine learning approach, to study a dataset of 836 indoor emission events that occurred over a 2-week period in 18 apartments in California. In this paper, we show model development and evaluate its performance as the sample size and source vary. We discuss the characteristics of the dataset that tended to help the source identification and why. For example, we show that data from many events and from different apartments are essential for the model to be suitable for analyzing a new separate dataset. We also show that longitudinal data appear to be more helpful than the time frequency of measurements within a given apartment. We use the resulting RF model to analyze PM2.5 data of an entirely separate dataset collected from 65 new homes in California. The RF model identifies 442 indoor emission events, with only a few misidentifications.

Air Pollutant Emission Rates for Dry Anaerobic Digestion and Composting of Organic Municipal Solid Waste.

Dry anaerobic digestion (AD) of organic municipal solid waste (MSW) followed by composting of the residual digestate is a waste diversion strategy that generates biogas and soil amendment products. The AD-composting process avoids methane (CH4) emissions from landfilling, but emissions of other greenhouse gases, odorous/toxic species, and reactive compounds can affect net climate and air quality impacts. In situ measurements of key sources at two large-scale industrial facilities in California were conducted to quantify pollutant emission rates across the AD-composting process. These measurements established a strong relationship between flared biogas ammonia (NH3) content and emitted nitrogen oxides (NOx), indicating that fuel NOx formation is significant and dominates over the thermal or prompt NOx pathways when biogas NH3 concentration exceeds ∼200 ppm. Composting is the largest source of CH4, carbon dioxide (CO2), nitrous oxide (N2O), and carbon monoxide (CO) emissions (∼60-70%), and dominate NH3, hydrogen sulfide (H2S), and volatile organic compounds (VOC) emissions (>90%). The high CH4 contribution to CO2-equivalent emissions demonstrates that composting can be an important CH4 source, which could be reduced with improved aeration. Controlling greenhouse gas and toxic/odorous emissions from composting offers the greatest mitigation opportunities for reducing the climate and air quality impacts of the AD-composting process.

Estimates of pollutant temporal and spatial variability in commercial buildings from the joint urban 2003 field experiments.

In this paper, we report on the indoor concentrations from a suite of full-scale outdoor tracer-gas point releases conducted in the downtown area of Oklahoma City in 2003. A point release experiment consisted of releases of sulfur hexafluoride (SF6 ) in multiple buildings and from different outdoor locations. From the measurements, we are able to estimate the concentration variations indoors for a building operating under "typical" operating conditions. The mean indoor spatial coefficients of variation are 30% to 45% from a daytime outdoor release are around 80% during an outdoor evening release. Having estimates of the spatial coefficient of variation provides stakeholders, including first responders, with the likely range of concentrations in the building when little is known about the building characteristics and operating behavior, such as developing urban-scale hazard and consequence analyses. We show differences in indoor measurements at different distances to the release points, floors of the building, and heating, ventilation, and air conditioning system (HVAC) operation. We also show estimates at different time resolutions. The statistics show that in the studied medium to large commercial buildings, spatial differences would result in peak indoor concentrations in certain parts of the buildings that may be substantially higher than the building average. To our knowledge, very few tracer gas measurements have been conducted in buildings of this scope, particularly with measurements on multiple floors and within a floor. The resulting estimates of spatial variability provide a unique opportunity for hazard assessment, and comparison to multi-zone models.

Prospects for finding Junge variability-lifetime relationships for micropollutants in the Danube river.

Persistence of chemical pollutants is difficult to measure in the field. Junge variability-lifetime relationships, correlating the relative standard deviation of measured concentrations with residence time, have been used to estimate persistence of air pollutants. Junge relationships for micropollutants in rivers could provide evidence that half-lives of compounds estimated from laboratory and field data are representative of half-lives in a specific system, location and time. Here, we explore the hypothesis that Junge relationships could exist for micropollutants in the Danube river using: (1) concentrations of six hypothetical chemicals modeled using the STREAM-EU fate and transport model, and (2) concentrations of nine micropollutants measured in the third Joint Danube Survey (JDS3) combined with biodegradation half-lives reported in the literature. Using STREAM-EU, we found that spatial and temporal variability in modeled concentrations was inversely correlated with half-life for the four micropollutants with half-lives ≤90 days. For these four modeled micropollutants, we found Junge relationships with slopes significantly different from zero in the temporal variability of concentrations at 88% of the 67 JDS3 measurement sites, and in the spatial variability of concentrations on 36% out of 365 modeled days. A Junge relationship significant at the 95% confidence level was not found in the spatial variability of nine micropollutants measured in the JDS3, nor in STREAM-EU-modeled concentrations extracted for the dates and locations of the JDS3. Nevertheless, our model scenarios suggest that Junge relationships might be found in future measurements of spatial and temporal variability of micropollutants, especially in temporal variability of pollutants measured downstream in the Danube river.

Tracking contributions to human body burden of environmental chemicals by correlating environmental measurements with biomarkers.

The work addresses current knowledge gaps regarding causes for correlations between environmental and biomarker measurements and explores the underappreciated role of variability in disaggregating exposure attributes that contribute to biomarker levels. Our simulation-based study considers variability in environmental and food measurements, the relative contribution of various exposure sources (indoors and food), and the biological half-life of a compound, on the resulting correlations between biomarker and environmental measurements. For two hypothetical compounds whose half-lives are on the order of days for one and years for the other, we generate synthetic daily environmental concentrations and food exposures with different day-to-day and population variability as well as different amounts of home- and food-based exposure. Assuming that the total intake results only from home-based exposure and food ingestion, we estimate time-dependent biomarker concentrations using a one-compartment pharmacokinetic model. Box plots of modeled R2 values indicate that although the R2 correlation between wipe and biological (e.g., serum) measurements is within the same range for the two compounds, the relative contribution of the home exposure to the total exposure could differ by up to 20%, thus providing the relative indication of their contribution to body burden. The novel method introduced in this paper provides insights for evaluating scenarios or experiments where sample, exposure, and compound variability must be weighed in order to interpret associations between exposure data.

A review of air exchange rate models for air pollution exposure assessments.

A critical aspect of air pollution exposure assessments is estimation of the air exchange rate (AER) for various buildings where people spend their time. The AER, which is the rate of exchange of indoor air with outdoor air, is an important determinant for entry of outdoor air pollutants and for removal of indoor-emitted air pollutants. This paper presents an overview and critical analysis of the scientific literature on empirical and physically based AER models for residential and commercial buildings; the models highlighted here are feasible for exposure assessments as extensive inputs are not required. Models are included for the three types of airflows that can occur across building envelopes: leakage, natural ventilation, and mechanical ventilation. Guidance is provided to select the preferable AER model based on available data, desired temporal resolution, types of airflows, and types of buildings included in the exposure assessment. For exposure assessments with some limited building leakage or AER measurements, strategies are described to reduce AER model uncertainty. This review will facilitate the selection of AER models in support of air pollution exposure assessments.

Siting samplers to minimize expected time to detection.

We present a probabilistic approach to designing an indoor sampler network for detecting an accidental or intentional chemical or biological release, and demonstrate it for a real building. In an earlier article, Sohn and Lorenzetti developed a proof of concept algorithm that assumed samplers could return measurements only slowly (on the order of hours). This led to optimal "detect to treat" architectures that maximize the probability of detecting a release. This article develops a more general approach and applies it to samplers that can return measurements relatively quickly (in minutes). This leads to optimal "detect to warn" architectures that minimize the expected time to detection. Using a model of a real, large, commercial building, we demonstrate the approach by optimizing networks against uncertain release locations, source terms, and sampler characteristics. Finally, we speculate on rules of thumb for general sampler placement.

Numerical solution of the Polanyi-DR isotherm in linear driving force models.

The Polanyi-Dubinin-Radushkevich isotherm has proven useful for modeling the adsorption of volatile organic compounds on microporous materials such as activated carbon. When embedded in a larger dynamic simulation--e.g., of whole-building pollutant transport--it is important to solve the sorption relations as quickly as possible. This work compares numerical methods for solving the Polanyi-DR model, in cases where transport to the surface is assumed linear in the bulk-to-surface concentration differences. We focus on developing numerically stable algorithms that converge across a wide range of inputs, including zero concentrations, where the isotherm is undefined. We identify several methods, including a modified Newton-Raphson search, that solve the system 3-4 times faster than simple bisection. Finally, we present a rule of thumb for identifying when boundary-layer diffusion limits the transport rate enough to justify reducing the model complexity.

Framework for evaluating anthrax risk in buildings.

If Bacillus anthracis (BA), the organism that causes anthrax, is known or suspected to have contaminated a building, a critical decision is what level of contamination is unacceptable. This decision has two components: (1) what is the relationship between the degree of contamination and the risk to occupants, (2) and what is an acceptable risk to occupants? These lead to a further decision: (3) how many samples must be taken to determine whether a building is unacceptably contaminated? We discuss existing data that bear on these questions, and introduce a nomogram that can be used to investigate the relationship between risk of contracting anthrax, the surface concentration of BA, the probability of detection, and the number of samples needed to ensure detection with a given degree of certainty. The same approach could be used for other agents that are dangerous due to resuspension of deposited particles.

Using information on uncertainty to improve environmental fate modeling: a case study on DDT.

Present and future concentrations of DDT in the environment are calculated with the global multimedia model CliMoChem. Monte Carlo simulations are used to assess the importance of uncertainties in substance property data, emission rates, and environmental parameters for model results. Uncertainties in the model results, expressed as 95% confidence intervals of DDT concentrations in various environmental media, in different geographical locations, and at different points in time are typically between 1 and 2 orders of magnitude. An analysis of rank correlations between model inputs and predicted DDT concentrations indicates that emission estimates and degradation rate constants, in particular in the atmosphere, are the most influential model inputs. For DDT levels in the Arctic, temperature dependencies of substance properties are also influential parameters. A Bayesian Monte Carlo approach is used to update uncertain model inputs based on measurements of DDT in the field. The updating procedure suggests a lower value for half-life in air and a reduced range of uncertainty for Kow of DDT. As could be expected, the Bayesian updating yields model results that are closer to observations, and model uncertainties have decreased. Sensitivity analysis and Bayesian Monte Carlo approach in combination provide new insight into important processes that govern the global fate and persistence of DDT in the environment.

Population physiologically based pharmacokinetic modeling for the human lactational transfer of PCB-153 with consideration of worldwide human biomonitoring results.

BACKGROUND: One of the most serious human health concerns related to environmental contamination with polychlorinated biphenyls (PCBs) is the presence of these chemicals in breast milk. OBJECTIVES: We developed a physiologically based pharmacokinetic model of PCB-153 in women, and predict its transfer via lactation to infants. The model is the first human, population-scale lactational model for PCB-153. Data in the literature provided estimates for model development and for performance assessment. METHODS: We used physiologic parameters from a cohort in Taiwan and reference values given in the literature to estimate partition coefficients based on chemical structure and the lipid content in various body tissues. Using exposure data from Japan, we predicted acquired body burden of PCB-153 at an average childbearing age of 25 years and compared predictions to measurements from studies in multiple countries. We attempted one example of reverse dosimetry modeling using our PBPK model for possible exposure scenarios in Canadian Inuits, the population with the highest breast milk PCB-153 level in the world. RESULTS: Forward-model predictions agree well with human biomonitoring measurements, as represented by summary statistics and uncertainty estimates. CONCLUSION: The model successfully describes the range of possible PCB-153 dispositions in maternal milk, suggesting a promising option for back-estimating doses for various populations.

Siting bio-samplers in buildings.

We describe a probabilistic approach to siting samplers for detecting accidental or intentional releases of biological material. In the face of uncertainty and variability in the release conditions, we place samplers in order to maximize the probability of detecting a release from among a suite of realistic scenarios. The scenarios may differ in any unknown, for example, the release size or location, weather, mode of building operation, etc. In an illustrative example, we apply the algorithm to a hypothetical 24-room commercial building, finding optimal networks for a variety of assumed sampler types. The results show how sampler characteristics, most importantly the detection limit, affect the network performance. This suggests using the probabilistic approach to guide the priorities of sampler designers, as well as to site samplers in specific buildings.

Standardized approach for developing probabilistic exposure factor distributions.

The effectiveness of a probabilistic risk assessment (PRA) depends on the quality and relevance of the output from exposure and risk models, which, in turn, depends on the critical inputs to the assessment. These critical inputs are often in the form of probabilistic exposure factor distributions that are derived for the given risk scenario. Deriving probabilistic distributions for model inputs can be time consuming and subjective. The absence of a standard approach for developing these distributions can result in PRAs that are inconsistent and difficult to review by regulatory agencies. We present an approach that reduces subjectivity in the distribution development process without limiting the flexibility needed to prepare relevant PRAs. The approach requires two steps. First, we analyze data pooled at a population scale to (i) identify the most robust demographic descriptors within the population for a given exposure factor, (ii) partition the data into subsets based on these variables, and (iii) construct archetypal distributions for each subpopulation. Second, we sample from these archetypal distributions according to site- or scenario-specific conditions to simulate exposure factor values and use these values to construct the scenario-specific input distribution. The archetypal distributions developed through Step 1 provide a consistent basis for developing scenario-specific distributions so risk assessors will not have to repeatedly collect and analyze raw data for each new assessment. We demonstrate the approach for two commonly used exposure factors--body weight (BW) and exposure duration (ED)--using data that are representative of the U.S. population. For these factors we provide a first set of subpopulation-based archetypal distributions and demonstrate methods for using these distributions to construct relevant scenario-specific probabilistic exposure factor distributions.