Source-Context Differences in Particulate Matter Removal Dynamics of Urban Forests: Evidence from Two-Year Field Measurements

Abstract

Urban forests (UFs) are increasingly promoted as a nature-based solution for mitigating particulate matter (PM) pollution, yet their removal performance can vary depending on surrounding emission sources and environmental conditions. Here, we quantified the particulate matter reduction efficiency (PMRE) of UFs located near roads, industrial complexes, and urban areas, together with background forests in South Korea, based on field observations during the late autumn–spring period across two consecutive years (November–May in 2021–2022 and 2022–2023). We applied vector autoregression (VAR) to examine the dynamic relationships between PMRE and meteorological and air pollutant variables across eight representative sites. The results revealed that PM mitigation dynamics were strongly particle-size-dependent and context-specific. Across all sites, ΔPM₁₀ RE was predominantly self-driven, explaining over 90% of its own variance, whereas fine-particle dynamics showed stronger interdependence. In particular, ΔPM_2.5 RE consistently acted as a key mediator, accounting for up to 70%–80% of the variation in ΔPM_1.0 RE depending on source context. Industrial-complex-adjacent UFs exhibited the strongest cross-variable interactions, while urban-core UFs were largely governed by intrinsic mitigation processes. Roadside UFs showed site-specific responses associated with CO and temperature variability. Notably, PMRE responses exhibited damped oscillation patterns across all source contexts, converging toward equilibrium over time, indicating stabilization of mitigation performance following disturbance events. These findings demonstrate that urban forest air-quality benefits are highly context dependent and governed by particle-size-specific dynamics. Our results provide evidence-based guidance for designing and managing urban forests, emphasizing the need for source-specific strategies and prioritization of PM_2.5-oriented mitigation, particularly in industrial and roadside environments where fine-particle interactions are strongest.

1. Introduction

Particulate matter (PM) pollution remains a persistent environmental and public health challenge in urban areas, particularly during recurrent high-concentration episodes. Fine particles, such as PM₁₀ and PM_2.5, can penetrate the human respiratory system and are associated with severe health outcomes, including cerebrovascular and cardiovascular diseases, lung cancer, and other respiratory complications [1,2]. Smaller particles also exhibit higher toxicity due to their larger surface area and stronger capacity to adsorb hazardous substances, increasing their retention time in the human body [3]. Although policy efforts have contributed to long-term improvements in air quality, a substantial share of urban air pollution continues to originate from major and persistent emission sources, highlighting the need for effective, practical mitigation strategies [4].

Urban forests (UFs) are widely implemented as nature-based solutions to address urban environmental and social challenges, including air quality improvement. Trees mitigate PM primarily through dry and wet deposition processes. During dry deposition, airborne particles accumulate on leaf surfaces depending on morphological traits (e.g., leaf structure) and physicochemical properties (e.g., wax layer composition), while deposited particles may be resuspended or removed through wind, leaf fall, or meteorological variability. Wet deposition further removes accumulated particles through precipitation wash-off [5,6,7,8,9,10]. Importantly, however, the PM mitigation performance of UFs is not uniform and can vary with vegetation characteristics, seasonality, meteorological conditions, and the surrounding emission environment.

Previous studies have examined PM mitigation by UFs across diverse geographic contexts, including roadside, industrial, and urban environments [11,12,13,14], and have further explored the influence of meteorology [15,16] and distance from forests [17,18]. These studies consistently indicate that mitigation capacity is shaped by multiple interacting factors. In cold-season conditions, for example, the removal of fine PM can be reduced in deciduous systems due to seasonal foliage loss, thereby changing deposition capacity and pollutant dynamics. Moreover, the composition and quantity of air pollutants differ across emission sources. Industrial activities commonly emit PM alongside sulfur dioxide (SO₂), nitrogen oxides (NO_X), and carbon monoxide (CO), and may also release heavy metals and other hazardous compounds depending on processes and materials [19]. Road traffic emissions arise from exhaust, tire wear, and road abrasion, contributing to VOCs, PM, NO_X, and CO [20], while dense urban areas also emit complex mixtures associated with residential, commercial, and transportation activities [21]. Such source-dependent differences suggest that the effectiveness of UFs in mitigating PM may vary substantially across land-use types and emission contexts. Recent observational studies have further advanced the understanding of urban forest contributions to particulate matter mitigation under real-world conditions. For example, Seo et al. [22] demonstrated that urban forests tend to exhibit higher removal efficiency for coarse particles than fine particles, with performance varying depending on seasonal and structural characteristics. These findings highlight the importance of considering particle-size-specific dynamics and contextual factors when evaluating urban forest mitigation performance.

Despite decades of research, current evidence remains limited in its ability to guide context-specific planning and management of UFs across diverse urban environments. Many studies have been restricted to specific regions or seasons, constraining the generalization of findings to recurrent high-PM episodes and contrasting emission settings. In addition, PM mitigation has often been assessed primarily through concentration changes, while fewer studies have examined the dynamic interactions between PM mitigation performance, co-pollutants, and meteorological drivers across different emission-source contexts. A deeper understanding of these interactions is essential for developing tailored UF strategies that maximize air-quality co-benefits near major pollution sources, such as roads and industrial complexes, while improving overall urban exposure conditions. Understanding time-lagged and cross-fraction responses is critical because urban forest management operates through repeated disturbances (pollution episodes, weather variability), and mitigation effectiveness must be interpreted as a dynamic service rather than a static concentration difference.

Therefore, in this study, we selected eight UF sites representing roadside, industrial-complex-adjacent, dense urban-core environments, along with background forests in South Korea, where fine PM concentrations are typically elevated from late autumn to spring. Using PM reduction efficiency (PMRE) as an indicator of UF mitigation performance during high-pollution periods, we investigated (1) the key influencing factors of PMRE across contrasting emission-source contexts, and (2) the dynamic relationships among PMRE, meteorological variables, and air pollutant variables, highlighting source-context-specific interactions relevant to urban forest design and management.

This study contributes to the current understanding of urban forest PM mitigation in several important ways. First, it applies a multivariate time-series framework (VAR) to examine dynamic interactions among particulate matter reduction efficiency, meteorological variables, and co-pollutants, moving beyond conventional static approaches. Second, it provides empirical evidence that PM mitigation dynamics are strongly particle-size dependent, with PM_2.5 acting as a central mediator linking coarse and ultrafine particle responses across emission-source contexts. Third, it offers context-specific insights for urban forest design and management by showing that mitigation performance varies systematically with surrounding emission environments.

2. Materials and Methods

2.1. Monitoring Network and Study Design

The National Institute of Forest Science established the Asian Initiative for Clean Air Networks (AICAN) to investigate the role of forests and urban green spaces in mitigating particulate matter (PM) pollution across South Korea. Between 2019 and 2023, AICAN strategically selected and operated 44 monitoring sites located near major pollution sources, enabling long-term observations of PM concentrations, meteorological conditions, and selected co-pollutants.

Site selection was conducted in two steps. First, candidate urban forests (UFs) were categorized according to four construction-background types. Second, candidate locations were evaluated using three-year meteorological data, atmospheric modeling, field surveys, and on-site evaluations to finalize monitoring locations. This network design provides a basis for comparing UF mitigation performance under contrasting emission environments.

For the present study, we selected study sites based on the following criteria:

(1)

availability of at least three years of long-term data, and

(2)

inclusion of at least two sites per emission-source context.

Accordingly, we analyzed eight sites representing four source contexts: roadside, industrial-complex-adjacent, dense urban-core, and background forest (control/reference) environments.

The selected sites were chosen to represent distinct emission-source contexts within the AICAN monitoring network, including roadside, industrial-complex-adjacent, urban-core, and background forest environments. Site classification was based on proximity to dominant emission sources and surrounding land-use characteristics, ensuring that each site reflects the typical environmental conditions of its respective context.

From the full set of monitoring sites, only those with sufficient data completeness and consistency were considered. Specifically, sites were required to have at least three years of continuous observations and complete datasets for particulate matter, meteorological variables, and co-pollutants. Several candidate sites were excluded due to incomplete time-series data, shorter monitoring periods, or inconsistencies in key variables, which could compromise the reliability of the VAR-based analysis.

2.2. Data Scope and Temporal Coverage

AICAN continuously measures PM concentrations and provides near-real-time information on air quality in forests and urban green spaces. For this study, UF monitoring data were acquired through the AICAN platform (publicly available via the national Open Data Portal). Pollution-source and ambient air-quality data were obtained from multiple sources depending on site context. Roadside and industrial-complex data were obtained from AICAN, whereas urban and background forest data were obtained from Air Korea (the national air-quality monitoring network). Meteorological data (temperature, pressure, humidity, wind speed, and wind direction) were obtained from the Korea Meteorological Administration (KMA) using the nearest station to each site.

All datasets were preprocessed and integrated based on the most recent period for which harmonized PM, meteorological, and co-pollutant observations were consistently available across all selected sites. We prioritized temporal completeness, cross-site comparability, and data quality control because the VAR framework requires a common and internally consistent time series across source contexts. The final dataset therefore covered two cold-season periods characterized by elevated PM concentrations in South Korea: November 2021–May 2022 and November 2022–May 2023. All datasets were processed at a daily resolution following temporal aggregation of high-frequency measurements to ensure consistency with the event-based analytical framework and comparability across sites.

Within the AICAN monitoring network, all variables were recorded at a 10 min temporal resolution, including particulate matter, meteorological parameters, and co-pollutants. These high-frequency measurements were subsequently aggregated into daily averages to ensure consistency across datasets and to align with the event-based analytical framework. This aggregation reduces short-term variability and measurement noise while preserving the overall temporal patterns relevant for the VAR-based time-series analysis.

The dataset includes particulate matter size fractions (PM₁₀, PM_2.5, and PM_1.0), meteorological variables (temperature, humidity, wind speed, and wind direction), and co-pollutants (SO₂, NO₂, CO, and O₃), allowing for a comprehensive assessment of UF–atmosphere interactions.

2.3. Study Sites and Source Contexts

Two roadside sites (Gomae and Yangjae) were located near the Gyeongbu Expressway and represented high-traffic environments, with an average traffic volume of approximately 192,000 vehicles per day [23]. Two industrial-complex-adjacent sites (Sihwa and Seongnam) were selected to represent industrial emission environments with contrasting surrounding land-use contexts. Two urban-core sites (Hongneung and Namsan) were located in Seoul, representing dense urban environments influenced by multiple local emission sources. Finally, two background forest sites (Pyeongchang and Hoengseong) were located in the eastern region of South Korea, where wintertime northwesterly winds and mountain ranges reduce pollutant transport from the western region [24]. These background sites were used to characterize reference atmospheric conditions in forested environments (Figure 1,

Table 1. Characteristics of study sites.

Division	Region	Location	Source Name (Control)	Urban Forest Name (Treatment)	Forest Type
Roadside	Gomae	Gomae Experimental Forest, Giheung-gu, Yongin-si, Gyeonggi-do	Gomae	Gomae	Deciduous
	Yangjae	Maehyeon Citizens’ Forest, Seocho-gu, Seoul	Yangjae	Yangjae	Mixed
Industrial complex	Sihwa	Sihwa Industrial Complex, Jeongwang-dong, Siheung-si, Gyeonggi-do	Sihwa Industrial	Sihwa blocking forest	Mixed
	Seongnam	Seongnam Industrial Complex, Seo-gu, Incheon	Seongnam Industrial	Seongnam buffer forest	Coniferous
Urban	Namsan	Namsan, Yejang-dong, Jung-gu, Seoul	Seoul station	Namsan	Coniferous
	Hongneung	Hongneung Forest, Hawolgok-dong, Seongbuk-gu, Seoul	Near office	Hongneung forest	Mixed
Background Forest	Pyeongchang	Jungwangsan Mountain, Jinbu-myeon, Pyeongchang-gun, Gangwon-do	Jeongseon Town	Pyeongchang	Mixed
	Hoengseong	Cheongtaesan Mountain, Dunnae-myeon, Hoengseong-gun, Gangwon-do	Pyeongchang town	Hoengseong	Mixed

).

2.4. Data Preprocessing and Event Selection

To analyze UF–pollution interactions during high-PM episodes, we conducted the following preprocessing steps:

(a)

Integration of multi-source datasets. For roadside and industrial-complex contexts, variables from AICAN (site information, observation time, and meteorological parameters) were integrated with Air Korea co-pollutant data (SO₂, NO₂, CO, and O₃). For urban and background forest contexts, datasets from AICAN, Air Korea, and KMA were integrated.

(b)

Wind-direction classification. Wind direction was used to distinguish periods when UFs were more likely to be influenced by their adjacent pollution sources. Each UF site was assigned a wind-direction range based on the relative location of nearby sources. For example, Gomae was assigned 180–360°, Sihwa 135–315°, and Seongnam 225–45° across north (i.e., 225–360° and 0–45°). In contrast, for urban-core sites (Namsan and Hongneung), all wind directions were included due to multi-source influences. Wind sectors were determined based on the relative geometry between the UF plot and the dominant upwind emission source to maximize source influence while minimizing mixed-source conditions.

(c)

High-PM episode definition. We focused on the cold-season period (November–May) when PM concentrations are typically elevated in South Korea. High-pollution events were defined based on the national episode classification thresholds: a day was classified as a high-PM event when either PM₁₀ exceeded 51 µg m⁻³ or PM_2.5 exceeded 36 µg m⁻³ (Ministry of Environment, South Korea). Extracted datasets were summarized at a daily resolution for subsequent analysis. These thresholds are substantially higher than the World Health Organization (WHO) air quality guideline for PM_2.5 (15 µg m⁻³, 24 h mean), indicating that the analyzed periods represent severe pollution conditions in a global context.

These preprocessing steps were designed to isolate the influence of dominant emission sources while ensuring temporal consistency across datasets. In particular, wind-direction filtering was applied to maximize source-specific signals and minimize mixed-source interference, thereby improving the interpretability of subsequent time-series analyses.

2.5. Variable Definition

To identify drivers of UF mitigation performance during high-pollution periods, the dependent variables were defined as the reduction efficiency of PM size fractions (PM₁₀, PM_2.5, and PM_1.0). Meteorological variables and ambient co-pollutant concentrations (SO₂, NO₂, CO, and O₃) were included as explanatory variables.

$$\begin{array}{c}\mathrm{Removal} \mathrm{efficiency} \mathrm{by} \mathrm{PM} \mathrm{type} \left(\mathrm{PMRE}\right)\hfill \\ =\left(\right(\mathrm{Control} \mathrm{PM} - \mathrm{Treatment} \mathrm{PM})/\mathrm{Control} \mathrm{PM})\times 100\hfill \end{array}$$

(1)

This formulation allows a relative comparison of PM reduction performance between urban forest (treatment) and pollution-source (control) sites, thereby minimizing site-specific baseline differences and enabling consistent cross-site comparison across different emission contexts (

Table 2. Used variable description in this study.

Variable Name	Description
PM RE	Remove efficiency by PM type ((Control − Treatment PM)/Control PM) × 100
d_Hum	Difference between Control and Treatment area
c/tTemp	Temperature of Control/Treatment
c/t WS	Wind speed of Control/Treatment
c/t Hum	Humidity of Control/Treatment
SO2, NO2, CO, O3	Air pollution at Air Korea stations near area

).

2.6. Time-Series Analysis Using Vector Autoregression

PM concentrations exhibit temporal dependencies and cross-variable interactions, particularly because PM_2.5 is a subset of PM₁₀ and PM_1.0 is a subset of PM_2.5. To account for particle-size interdependence, time-lagged responses, and multivariate interactions among PM reduction efficiency, meteorology, and co-pollutants, we employed a vector autoregression (VAR) framework [25].

We used a VAR(p) model of the form:

$$y_t=C+A_1{y}_{t-1}+\cdots +A_p{y}_{t-p}+{\epsilon }_t$$

(2)

where $y_t$ denotes the vector of dependent and explanatory variables at time $t$; $C$ is a constant vector; $A_i$ are coefficient matrices; and ${\epsilon }_t$ is the error term. We interpreted dynamic relationships using impulse response functions (IRFs) and forecast error variance decomposition (FEVD), which quantify how shocks to one variable propagate through the system over time and the relative contributions of each variable to variance in PM reduction efficiency.

The VAR framework was adopted because it enables the joint modeling of multivariate time-series data while capturing interdependencies, feedback mechanisms, and time-lagged effects among variables. This is particularly appropriate for urban air quality systems, where PM fractions, meteorological conditions, and co-pollutants interact dynamically rather than operating in isolation. The robustness of the impulse response functions was assessed through standard VAR diagnostics, including lag-length selection based on information criteria (AIC/BIC), stationarity testing, and residual analysis. These procedures ensure the stability and reliability of the estimated dynamic responses.

2.7. Model Specification and Analytical Workflow

Before fitting VAR models, we conducted standard diagnostic and specification procedures:

(1)

PM reduction efficiency generation and normalization. Difference variables for PM reduction efficiency and meteorological factors were generated between pollution-source (control) and UF (treatment) sites. Variables were normalized to a common scale using min–max normalization.

(2)

Outlier detection. Multivariate outliers were identified using Mahalanobis distance for joint distributions of PM variables. Normality was evaluated using the Jarque–Bera test.

(3)

Stationarity and differencing. Unit-root tests were performed, and first differences were applied to dependent variables to improve stability and ensure consistent interpretation across sites.

(4)

Granger causality testing. Granger causality tests were conducted to assess temporal dependencies and inform variable ordering.

(5)

Lag selection and model fitting. Lag length was selected based on information criteria (AIC and BIC).

(6)

Cointegration testing. Cointegration tests were conducted to examine potential long-term relationships among variables.

(7)

IRF and FEVD analyses. Impulse responses and variance decompositions were used to interpret dynamic interactions among variables and identify key drivers of PM reduction efficiency across source contexts.

These procedures were implemented sequentially as an integrated analytical pipeline to ensure model stability, reduce statistical bias, and enhance the interpretability of dynamic relationships. The stepwise framework also improves reproducibility by providing a transparent linkage between data preprocessing, model specification, and result interpretation (Figure 2).

3. Results

3.1. VAR Model Specification and Diagnostics

Prior to fitting the VAR models, we evaluated time-series properties using unit-root testing, cointegration analysis, Granger causality testing, and lag-length selection. Unit-root tests indicated that some variables were non-stationary in several regions; therefore, first-order differencing was applied to all variables to ensure stationarity and comparability across sites. The Johansen cointegration test confirmed cointegration relationships in all study regions, suggesting a long-term equilibrium relationship between UF PM reduction efficiency (PMRE) and explanatory variables. Although cointegration typically motivates the use of a vector error correction model, a standard VAR framework was adopted here to avoid overfitting, particularly in regions with limited sample sizes.

Granger causality results were used to support variable ordering (

Table A1. Granger casual test results on each site. Null hypothesis: A particular variable does not have Granger causality with the rest of the set of variables except for it.

Region	Variable	F-Test	p-Value	Region	Variable	F-Test	p-Value
Gomae	ΔcTemp	1.6082	0.09838	Yangjae	ΔcTemp	1.8826	0.04334
	ΔcHum	1.8938	0.04185		ΔPM2.5 RE	1.6259	0.09353
	ΔPM10 RE	2.5119	0.005356		ΔPM1.0 RE	2.0509	0.02541
	ΔPM2.5 RE	1.702	0.07492		ΔSO2	1.7529	0.06439
Sihwa	ΔcTemp	1.866	0.04561	Seongnam	ΔcTemp	2.4331	0.008086
	ΔNO2	1.9274	0.03766		ΔcWS	2.4054	0.00886
	-	-	-		ΔcHum	3.1152	0.0007826
					ΔO3	3.5564	0.0001622
					ΔNO2	2.3394	0.011
Hongneung	-	-	-	Namsan	ΔcTemp	2.2455	0.0171
					ΔPM10 RE	1.8242	0.05946
					ΔNO2	1.9569	0.04065
Pyeongchang	Δd_Hum	2.9608	0.00166	Hoengseong	ΔPM10 RE	1.7498	0.07315
	ΔcWS	5.4534	1.89 × 10−7		ΔcWS	2.0402	0.03177
	ΔcHum	5.1878	5.20 × 10−7		ΔCO	1.8595	0.05374

). The optimal lag length was primarily determined using the Schwarz criterion (SC), supported by residual diagnostics based on the autocorrelation and partial autocorrelation functions (

Table A2. The selection of VAR model Time lag length for each region.

Region	AIC (n)	HQ (n)	SC (n)	FPE (n)	Selection of Time Lag Length
Gomae	10	1	1	3	1
Yangjae	10	1	1	2	1
Sihwa	10	1	1	3	1
Seongnam	3	3	3	3	1
Namsan	5	1	1	5	1
Hongneung	8	8	1	8	1
Pyeongchang	4	2	1	4	1
Hoengseong	2	1	1	2	1

). While SC suggested a lag of 3 for Seongnam, residual diagnostics indicated that autocorrelation was not persistent beyond lag 1; therefore, a lag length of 1 was consistently applied across all regions. In addition, a trend-included VAR specification was used to account for underlying trends in the air quality and meteorological datasets.

3.2. Source-Context Differences in PMRE Dynamics (Impulse Response Functions)

Impulse response functions (IRFs) were used to quantify how PMRE variables responded over time to shocks in PMRE itself, meteorological factors, and co-pollutants. Specifically, we examined the temporal evolution of ΔPM₁₀ RE, ΔPM_2.5 RE, and ΔPM_1.0 RE in response to a one-unit shock in impulse variables at the previous time step (t−1), including ΔPM₁₀ RE, ΔPM_2.5 RE, ΔPM_1.0 RE, ΔdHum, ΔcTemp, ΔcHum, ΔcWS, ΔO₃, ΔSO₂, ΔCO, and ΔNO₂. Shocks were defined as one-unit increases corresponding to the square root of the diagonal elements of the forecast error variance matrix (

Table A3. The square root value of forecast error.

	ΔPM10 RE	ΔPM2.5 RE	ΔPM1.0 RE	ΔcWS	ΔcTemp	ΔcHum	Δd_Hum	ΔO3	ΔNO2	ΔSO2	ΔCO
Gomae	0.08	0.09	0.1	0.17	0.11	0.19	0.17	0.15	0.21	0.14	0.16
Yangjae	4.53	4.36	4.6	1.71	3.12	12.32	2.37	8.57	12.51	6.49	2.15
Sihwa	9.44	9.02	9.92	36.32	3.03	11.69	7.78	13.11	12.83	0.752	0.16
Seongnam	0.11	0.08	0.07	0.15	0.12	0.14	0.2	0.13	0.23	0.21	0.15
Hongneung	0.1	0.17	-	0.21	0.13	0.19	0.16	0.13	0.2	0.21	0.15
Namsan	0.08	0.09	-	0.14	0.11	0.21	0.09	0.11	0.25	0.16	0.19
Pyeongchang	0.22	0.15	-	0.14	0.08	0.19	0.21	0.09	0.14	0.15	0.14
Hoengseong	0.2	0.14	-	0.19	0.2	0.15	0.13	0.19	0.15	0.14	0.19

,

Table A4. Summary statistics of data.

Variable	Region	N	Mean	SD	Median	Min	Max	Region	N	Mean	SD	Median	Min	Max
cPM10	Gomae	169	65.38	16.22	60.37	42.11	129.17	Yangjae	167	67.81	19.35	62.23	39.09	166.43
cPM2.5			48.33	12.54	45.62	20.9	109.07			51.84	14.56	48.27	21.76	105.27
cPM1.0			43.05	12.29	41.16	11.57	100.64			45.95	14.07	42.85	8.55	96.91
cTemp			5.85	6.73	5.34	−9.78	21.89			6.46	6.13	6.1	−7.49	19.76
cHum			66.78	12.39	67.05	29.77	93.46			64.45	13.04	65.36	31.84	90.65
cWS			0.6	0.3	0.6	0	1.74			0.33	0.16	0.31	0.03	0.94
CO			0.4	0.1	0.4	0.2	0.79			0.88	0.3	0.85	0.25	1.62
O3			0.03	0.02	0.03	0	0.09			0.02	0.01	0.01	0	0.05
SO2			0.003	0.001	0.003	0	0.01			0.004	0.001	0.004	0.001	0.01
NO2			0.02	0.01	0.02	0.01	0.05			0.04	0.01	0.04	0	0.07
tPM10			57.12	12.75	54.4	37.3	121.3			58.46	15.37	54.7	37.97	134.99
tPM2.5			47.04	11.49	45.51	19.83	112.52			50.08	12.68	46.33	21.78	95.14
tPM1.0			42.69	11.35	40.66	11.66	103.55			45.79	12.78	42.75	8.43	90.94
tTemp			8.06	6.77	7.87	−8.36	22.72			6.64	6.15	6.36	−7.66	21.75
tHum			58.72	15.45	57.91	26.78	93.37			64.28	13.27	65.25	32.84	90.55
tWS			0.8	0.46	0.75	0	2.23			0.32	0.18	0.3	0.01	0.97
cPM10	Sihwa	171	64.9	19.91	60.47	39.06	213.54	Seongnam	47	59.44	11.7	55.38	42.4	107.44
cPM2.5			52.1	14.27	48.34	17.15	103.09			49.31	12.28	46.58	28.23	98
cPM1.0			46.59	13.94	43.95	8.73	96.85			46.09	12.04	42.9	23.19	92.63
cTemp			7.42	6.06	8.09	−8.52	20.87			3.29	4.78	2.81	−5.34	15.2
cHum			69.37	11.74	69.69	37.53	92.53			61.98	11.59	60.06	39.84	96.1
cWS			0.61	0.33	0.55	0.06	1.93			0.59	0.36	0.53	0.11	1.71
CO			0.63	0.18	0.61	0.21	1.09			0.76	0.15	0.75	0.5	1.06
O3			0.03	0.02	0.02	0	0.08			0.01	0.01	0.01	0	0.03
SO2			0.003	0.001	0.003	0.001	0.01			0.004	0.001	0.004	0.003	0.01
NO2			0.03	0.01	0.03	0.01	0.06			0.04	0.01	0.04	0.01	0.06
tPM10			60.27	19.81	54.16	38.53	224.22			62.3	11.9	58.52	45.27	105.21
tPM2.5			46.18	13.14	43.13	15.4	96.51			46.39	12.44	43.27	23.26	92.66
tPM1.0			40.83	12.79	38.59	7.38	93.1			41.85	11.91	39	18.12	86.36
tTemp			9.26	6.13	9.54	−6.4	22.8			3.43	4.68	2.65	−4.56	13.93
tHum			61.25	14.31	61.2	26.02	91.8			63.01	11.78	62.64	38.98	95.25
tWS			0.55	0.19	0.55	0.2	1.39			0.29	0.21	0.26	0.01	0.96
cPM10	Hongneung	211	56.44	26.37	51.29	14.62	172.12	Namsan	104	56.79	29.3	50.98	15.83	196.83
cPM2.5			24.77	13.26	21.79	6.46	95.92			25.43	15.14	21.4	5.92	84.58
cTemp			7.98	8.58	9.43	−14.13	22.64			15.25	5.89	16.25	−3.23	24.34
cHum			56.21	14.76	54.63	14.58	99.22			49.78	11.3	47.97	15.17	80.84
cWS			2.5	1.13	2.3	0.46	8.68			1.82	0.58	1.79	0.82	3.7
CO			0.6	0.16	0.56	0.36	1.15			0.49	0.19	0.43	0.16	1.12
O3			0.02	0.01	0.02	0	0.07			0.04	0.01	0.04	0.01	0.07
SO2			0.003	0	0.003	0.002	0.005			0.003	0.001	0.003	0.002	0.004
NO2			0.03	0.01	0.03	0.01	0.06			0.02	0.01	0.02	0.01	0.05
tPM10			49.62	23.56	45.27	13.39	161.57			51.54	26.15	46.79	14.56	172.8
tPM2.5			33.36	17.22	28.54	10.1	110.25			36.13	19.9	30.93	9.8	109.6
tTemp			9.18	8.3	10.52	−12.17	23.71			14.5	5.97	15.61	−4.8	23.75
tHum			54.32	13.79	52.84	15.06	96.79			53.59	11.51	51.7	17.35	84.78
tWS			1.05	0.38	0.95	0.47	3.23			0.44	0.15	0.4	0.17	1.23
cPM10	Pyeongchang	273	42.13	23.78	36.62	7.25	210.62	Hoengseong	202	37.84	16.97	35.62	6.08	105.42
cPM2.5			18.95	10.37	16.12	2.54	54.04			21.48	11.39	19.69	3.58	67.88
cTemp			2.03	7.94	0.8	−14.86	20.35			1.53	8.09	0.21	−14.21	19.63
cHum			60.06	12.54	59.64	29.51	89.61			68.2	12.23	67.76	36.94	97.97
cWS			1.04	0.48	0.97	0.17	3.07			1.31	0.6	1.11	0.42	3.48
CO			0.39	0.12	0.38	0.15	0.79			0.4	0.15	0.4	0.12	0.95
O3			0.03	0.01	0.03	0.01	0.1			0.03	0.01	0.03	0.01	0.06
SO2			0.002	0	0.002	0.002	0.004			0.002	0	0.002	0.001	0.003
NO2			0.01	0	0.01	0	0.03			0.01	0	0.01	0	0.03
tPM10			29.04	19.37	24.98	2.89	177.95			27.08	14.76	24.73	1.69	77.26
tPM2.5			20.26	11.67	17.86	2.39	64.33			21.38	12.87	19.21	1.27	72.12
tTemp			0.18	7.95	−0.97	−18.91	19.87			−2.71	8.37	−2.99	−20.89	14.39
tHum			54.89	17.41	52.76	19.49	97.37			64.99	15.64	64.63	23.19	99.22
tWS			0.6	0.27	0.54	0.19	1.75			0.41	0.22	0.35	0.11	1.41

and

Table A5. Regression coefficient estimation results.

Roadside	Gomae	PM10 RE	PM2.5 RE	PM1.0 RE	Yangjae	PM10 RE	PM2.5 RE	PM1.0 RE
	R2	0.181	0.281	0.277	R2	0.233	0.093	0.102
	Adjusted R2	0.123	0.23	0.226	Adjusted R2	0.173	0.022	0.031
	Residual Std. Error (df = 168)	0.084	0.097	0.104	Residual Std. Error (df = 153)	4.694	4.512	4.767
	F Statistic (df = 12; 168)	3.095 ***	5.478 ***	5.375 ***	F Statistic (df = 12; 153)	3.881 ***	1.308	1.446
Industrial	Sihwa	PM10 RE	PM2.5 RE	PM1.0 RE	Seongnam	PM10 RE	PM2.5 RE	PM1.0 RE
	R2	0.246	0.304	0.313	R2	0.397	0.602	0.521
	Adjusted R2	0.188	0.251	0.26	Adjusted R2	0.177	0.457	0.347
	Residual Std. Error (df = 157)	9.764	9.327	10.259	Residual Std. Error (df = 33)	0.124	0.095	0.078
	F Statistic (df = 12; 157)	4.263 ***	5.725 ***	5.960 ***	F Statistic (df = 12; 33)	1.808 *	4.158 ***	2.990 ***
Urban	Hongneung	PM10 RE	PM2.5 RE		Namsan	PM10 RE	PM2.5 RE
	R2	0.129	0.093		R2	0.14	0.267
	Adjusted R2	0.024	−0.016		Adjusted R2	0.085	0.221
	Residual Std. Error (df = 91)	0.102	0.179		Residual Std. Error (df = 175)	0.081	0.094
	F Statistic (df = 11; 91)	1.226	0.852		F Statistic (df = 11; 175)	2.580 ***	5.801 ***
Background	Pyeongchang		PM2.5 RE		Hoengseong	PM10 RE	PM2.5 RE
	R2	0.2	0.189		R2	0.199	0.239
	Adjusted R2	0.166	0.155		Adjusted R2	0.152	0.195
	Residual Std. Error (df = 262)	0.228	0.132		Residual Std. Error (df = 189)	0.234	0.145
	F Statistic (df = 11; 262)	5.942 ***	5.541 ***		F Statistic (df = 11; 189)	4.270 ***	5.393 ***

Note: Significance levels are denoted as follows: * p < 0.05, and *** p < 0.001.

). Standardized IRFs were normalized by the standard deviation of each response variable to enable comparison across variables (Figure 3).

3.2.1. Background Forests

Background forest sites (Hoengseong and Pyeongchang) showed strong self-responses in both ΔPM₁₀ RE and ΔPM_2.5 RE. When cross-responses were examined, ΔPM₁₀ RE reacted more strongly to shocks in ΔPM_2.5 RE than the reverse, indicating an asymmetric interaction between these two PM fractions under background conditions.

3.2.2. Roadside Urban Forests

At roadside sites (Gomae and Yangjae), ΔPM₁₀ RE responded significantly to shocks from all PMRE variables. In contrast, ΔPM_2.5 RE exhibited responses primarily to its own shocks, whereas ΔPM₁₀ RE responded to both its own shocks and ΔPM_2.5 RE shocks. These patterns indicate that roadside UF mitigation dynamics are strongly structured around ΔPM₁₀ RE responses but show more limited cross-coupling for fine PM fractions.

3.2.3. Industrial-Complex-Adjacent Urban Forests

The two industrial sites represent distinct emission environments, with Sihwa characterized by large-scale heavy industry and relatively strong, continuous emission sources, whereas Seongnam is situated within a densely urbanized area with smaller-scale and mixed urban–industrial activities, implying more heterogeneous emission patterns.

Industrial sites (Sihwa and Seongnam) exhibited the strongest and most complex cross-interactions among PMRE variables, but the patterns differed between the two locations. In Sihwa (a mechanical parts and materials industrial complex), ΔPMRE variables responded to shocks from all PMRE variables, and finer particle fractions (ΔPM_2.5 RE) exhibited more pronounced responses than ΔPM₁₀ RE, indicating greater sensitivity to interactions among smaller particles. In Seongnam (a wood-processing industrial area), ΔPM₁₀ RE responded to shocks from all PMRE variables; however, fine particle responses (ΔPM_2.5 RE and ΔPM_1.0 RE) to ΔPM₁₀ RE shocks were minimal. Instead, ΔPM_2.5 RE and ΔPM_1.0 RE showed strong mutual interactions, suggesting a distinct fine-particle-driven dynamic structure in this region.

Figure 3. Impulse response functions (IRFs) across emission-source contexts. (a,b) roadside environments; (c,d) industrial-complex-adjacent environments; (e,f) urban-core environments; (g,h) background forest environments.

Figure 3. Impulse response functions (IRFs) across emission-source contexts. (a,b) roadside environments; (c,d) industrial-complex-adjacent environments; (e,f) urban-core environments; (g,h) background forest environments.

3.2.4. Dense Urban-Core Urban Forests

For urban-core sites (Hongneung and Namsan), PMRE variables predominantly responded to their own shocks, with minimal cross-interactions among different PMRE fractions. This indicates that PM mitigation dynamics in dense urban cores were largely driven by intrinsic UF processes rather than strong coupling among PM fractions.

3.2.5. Key Cross-Cutting Patterns

Across all source contexts, five main findings emerged.

(1)

ΔPM₁₀ RE responded strongly to PMRE shocks, but its influence on other ΔPM₁₀ RE variables was limited.

(2)

Responses triggered by fine PM shocks (ΔPM_2.5 RE and ΔPM_1.0 RE) persisted longer than those induced by ΔPM₁₀ RE shocks, indicating greater temporal persistence for finer particles.

(3)

Industrial-complex-adjacent UFs showed the clearest evidence that reductions in one PM fraction were associated with reductions in others, suggesting strong interdependence among PM fractions.

(4)

ΔPM_2.5 RE acted as a central mediator linking PM fractions, particularly for ΔPM₁₀ RE–ΔPM_2.5 RE and ΔPM_2.5 RE–ΔPM_1.0 RE interactions, indicating the importance of PM_2.5-driven mitigation dynamics.

(5)

PMRE responses commonly exhibited damped oscillation patterns and gradually converged toward zero over time, suggesting stabilization of mitigation performance following disturbances. This behavior indicates that shocks to the system initially propagate through short-term fluctuations but progressively diminish over time, reflecting a stabilizing adjustment process and dynamic equilibrium in PM reduction dynamics.

3.2.6. Responses to Meteorological Factors and Co-Pollutants

Responses to meteorological variables and co-pollutants differed across sites, even within the same source-context category (

Table A6. Response variables other than $\Delta PMs RE$ by pollution source.

	Region	Variable		Region	Variable
		Response	Impulse		Response	Impulse
Roadside	Gomae	$\Delta {PM}_{10}RE$	-	Yangjae	$\Delta {PM}_{10}RE$	ΔCO
		$\Delta {PM}_{2.5}RE$			$\Delta {PM}_{2.5}RE$	ΔcTemp
		$\Delta {PM}_{1.0}RE$			$\Delta {PM}_{1.0}RE$	cTemp
Industrial	Sihwa	$\Delta {PM}_{10}RE$	$\Delta S{O}_2,\Delta cTemp$	Seongnam	$\Delta {PM}_{10}RE$	ΔcHum
		$\Delta {PM}_{2.5}RE$	$\Delta S{O}_2,\Delta CO,\Delta cTemp$		$\Delta {PM}_{2.5}RE$	ΔcHum, cWS
		$\Delta {PM}_{1.0}RE$	$\Delta S{O}_2,\Delta CO,\Delta cTemp$		$\Delta {PM}_{1.0}RE$	ΔcHum, cWS
Urban	Hongneung	$\Delta {PM}_{10}RE$	ΔcTemp	Namsan	$\Delta {PM}_{10}RE$	ΔcWS
		$\Delta {PM}_{2.5}RE$	$\Delta O_3$		$\Delta {PM}_{2.5}RE$	$\Delta O_3$
Background	Pyeongchang	$\Delta {PM}_{10}RE$	ΔcTemp	Hoengseong	$\Delta {PM}_{10}RE$	$\Delta N{O}_2,\Delta CO$
		$\Delta {PM}_{2.5}RE$	ΔcWS		$\Delta {PM}_{2.5}RE$	$\Delta N{O}_2,\Delta CO$

). In Gomae, where one side of the UF bordered a high-traffic roadside and tree density was high, the influence of meteorological and air pollutant variables on PMRE was relatively small. In contrast, Yangjae, a roadside-encircled urban park forest, exhibited stronger responses to CO and temperature fluctuations.

Industrial-complex-adjacent UFs showed the largest magnitude of responses to non-PMRE variables. Despite both being located in western South Korea, Sihwa responded more strongly to air pollutants and temperature, whereas Seongnam was more sensitive to meteorological factors, reflecting differences in industrial activities and geographic context. In dense urban-core sites (Hongneung and Namsan), responses differed by particle size: larger particles tended to respond more strongly to meteorological factors, whereas finer particles exhibited greater sensitivity to ozone (O₃). However, consistent with Figure 3, strong non-PMRE responses were most prominent in industrial settings, while meteorological and co-pollutant influences were comparatively limited in other contexts.

3.3. Long-Term Contributions of Drivers (Forecast Error Variance Decomposition)

Forecast error variance decomposition (FEVD) was used to quantify the relative contributions of PMRE variables, meteorological factors, and co-pollutants to overall variance in mitigation dynamics [25]. Because individual meteorological and air pollutant contributions were relatively small, they were aggregated into broader categories for clearer comparison (

Table 3. Forecast Error Variance Decomposition for $\Delta \mathrm{PMs} \mathrm{RE}$ for each pollution source.

Response Variables	Region	Time Lag	Impulse Variable					Region	Time Lag	Impulse Variable
			$\Delta \mathit{P}{\mathit{M}}_{10}$ $\mathit{R}\mathit{E}$	$\Delta \mathit{P}{\mathit{M}}_{2.5}$ $\mathit{R}\mathit{E}$	$\Delta \mathit{P}{\mathit{M}}_{1.0}$ $\mathit{R}\mathit{E}$	Sum of Variables				$\Delta \mathit{P}{\mathit{M}}_{10}$ $\mathit{R}\mathit{E}$	$\Delta \mathit{P}{\mathit{M}}_{2.5}$ $\mathit{R}\mathit{E}$	$\Delta \mathit{P}{\mathit{M}}_{1.0}$ $\mathit{R}\mathit{E}$	Sum of Variables
						Meteorological	Pollutant						Meteorological	Pollutant
$\Delta P{M}_{10}RE$	Gomae (Roadside)	1	100	0	0	0	0	Yangjae (Roadside)	1	100	0	0	0	0
		3	93	0	2	2	0		3	91	0	2	2	4
		5	92	1	2	3	0		5	90	1	2	2	4
$\Delta P{M}_{2.5}RE$		1	29	71	0	0	0		1	26	74	0	0	0
		3	29	67	0	1	3		3	27	70	0	2	1
		5	29	67	0	1	3		5	27	70	0	2	1
$\Delta P{M}_{1.0}RE$		1	16	82	2	0	0		1	11	84	5	0	0
		3	18	76	2	1	3		3	12	79	5	1	2
		5	18	75	2	1	3		5	12	79	5	1	2
$\Delta P{M}_{10}RE$	Sihwa (Industrial)	1	100	0	0	0	0	Seongnam (Industrial)	1	100	0	0	0	0
		3	92	0	1	2	4		3	81	2	1	10	4
		5	90	0	2	2	4		5	76	2	1	15	5
$\Delta P{M}_{2.5}RE$		1	62	38	0	0	0		1	5	95	0	0	0
		3	52	40	3	2	3		3	7	58	2	25	8
		5	51	39	4	2	3		5	6	51	2	30	10
$\Delta P{M}_{1.0}RE$		1	44	53	2	0	0		1	0	95	5	0	0
		3	37	54	3	3	2		3	2	64	5	23	7
		5	36	54	4	4	2		5	2	58	5	26	8
$\Delta P{M}_{10}RE$	Hongneung (Urban)	1	100	0	0	0	0	Namsan (Urban)	1	100	0	0	0	0
		3	89	2	5	4	0		3	94	0	5	2	4
		5	88	3	5	4	0		5	94	0	5	2	4
$\Delta P{M}_{2.5}RE$		1	15	85	0	0	0		1	25	75	0	0	0
		3	15	79	3	3	3		3	19	72	6	2	1
		5	15	78	3	4	3		5	19	71	6	3	1
$\Delta P{M}_{10}RE$	Pyeongchang (Background)	1	100	0	0	0	0	Hoengseong (Background)	1	100	0	0	0	0
		3	97	2	0	0	4		3	97	0	0	3	4
		5	96	2	0	0	4		5	96	0	0	3	5
$\Delta P{M}_{2.5}RE$		1	70	30	0	0	0		1	89	11	0	0	0
		3	66	31	2	0	3		3	87	9	1	2	8
		5	66	31	2	1	3		5	87	9	1	3	10

).

Forecast horizons represent the number of time steps ahead (i.e., days in this study) over which the contribution of each variable to forecast error variance is assessed. The percentage values reported in Table 3 indicate the relative contribution of each variable to the total variance of PMRE at each horizon. Higher percentages denote a stronger influence of a given variable on the system’s variability, thereby reflecting the relative importance and persistence of dynamic interactions over time. Accordingly, FEVD provides a dynamic perspective on how different drivers influence the evolution of PMRE over time.

Across all emission-source contexts, ΔPM₁₀ RE generally exhibited the highest explanatory power over time, indicating that its past values substantially influenced current ΔPM₁₀ RE behavior. However, fine-particle dynamics (ΔPM_2.5 RE and ΔPM_1.0 RE) varied by context, requiring source-specific interpretation.

In background forests, ΔPM₁₀ RE predominantly explained variance in ΔPM_2.5 RE, indicating strong dependency between these PM fractions. At roadside sites (Gomae and Yangjae), the relative contribution of ΔPM₁₀ RE and ΔPM_2.5 RE to fine particle variance was approximately 3:7 and 2:8, respectively, suggesting that ΔPM_2.5 RE and ΔPM₁₀ RE were the primary contributors to fine particle variation. For ΔPM_1.0 RE, the contribution of ΔPM₁₀ RE was relatively low, whereas ΔPM_2.5 RE accounted for a larger share, indicating that ΔPM_2.5 RE served as a key explanatory variable for ultrafine particle dynamics.

In industrial settings, the relative contributions of ΔPM₁₀ RE and ΔPM_2.5 RE shifted across sites and time horizons. In Sihwa, the ratio of ΔPM₁₀ RE to ΔPM_2.5 RE contributions was approximately 6:4 for ΔPM_2.5 RE and 4:5 for ΔPM_1.0 RE, with increasing contributions over time from ΔPM_1.0 RE, meteorological variables, and co-pollutants. In Seongnam, ΔPM_2.5 RE initially dominated variance in both ΔPM_2.5 RE and ΔPM_1.0 RE, but its dominance diminished over time, with increasing contributions from meteorological and air pollutant variables. These results indicate strong PM–fraction interactions in industrial-complex-adjacent UFs, while the balance of drivers varies across industrial contexts.

Overall, the FEVD results indicate that the dominant drivers of PM reduction efficiency differ by particle size. For PM_2.5 RE, variability is primarily influenced by interactions with co-pollutants and meteorological factors, reflecting the complex and dynamic nature of fine particle behavior. In contrast, PM₁₀ RE is more strongly associated with its own past values and exhibits relatively more stable dynamics over time. Across both particle sizes, meteorological variables contribute consistently but to a lesser extent, suggesting that while environmental conditions play a supporting role, the core variability is driven by particle-specific dynamics and their interactions with surrounding atmospheric components.

In dense urban-core sites (Hongneung and Namsan), each PMRE variable was primarily explained by its own past values, and contributions gradually shifted toward meteorological and air pollutant variables over time. This pattern suggests that intrinsic UF mitigation processes play a significant role in shaping PMRE dynamics in urban cores, while external drivers exert comparatively smaller effects.

Finally, consistent with source-specific ratio characteristics, the PM_2.5/PM₁₀ ratio in background UFs was higher than those observed near roads, industrial complexes, and urban areas (

Table 4. Characteristics of particulate matter ratios by pollution source.

Division	Region	Ratio (Median)	Region	Ratio (Median)
PM2.5/PM10	Gomae (Roadside)	0.74	Yangjae (Roadside)	0.78
PM1.0/PM10		0.62		0.66
PM1.0/PM2.5		0.84		0.86
PM2.5/PM10	Sihwa (Industrial)	0.77	Seongnam (Industrial)	0.75
PM1.0/PM10		0.65		0.64
PM1.0/PM2.5		0.86		0.86
PM2.5/PM10	Hongneung (Urban)	0.75	Namsan (Urban)	0.69
PM2.5/PM10	Pyeongchang (Background)	0.80	Hoengseong (Background)	0.80

), indicating differences in PM composition between background forests and source-influenced environments.

4. Discussion

4.1. Overview: Why Source Context Matters for Urban Forest PM Mitigation

South Korea frequently experiences elevated PM concentrations during cold seasons due to a combination of local emissions and regional atmospheric transport. During such episodes, PM_2.5 levels often exceed international guideline values, including those recommended by the World Health Organization, placing urban populations at increased health risk. Therefore, understanding the performance of urban forests under these high-pollution conditions is critical not only for national policy but also for informing urban air quality management in other regions experiencing similar pollution regimes.

This study used multivariate time-series modeling to examine how urban forests (UFs) mitigate particulate matter (PM) near contrasting emission environments (roadside, industrial-complex-adjacent, dense urban-core) relative to background forest conditions. By focusing on recurrent high-pollution periods (late autumn–spring) and comparing eight sites across two cold seasons, we show that PM reduction efficiency (PMRE) is strongly context dependent and that mitigation dynamics differ by particle size fraction. These results support a key implication for urban forestry practice: “one-size-fits-all” greening is unlikely to maximize air-quality co-benefits, and UF planning and management should be tailored to local emission environments and particle-size-specific dynamics. From an ecosystem-service perspective, PM mitigation should be interpreted as a dynamic regulating service that responds to repeated disturbances rather than a fixed removal rate. While previous studies have often focused on static concentration differences or site-specific assessments, this study contributes to the literature by explicitly quantifying dynamic interactions among particulate matter fractions and environmental drivers across multiple emission-source contexts using a unified time-series framework.

Taken together, these findings demonstrate the value of integrating dynamic time-series analysis with source-context-specific evaluation of urban forest PM mitigation, offering insights into underlying mechanisms and implications for urban environmental management.

4.2. Particle-Size-Dependent Dynamics and Stabilization of Mitigation Performance

Across source contexts, ΔPM₁₀ RE was largely explained by its own past values, whereas fine-particle dynamics (ΔPM_2.5 RE and ΔPM_1.0 RE) showed stronger interdependencies, with ΔPM_2.5 RE repeatedly emerging as a central mediator for ΔPM_1.0 RE. This pattern indicates that fine and ultrafine particle mitigation should not be interpreted as independent outcomes, but rather as coupled processes in which PM_2.5 plays a critical bridging role. The longer persistence of responses to fine-particle shocks further suggests that fine PM mitigation can exert sustained effects across time, highlighting the importance of prioritizing PM_2.5-oriented strategies when the goal is to reduce ultrafine PM exposure. This interpretation is consistent with evidence that fine and ultrafine particles may originate from combustion byproducts and industrial emissions and can remain dynamically linked over time [26], as well as prior time-series studies reporting coupled dynamics among PM_2.5 and co-pollutants and meteorology [27,28]. This particle-size-dependent behavior is consistent with previous findings that fine particles tend to exhibit stronger atmospheric persistence and more complex interactions with co-pollutants and meteorological variables than coarse particles. In particular, the central role of PM_2.5 observed in this study supports the interpretation that fine particles act as key intermediates in transformation and transport processes within urban atmospheric systems [29]. This interpretation is consistent with recent observational evidence indicating that the effectiveness of urban forests in mitigating particulate matter differs by particle size, with stronger removal observed for coarse particles compared to fine particles under certain environmental conditions [22].

Notably, the impulse-response patterns commonly exhibited damped oscillations, where responses initially deviated upward or downward after a shock and gradually converged toward zero. From an urban forestry perspective, this indicates that UF mitigation effects may stabilize over time following disturbances, which is consistent with the interpretation that UFs can sustain a baseline level of PM reduction function during repeated high-pollution conditions. However, because stabilization patterns and cross-coupling differed between PM size fractions, UF strategies should be particle-size specific, rather than relying on a single generalized “PM reduction” objective.

4.3. Structural Mediation of PM Mitigation Dynamics in Urban Forests

Although the present study focused on dynamic interactions among PM fractions and environmental variables, the observed patterns may be influenced by the structural and compositional characteristics of urban forests. Particulate matter interception is not solely an atmospheric process but is likely to occur at the canopy–atmosphere interface, where vegetation structure can regulate deposition, retention, and resuspension processes. However, it should be noted that canopy structural variables were not directly measured in this study; therefore, these mechanisms are proposed as plausible interpretations rather than empirically tested factors. Future studies incorporating direct measurements of vegetation structure (e.g., canopy density, leaf area index, and species composition) would be necessary to quantitatively evaluate these mechanisms.

The differentiated response patterns identified across pollution-source contexts can be ecologically interpreted through forest structure. In roadside and industrial-complex environments—where cross-fraction coupling between PM₁₀ and PM_2.5 was pronounced—urban forests likely function as semi-permeable filtration systems. Dense canopies, multilayered vegetation (arbor–shrub–herb structures), and edge-facing leaf area can enhance aerodynamic roughness, reduce wind speed locally, and increase particle interception surfaces. These structural attributes may strengthen the dynamic linkage between coarse and fine particle removal observed in the VAR impulse responses. These interpretations are consistent with previous studies indicating that vegetation structure, canopy density, and leaf characteristics play important roles in regulating particulate matter interception, retention, and resuspension processes in urban forests [30,31].

In contrast, in urban-core environments where PM reduction efficiencies were predominantly self-driven and exhibited weaker cross-fraction interactions, structural constraints such as limited patch size, fragmented canopy continuity, and restricted airflow between built structures may reduce turbulence mixing and limit inter-fraction transformation processes. Under such conditions, PM mitigation appears to operate more as a localized retention process rather than an integrated particle-size interaction system.

The damped oscillation pattern consistently observed across sites further suggests that urban forests provide a stabilizing ecological function. Canopy structure likely buffers short-term pollution shocks by temporarily storing particles on foliar and bark surfaces before gradual resuspension or wash-off, thereby contributing to temporal stabilization of PM reduction efficiency.

Taken together, these findings indicate that PM mitigation in urban forests should be conceptualized as a structurally mediated ecosystem service. Forest configuration—including canopy density, vertical layering, spatial buffering distance from emission sources, and edge orientation—may regulate not only the magnitude of PM removal but also the dynamic coupling among particle-size fractions. Therefore, urban forest design strategies should move beyond species selection alone and incorporate structural optimization tailored to pollution-source characteristics.

4.4. How Mitigation Dynamics Differed by Source Context

4.4.1. Background Forests: Baseline Dynamics and Natural Particle Processes

In background forest settings with limited nearby emission sources, ΔPM₁₀ RE explained a substantial share of ΔPM_2.5 RE variation. This suggests that PM_2.5 formation and removal under relatively clean conditions may be influenced by natural processes and particle transformation pathways. Prior studies have reported that larger particles can break down into smaller fractions through mechanical fragmentation or chemical degradation [32,33], which may be consistent with the observed dependency pattern. Importantly, the background sites provide a useful baseline for interpreting the incremental benefits of source-adjacent UFs and for distinguishing source-driven dynamics from more natural background behavior.

4.4.2. Roadside Urban Forests: Site Sensitivity and Local Configuration Effects

For roadside contexts, ΔPM₁₀ RE and ΔPM_2.5 RE were largely explained by their own past values, suggesting relatively independent mitigation behavior of each fraction within roadside UFs. However, responses to non-PMRE shocks differed markedly across roadside sites, implying strong sensitivity to local configuration (e.g., the number and arrangement of nearby roads, surrounding built form, and UF stand structure). For example, Gomae—characterized by high tree density and proximity to a primary roadside source—showed minimal responses to meteorological and co-pollutant variables, whereas Yangjae—an urban park surrounded by roads—responded more strongly to CO and temperature variability. These results indicate that roadside UF performance should be interpreted through both the emission environment and local UF configuration, reinforcing the need for site-specific design and management even within the same “roadside” category.

4.4.3. Industrial-Complex-Adjacent Urban Forests: Strongest Coupling and Fine-Particle Centrality

Industrial contexts exhibited the strongest PM–fraction interactions, but the balance of drivers varied across locations. In Sihwa, both ΔPM₁₀ RE and ΔPM_2.5 RE contributed substantially to fine-particle variation, suggesting active coupling between coarse and fine fractions. In Seongnam, ΔPM_2.5 RE dominated fine-particle dynamics initially, but its explanatory power gradually dispersed as meteorological and co-pollutant influences increased over time. These differences likely reflect heterogeneity in industrial emission profiles and geographic settings. More broadly, the consistent centrality of ΔPM_2.5 RE across industrial settings implies that PM_2.5 reduction should be a primary management objective near industrial complexes, particularly because PM_1.0 constitutes a substantial portion of PM_2.5 and the ΔPM_2.5 RE–ΔPM_1.0 RE linkage remained strong across contexts [34,35]. Such strong interactions among particle fractions in industrial contexts are in line with previous findings that complex emission mixtures and secondary particle formation processes can enhance interdependence among pollutants. However, our results further demonstrate that these interaction patterns are not uniform but vary depending on site-specific emission characteristics and environmental conditions.

4.4.4. Dense Urban-Core Urban Forests: Intrinsic Dynamics and Exposure-Reduction Priorities

Urban-core sites (Namsan and Hongneung) were largely self-driven, with limited cross-interactions among PM fractions. This pattern may reflect the presence of multiple small, localized sources (traffic, heating, biomass combustion) and restricted dispersion due to dense built form, which can reduce cross-fraction coupling at the monitoring scale [36,37]. The observed particle-size-specific sensitivity—larger particles responding more to meteorology and finer particles showing stronger responses to ozone—supports the interpretation that mechanisms and management levers differ by particle size in dense urban areas.

4.5. Planning and Management Implications for Maximizing PM Mitigation

Together, these findings suggest that UF strategies should be tailored to both source context and particle size. These recommendations are grounded in the empirical findings of this study, particularly the FEVD results (Section 3.3), which reveal particle-size-dependent mitigation dynamics and context-specific differences in the relative contributions of co-pollutants and meteorological factors.

Roadside contexts. Given strong site sensitivity, roadside UFs should adopt adaptive structural designs that consider the number and arrangement of nearby roads, the UF’s stand composition, and its interface with pedestrian spaces. This is supported by the VAR and FEVD results, which indicate that PM_2.5 RE plays a central role in mediating interactions among pollutants, particularly in industrial and source-influenced contexts, highlighting the need for strategies that explicitly address fine particle dynamics. Prior evidence highlights that green infrastructure effectiveness depends on type and placement and is most effective when paired with emission reduction and strategic placement near roads [38]. Multi-layer vegetation structures (arbor–shrub–herb) can enhance deposition and interception [39], and plant community attributes such as canopy area and stem size may support PM reduction [40]. Because PM₁₀ and PM_2.5 exhibited different response characteristics, roadside designs should explicitly plan for both barrier/interception functions for coarse particles and enhanced capture of fine particles within the UF interior. Where CO and temperature sensitivity is high (e.g., Yangjae-type settings), species selection and maintenance should also consider tolerance to pollutant stress and thermal stress to ensure sustained performance and resilience.

Industrial-complex-adjacent contexts. The strong coupling among fine particles indicates that PM_2.5-oriented mitigation should be prioritized. In settings where PM₁₀ also contributes substantially to fine particle dynamics (e.g., Sihwa), designs should aim to reduce the interaction pathways between coarse and fine fractions by optimizing species composition and stand structure for both interception and retention. In settings where meteorology and co-pollutants increasingly contribute over time (e.g., Seongnam), complementary strategies that leverage airflow and deposition processes may be beneficial, such as enhancing buffer continuity, linking nearby green spaces to improve ventilation pathways, and using surface designs that support wet deposition and wash-off processes [5,40]. Coniferous species have been suggested as effective for PM blocking and retention in some contexts [40,41], but selection should be balanced with local climate and management feasibility.

Dense urban-core contexts. Because urban-core emissions are diverse and green-space availability is limited, maximizing air-quality co-benefits requires optimizing small green spaces and street greenery as exposure-reduction infrastructure. Multi-layer structures in parks and plazas can enhance deposition and sustain purification functions [40], while targeted planting along roadside edges can provide biofiltration benefits for pedestrian corridors [39]. Where biomass burning or localized fine PM sources are relevant, vertical greening (green walls, vertical gardens) and distributed greening systems may provide supplementary mitigation opportunities for PM_2.5 and associated pollutants.

Overall, these results underscore that UF planning should move beyond generic greening targets toward context-specific prescriptions that explicitly consider local emission environments and particle-size-dependent mitigation dynamics.

Our results suggest that design levers such as canopy density, vertical layering, and edge-to-interior configuration may be especially relevant in roadside and industrial contexts, where fine-particle coupling and cross-variable responses were most pronounced. Recent studies have also emphasized the broader implications of urban forest-based particulate matter mitigation in terms of long-term environmental and health benefits, reinforcing the importance of context-specific management strategies [42].

4.6. Limitations and Future Research

This study has several limitations. First, although two representative sites per source context enabled comparative analysis across contrasting emission environments, site-specific features (e.g., local built form, stand structure, and microclimate) may influence observed dynamics. Second, some regions showed narrower variation in response variables and higher residual errors (e.g., Yangjae and Sihwa), and one urban site (Hongneung) showed limited model significance, indicating that time-series models may not fully capture all patterns at every site. Third, VAR outcomes can be sensitive to lag selection and variable ordering, representing an inherent methodological limitation. Finally, incorporating multiple variables increases model complexity and parameter estimation burden, potentially increasing overfitting risk and reducing degrees of freedom [25]. Future studies should therefore expand multi-site monitoring, strengthen site characterization (e.g., canopy structure, species composition, and spatial configuration), and develop context-specific models that better represent local emission mixtures and seasonal dynamics. Extending observations to leaf-on periods and additional seasons would further improve generalizability and support year-round UF management guidance. Because observations focus on late autumn–spring, mitigation dynamics may differ in leaf-on periods when deposition surfaces and BVOC-related secondary aerosol processes can change.

5. Conclusions

In this study, we quantified particulate matter reduction efficiency (PMRE) of urban forests (UFs) across contrasting emission-source contexts in South Korea (roadside, industrial-complex-adjacent, dense urban-core, and background forests) using field observations during recurrent high-pollution periods (late autumn–spring) over two consecutive years. By applying vector autoregression, we compared region-specific characteristics and dynamic interactions among PM size fractions, meteorological variables, and co-pollutants to derive implications for urban forest planning and management.

Three key insights emerged. First, interactions among PM size fractions were particle-size dependent and varied by site context. While ΔPM₁₀ RE and ΔPM_2.5 RE were largely explained by their own past values, ΔPM_1.0 RE was strongly influenced by ΔPM_2.5 RE, indicating a persistent linkage between fine and ultrafine fractions. This finding supports the need to differentiate UF strategies by particle size, particularly by prioritizing PM_2.5-oriented management when ultrafine particle reduction is a key objective. Second, PM mitigation characteristics differed markedly across emission-source contexts. In background forests, ΔPM_2.5 RE was predominantly explained by ΔPM₁₀ RE, suggesting a stronger influence of primary particle processes under relatively clean conditions. In roadside environments, ΔPM₁₀ RE and ΔPM_2.5 RE behaved more independently, yet local sensitivity to external factors (e.g., CO and temperature) varied by site configuration. In industrial-complex-adjacent contexts, interactions between ΔPM₁₀ RE and ΔPM_2.5 RE were prominent or ΔPM_2.5 RE dominated, depending on emission profiles; in coastal industrial areas, the influence of meteorological variables and co-pollutants increased over time, highlighting geographic modulation of mitigation dynamics. In dense urban-core environments, PMRE was primarily self-driven, with notable responses to O₃ and other pollutants, indicating that mitigation pathways and levers may differ by particle size and co-pollutant regime. Third, the predominance of self-driven dynamics in PMRE, together with damped oscillation patterns, suggests that UF mitigation performance can stabilize and persist over time, reinforcing the value of UFs as long-term air-pollution mitigation infrastructure.

Overall, these findings corroborate that UFs can contribute substantially to PM mitigation, but that benefits are strongly context dependent. Accordingly, UF design and management should be tailored to local emission environments and particle-size-specific dynamics to maximize air-quality co-benefits. Future research should strengthen site characterization (e.g., stand structure, species composition, and spatial configuration), further examine interactions between meteorology and co-pollutants across seasons, and develop empirical, context-specific models that translate monitoring evidence into practical planning guidance for optimizing UF PM mitigation.