Introduction
The International Agency for Research on Cancer classified particulate matter as a Group 1 carcinogen in 2013 (Hamra et al., Reference Hamra, Guha, Cohen, Laden, Raaschou-Nielsen, Samet, Vineis, Forastiere, Saldiva and Yorifuji2014; IARC, 2015). Particulate matter’s small size allows it to penetrate the brain, contributing to respiratory diseases, cardiovascular disease, diabetes and mental illnesses (Block and Calderón-Garcidueñas, Reference Block and Calderón-Garcidueñas2009; Calderón-Garcidueñas et al., Reference Calderón-Garcidueñas, Calderón-Garcidueñas, Torres-Jardón, Avila-Ramírez, Kulesza and Angiulli2015). Green spaces have been proposed as a way to mitigate the harmful impacts of particulate matter. These spaces offer the potential to mitigate particulate matter through various mechanisms, such as adsorption, blocking and precipitation (Bernatzky, Reference Bernatzky1983; Choi et al., Reference Choi, Yoo, Yeo and Park2022; Janhäll, Reference Janhäll2015; Jin et al., Reference Jin, Guo, Wheeler, Kan and Che2014; Liu et al., Reference Liu, Zhu, Wang, Yang, Liu, Qiu, Ma, Zhang and Liu2016; Sgrigna et al., Reference Sgrigna, Sæbø, Gawronski, Popek and Calfapietra2015; Vesala et al., Reference Vesala, Suni, Rannik, Keronen, Markkanen, Sevanto, Grönholm, Smolander, Kulmala and Ilvesniemi2005). The theory of biophillia suggests that green spaces provide benefits by promoting social interaction and social cohesion, which are associated with mental health (White and Heerwagen, Reference White, Heerwagen and Lundberg1998; Wilson, Reference Wilson1984). The presence of green spaces, acting as buffers, may mitigate the adverse effects of air pollution on mental health by improving air quality and providing psychological benefits (Gascon et al., Reference Gascon, Triguero-Mas, Martínez, Dadvand, Rojas-Rueda, Plasència and Nieuwenhuijsen2016; James et al., Reference James, Banay, Hart and Laden2015).
Depression is the most common mental disorder globally and could lead to suicide if it worsens, according to the WHO (WHO, 2023). A recent epidemiological study on mental health revealed that depression prevalence has been gradually rising from 4.0% in 2001 to 6.7% in 2011 in Korea as well, where it has also emerged as a significant health issue (Koo, Reference Koo2018; Lee and Park, Reference Lee and Park2014). From 2000 to 2010, the disability-adjusted life years per 100,000 people with major depressive disorder in Korea rose from 427 to 1508 (Lee and Park, Reference Lee and Park2014). Furthermore, studies suggested a depression prevalence of 26% among individuals aged 60–64 (Park et al., Reference Park, Park and Jun2013), which increases to an estimated 35.4% among those older than 80 (Park and Kim, Reference Park and Kim2011). Specifically, in 2014, 46.3% of older adults in South Korea living in relative poverty experienced depression (Kim et al., Reference Kim, Kim, Ahn and Juon2022). The impact of depression might also be underestimated because it is more likely for patients with depression to perform poorly in school, have unstable marriages and relationships, have reduced birth outcomes, experience unemployment, and even have suicidal thoughts, all of which further reduce their quality of life (Kessler, Reference Kessler2012; Ribeiro et al., Reference Ribeiro, Huang, Fox and Franklin2018).
Given the high prevalence and severity of depression, finding modifiable environmental features such as air pollution and greenness cannot be disregarded in the quest to better handle depression. Some epidemiological studies imply that air pollution may be related to mental health, including depression (Braithwaite et al., Reference Braithwaite, Zhang, Kirkbride, Osborn and Hayes2019), while negative associations have been shown between greenness and self-perceived stress (Banay et al., Reference Banay, James, Hart, Kubzansky, Spiegelman, Okereke, Spengler and Laden2019; Roe et al., Reference Roe, Thompson, Aspinall, Brewer, Duff, Miller, Mitchell and Clow2013) or mental health (Gascon et al., Reference Gascon, Sánchez-Benavides, Dadvand, Martínez, Gramunt, Gotsens, Cirach, Vert, Molinuevo and Crous-Bou2018). In a longitudinal paper that studied depression, residential greenness and particulate matter, it was found that the protective effect of residential green space was partially mediated by PM2.5 and PM10 (Zhang et al., Reference Zhang, Wei, Yu, Guo, Wang, Jin, Shui, Lin, Tang and Chen2022). Previous studies focused on the effects of each factor of air pollution and green space or on the effects of simultaneous exposure on depression, and the results of studies on the interaction of the two variables were not sufficient, limiting the interpretation of the study results.
This study used a longitudinal data (Korean Longitudinal Study of Aging [KloSA]) to examine the relationship between air pollution and depressed symptoms as well as how greenness was affected.
Method
Population
This study was based on the KLoSA, which was conducted by the Ministry of Employment and the Korea Employment Information Service. Using the stratified systematic sampling method, population survey districts stratified by 15 metropolitan cities, 2 provinces (dong and eup/myeon) and 2 housing types (house and apartment) were arranged by administrative code, and the assigned number of survey districts was extracted by the systematic sampling method. The interviewers visited the extracted households, selected the people aged 45 or older among the household members, and then conducted the Population Research Panel Survey through a computer-assisted personal interview. The KLoSA, which has been conducted every other year since 2006, is a longitudinal survey of the older adults residing in the Korea regions excluding islands, and the eighth survey was finished in 2020. The main purpose of the KLoSA was to produce basic information on older people, including their socioeconomic characteristics, family relationships, living area, psychological health, and health status. The number of participants in local communities nationwide was 10,254 in 2006, and 920 more were registered in 2014. A detailed user’s guide is available on the website (https://www.Klosa.re.kr).
In this analysis, only three waves were used: the sixth wave (n = 7490), the seventh wave (n = 6940) and the eighth wave (n = 6488). The overall weighted response rate was 79.6%, 78.8% and 78.1% for waves 6, 7 and 8, respectively. Among them, those who didn’t complete the questionnaire on depression or socioeconomic characteristics were excluded. The total study population was 7376, 6730 and 6404 in 2016, 2018 and 2020, respectively.
CES-D 10 (Centre for Epidemiological Studies Depression Scale)
Individual-level time-variant depression was assessed using the CES-D 10 of the Boston Form (Bae et al., Reference Bae, Kim, Doh, Kim and Park2020; Radloff, Reference Radloff1977). The CES-D 10 is composed of 10 different items from the original 20 (Lewinsohn et al., Reference Lewinsohn, Seeley, Roberts and Allen1997) and is used to assess respondents’ recent (the past week or so) depression symptoms on a 4-point Likert scale with a cut-off of 20, especially in the elderly population, while the total score of CES-D 10 ranges from 10 to 40, with higher scores indicating more severe depressive symptoms (Lee and Lee, Reference Lee and Lee2014; Pun et al., Reference Pun, Manjourides and Suh2018; Zhang et al., Reference Zhang, O’Brien, Forrest, Salters, Patterson, Montaner, Hogg and Lima2012). The questionnaire includes three questions about depressed affect, two questions about positive affect, three questions about somatic symptoms, and two questions about interpersonal relationships. The CES-D 10 exhibited strong internal consistency, as indicated by Cronbach’s alpha coefficients of 0.87, 0.89 and 0.87 in 2016, 2018 and 2020, respectively.
Air pollution data
We employed monthly concentration predictions for ambient levels of PM2.5 and PM10 at 1 km2 spatial resolution from a machine-learning-based validated model with high prediction performance across 226 districts in contiguous South Korea (cross-validated R 2: 0.87). This predicted air pollution concentration was previously used in a previous paper (Park et al., Reference Park, Kang, Min, Kim, Song, Jang, Kwon, Oh, Moon and Kim2023). The prediction model is described in detail in the supplementary material (see Appendix, 1. Air Pollution Prediction Model). We estimated the annual mean concentration of PM2.5 and PM10 for each of the 217 districts using this data by averaging the concentrations of grids with the centroid point inside the district boundary. Annual district-specific air pollution concentrations for the years 2016, 2018 and 2020 are then linked to the individuals, representing long-term air pollution exposure levels corresponding to their residential district each year. We also calculated annual concentrations of NO2 and 8-hour maximum O3 (cross-validated R 2: 0.76 and 0.84, respectively) to utilize other pollutants as confounders to identify the PM10 and PM2.5 effects after controlling for them.
Greenness data
The level of residential greenness was assessed by two satellite-image-based metrics: Moderate Resolution Imaging Spectroradiometer (MODIS) normalized difference vegetation index (NDVI) and Enhanced Vegetation Index (EVI). The data, collected at 16-day intervals on a 250 m grid unit basis, involved averaging NDVI and EVI values within each district (si/gun/gu) polygon, utilizing the district shape file. We acquired the datasets from the latest version of the Tropical Rainfall Measuring Mission (TRMM) (https://developers.google.com/earth-engine/datasets/catalog/MODIS_061_MOD13A2, accessed on 8 October 2022) covering the period from 2016 to 2020 for the catchment area. A detailed description is available on the website.
NDVI calculates the ratio of the difference of near-infrared region and red reflectance to the sum of these two and ranges from −1 to +1 (Song et al., Reference Song, Lane, Kim, Kim, Byun, Le, Choi, Park and Lee2019). The estimate of EVI is based on NDVI and further controls canopy background, aerosol resistance and gain factor, which means it functions better in densely vegetated areas (Nagler et al., Reference Nagler, Glenn, Nguyen, Scott and Doody2013) and has higher sensitivity over high biomass regions. High NDVI and EVI indicate high greenness. The time-varying annual NDVI and EVI of 2016–2020 were matched to the panel data of the same period. The core model and description were based on the NDVI to study the beneficial effect of greenness.
Controlled variables
Controlled variables were used in this data set, including socioeconomic characteristics and physical condition. Sex was coded as a two-factor variable (female; male). Age was coded as a continuous variable (years). Current smoking was coded as a two-factor variable (no or yes in the past; yes). Current drinking was coded as a two-factor variable (no or yes in the past; yes). Education attainment was coded as a three-factor variable (primary school and below; middle and high school; college and above). Marital status was coded as a two-factor variable (single, divorced and widowed, married and not living with spouse; married and living with spouse). Social contact was coded as a three-factor variable (every other month or less often; once a month or more often and less than once a week; once a week or more often). Self-reported health status was coded as a two-factor variable (normal or bad; good). Private medical insurance was coded as a two-factor variable (no medical insurance; having medical insurance). Exercise was coded as a two-factor variable (no; yes). Household net worth (10000 Korean won, 7.59 USD) was coded as a five-factor variable (∼7000; 7001–15000; 15001–23000; 23001–39000; 39001∼). Density of population (/km2) was coded as a five-factor variable (∼280.4; 280.5–1125.5; 1125.6–4745.7; 4745.8–10891.2; 10891.3∼). Longitude, latitude, number of beds in hospitals per 1,000 persons, number of national basic livelihood beneficiaries, independent rate of finance of local government, and proportion of basic pension beneficiaries were coded as continuous variables. We chose covariates to account for possible confounding variables. As basic demographic information, age and sex, behavioural factors such as current smoking, current drinking, socio-economic factors such as private medical insurance, household net worth, education attainment and marital status, known to be closely related to depression were selected (Kim et al., Reference Kim, Cho, Isehunwa, Noh, Noh, Oh, Koh and Kim2020, Reference Kim, Lim, Bae, Kim, Jung and Hong2016; Shi et al., Reference Shi, Li, Zhang, Sun, Chen, Wang, Fang, Zhao, Du and Shi2020; Zhang et al., Reference Zhang, Zhao, Hong, Chang, Ryu, Kang, Monteiro, Shin, Guallar and Cho2019). Social contact, exercise (Wang et al., Reference Wang, Liu, Xue, Yao, Liu and Helbich2019b, Reference Wang, Yang, Yao, Bloom, Feng, Yuan, Zhang, Liu, Wu and Lu2020), self-reported health status (Shi et al., Reference Shi, Li, Zhang, Sun, Chen, Wang, Fang, Zhao, Du and Shi2020) are important confounders potentially affecting to particulate matters, greenness, and depression, and were also considered. Lastly, we selected regional factors (density of population, number of beds in hospitals per 1,000 persons, number of national basic livelihood beneficiaries, independent rate of finance of local government, proportion of basic pension beneficiaries) because there is evidence that the impact of greenness differs by urbanicity (Huang et al., Reference Huang, Huang, Feng, Pearce, Zhao, Pan and Liu2021). We summarised potential confounding factors from the literature using a directed acyclic graph (Supplementary Figure S1).
Statistical analysis
The study involved the categorization of participants into four groups based on the quantiles of NDVI due to ease of epidemiological interpretation, given the heavily skewed distribution (Banay et al., Reference Banay, James, Hart, Kubzansky, Spiegelman, Okereke, Spengler and Laden2019; White et al., Reference White, Elliott, Grellier, Economou, Bell, Bratman, Cirach, Gascon, Lima and Lõhmus2021). Descriptive statistics were used to present continuous variables as mean and standard deviation across the four groups, while categorical variables were presented as frequency and proportion. Statistical tests such as Pearson’s chi-squared test and ANOVA were employed to compare categorical and continuous variables, respectively.
To investigate the association between long-term exposure to air pollution (specifically PM2.5 and PM10) and the score of CES-D 10, as well as the potential protective impact of NDVI on air pollution, generalized estimating equation (GEE) models were utilized. This approach allowed us to analyse longitudinally measured data with binary outcomes, accommodating within-subject variability and changes in outcome status over time within the model. Air pollution variables were treated as continuous, CES-D 10 scores as dichotomous (depression: ≥20, without depression: <20) (Fu et al., Reference Fu, Si and Guo2022; Lee and Lee, Reference Lee and Lee2014), and NDVI as a four-factor variable. The particulate matter data were adjusted for 10 
${\mu\text{ g/}}{{\text{m}}^{\text{3}}}$. Odds ratios (ORs) and 95% confidence intervals were calculated and reported. The epidemiological perspective would find the OR of depression linked to air pollution more comprehensible. Additionally, interaction effects were examined to identify the potential beneficial effects of NDVI on depression. Interaction terms of NDVI quantiles and particulate matter were included in two exposure models to explore the interaction effects of combined exposures.
The primary hypothesis of the study posited that NDVI would mitigate the adverse impacts of air pollution on depressive symptoms. Three models were employed: an unadjusted model that considered particulate matter (PM10 or PM2.5), NDVI quantiles, and interaction term of particulate matter and NDVI quantiles; a minimally adjusted model that considered particulate matter, NDVI quantiles, and interaction term of particulate matter and NDVI quantiles, adjusted for year, longitude, latitude, interaction term of longitude and latitude; and a fully adjusted model that considered particulate matter, NDVI quantiles, and interaction term of particulate matter and NDVI quantiles, adjusted for year, longitude, latitude, interaction term of longitude and latitude, age, sex, current smoking, current drinking, education attainment, marital status, social contact, self-reported health status, exercise, private medical insurance, density of population quintiles, number of beds in hospitals per 1,000 persons, number of national basic livelihood beneficiaries, independent rate of finance of local government, proportion of basic pension beneficiaries.
Thus, the stratified group analyses (based on sex, age and exercise) were conducted, revealing that women (Park et al., Reference Park, Kang, Team and Kim2024; Zhang et al., Reference Zhang, Zhao, Hong, Chang, Ryu, Kang, Monteiro, Shin, Guallar and Cho2019), elderly individuals (Liu et al., Reference Liu, Wang, Gu, Deng, Wang, Lin, Guo and Wu2021), and those with low exercise levels (Zhang et al., Reference Zhang, Zhao, Hong, Chang, Ryu, Kang, Monteiro, Shin, Guallar and Cho2019) are particularly susceptible to the impacts of particulate matter-induced depression, as previously demonstrated. Further stratified analyses were performed to explore changes across socioeconomic status, considering sex, age group, exercise, current smoking, current drinking, education attainment, marital status, social contact, self-reported health status, and private medical insurance.
Sensitivity analysis was conducted to assess the robustness of the findings. First, we conducted a generalized linear mixed model with a binomial distribution and logit link, mirroring the specifications of the GEE model, while incorporating the random intercept of each subject. Second, we treated depression as a continuous scale rather than a binary one, employing a linear mixed model to investigate the relationship between air pollutants and the continuous CES-D 10 scale. Third, this involved evaluating EVI as an alternative representation of greenness in the models and investigating the effects of a 6-month exposure to air pollution. Moreover, supplementary exposure models were included for other air pollutants (NO2 and O3) and analysed separately from the main models to assess whether particulate matter demonstrates an impact even in the presence of other air pollutants (Borroni et al., Reference Borroni, Pesatori, Bollati, Buoli and Carugno2022; Lim et al., Reference Lim, Kim, Kim, Bae, Park and Hong2012). Variance inflation factors (VIFs) were assessed to detect multicollinearity among variables, with a threshold of VIFs <2.5 indicating acceptable levels.
All statistical analyses were performed using SAS version 9.4 (SAS Institute Inc.) and R version 4.1.3. The significance level was set at p <0.05 for all analyses.
Results
Table 1 shows the descriptive characteristics of the variables in the baseline, the first wave of each participant, by the quantiles of NDVI. The number of participants was 7657, 1953, 1947, 1687 and 2070 in total, quantile 1–4, respectively. The mean score of CES-D 10 was 16.20 (SD: 5.24). Overall 1700 (22.20%) participants in the sample had a depressive symptom in the baseline. Table 2 shows the descriptive and correlation analysis table of air pollution and greenness levels (see also supplementary Table S1). Almost two thirds of Korea’s topography is composed of mountains (Choung et al., Reference Choung, Lee, Cho, Lee, Jang, Kim, Hong, Jung and Choung2004), and the result is reflected in the NDVI figures (mean: 0.50, SD: 0.11, Supplementary Figure S2, Supplementary Table S2).
Table 1. Descriptive characteristics of the variables in the baseline of participants (Mean (SD) or proportion)
Note: Table 1 presents descriptive statistics of the study participants, including proportions, means and standard deviations, categorized by the baseline year (i.e., the year when participants initially entered the study, ranging from 2016 to 2020). ANOVA or chi-squared test p-value was presented to compare quantile groups of NDVI.
Abbreviation: CES-D, Center for Epidemiology Studies of Depression scale; PM10, particulate matter of 10 microns in diameter or smaller; PM2.5, particulate matter of 2.5 microns in diameter or smaller; NDVI, Normalized Difference Vegetation Index; SD, standard deviation.
Table 2. Descriptive and correlation table of air pollution and greenness averaged between 2016, 2018 and 2020
PM10, particulate matter of 10 microns in diameter or smaller; PM2.5, particulate matter of 2.5 microns in diameter or smaller; NDVI, Normalized Difference Vegetation Index; EVI, Enhanced Vegetation Index; SD, standard deviation.
*** p < 0.001
Figure 1 shows the minimally or fully adjusted GEE model results of air pollution effects by NDVI quantile level on CES-D 10 score. The ORs are the ORs of air pollution on depressive symptoms considering the NDVI quantile effect. In the fully adjusted model, the ORs for depression with a 10 
${\mu\text{ g/}}{{\text{m}}^{\text{3}}}$ increment of PM10 were 1.88 (95% CI: 1.58, 2.25) in NDVI Q1 and 1.29 (95% CI: 1.06, 1.58) in NDVI Q4 (p for interaction < .0001). With a 10 
${\mu\text{ g/}}{{\text{m}}^{\text{3}}}$ increment of PM2.5, the ORs for depression were 3.75 (95% CI: 2.75, 5.10) in NDVI Q1 and 1.78 (95% CI: 1.30, 2.44) in NDVI Q4 (p for interaction < .0001). The ORs in the minimally adjusted generalized estimation equation model were consistent with the ORs in the fully adjusted model. The ORs for depression showed a positive association with PM10 and PM2.5 over all NDVI quantiles, but compared to the ORs in NDVI Q1, the ORs significantly decreased in NDVI Q4.
Figure 1. Associations between air pollution exposure (per 10 
${\mu\text{ g/}}{{\text{m}}^{\text{3}}}$ increase) and depression by NDVI quantiles.
Several sensitivity analyses were also performed. The generalized linear mixed models and linear mixed models were shown in Supplementary Table S3–4. Thus, instead of NDVI, EVI was added as a green effect and analysed with PM10 (OR: 1.92 [95% CI: 1.61, 2.29] in EVI Q1, 1.25 [1.02, 1.54] in EVI Q4, p for interaction < .0001) and PM2.5 (4.02 [2.96, 5.47] in EVI Q1, 1.53 [1.11, 2.12] in EVI Q4, p for interaction < .0001) (Supplementary Table S5). These findings indicate that the OR for particulate matter in EVI Q4 was lower than that in EVI Q1, a pattern consistent with the results of the main model depicted in Fig. 1. One year exposure to air pollution was performed in the main model, we also performed 6 months exposure to air pollution (Supplementary Table S6). Both PM10 and PM2.5 of the models showed the odd ratios in NDVI Q4 were significantly decreased compared to NDVI Q1. Additionally, the observed relationships between particulate matter and NDVI interaction with depression remained consistent even after controlling for other pollutants such as NO2 and O3, as demonstrated by the two-pollutant models (Supplementary Table S7).
Figure 2 shows the results of stratified analyses showing the estimated association (OR) between air pollution and depression by NDVI quantiles among separate subgroups such as sex, age group and exercise. In NDVI Q1, the ORs for depression per 10 
${\mu\text{ g/}}{{\text{m}}^{\text{3}}}$ increase in PM10 were higher among the stratified groups compared to the ORs in NDVI Q4. The results of PM2.5 were consistent with those of PM10, and additional stratified analysis results are provided in Supplementary Table S8.
Figure 2. (a) Subgroup analysis of associations between air pollution (PM10) exposure (per 10 increase) and depression by NDVI quantiles. (b) Subgroup analysis of associations between air pollution (PM2.5) exposure (per 10 
${\mu \text{g/}}{{\text{m}}^{\text{3}}}$ increase) and depression by NDVI quantiles.
Discussion
The association between greenness (NDVI) and air pollution in the KLoSA has been found to significantly alleviate the negative impacts of air pollution on depression. This finding particularly holds true for the highest quantile (Q4) of NDVI compared to the lowest quantile (Q1). PM2.5 exhibited higher ORs compared to PM10. Subgroup analysis results demonstrated beneficial effects of greenness in the relationship between particulate matter and depression.
This study examines the association of greenness with air pollution on depression, employing the CES-D 10 questionnaire as an outcome measure. The findings of this study align with previous studies that also employed a questionnaire or utilized antidepressant use as an outcome measure. The Yinzhou Cohort, a Chinese study that followed participants for five years, revealed decreased hazard ratios for depression per interquartile range increase in NDVI after adjusting for PM2.5 and PM10, respectively (Zhang et al., Reference Zhang, Wei, Yu, Guo, Wang, Jin, Shui, Lin, Tang and Chen2022). Another study, a cross-sectional study based on the Dutch national health survey, demonstrated a significant association between greenness, antidepressant use, and PM2.5 exposure. The ORs for NDVI exposure and PM2.5 exposure were 0.96 (95% CI: 0.94, 0.98) and 1.01 (95% CI: 0.99, 1.03), respectively in the adjusted model accounting for two exposures (Klompmaker et al., Reference Klompmaker, Hoek, Bloemsma, Wijga, van den Brink, Brunekreef, Lebret, Gehring and Janssen2019). A study conducted in Guangzhou, utilizing a cross-sectional survey design, discovered a beneficial association between green spaces, such as street trees, and psychological well-being, which was mediated by PM2.5 levels (Wang et al., Reference Wang, Yang, Yao, Bloom, Feng, Yuan, Zhang, Liu, Wu and Lu2020). A systematic review was conducted to assess the impact of residential greenness on the relationship between air pollution and various health outcomes, including mental health. However, this review found only a limited number of studies available for direct comparison with the present study (Son et al., Reference Son, Choi, Fong, Heo, Lim and Bell2021).
Several studies have explored the relationship between greenness and depression, yielding noteworthy findings. In a cross-sectional study conducted in the United Kingdom, participants aged 37–73 years demonstrated an OR of 0.96 (95% CI: 0.93, 0.99) for depression, as assessed using a questionnaire on probable major depressive disorder experiences, in relation to NDVI (Sarkar et al., Reference Sarkar, Webster and Gallacher2018). An American Nurses’ Health Study, which followed a cohort of individuals aged 54–91 years, revealed that those residing in areas with the highest quintile of greenness within a 250-meter radius exhibited a lower risk of depression compared to those in the lowest quintile (HR: 0.87, 95% CI: 0.78, 0.98) (Banay et al., Reference Banay, James, Hart, Kubzansky, Spiegelman, Okereke, Spengler and Laden2019). A study conducted in Henan Province, China, employing the Patient Health Questionnaire-2, indicated a negative association between an interquartile range increase in NDVI and depression in a mixed model adjusted for covariates (−0.024, 95% CI: −0.041, −0.006) (Di et al., Reference Di, Li, Xiang, Xie, Mao, Hou, Liu, Huo, Yang and Dong2020). However, contrary to previous findings, some studies conducted in the United States did not establish a significant association between greenness and depression (Pun et al., Reference Pun, Manjourides and Suh2018).
The reduction of particulate matter within green spaces has been proposed as a physical mechanism (Choi et al., Reference Choi, Yoo, Yeo and Park2022), involving processes such as adsorption, absorption, blocking and sedimentation of particulate matter. The microstructure of trees provides an ideal surface for the attachment of particulate matter (Sgrigna et al., Reference Sgrigna, Sæbø, Gawronski, Popek and Calfapietra2015), while the leaves of tress have the capacity to absorb particulate matter (Liu et al., Reference Liu, Zhu, Wang, Yang, Liu, Qiu, Ma, Zhang and Liu2016). The presence of green space creates turbulence in the surrounding air, which can lead to the reduction or obstruction of particulate matter (Bernatzky, Reference Bernatzky1983; Jin et al., Reference Jin, Guo, Wheeler, Kan and Che2014; Vesala et al., Reference Vesala, Suni, Rannik, Keronen, Markkanen, Sevanto, Grönholm, Smolander, Kulmala and Ilvesniemi2005). Particulate matter tends to settle on the leaves, stems, and branches of tress, highlighting the effectiveness of green spaces in reducing particulate matter levels (Janhäll, Reference Janhäll2015). Beyond minimizing harm, greenness has been associated with various protective mechanisms that contribute to mental health. These mechanisms include the restoration and enhancement of cognitive capacities (Markevych et al., Reference Markevych, Schoierer, Hartig, Chudnovsky, Hystad, Dzhambov, De Vries, Triguero-Mas, Brauer and Nieuwenhuijsen2017), and an increase in physical activity levels (James et al., Reference James, Banay, Hart and Laden2015). The biophila hypothesis, developed by Wilson, proposes a natural affinity between plants and people (White and Heerwagen, Reference White, Heerwagen and Lundberg1998; Wilson, Reference Wilson1984), and it has been frequently mentioned in studies investigating the relationship between greenness and mental health. The availability of green spaces has the potential to foster social cohesiveness and encourage social interactions through recreational or regular physical activities, which are strongly linked to mental well-being (Berkman et al., Reference Berkman, Kawachi and Glymour2014).
The group-specific analysis results uniformly indicated a consistent beneficial role of greenness in the association between particulate matter and depression. Previous studies into the association between greenness and depression have yielded significant that were sex-specific, appearing exclusively among males (Di et al., Reference Di, Li, Xiang, Xie, Mao, Hou, Liu, Huo, Yang and Dong2020) or exclusively among the females (De Vries et al., Reference De Vries, Verheij, Groenewegen and Spreeuwenberg2003; Sarkar et al., Reference Sarkar, Webster and Gallacher2018; Triguero-Mas et al., Reference Triguero-Mas, Dadvand, Cirach, Martínez, Medina, Mompart, Basagaña, Gražulevičienė and Nieuwenhuijsen2015), as well as age-specific, with effects observed among participants aged over 65 years (Sarkar et al., Reference Sarkar, Webster and Gallacher2018) or those below 65 years (Di et al., Reference Di, Li, Xiang, Xie, Mao, Hou, Liu, Huo, Yang and Dong2020; Wang et al., Reference Wang, Meng, Lam and Ponce2019a).
Our study has several limitations. First, the assessment of people’s depression was reliant on self-questionnaires rather than utilizing formal diagnoses from medical professionals, which could lead to outcome misclassification. However, it is crucial to highlight that the CES-D 10, the depression score employed in our study, has been extensively validated and demonstrated strong reliability in previous research studies (Boey, Reference Boey1999; Cole et al., Reference Cole, Rabin, Smith and Kaufman2004; Malakouti et al., Reference Malakouti, Pachana, Naji, Kahani and Saeedkhani2015). Secondly, apart from the inclusion of air pollutants such as NO2, CO and O3, our analysis did not account for other conceivable environmental factors, such as noise pollution, which have the potential to influence the manifestation of depression (van den Bosch and Meyer-Lindenberg, Reference van den Bosch and Meyer-Lindenberg2019). Thirdly, it is important to acknowledge that people’s exposure to air pollution was estimated on a district (si/gun/gu) basis rather than an individual basis. We made the assumption that people residing within the same district would experience comparable levels of air pollution concentration and greenness level. Despite being second-level local authority areas within metropolitan cities and provinces in South Korea, districts are characterized by a median area size of approximately 397 km2, which is roughly 1.7 times larger than a ZIP code in the United States (233 km2). Consequently, we hypothesize that the effects of air pollution can be adequately captured in nationwide-scale environmental epidemiological studies (Di et al., Reference Di, Wang, Zanobetti, Wang, Koutrakis, Choirat, Dominici and Schwartz2017; Lee et al., Reference Lee, Seo, Myung, Prifti, Kang, Jang, Park, Bell and Kim2021; Park et al., Reference Park, Kang, Min, Kim, Song, Jang, Kwon, Oh, Moon and Kim2023), despite the challenge of accurately reflecting individual-level exposures. However, this approximation may oversimplify the actual exposure variations experienced by people within the district boundaries. These limitations underscore the need for further research endeavours to address the aforementioned gaps and enhance the precision and comprehensiveness of our understanding of the complicated relationship between air pollution, environmental factors and depression.
Above these limitations, our study draws upon a high-quality longitudinal study conducted in Korea, characterized by extensive national coverage, specifically focusing on the elderly population across diverse regions. This study represents the pioneering investigation in Korea examining the association between greenness, particulate matters and depression in older individuals, while effectively adjusting for socioeconomic characteristics, social relationships and health-related variables. Given Korea’s rapid progression towards a super-aging society, the findings of this study hold substantial utility for addressing depressive conditions in the elderly population (Park et al., Reference Park, Yang, Ha and Lee2014). This study could contribute to enhancing our comprehension of depression in elderly Koreans by elucidating the environmental factors that can either protect or exert detrimental effects. The findings can be instrumental in formulating policy implications for the betterment of this population. Furthermore, in addition to utilizing NDVI, this study also incorporates EVI as a measure of residential greenness in the supplementary results. The utilization of EVI, a vegetation index that accounts for seasonal and interannual variations in vegetation production with greater precision, further enhances the methodological robustness of the study (Leng et al., Reference Leng, Huete, Cleverly, Yu, Zhang and Wang2022).
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S2045796024000684.
Availability of data and materials
The data used in this study are not publicly available due to restrictions imposed by the Korea Employment Information Service (KEIS) and AiMS-CREATE, from where the data were obtained under license. However, the authors can provide the data upon reasonable request and with permission from KEIS and AiMS-CREATE.
Acknowledgements
The KloSA used in this study were gathered from the Korea Employment Information Service (KEIS) conducted by the Korea Ministry of Employment and Labor.
Financial support
This work was supported by Korea Environment Industry & Technology Institute (KEITI) through ‘Climate Change R&D Project for New Climate Regime’, funded by Korea Ministry of Environment (MOE) (2022003570006).
Competing interests
None.
Ethical standards
The Institutional Review Board of Seoul National University granted ethics approval for the study protocol and data analysis (IRB No. E2212/002-001). Informed consent was not required because a prospectively anonymized database was employed.
Open access
