Spatial and Temporal Variability of Near-Surface CO₂ and Influencing Factors in Urban Communities

Abstract

CO₂ is the primary contributor to global warming, and also the most significant anthropogenic emission gas in cities. This study investigates near-surface CO₂ spatiotemporal variability patterns at the community scale to address the critical gap in urban CO₂ high-resolution measurement and promote urban carbon neutrality. Combining fixed and mobile monitoring across five representative communities (1-km² coverage) with two-hour temporal precision and 20 m spatial resolution, results revealed average CO₂ concentrations of 440–480 ppm, exhibiting bimodal diurnal cycles and highlighting spatiotemporal divergent emission behaviors. Three communities peaked during 17:00–19:00 LT, while two peaked during 08:00–10:00 LT. Spatial correlation analysis identified two dominant patterns: road-adjacent “externally dominated” hotspots and “internally dominated” zones with elevated intra-community levels. Spearman correlation analysis, Random Forest, and Geographically and Temporally Weighted Regression models quantified spatial morphology and element contributions, demonstrating that building morphology exerted time-varying impacts across communities. Meanwhile, external traffic contributed 18–39% to concentration variability, while internal traffic and energy consumption drove localized peaks. The findings indicated that apart from the emission sources, the micro-scale urban spatial design elements also regulate the near-surface CO₂ distribution. This high-resolution approach provides actionable insights for optimizing community layouts and infrastructure to mitigate localized emissions, advancing carbon neutrality targeted urban planning.

1. Introduction

CO₂ is a non-self-generating greenhouse gas that comes mainly from fossil energy consumption in human production and living processes. Additionally, the gaseous molecular properties of CO₂ enable it to absorb infrared radiation, causing and exacerbating the greenhouse effect. In high-density urban environments, urban communities serve as spatial carriers of residential activities and direct CO₂ emissions [1]. Prior studies have indicated that urban communities, as important sites of human activity, account for approximately 30–40% of the total anthropogenic greenhouse gases globally [2]. And this proportion exceeds 40% in China, indicating a significant potential for emission reduction [3]. Furthermore, as atmospheric CO₂ concentrations rise, they lead to increased heat absorption and re-radiation, intensifying the greenhouse effect in urban areas and serving as a catalyst for extreme climatic hazards.

Understanding the spatial and temporal characteristics of near-surface CO₂ concentrations in urban communities is crucial for effective in situ sequestration mitigation strategies, which are essential for achieving carbon neutrality in built environments [4,5]. Therefore, obtaining high-precision measurements of near-surface CO₂ and analyzing its distribution has become a priority for identifying relevant influencing factors. In recent years, multiple measurements of urban CO₂ levels worldwide have shown that urban areas exhibit a pronounced CO₂ dome effect, with concentrations in high-density urban regions generally exceeding those in suburban or rural areas by 20–50 ppm [6,7]. In high-density urban environments, CO₂ concentrations display relatively regular patterns. CO₂ concentrations on weekdays follow a specific fluctuation pattern, peaking in the early morning and declining their lowest levels in the afternoon [8,9]. However, observations indicate notable spatial and temporal distribution differences in CO₂ levels across different space types of cities and communities that do not precisely follow the “peak–valley” pattern [10,11,12]. This variability suggests that the near-surface CO₂ in urban residential communities is influenced by a range of spatial factors, including anthropogenic and geogenic emissions. Therefore, investigating the spatial and temporal distribution characteristics and factors influencing near-surface CO₂ emissions is vital for identifying high-concentration zones and analyzing the underlying causes of their formation.

Table 1. Summary of existing studies on measuring urban carbon concentrations.

Municipality	Area	Space Type	Measuring Height	Measurement Method	Time Resolution	Spatial Resolution	Reference
Palermo	Urban (with a nearby landfill)	Middle density urban area	13.60 m above the ground floor	Delta Ray Isotope Ratio Infrared Spectrometer	5 min	--	[13]
Vancouver	Urban; suburban	Low density low-rise residential area	Near-surface (1.2 m above ground level)	Continuous measurements at a common point combined with vehicle moving measurements	1 h	20 m	[14]
Shanghai	Urban; suburban	High-density low-rise/high-rise residential area	Near-surface (2 m above ground level)	Multi-point urbanization gradient observation	1 h	5 km	[15]
Syracuse	Urban; suburban	High-rise commercial area; low-density low-rise residential area	Roof of building (46 m above ground level)	Continuous measurements on multiple points	30 min	--	[16]
Nanjing	Urban; suburban; rural	High-density Mixed Forms	Roof of building (110 m above ground level)	Continuous measurements on multiple points; vertical profile measurements	1 h	--	[17]
London	Urban	School area	Building walls and roofs	Continuous measurements on multiple points; vertical profile measurements	30 min	--	[18]
Shanghai	Urban	High-density high-rise residential area	Near-surface (1.2 m above ground level)	Moving measurement	75 min	20 m	[19]
Poland	urban	School area	Roof of building (20 m above ground level)	Continuous measurements on multiple points	2 h	--	[20]
Singapore	Urban	High-density low-rise residential area	Near-surface (1.7 m above ground level)	Continuous measurements at a common point combined with moving measurements	1 h	20 m	[21]

summarizes the measurement methods researchers have adopted to analysis the spatial and temporal characteristics of urban CO₂ levels, including continuous measurement methods of both near-surface and roof of building height, and mobile measurement methods.

• The continuous measurement method relies on CO₂ concentration data collection at common or multiple points [22], which is unable to assess the distribution of CO₂ at a high spatial resolution because of the sparsity of fixed measurement points [23].
• The mobile measurement allows for the study of highly fine-grained CO₂ spatial distribution characteristics, yielding reliable, reproducible, and representative results [21]. A comparison of urban canopy cross sections and near-surface CO₂ monitoring results indicated that urban CO₂ emissions mainly stem from near-surface human activities [24], and the moving measurement is more adaptable for identifying the sources of CO₂ at a height of 1–3 m above the ground [25].

However, the limitations of the measurement methods in the existing studies result in CO₂ measurement outcomes that cannot simultaneously achieve high temporal and spatial resolutions. Therefore, current research lacks attention to near-surface CO₂ distribution at the neighborhood scale, making it difficult to resolve the influence of spatial elements on near-surface CO₂ levels.

This study aims to explore the spatial and temporal variability of near-surface CO₂ in typical urban communities and the impact of urban spatial factors on near-surface CO₂. The three main focused research questions are as follows:

RQ1: how can high-temporal and spatial-resolution near-surface CO₂ measurements be conducted in urban communities?

RQ2: what are the spatiotemporal characteristics of near-surface CO₂ concentrations in representative urban communities?

RQ3: which spatial factors impact on the variability of near-surface CO₂ in urban communities?

2. Materials and Methods

2.1. The Framework of the Study

The measurement and analysis of near-surface CO₂ spatial and temporal variability for different types of communities will follow a five-part logic that includes the selection of representative community samples, the near-surface CO₂ concentration measurements, the spatial and temporal characterization evaluations, and spatial influencing factor analyses (Figure 1). Step 1 establishes representative sampling through morphological clustering analysis of urban form parameters, including building geometry, morphology, density, and layout. Five distinct community types were systematically selected using machine learning-assisted categorization.

Step 2 implements a dual-mode measurement protocol combining fixed-point and mobile monitoring. High-precision CO₂ sensors were deployed in a 20 × 20 m grid configuration, synchronized with GPS-tracked mobile measurements along predefined routes. Continuous data collection spanned 24 h periods to capture diurnal variations.

Step 3 incorporates multi-scale analytical techniques: (1) temporal trends were decomposed using Regression Moving Average (RMA) modeling; and (2) spatial heterogeneity was quantified through Getis-Ord Gi* hotspot analysis.

Step 4 evaluates driving mechanisms through Spearman’s rank correlation, Random Forest feature importance ranking (500 trees, R² = 0.82, RMSE = 0.18), and Geographically and Temporally Weighted Regression (GTWR) with adaptive bandwidths.

Step 5 synthesizes findings through comparative analysis across neighborhood typologies, establishing mechanistic links between urban morphology and CO₂ dispersion patterns. This hierarchical framework enables systematic integration of urban form data, microenvironmental monitoring, and spatiotemporal modeling, addressing critical challenges in urban carbon emission characterization.

2.2. Sample Selection

Based on the research content, Nanjing, which is characterized by a typical high-density urban environment, was chosen as the sample for the study. Nanjing is located in an economically developed region of the Yangtze River Delta in China, and is a typical Chinese megacity. This study focuses on the central area of Nanjing city (latitude 32°3′41.6″ N, longitude 118°47′29.6″ E), which is the core region of this important eastern Chinese city. This area is characterized by a high population density, high spatial density, and frequent economic activities, exhibiting typical features of a high-density urban environment and encompassing various typical urban residential types found globally.

To select the spatial scope of the research samples, we first employed the k-means unsupervised clustering method to classify the indicators of urban communities. Five typical spatial types of urban communities were identified (see

Table 2. Space morphology indicators for the classification of high-density urban communities.

Type	Parameter	Formula	Notes
Plot indicator	Plot area	PA
	Plot Perimeter	PP = ∑Li
Morphological indicator	Building area	BA = ∑Si
	Total building area	$\mathrm{T}\mathrm{B}\mathrm{A}=\sum \mathrm{S}\mathrm{i}\times \mathrm{L}\mathrm{Y}\mathrm{i}$
	Average height	$AH={\displaystyle \frac{\sum {\mathrm{H}}_{\mathrm{i}}}{\mathrm{n}}}$
	Max height	$MH=$HAVE + DD
Density indicator	Building density	$\mathrm{B}\mathrm{D}={\displaystyle \frac{\sum {\mathrm{s}}_{\mathrm{i}}}{{\mathrm{S}}_{\mathrm{B}}}}$
	Floor area ratio	$\mathrm{F}\mathrm{A}\mathrm{R}={\displaystyle \frac{\sum {\mathrm{S}}_{\mathrm{i}}\times {\mathrm{L}\mathrm{Y}}_{\mathrm{i}}}{{\mathrm{S}}_{\mathrm{B}}}}$
	Open space ratio	$\mathrm{O}\mathrm{S}\mathrm{R}={\displaystyle \frac{\sum {\mathrm{s}}_{\mathrm{i}}}{{\mathrm{S}}_{\mathrm{B}}}}$
Layout indicator	Enclosure degree	$\mathrm{E}\mathrm{D}={\displaystyle \frac{\sum {\mathrm{s}}_{eni}}{{\mathrm{S}}_{\mathrm{B}}}}$

for space morphology indicators and their calculation formulas, and Appendix A shows details about data acquisition, clustering methods, and results).

Subsequently, to ensure the validity of the CO₂ measured results, five adjacent urban communities in the central area of Nanjing city were selected as the research samples (Figure 2). These samples represented typical community types, with inter-sample distances not exceeding 1 km. Previous studies have indicated that selecting samples within this range allows for the investigation of the spatial and temporal variability of near-surface CO₂ distribution and its influencing factors, while controlling for external factors such as urban traffic flow, population density, and urban microclimate [26].

Five selected urban communities exhibited typical spatial characteristics (

Table 3. Spatial information of community samples.

Community	A	B	C	D	E
Plan
Model
Household number	1155	111	670	62	967
Building number	12	21	15	62	15
PP (m)	533.0	1014.4	553.6	507.7	908.7
PA (m2)	12,856.2	32,006.3	13,641.6	15,978.8	50,505.8
BD (m2/m2)	0.5	0.5	0.4	0.6	0.3
FAR (m2/m2)	5.8	1.0	2.4	0.8	2.1
AH (m)	96	7	21	6	39
ED (m2/m2)	0.9	0.2	0.7	0.8	0.6
OSR (m2/m2)	0.17	0.03	0.01	0.02	0.20

). Community A (Cluster1, 6.52%) was characterized as a high-rise apartment with a high plot ratio. Community B (Cluster2, 6.42%) represented villa communities, which are characterized by large areas, consisting of low-rise (2–4 stories) detached houses. Community C (Cluster 3, 46.16%) was identified as row houses built during 1980s–2000s, which are the most common type in Chinese cities [27], and characterized by medium heights (7–13 stories). Community D (Cluster 4, 19.54%) is mainly distributed within the old city, and represents typical urban village communities, which are characterized by small plot sizes, low plot ratios, and extremely high building coverages, predominantly consisting of single-story bungalow [28]. Community E belongs to Cluster 5 (21.35%) and is a newly built residential district, which is characterized by a large plot size, relatively high average height (16–30 stories), and high open space ratio.

2.3. CO₂ Measurement Scheme

The measurement scheme consisted of three phases (Figure 3). From 14 to 16 May 2024, measurements were conducted on five samples, including two days of common weather (average wind speed: 1.9 m/s; maximum wind speed: 3.3 m/s), and one day of windy weather (average wind speed: 2.9 m/s; maximum wind speed: 7.5 m/s). The selection of these measurement days was based on their ability to represent common carbon emission conditions observed on working days throughout the year. Furthermore, during the spring season (April to June), near-surface atmospheric turbulence remains relatively stable, and near-surface CO₂ concentrations are less influenced by non-spatial climatic factors. This makes it more suitable to reflect the impact of community spatial configurations on CO₂ concentrations [29,30].

A combination of fixed and mobile measurement methods was employed, collect two sets of measured data that complement and verify each other, which allowed high temporal and spatial precision. This approach provides better adaptability for measuring CO₂ at heights of approximately 1–3 m above the ground, which effectively represents atmospheric conditions at this scale, and helps avoid potential turbulence and obstacles near the ground [21].

To ensure the reliability and validity of the CO₂ measurement scheme, we implemented a systematic approach to the selection of both mobile measurement routes and fixed measurement points. The selection process was guided by several key criteria: Firstly, the measurement points were chosen to capture the diversity of spatial features within each urban community. We ensured that the selected points represented various land use types, such as residential, commercial, and green spaces, to provide a comprehensive assessment of CO₂ distribution. Secondly, the routes for mobile measurements were planned to maximize accessibility. By dividing the communities into grids, we ensured adequate spatial coverage, allowing for detailed mapping of CO₂ concentrations. Thirdly, to account for temporal variations in CO₂ emissions, mobile measurements were conducted at four distinct time periods throughout the day, allowing to capture diurnal variations in CO₂ levels. Lastly, the fixed measurement points were strategically located in open spaces at the center of each urban community.

This positioning minimized interference from surrounding structures and ensured that the measurements were representative of the broader community environment. The 24 h continuous data collection at these points provided a stable reference for validating mobile measurements.

2.4. Measuring Methods

2.4.1. Measuring Instruments

The primary measurements in this study were CO₂ concentration. We used dual-wavelength non-infrared diffusion (NDIR) measuring instruments, specifically the TSI-7515 and TSI-7525 models (

Table 4. Parameters of measurement instruments.

Type	Parametric	Instrument Type	Range	Accuracy	Resolution	Unit	Sampling Time	Time Resolution	Spatial Resolution
Fixed	CO2 concentration	Qing Ping	400–9999	±15%	1	ppm	30 s	10 min	--
Mobile	CO2 concentration	TSI-7515, TSI-7525	0–5000	±3%	1	ppm	30 s	2 h	20 m
	Air velocity	PLC-16025	0.8–30	±2%	0.01	m/s	30 s	2 h	20 m

). These devices are characterized by high precision and stability, which ensure the validity of the measurement results [30,31,32].

For fixed measurements, a non-infrared diffusion CO₂ detector equipped with Wi-Fi functionality was employed, allowing continuous outdoor monitoring. The instrument was positioned at a height of 1.2 m in an open area.

To address potential discrepancies between the fixed sensor (Qingping, the equipment was sourced by Qingping Technology (Beijing) Co., Ltd., Shenzhen Branch, Shenzhen, China) and the mobile sensor (TSI-7515/7525, the equipment was sourced from TSI Inc., located in Shoreview, MN, USA), a series of calibration and consistency control measures were implemented. The main procedures included preheating and zero calibration, followed by cross-validation of the two datasets. First, the measurement instruments were zero-calibrated using high-purity nitrogen (CO₂ ≈ 0 ppm), and all instruments were preheated for 10–15 min prior to each measurement session. After the measurements, the fixed-point data were spatiotemporally matched with the mobile data, and the relative deviation between the two datasets was calculated, with the mean deviation controlled within <5 ppm. Outliers exceeding 3σ were excluded from further analysis. These calibration steps effectively minimized potential discrepancies between instruments, enhancing the overall accuracy and consistency of data collected in both fixed and mobile measurement scenarios. Additionally, GPS logging software (2bulu APP V7.9.7) and spatiotemporal data-matching techniques were employed to ensure data traceability.

2.4.2. Fixed and Mobile Measurements

To investigate CO₂ concentration variations in urban communities, data collection was divided into two parts: fixed measurements and mobile measurements (Figure 3 illustrates the specific workflow). The core steps are as follows:

(1)

Sample selection: five typical community types were selected through morphological cluster analysis.

(2)

Design of measurement points: Communities were divided into 20 m × 20 m grids. Points completely obstructed by buildings were excluded, resulting in the identification of 285 measurement points (29.8% of which were near roads). The grid centers were used as mobile measurement points, while fixed measurement points were located in the central open area of each community, suitable for 24 h instrument placement.

(3)

Data collection:

• Fixed measurements: instruments were placed at a height of 1.2 m, continuously recording CO₂ concentrations for 24 h, with data collected at 10 min intervals.
• Mobile measurements: Conducted in four time periods each day (LT8–10, LT11–13, LT14–16, LT17–19) along preset routes. Each point was sampled for 30 s with the instrument, and all points were measured within approximately two hours [33].

(4)

Data calibration: All instruments were calibrated with standard gases prior to measurements. Cross-validation was conducted between mobile and fixed data, achieving an average measurement accuracy of 96.8%.

2.5. Data Analysis Methods

2.5.1. Space Influencing Indicators

The spatial impact indicator system comprises both spatial morphology indicators and spatial distribution indicators. Spatial morphology indicators quantify the geometric characteristics of urban communities. Based on the morphological clustering results obtained during the sample selection phase described in Section 2.2, this study selected seven representative parameters: Plot Area (PA), Plot Perimeter (PP), Building Density (BD), Far Area Ratio (FAR), Average Height (AH), and Enclosure Degree (ED), with their calculation formulas detailed in Table 2.

Spatial distribution indicators characterize the topological relationships of internal spatial elements, reflecting the interaction between human activities and natural processes. Existing research demonstrates that road networks, building layouts, and green space systems influence near-surface CO₂ distribution through differentiated carbon source/sink mechanisms [34,35,36]. This study established multi-level road impact indicators:

• Distance to External Transportation (DO): shortest path from monitoring points to boundary roads of communities;
• Distance to Internal Streets (DS): Euclidean distance from monitoring points to community-level roads;
• Distance to Intersections (DI): topological distance from monitoring points to nearest road intersections.

Building impact is characterized by the Distance to Building Energy Consumption (DB), defined as the weighted distance from monitoring point centers to the nearest residential buildings (weight = building volume × occupancy rate). The green space effect was quantified through the Normalized Difference Vegetation Index (NDVI)-adjusted Green Space Ratio (GR),

$$GR={\displaystyle \frac{{\sum }_{i=1}^n\left({NDVI}_i>0.3\right)·A_i}{PA}} (0\le GR\le 1)$$

(1)

where Ai represents the area of raster cells meeting vegetation coverage criteria (NDVI > 0.3), and PA denotes the total area of the community. This metric simultaneously captures both the spatial extent and ecological efficacy of green spaces, demonstrating greater biophysical significance compared to conventional area-based ratios.

2.5.2. Spearman Correlation Analysis

This study employed Spearman’s correlation coefficient as a bivariate correlation method for univariate analysis to assess the monotonic relationship between near-surface CO₂ in urban communities and spatial factor variables. Compared with the Pearson correlation analysis, the Spearman correlation analysis is more suitable for evaluating nonlinear relationships between variables, and does not impose requirements on the overall distribution pattern of the sample, making it a more appropriate method for the data conditions of this study [37].

2.5.3. Random Forest Model

A Random Forest (RF) model was used to capture the complex nonlinear relationships between spatial elemental variables and CO₂ concentrations in different types of urban communities to assess the contribution of each spatial elemental variable. The random forest model has demonstrated high accuracy in multiparameter predictions, and has been widely used in urban studies [38,39]. Compared with other machine learning models, the Random Forest model offers better interpretability, and can provide an intuitive understanding of the contribution of each underlying spatial element variable to the distribution of CO₂ concentrations [40].

2.5.4. Geographically and Temporally Weighted Regression Model

Due to the significant spatial heterogeneity and temporal dynamics of CO₂ driving factors in urban communities, traditional regression models struggle to capture these variations. The Geographically and Temporally Weighted Regression (GTWR) model addresses this limitation by integrating a temporal weighting function into the classical Geographically Weighted Regression (GWR) model [41]. The GTWR model is expressed as follows:

$$y_i={\sum }_{k=0}^p{\beta }_k\left(u_i,v_i,t_i\right)x_{ik}+{\epsilon }_i$$

(2)

where $\left(u_i,v_i,t_i\right)$ denotes the spatiotemporal coordinates of observation i, ${\beta }_k\left(u_i,v_i,t_i\right)$ represents the local regression coefficient for variable k, and ${\epsilon }_i$ is the error term. The dual bandwidth optimization process accounts for urban mobility patterns (spatial) and diurnal emission cycles (temporal), capturing the metastable nature of near-surface CO₂ dispersion.

The GTWR model’s strength lies in its ability to simultaneously consider spatial and temporal effects on CO₂ concentrations. For example, traffic impacts may intensify during rush hours [42], while building distributions may affect CO₂ emission variations according to daytime or nighttime meteorological conditions [43]. By adaptively optimizing spatial and temporal bandwidths, the GTWR model accurately captures these dynamic relationships.

In this study, we tested Ordinary Least Squares (OLS), GWR, and GTWR models. After testing for multicollinearity, results revealed that GTWR significantly outperformed both OLS and GWR (

Table 5. Evaluation results of the OLS, GWR, and GTWR models.

MODEL	R2	R2 Adjusted	AICc	Spatiotemporal Distance Ratio
GTWR	0.512	0.509	9562	0.642
GWR	0.243	0.239	9885.63	--
OLS	0.055	--	10,055.64	--

). Using the golden section search method, GTWR’s spatial bandwidth was set to $h_s=0.3 \mathrm{k}\mathrm{m}$ and its temporal bandwidth to $h_t=0.7 \mathrm{h}$. The superior performance of GTWR (R² = 0.512, AICc = 9562) demonstrates its reliability in analyzing spatiotemporal CO₂ variability in urban communities. Therefore, we used GTWR to compute and visualize the regression coefficients of five spatial variables, with 0 as the critical value to distinguish between positive and negative effects.

3. Results

3.1. General Characteristics

According to the measured results, near-surface CO₂ concentrations in urban communities exhibited distinct common characteristics in both temporal and spatial distributions. Under typical weather conditions, the average near-surface CO₂ concentration in these communities ranged from 440 to 480 ppm, which is consistent with observations in the urban communities of Shanghai [30]. In comparison, the global monthly mean CO₂ concentration was 425.19 ppm in December 2024 [44], indicating that urban areas typically exhibit higher CO₂ levels due to local anthropogenic emissions.

In terms of temporal distribution, as illustrated in Figure 4a, CO₂ levels across the five urban communities displayed a similar bimodal pattern over a 24 h period on weekdays. This pattern was further refined using linear regression to produce a combined curve in Figure 4b. The results revealed a significant increase in the concentration during the early morning and evening hours, with an average growth rate of 0.04%, peaking at 479.55 ppm. Conversely, the concentrations declined during the noon and afternoon periods.

On a typical week day, near-surface CO₂ concentrations in the urban communities reached their lowest point, but showed the greatest fluctuations during the LT10–15 time period, with an hourly difference of up to 8.98 ppm. This fluctuation reflects variations in the direct CO₂ emission intensity in urban communities across different times. These variations align with urban mobility patterns during the daytime, as supported by statistical findings indicating that afternoon dynamic and traffic density fluctuates significantly across various urban areas [45].

Additionally, Figure 5 indicates that CO₂ concentrations at measurement points within urban neighborhoods that are not adjacent to roads are generally lower than the overall average, with minor differences ranging from −0.94 ppm to 5.90 ppm. This suggests that external emission sources significantly influence CO₂ levels. To facilitate a more comprehensive analysis of the temporal and spatial differentiation characteristics of near-surface CO₂, this study further explored both temporal and spatial variability.

3.2. Temporal Variability

The results from mobile and fixed measurements in the five sampled communities revealed significant temporal variations in CO₂ levels, including distinct daily fluctuation patterns and peak characteristics across different urban communities. The mobile measurement data indicated differentiated average CO₂ values and daily variation trends (Figure 6a). Community A consistently exhibited the highest near-surface CO₂ levels across all four time periods, with a peak value of 485.59 ppm recorded during the LT17–19 period. In contrast, Community B maintained relatively low CO₂ throughout the LT11–19 period, reaching a minimum of 426.64 ppm during the LT11–13 period. The lowest value in the LT8–10 period was found in Community D, at 447.22 ppm.

Regarding daily variation patterns, Communities A, C, and D showed consistency, with the highest average CO₂ levels recorded during the LT17–19 period at 485.59 ppm, 480.63 ppm and 475.78 ppm, respectively. Subsequently, CO₂ levels were lower during the LT14–16 and LT8–10 periods, with the minimum levels occurring during the LT11–13 period. In contrast, Communities B and E exhibited a different pattern, with peak near-surface CO₂ levels occurring during the LT8–10 period, measuring 461.86 ppm and 469.27 ppm, respectively.

The fixed measurement results illustrated the continuous distribution and variations in near-surface CO₂ levels across central open spaces in the five communities, highlighting the differences in peak fluctuations (Figure 6b). Specifically, Community A exhibited higher near-surface CO₂ levels, greater fluctuations, and elevated peaks during LT8–13 period, with the peak value reaching 493.3 ppm. Community C displayed abnormally high near-surface CO₂ levels and significant fluctuations during the LT7–8 and LT16–22 periods, notably recording two anomalous peaks at LT11 and LT13, measuring 477.3 ppm and 502.3 ppm, respectively. Meanwhile, during the LT13–17 period, Community E experienced greater fluctuations in the near-surface CO₂ levels, peaking at 470 ppm. Additionally, Community D demonstrated relatively low near-surface CO₂ levels with minor fluctuations during the LT10–15 period.

Compared with the mobile measurement results, the findings from fixed measurements further indicate that the temporal differentiation of near-surface CO₂ levels among different communities primarily occurs during periods of human activity (LT 7–22). Moreover, the five-minute temporal resolution of fixed measurements effectively captured short-term CO₂ fluctuations induced by human activities.

3.3. Spatial Variability

Because of the limitations of fixed measurements in capturing CO₂ distribution differences at various locations within communities [46], this study primarily employed mobile measurement results to analyze the spatial differentiation characteristics of near-surface CO₂ in communities, while utilizing fixed measurement methods for multi-point calibration of the mobile measurement results. The results indicated that, regarding the average concentration differences in near-surface CO₂ across communities (Figure 7a), Communities A and E had noticeably higher average CO2 levels compared to others, at 474.39 ppm and 463.28 ppm, respectively, whereas Community B had the lowest average concentration of 442.18 ppm. In terms of spatial differences in near-surface CO₂ (Figure 7b,c), Community C showed the greatest overall concentration difference compared with its internal concentrations (5.80 ppm, 1.28%), and the difference between maximum and minimum CO₂ values for Communities C was relatively large, indicating an uneven CO₂ spatiotemporal distribution.

Based on these results, the study further visualized the gridded CO₂ concentrations spatially (Figure 8a). It was found that near-surface CO₂ levels across the five communities generally exhibited two types of spatial distribution patterns (Figure 8b). The external high-concentration dominant type is characterized by higher CO₂ levels along the outer edges of the communities, with lower internal CO₂ levels. CO₂ hotspots tend to be adjacent to external roads and intersections, such as in Communities A and C. The second spatial distribution pattern was the internal high-concentration dominant type, where high CO₂ concentration areas were found internally rather than along the outer edges, as seen in Communities B, D, and E. Furthermore, fixed measurement data revealed the presence of localized anomalous hotspots in internally high-concentration dominant communities, whereas the riverside area of Community B showed significantly low concentrations, supporting the notion that water bodies serve as important factors in carbon sinks within high-density built environments [43].

Combined with the temporal variation analysis (Figure 8c), it was observed that during the LT 8–10 and LT 14–16 periods, CO₂ concentrations along the roadsides were generally higher, with near-surface CO₂ concentration peaks aligned with morning and evening traffic peaks [47,48]. Conversely, during the LT 11–13 and LT 17–19 periods, CO₂ hotspots emerged within the communities, potentially indicating the impact of household energy use on near-surface CO₂ levels. To further investigate the influencing factors of temporal and spatial differentiation characteristics of near-surface CO₂ in communities, this study discussed the causes of these spatiotemporal differentiation characteristics from the perspectives of spatial factors influences.

4. Analysis of Spatial Influencing Factors

As specified in Section 2.5.2, this study employed six spatial morphology metrics (PP, PA, BA, FAR, AH, and ED) and five spatial distribution indicators (DO, DS, DI, DB, and GR). Three analytical approaches were integrated to investigate spatial impacts on near-surface CO₂ concentrations at the community scale: Spearman’s rank correlation coefficient for nonparametric relationship detection, Random Forest (RF) machine learning for nonlinear pattern identification, and Geographically and Temporally Weighted Regression (GTWR) for spatiotemporal heterogeneity modeling.

4.1. Correlation Between Spatial Influencing Factors and Near-Surface CO₂ Levels

The Spearman correlation analysis between spatial morphology metrics and CO₂ values revealed significant diurnal patterns in spatial–CO₂ interactions (

Table 6. Spearman correlation between spatial morphology indicators and CO₂ concentration values (n = 5).

	CO2 (Average)	CO2 (LT 8–10)	CO2 (LT 11–13)	CO2 (LT 14–16)
PP	−0.3	0.1	−0.5	−0.9 *
PA	−0.4	0.0	−0.3	−0.8
BD	−0.6	−0.2	−0.3	0.2
FAR	0.8	0.4	0.5	0.4
AH	0.9 *	0.7	0.7	0.3
ED	0.5	0.9 *	0.6	−0.1

* p < 0.05.

). Building height (AH) demonstrated a strong positive correlation with daytime-average CO₂ levels (ρ = 0.900 *, p < 0.05), while community enclosure showed contrasting temporal effects with morning positive correlation (ρ = 0.900 *, p < 0.05) and afternoon negative trend (ρ = −0.100).

The Spearman correlation coefficients between the spatial elements and near-surface CO₂ levels across communities are shown in Figure 9. In Community A, CO₂ concentrations showed significant negative correlations with DO (ρ = −0.54, p < 0.05) and GR (ρ = −0.51, p < 0.1), indicating closer proximity to arterial roads and reduced vegetation coverage correspond to elevated CO₂ levels. Community B exhibited contrasting diurnal effects—positive correlation with road distance during LT8–10 (morning peak hours, ρ = 0.49), but negative correlation at LT14–16 (afternoon period, ρ = −0.38).

Notably, Community C demonstrated a tri-modal relationship: negative correlation with DO (ρ = −0.43), positive associations with DB (ρ = 0.51) and GR (ρ = 0.33), suggesting complex urban canyon effects. In contrast, Community D revealed counterintuitive positive correlations between CO₂ levels and DS (ρ = 0.54) and DI (ρ = 0.44)—potentially indicating traffic flow hysteresis effects in these community types. The results revealed a negative correlation between CO₂ levels and distance to both internal and external roads (n = 1115), with distance to external roads having a particularly pronounced effect on communities A and C. Although the impact of road traffic on near-surface CO₂ levels in urban communities is well-documented [30], further analysis is required to dynamically quantify the role of other spatial elements in shaping CO₂ distribution.

4.2. Contribution of Spatial Elements to Community Near-Surface CO₂ Levels

To accurately assess the impact of various spatial factors on CO₂ concentrations, this study employed the Random Forest (RF) machine learning model. Using a cross-validation approach, grid data covering all communities were constructed, and the datasets of each community were divided into training and testing sets as model input. The results showed that the RF model demonstrated excellent predictive performance (R² = 0.82, RMSE = 0.18). A higher R² value (closer to 1) indicates a stronger ability of the model to explain data variation, while a lower RMSE value reflects smaller discrepancies between predicted and actual values, indicating higher model performance.

Analysis of the RF results revealed the contributions of different spatial factors to near-surface CO₂ concentrations (see Appendix B). External and internal traffic factors were identified as the primary drivers, exerting the most significant influence on CO₂ levels. Road intersections and household energy consumption were secondary contributors, with moderate impacts on CO₂ variability. In contrast, blue and green space elements, such as vegetation coverage and waterbody distribution, had weaker overall effects on CO₂ concentrations; however, localized variations were observed in certain communities.

Moreover, notable spatial heterogeneity was evident in the influence of these factors across different communities. For instance, external traffic factors exhibited a pronounced impact in Community A and Community C, whereas internal traffic factors played a dominant role in shaping CO₂ levels in Community D and Community E. These findings highlight the complex interplay between urban spatial factors and CO₂ variability, as well as the importance of considering localized dynamics when addressing urban carbon emissions.

4.3. Influence of Spatial Elements on the Spatial and Temporal Variability to Community Near-Surface CO₂ Levels

To explore the impact of spatial element variables on the temporal and spatial differentiation of near-surface CO₂ in residential communities, this study further introduced a GTWR model to analyze the degree of influence of various factors on near-surface CO₂ at different locations within communities over different time periods. Through map visualization, the local regression coefficients at different locations are displayed, with darker colors indicating a more significant influence of the spatial variables (Figure 10). The results suggested that the influence levels calculated by the GTWR model were generally consistent with those predicted from the RF machine learning model, illustrating the good results of the model.

The quantitative analysis revealed significant traffic-related impacts on community-scale CO₂ dynamics. Distance to Intersections (DI) demonstrated marked regression coefficients during LT8–10 (morning peak) in Communities A, B, and E (β = −0.31–0.1, p < 0.01), indicating substantial impacts of internal road proximity on CO₂ concentrations, highlighting residents’ commuting behaviors. Distance to External Roads (DO) emerged as a critical predictor for Community C during LT8–10 (β = −0.62) and Community A at LT14–19 (evening peak, β = −0.82), suggesting elevated sensitivity to through-traffic emissions during rush hours.

Notably, DB exhibited dual-phase influences: positive correlations during LT11–13 in Community B (β = −1.45) and Community D (β = −0.75). Spatial analysis identified persistent dark hotspots (Figure 9) where DB-associated CO₂ emissions peaked during lunch (12:00–14:00) and dinner (18:00–20:00) periods. This phenomenon correlates with the prevalent use of centralized exhaust ducts in these communities’ residential kitchens (85.7% adoption rate).

The Green Space Ratio (GR) exhibited dual-phase regulation of near-surface CO₂ concentrations, demonstrating suppression effects during daytime and enhancement effects during evening-night period. This diurnal phasic alternation aligns with vegetation’s photosynthetic assimilation and respiratory release processes [49,50,51,52]. Meanwhile, this result also indicates that the green spaces within cities should not be considered merely as carbon sink elements. In fact, the green spaces undergo a transformation process from carbon source to carbon sink over time. Vegetation coverage displayed elevated regression coefficients (β = 130.1–208.6, p < 0.05) in Communities C and D, suggesting that while the carbon sequestration capacity of green-blue infrastructure remains limited compared to traffic-related sources (β = −0.18 ± 0.64) and household energy emissions (β = 0.65 ± 2.11), strategic greening interventions can amplify CO₂ mitigation efficiency.

5. Discussion

5.1. Spatiotemporal Pattern Typology

The above findings revealed significant spatiotemporal heterogeneity in near-surface CO₂ distributions across the five community types. Community A (high-rise apartments) exhibited the highest mean CO₂ concentration, peaking in the evening with pronounced diurnal fluctuations (LT8–13), classified as an externally dominated high-concentration community. Its elevated CO₂ levels were primarily influenced by external traffic and green space ratio. Community E (newly built residential complexes) showed the second-highest mean concentration but displayed a bimodal temporal pattern with morning peaks (LT7–10 and LT13–17), representing an internally dominated high-concentration type. Its CO₂ dynamics were strongly associated with road intersections. Communities C (row houses) and D (urban villages) demonstrated similar mean concentrations with evening maxima and afternoon minima. Community C exhibited significant spatial heterogeneity with elevated concentrations near roads, suggesting limited vertical/horizontal diffusion due to external traffic impacts. Notably, green spaces in Community C paradoxically enhanced CO₂ levels rather than mitigating them, potentially due to restricted airflow from shrub vegetation. Conversely, Community D displayed concentrated hotspots in interior zones, primarily influenced by internal traffic and domestic energy consumption patterns. Community B (villa district) maintained the lowest mean concentrations, particularly during LT11–13, with morning peaks and evening secondary elevations. Its riverside green spaces showed the minimum CO₂ concentrations spatially.

5.2. Differential Driver Mechanisms

The analysis identified AH as a significant factor positively correlated with near-surface CO₂ concentrations, explaining why high-rise apartments (Type A) and commercial complexes (Type E) exhibited the highest CO₂ levels. This finding corresponds with Tian et al.’s observation that building height strongly influences residential and traffic-related CO₂ emissions [53]. Additionally, the ED index demonstrated robust correlations with afternoon and evening CO₂ levels, highlighting that enclosed urban forms obstruct effective carbon dispersion.

While prior studies primarily focused on the impact of road traffic on residential CO₂ emissions [54], our findings shed light on how spatial drivers vary across different community types. In Type A and Type C communities, external traffic acted as the primary contributor to CO₂ levels, whereas in Types B, D, and E, internal traffic and household energy use were the dominant sources. These findings align with earlier research identifying home heating and vehicular traffic as major anthropogenic sources of urban CO₂ emissions [55], providing a more nuanced perspective on how spatial factors drive CO₂ variation in diverse urban communities.

Contrary to the widely held perception of green spaces as universal carbon sinks, their carbon sequestration efficiency varies significantly between community types. While effective in reducing CO₂ levels in Type A and Type B communities, green spaces in Type C communities paradoxically contributed to elevated CO₂ concentrations. This unexpected result can be attributed to the limited carbon sequestration capacity of urban vegetation, intensified by nocturnal plant respiration and restricted CO₂ diffusion caused by dense vegetation structures.

Urban green spaces generally exhibit limited carbon sequestration capability, offsetting less than 2% of total CO₂ emissions at the community level [3]. In high-density areas, green spaces are often small, overly compact, and sparsely distributed [56], resulting in inherently low sequestration rates. The issue is further compounded by high anthropogenic emissions from building energy use and human respiration, which can overpower the carbon absorption function of urban vegetation. Nocturnal plant respiration also significantly impacts localized CO₂ levels. In communities with dense or poorly planned vegetation arrangements, such as Type C, nocturnal respiration contributes substantially to CO₂ emissions. For instance, data from Zurich’s CO₂ monitoring network revealed that the highest CO₂ concentrations coincided with areas experiencing intense nocturnal plant respiration [57].

Furthermore, studies have shown that the spatial configuration and structural characteristics of vegetation critically influence its carbon absorption capacity [58]. In Type C communities, dense shrub vegetation often creates physical barriers to both horizontal and vertical air circulation. Computational Fluid Dynamics (CFD) simulations [59] indicate that dense vegetation, including evergreen shrubs, vines, and lawns, can hinder airflow under calm or low-wind conditions, exacerbating the localized accumulation of CO₂.

5.3. Practical Mitigation Pathways

The findings strongly suggest the necessity for developing tailored carbon management strategies for different community types to mitigate near-surface CO₂ concentrations and advance sustainable urban development. The recommendations focus on three aspects: spatial organization, green landscape design, and policy guidelines.

(1)

Optimization of spatial organization

Urban planning and community design are expected to prioritize the optimization of spatial configurations to facilitate airflow and reduce CO₂ concentrations. This involves controlling critical spatial parameters that influence air movement and carbon emission dynamics.

Firstly, improving spatial morphology is essential for enhancing air circulation. In the context of urban renewal or the design of new communities, attention should be given to building layouts to control levels of enclosure and openness. For example, reducing the enclosure degree can promote optimal airflow near the ground surface. In high-concentration areas such as Type A and Type E communities, lowering building heights can minimize obstructions to air circulation and encourage better natural air convection, enabling the effective dispersion of CO₂.

Additionally, the internal organization of community functions plays a crucial role in reducing emissions from transportation. Communities can adopt the “15 min walkable community” principle to enhance the functional diversity and accessibility of daily services, reducing resident dependency on motorized commuting [60]. Shorter commutes not only decrease travel distances and time but also mitigate carbon emissions directly at their source. Specific measures, such as traffic flow regulation during peak hours (e.g., at LT7–9 and LT17–19 in Type A communities), can further alleviate the carbon footprint of urban transportation.

(2)

Green landscape design and carbon sequestration enhancement

Green landscape design serves as a critical component of community carbon management, contributing to both carbon sequestration and localized CO₂ dispersion. Various strategies can be employed to maximize its effectiveness.

The first recommendation is to allocate dedicated carbon sink spaces. For high-density urban renewal initiatives and new community developments, reserving 10–15% of the total community area for carbon sink landscapes is advised. In Type A and Type B communities, implementing strategic green coverage of at least 25% can not only offset emissions from traffic and daily activities, but also enhance localized CO₂ dispersion conditions [48].

Strategic vegetation planning is another critical measure. Urban green spaces should prioritize plant species with high carbon absorption efficiency [61], customized to the local climatic conditions. In Type C communities, planners should avoid overly dense shrub vegetation, which can impede airflow. Instead, adjustments to the ratio of grassland to trees can improve CO₂ dispersion at the ground level while enhancing carbon storage capacity [62].

Integrating vertical and rooftop greening into architectural design represents a further step toward more effective community-wide carbon sequestration. Vertical greening of building facades, combined with rooftop greening systems, can not only increase vegetation’s ability to absorb CO₂, but also enhance airflow along building surfaces [63,64]. This synergy improves near-surface CO₂ diffusion, leading to substantial enhancements in localized air quality and overall carbon management efficiency.

(3)

Policy guidelines and regulatory frameworks

To ensure implementation of these carbon management strategies, it is vital to establish clear policy guidelines and regulatory standards. Near-surface CO₂ concentration thresholds (e.g., less than 450 ppm) should be integrated into community planning and design standards as mandatory requirements for new construction and redevelopment projects. These thresholds would ensure stricter adherence to carbon reduction goals across communities.

5.4. Limitations

This study has three primary limitations. One limitation of this research lies in its restricted analysis of seasonal variation in near-surface CO₂ concentrations. The CO₂ measurement data were primarily collected during the spring, a season characterized by relatively stable atmospheric conditions near the surface. This ensured the reliability and representativeness of the data, providing a sound basis for analyzing the relationship between spatial forms and CO₂ concentrations. However, the lack of data from other seasons prevents the full characterization of seasonal dynamics in CO₂ concentrations. For instance, during the winter, heating activities generate additional emissions that may lead to significantly elevated CO₂ concentrations. Consequently, the study might underestimate the specific concentration patterns of winter, limiting the generalizability of its conclusions.

This study also has some methodological constraints. For instance, the study employed both mobile measurement devices (TSI-7525) and fixed sensors (Qingping), which have a measurement error margin of ±15 ppm. Despite implementing several calibration and data control strategies, this measurement variability introduces potential uncertainties in data precision. Meanwhile, the study’s spatial scope was confined to a 1 km² survey area, with typological analyses covering only five representative community types (A through E). The diversity of community structures in other urban settings may not be fully captured within this study’s scope. Thus, the applicability of the conclusions might be limited in cities with a broader range of community typologies and climatic conditions.

Finally, there were some certain extraneous variables that were beyond the scope of quantification might have influenced the measured near-surface CO₂ concentrations. For example, temporary human activities, such as nighttime street markets in Type A communities, can significantly elevate local emission levels. However, due to their transient and dynamic nature, these factors were not systematically accounted for in the analysis.

To address the above limitations and enhance the robustness of the findings, future studies will focus on incorporating seasonal and temporal variations, as well as integrating remote sensing and big data analysis. We will include CO₂ measurements conducted during different seasons, combining routine weekdays and weekends. This approach aims to capture the spatiotemporal variations in near-surface CO₂ concentrations in high-density urban environments, thereby improving the generalizability and representativeness of the study conclusions. Furthermore, remote sensing technologies will be integrated with urban big data analytics to enhance the spatial and temporal resolution of CO₂ monitoring.

6. Conclusions

This study uncovers three fundamental characteristics of near-surface CO₂ distributions in heterogeneous urban communities and provides important insights into localized carbon management. By employing high spatiotemporal resolution measurements, we analyzed CO₂ concentrations across five distinct urban residential typologies, advancing the understanding of how spatial factors influence urban carbon dynamics. Specifically, our investigation achieved a temporal resolution of hourly intervals and a spatial resolution of 20 m grid cells for CO₂ monitoring. Compared to traditional urban CO₂ monitoring techniques utilizing eddy covariance (EC) systems—typically with spatial footprints ranging from 100 m to 1 km—our study significantly enhances both the temporal and spatial precision of near-surface CO₂ data.

In terms of carbon emission drivers, prior research has predominantly examined the macro-level phenomenon in which residential proximity to external roads correlates with elevated CO₂ concentrations. However, our findings reveal pronounced differences in the impact of spatial elements across various community typologies. For high-rise apartment complexes and row-house residential areas, external traffic emissions predominantly drive CO₂ concentrations. In contrast, internal circulation of vehicles and household energy consumption are stronger determinants of carbon emissions in villa districts, urban villages, and residential complexes. Additionally, the effects of green space ratio vary across different configurations. While green spaces significantly reduce CO₂ concentrations in well-ventilated community, they paradoxically contribute to concentration increases in other community types, underscoring the importance of context-specific green infrastructure planning.

These findings provide an empirical foundation for spatially explicit carbon management, highlighting the urgent need for typology-specific carbon mitigation strategies. Optimizing spatial configurations—such as adjusting building heights, managing enclosure degrees, and strategically placing green infrastructures—can effectively reduce CO₂ concentrations across diverse community types. Complementary measures, such as integrated traffic flow management during peak hours (07:00–09:00 and 17:00–19:00 LT) and energy-efficient retrofitting of residential infrastructure, present significant potential for driving progress toward urban carbon neutrality.

The holistic integration of spatial optimization, green landscape design, and policy innovation offers a comprehensive pathway for effective carbon management across urban communities. By addressing both natural and anthropogenic sources of CO₂, these measures not only reduce CO₂ levels but also facilitate long-term sustainable urban development. Tailored strategies designed for the unique spatial characteristics of different communities provide cities with the tools to construct adaptive and robust decarbonization frameworks, thereby laying the foundation for future low-carbon urban growth.