## Introduction

Inhabitants of developed countries spend up to 90% of their lives indoors [1]. As such, the quality of indoor air is critical to human exposure to pollution. Indoor pollution is composed of myriad constituents, which include oxidants and irritants, volatile organic compounds (VOC), and particulate matter (PM) [2,3,4,5,6,7,8,9,10]. Much, though not all, of indoor pollution is sourced directly from the indoor environment itself. VOC concentrations particularly are driven by indoor emissions, traceable to building materials and furnishings [11], use of consumer products and air fresheners [12], and cooking [13], among others. VOCs may be a primary cause of many sick building syndrome (SBS) symptoms and other health problems associated with indoor air [14,15,16,17,18]. Oxidation of VOCs can also produce secondary organic aerosols [19,20,21,22,23,24,25], which compound the PM burden and may pose harmful health risks themselves [26,27,28].

To reduce VOCs and other indoor-sourced pollutants from the indoor environment, buildings traditionally make use of infiltration and natural or mechanical ventilation air exchange [29], which is the replacement of stale indoor air with fresh air from the outdoors. Higher ventilation rates have been correlated with lower absenteeism and SBS symptom incidences, reductions in perceptions of odors, and increased task performance [30,31,32,33,34,35]. However, increased ventilation may augment the indoor concentration of outdoor-sourced pollutants, such as ozone and PM [9, 10, 36,37,38]. Increased ventilation also typically uses more energy [39,40,41], as outdoor air must be conditioned to be thermally comfortable. To address these drawbacks, alternative means of purifying indoor air to replace or supplement ventilation air are being investigated.

Experiments have demonstrated the ability of potted plants to reduce airborne VOC concentrations within sealed chambers. Many studies which carried out these experiments subsequently draw conclusions that potted plants may improve indoor air quality, spurring a presence of nonacademic resources (predominantly online) touting the use of houseplants as a sustainable means of cleaning indoor air. However, the experimental results of the underlying scientific works are often reported in ways such that they cannot simply be extrapolated into impacts in real indoor environments. Typical for these studies, a potted plant was placed in a sealed chamber (often with volume of ~1 m3), into which a single VOC was injected, and its decay was tracked over the course of many hours or days [42,43,44,45,46,47,48,49,50,51,52]. In contrast, building volumes are much larger than that of an experimental chamber, and VOC emissions are persistent. Also, indoor air is continuously exchanged with the outdoors. For instance, the median of measured residence times for air in US offices is about 50 min [53], and 80 min for US homes [19, 54, 55], corresponding to air exchange rates (AER) of 1.2 and 0.75 h−1, respectively, contrasting sharply with the long timescales needed for the chamber experiments to produce meaningful VOC reductions.

Some endeavors to minimize these differences between chambers and indoor environments have been pursued in studies, though not all issues have been resolved. For instance, Xu et al. [56] attempted to mirror more realistic conditions in what they referred to as a “dynamic” chamber, but no mention of air exchange was explicitly found in their work. Liu et al. [57] incorporated continuous airflow into their experiments, with constant upstream benzene concentrations of about 150 ppb. However, they maintained a very small chamber volume, inflating the relative influence of the plants. Sorption of VOCs onto the surfaces of the chamber is sometimes, but not always considered by these studies, which may be the cause of some of the observed VOC decay, rather than uptake by the plants. Other studies have proposed improvements to the design of plant chamber experiments, but they focused on conditions such as temperature, humidity, and carbon dioxide concentrations (all of which may impact plant health), instead of parameters which affect pollutant-building interactions [58, 59].

A few field campaigns have tried to measure the impact of plants within indoor environments, although Girman et al. [60] documented in detail the likely inaccuracies of the measuring equipment used in these studies. More importantly, none of them controlled or measured the outdoor air exchange rate. Conclusions can therefore not be drawn about the influence of plants versus the influence of VOC removal by air exchange. Of these studies, however, Dingle et al. [61] found no reduction in formaldehyde until plant density reached 2.44 plants/m2, at which point only a 10% reduction was seen. Wood et al. [62] claimed to observe VOC reductions of up to 75% within plant-containing offices at high VOC loadings, but they only sampled 5-min measurements once each week and neglected to report air exchange.

Only two publications were found that not only acknowledge these issues, but explicitly refute the notion that common houseplants improve indoor air quality. They were written by Girman et al. [60] and Levin [63]. Those works, authored by indoor air and building scientists, discuss in detail the history and limitations of the chamber and field studies, and provide a mass balance calculation that highlights the predicted ineffectiveness of using potted plants to remove VOCs from indoor air. Building upon that foundation, the work herein presents a review and impact analysis of removal rates reported by 12 cited works, most of which were conducted after the 1992 publication by Levin [63]. Among these works, the metrics used to report VOC removal are inconsistent, so comparisons and reproducibility are difficult to assess, as is predicting indoor air impacts. The present analysis thus first standardizes 196 experimental results into a metric useful for measuring indoor air cleaning, and then uses those standardized results to assess the effectiveness of using potted plants to remove VOCs and improve indoor air quality.

## Methodology

### Standardization of reported VOC removal

Within the building sciences, the indoor air-cleaning potential of a standalone device is parameterized with the clean air delivery rate (CADR). The CADR is the effective volumetric flow rate at which “clean” air is supplied to the environment, reflecting the rate at which the air cleaner removes pollutants. It is the product of the flow rate of air through the air cleaner (Qac., m3/h) and its removal efficiency (η), so CADR = Qacη (m3/h). The same air cleaner will have a greater impact in a smaller environment, so to gauge the impact of an air cleaner within the context of the indoor space it occupies, CADR must be normalized by the relevant indoor volume (V, m3). This CADR/V (h−1) parameter corresponds to a first-order loss rate constant (i.e., rate of pollutant removal is proportional to pollutant concentration).

Given that sufficient information is provided by a chamber study (e.g. physical chamber characteristics, experimental parameters), a CADR-per-plant (CADRp, m3 h−1 plant−1) can be computed using its results. The experimental procedures of the 12 considered studies used one of two general experimental setups. The first setup (setup I) assumes a perfectly sealed chamber with no VOC sources with uptake by the plant being the only loss mechanism, with a corresponding differential mass balance equation being:

$$V_{\mathrm{c}}\frac{{{\mathrm{d}}C}}{{{\mathrm{d}}t}} = - {\mathrm{CADR}}_{\mathrm{p}}C,$$
(1)

where C represents the VOC concentration in the chamber; Vc (m3) is the volume of the chamber; and t (h) is time. By integrating Eq. 1:

$$C_t = C_0e^{ - \left( {\frac{{{\mathrm{CADR}}_{\mathrm{p}}}}{{V_{\mathrm{c}}}}} \right)t},$$
(2)

where C0 is the initial concentration within the chamber; and Ct is the concentration chamber after t hours have elapsed. Using data provided by the chamber studies, the CADRp can be computed by rearranging Eq. 2:

$${\mathrm{CADR}}_{\mathrm{p}} = - \frac{{V_{\mathrm{c}}}}{t}\ln \left( {\frac{{C_t}}{{C_0}}} \right).$$
(3)

The second experimental setup (setup II) consists of steady state conditions in a flow-through chamber, instead of pollutant decay occurring in a sealed chamber. Equeations 13 no longer apply to this condition. In this case, the differential mass balance is described by the difference between the source terms (inlet flow) and loss terms (outlet flow + plant filtration):

$$V_{\mathrm{c}}\frac{{{\mathrm{d}}C}}{{{\mathrm{d}}t}} = Q_{\mathrm{c}}C_{{\mathrm{inlet}}} - \left( {Q_{\mathrm{c}} + {\mathrm{CADR}}_{\mathrm{p}}} \right)C_{{\mathrm{outlet}}},$$
(4)

where Qc (m3/h) is the flow rate through the chamber; Cinlet is the VOC concentration entering the chamber through its inlet; and Coutlet is the VOC concentration exiting the chamber (where C = Coutlet). Solving for CADRp under steady state conditions yields:

$${\mathrm{CADR}}_{\mathrm{p}} = \frac{{Q_{\mathrm{c}}}}{{1 - \left( {C_{{\mathrm{outlet}}}/C_{{\mathrm{inlet}}}} \right)}} - Q_{\mathrm{c}}.$$
(5)

The biases produced by neglecting surface sorption (in both setups) and chamber leakage (in setup I) from the mass balance equations (Eqs. 1 and 4, respectively) implicitly favor the efficacy of the plant removal, thereby providing absolute best-case estimates of the CADRp for the reviewed chamber studies.

### Description of considered chamber experiments

A CADRp dataset was developed using results of 12 published studies, comprising 196 potted plant chamber experiments. The experimental details of the 12 publications are summarily presented in Table 1, with further experimental detail and CADRp calculation results provided in the supplementary information (SI). All experiments measured VOC removal by a single plant within a controlled chamber, and one CADRp was computed for each experiment per plant per VOC species removed. However, the 12 studies reported their results in a variety of inconsistent metrics, as follows. Some studies only displayed plots of pollutant decay. Others included tables listing an initial concentration and the concentration after a certain amount of time (e.g. 24 h). Some reported drop in concentration per hour (in reality, the concentration reduction each hour will not be constant, because removal is likely first order, not linear). Furthermore, some normalized their results by surface area of plant leaf, while others did not measure leaf area at all—though if anything, large leaf surface areas may hinder VOC uptake, as the leaves serve to block air from passing over the growth substrate, which can dominate VOC removal [44, 64]. Table 1 broadly categorizes the studies into three groups based on their experimental setups and how their results were reported, each necessitating a different approach to determining CADRp values, including:

1. (1)

A sealed chamber (setup I) presenting only initial and final concentration measurements (or their ratios), for a certain duration of time.

2. (2)

A sealed chamber (setup I) presenting a timeseries of concentration measurements.

3. (3)

A flow-through chamber (setup II) presenting Cinlet and Coutlet measurements.

For the first category, Eq. 3 was used to compute CADRp values for the experiments. Aydogan and Montoya [42] tabulated the time taken for two-thirds of initial formaldehyde to be removed for four different plant species. Orwell et al. [47] tabulated average 24-h removal of benzene (C0Ct) from an initial dose (C0) for seven plant species, while Orwell et al. [48] tabulated the required time to reach Ct/C0 = 0.5 for various combinations of plant species, toluene, xylene. Wolverton et al. [49] tabulated percent removed after 24 h of formaldehyde, benzene, and trichloroethylene (TCE) for several plant species. Yoo et al. [51] reported removal per hour per leaf area (ng m−3 h−1 cm−2) for four plants removing benzene and toluene, providing initial concentrations and leaf surface areas. This CADRp calculation was carried out assuming their reported numbers corresponded to the first hour of the chamber experiment. Yang et al. [50] presented results similarly for five VOCs across several plant species organized qualitatively by performance (i.e., “superior,” “intermediate,” and “poor” performing plants). Zhang et al. [52] used a genetically modified version of Pothos Ivy, designed to enhance VOC uptake, and provided a percent reduction of concentration achieved over the timespan of days. The CADRp results for these studies are detailed in Table S1.

For the second category, a CADRp value was computed using Eq. 3 for each reported point in the timeseries. Their average was taken as the overall CADRp for that experiment. Irga et al. [43] plotted percent of benzene removed for two plant setups over the course of four days. Kim et al. [45] took hourly measurements over a 5-h period of cumulative concentration reduction of formaldehyde normalized by leaf area (µg m−3 cm−2) for dozens of plant species spanning four categories. Their 36 woody and herbaceous foliage plants were used for this dataset. Given the leaf area of all plant species and an initial concentration in the chamber, conversion to CADRp was possible. Kim et al. [46] plotted concentration over time for two distinct plant species removing three different VOCs. The CADRp results for these studies are detailed in Table S2.

For the third category, computing CADRp necessitated the use of Eq. 5. The Coutlet/Cinlet expression within Eq. 5 may equivalently be thought of as the fractional VOC removal, which Liu et al. [57] reported using setup II for benzene. Three of their plant species yielded 60–80% removal, 17 species yielded 20–40%, another 17 yielded 10–20%, 13 removed less than 10%, and 23 did not yield any benzene removal. These CADRp results are detailed in Table S3.

### Assessing effectiveness of potted plants as indoor air cleaners

The most prominent way by which VOCs are removed from indoor spaces is by outdoor-to-indoor air exchange. Air flows through a building at a certain flow rate (Qb, m3/h), which may be a combination of mechanical ventilation, natural ventilation, and uncontrolled infiltration through the building envelope. Typically, Qb scales with building size, so the volume-normalized flow, which is the air exchange rate (called AER or λ, h−1), is used to parameterize building airflow, where λ = Qb/V. This metric, as with CADR/V, is a first-order loss rate constant. Consequently, λ and CADR/V can be directly compared to assess the relative efficacy of each removal type. For air cleaning to be considered effective, the loss rate due to the air cleaner (CADR/V) must be on the same order or higher as that of the air exchange (λ) loss rate. So, if λ CADR/V, most of the pollution removal is accomplished via air exchange alone. If λ CADR/V, the air cleaner is responsible for the most removal. If λ = CADR/V, the two loss mechanisms have the same influence.

For the case of multiple indoor potted plants combining their individual CADRp to remove VOCs from an indoor environment, the net CADR/V loss rate may be computed given the density of plants in a given floor area (ρp, plants/m2), and the volume of the considered building in terms of the product of an average ceiling height (h, m) and the given floor area (A, m2) by:

$$\frac{{{\mathrm{CADR}}}}{V} = \frac{{\left( {{\mathrm{CADR}}_{\mathrm{p}}\rho _{\mathrm{p}}A} \right)}}{{\left( {hA} \right)}} = \frac{{{\mathrm{CADR}}_{\mathrm{p}}\rho _{\mathrm{p}}}}{h}$$
(6)

so that CADR/V depends on CADRp, ρp, and h. Since the ceiling height h is likely far less varied than CADRp or ρp throughout the US building stock, excluding atriums, it is taken as a constant h = 2.5 m ≈ 8 ft throughout the following analysis.

Comparisons of plant and AER loss mechanisms may be quantified by the effectiveness parameter (Γ), defined as the fraction of VOC removal by which plant-induced air cleaning alone is responsible:

$$\Gamma = \frac{{({\mathrm{CADR}}/V)}}{{\lambda + \left( {{\mathrm{CADR}}/V} \right)}}$$
(7)

Thus, Γ is bounded by 0 and 1. If Γ 0 (λ CADR/V), the air cleaner is wholly ineffective compared to air exchange loss; if Γ 1 (λ CADR/V), the air cleaner dominates removal; and if Γ = 0.5 (λ = CADR/V), the air cleaner and air exchange losses contribute equally to total removal. Substituting the right-hand-side of Eq. 6 into (CADR/V) in Eq. 7 facilitated a simulation-based parametric analysis of the effectiveness of VOC removal by potted plants indoors.

## Results and discussion

### CADR of potted plants in reviewed studies

In total, 196 CADRp values were computed from the 12 reviewed chamber studies. A histogram expressing this entire dataset is provided in Fig. 1a, which possesses a wide spread of nearly four orders of magnitude (ranging from 0.0004–0.2 m3 h−1 plant−1 at 10th and 90th percentiles), a median CADRp = 0.023 m3 h−1 plant−1, and a mean (standard deviation) of 0.062 (0.089) m3 h−1 plant−1. Even though these CADRp values represent best-case scenarios (as they were computed assuming negligible chamber sorption and leakage), their magnitudes are exceedingly small. For context, typical gas or particle air cleaners possess average CADR values on the approximate order of ~100 m3/h [65,66,67].

Figure 1b resolves all 196 datapoints contributing to the Fig. 1a histogram by type of VOC measured, labeled by the study‘s first author and reference number. This figure thus explores the possibility of constraining CADRp for each VOC. Some of the data preliminarily indicates that certain VOCs may be more efficiently removed by potted plants; for instance, Kim et al. [44,45,46] observed better formaldehyde removal than for xylene, and Wolverton et al. [49] observed a much lower TCE removal than for formaldehyde and benzene. However, these trends are not consistent throughout all studies; for instance, Yang et al. [50] observed similar removal of TCE, benzene, and toluene. Also, not enough studies assessed the same combinations of VOCs sufficient for a definitive trend to be established. Furthermore, some results vary largely from study-to-study even for the same VOC.

More notably, however, the variance of CADRp values belonging to a particular study is much smaller than the variance of the dataset as a whole (intra-study values range 1–2 orders of magnitude, as compared to the total CADRp range of ~4 orders of magnitude). For example, of the 46 CADRp values calculated from Kim et al. [44,45,46], 32 of them (70%) reside above 0.1 m3 h−1 plant−1, making up 84% of the total 38 CADRp greater than 0.1 m3 h−1 plant−1. On the other end of this spectrum, all CADRp values belonging to Irga et al. [43] and Yang et al. [50] were less than 0.001 m3 h−1 plant−1, making up all but one other CADRp below 0.001 m3 h−1 plant−1. The one remaining CADRp existing in this lowest-performing interval belongs to Zhang et al. [52], who also conducted an experiment with chloroform, despite their use of genetically modified plants shown to enhance VOC uptake. We believe these trends suggest that the varying VOC removal performance among different research studies may be an indicator of differences among removal measurement methodologies, which should be further investigated. These perhaps include measurement techniques, plant and rhizosphere health, and other characteristics and relative sizes of the chamber, soil, pot, or the plant itself (e.g. VOC sorption onto competing surfaces).

### Effectiveness in typical buildings

Using the entire CADRp dataset (Fig. 1a), Eq. 6 was used to compute four sets of total CADR/V loss rates, binned into four distinct plant density (ρp) cases separated at logarithmic intervals (0.1, 1, 10, and 100 plants/m2). In Fig. 2, these loss rates are compared directly to a distribution representing the AER typical of US residences [54, 55] and another representing AERs typical of US offices [53]. Again, these two types of loss rates can be directly compared to demonstrate their relative impacts on VOC removal. The two boxes corresponding to ρp values of 0.1 and 1 plants/m2 are barely visible, so their corresponding loss rates are almost certain to be negligible, even if plants exhibiting the highest plausible CADRp are used. For a ρp = 10 plants/m2, some of the loss rates due to VOC removal by the plants from the upper end of the CADRp distribution may comparable to air exchange losses in particularly tight buildings, but the median CADR/V is still negligible compared to the median AER for both residences and offices.

This assessment is in strong agreement with the conclusions of Girman et al. [60] and Levin [63]. Using similar mass balance calculations and the most generous selection of the early published Wolverton et al. [49] data, Levin [63] determined that a ~140 m2 house (1500 ft2) would require 680 houseplants (i.e., ρp = 4.9 plants/m2) for the removal rate of VOCs by plants indoors to just reach 0.096 h−1. Achieving these rates of plant density throughout a building is obviously not attainable. Even ρp = 1 plants/m2 would rule out any useful occupant-driven architectural programming being applied to a building, and it would take a theoretical ρp = 100 plants/m2 for the entire CADR/V loss rate distribution to be comparable to the AER distributions on a whole.

A parametric analysis was used to predict the required ρp necessary to achieve a desired effectiveness for various combinations of AER and representative CADRp. The analysis computed ρp required for varied Γ between 0 to 1 and AER between 0.1 and 10 h−1, thus exhausting all Γ possibilities and all reasonably expected indoor AERs in typical buildings. The CADRp was set at one of three discrete cases. The first was a low CADRp case, corresponding to the 10th percentile of the complete CADRp dataset (0.00014 m3 h−1 plant−1); the second used the median of the CADRp dataset (0.023 m3 h−1 plant−1); while the third used the 90th percentile (0.19 m3 h−1 plant−1). The ρp predictions are presented as contour plots in Fig. 3, which are binned at factor-of-ten intervals from ρp < 1 to ρp > 10,000 plants/m2.

At the strongest-case CADRp assumptions (Fig. 3c), an effectiveness of ~20% may be realized in an extremely low-AER building (e.g. λ < 0.2 h−1) if one potted plant is used per square meter of the indoor floor area. This effectiveness quickly falls off if an even slightly higher air exchange rate is experienced. But, as was stated, this ρp = 1 plants/m2 is too dense to be practical within a building, and it barely registers as effective under the most generous CADRp and AER assumptions. Under the more likely plant-removal characteristics (Fig. 3a, b), any legitimate effectiveness, even in buildings with the lowest air exchange, would require ρp values that are not only impractical or infeasible indoors, but are ludicrously large. Note again that the analyses in this section were carried out with a best-case CADRp dataset, which computed CADRp assuming neither chamber leakage nor surface sorption contributed to observed losses, so even these impossibly large ρp values essentially represent a lower bound.

### Other considerations

The conditions within sealed chambers do not scale up to the conditions of real indoor environments, which have high AER, large volumes, and persistent VOC emissions. Our conclusion that plants have negligible impact on indoor VOC loads is consistent with the results of field studies that did not observe real VOC reductions when plants were placed in buildings. Despite potted plants not appreciably affecting indoor VOC concentrations, conducting chamber experiments on plants can remain a consequential effort. There is much to still be learned pertaining to the mechanisms of botanical uptake of VOCs. And, other applications of botanical filtration do exist (although passively cleaning indoor air is not one of them). Potential usefulness for further research perhaps lies in plant-assisted botanical bio-trickling purifiers (colloquially, “biowalls” or plant walls), which mechanically pull air through a porous substrate supporting plants and their root ecosystems [68,69,70]. These may create a more effective means of VOC removal because of their size, exposed rhizosphere, and controlled and continuous airflow. Some recent studies suggest that biowalls may yield CADRs on orders of 10–100 m3/h for certain VOCs [71, 72], with the potential to make worthy contributions to indoor VOC removal. However, more biowall field assessments and modeling endeavors are required to better hone our understanding of their true air cleaning and cost effectiveness.

Regardless of application, more rigor is required in future chamber experiments to remove methodological ambiguities. First-order loss must be used to interpret results, and chamber leakage and surface sorption (to the chamber walls as well as to the pot and soil) must be accounted for. A standardized metric to be used in mass balance calculations, such as the CADR, should also be a critical aspect of future experimental reporting. Research also suggests that the plant itself is less crucial to VOC removal than the microbial community which resides within the rhizosphere/soil system of the plant [73, 74].

The issue of bringing plant life into the indoor environment is also a complex one, not settled by a potted plant’s (in)ability to reduce airborne VOCs. Indoor plants, by helping to create a more biophilic indoor environment, may have a positive impact on occupant well-being [75], which may also translate into productivity improvements for businesses. However, plant introduction may also come with certain costs or trade-offs. One potential associated downside of plants indoors may be increased humidity. Also, plants have been shown to produce certain VOCs under particular conditions [76, 77]. So even if a potted plant works to slightly reduce, for instance, the persistence of formaldehyde indoors, its net impact on total VOC concentrations and overall indoor air quality is less clear. Spores and other bioparticle emissions may also be produced by plants, which have been observed from biowall systems [65, 74, 75]. Continued rigorous laboratory and field studies are required to develop a more complete and nuanced understanding of the interplay between plants and indoor environmental outcomes.