Re-Evaluating the Value of Ecosystem Based on Carbon Benefit: A Case Study in Chengdu, China

Abstract

Despite the acknowledged importance of terrestrial ecosystems in achieving carbon neutrality, current carbon accounting predominantly focuses on CO₂ uptake, neglecting indirect contributions from ecosystem services, such as temperature regulation and air purification. We established a carbon benefit (C benefit) accounting framework that integrated these services and analyzed the drivers influencing the spatial and temporal changes in the C benefit. It was found that the average annual growth rate of C benefits in Chengdu over the past 20 years was 0.91 Tg/a, and the CO₂ emissions reduction due to ecosystem services was 22.47 times that of carbon sinks. Therefore, the contribution of ecosystem regulating services to carbon neutrality cannot be ignored. In addition, the elevation, gross domestic product (GDP), and normalized differential vegetation index (NDVI) are key factors affecting C benefits. It is worth noting that the intensive management of constructed ecosystems can result in significant reductions in ecosystem C benefits. Finally, our findings underline the need for low-carbon policies to not only promote carbon sink projects but also enhance the overall capacity of ecosystem services, which could substantially mitigate global climate change.

1. Introduction

To mitigate climate change and achieve sustainable development, more than 130 countries around the world have made “carbon neutrality” commitments [1]. The path to carbon neutrality can be generally divided into two methods: reducing carbon emissions from the source and enhancing carbon sinks [2]. Terrestrial ecosystems have been a significant carbon sink, which can neutralize CO₂ emissions from hard-to-abate industries through the photosynthesis of plants. Therefore, the role of terrestrial ecosystems in carbon neutrality has been widely discussed [3,4].

Currently, many researchers have conducted numerous studies on ecosystem carbon sinks using various methods, such as the field survey [5,6,7,8], flux measurement [9,10,11,12], atmospheric inversion [3,13,14,15], and model simulation [16,17,18]. Field surveys and flux measurement methods can achieve accurate and reliable results. However, these methods impose high requirements on sampling and instrumentation and involve complicated operations. Therefore, they are not applicable to large-scale areas. The atmospheric inversion method fails to fully consider the non-CO₂ compounds released from fossil fuels and biomass fuels and overestimates corresponding CO₂ emissions [19]. Therefore, the carbon sequestration level estimated using this method is likely higher than the actual carbon sequestration level. Furthermore, the model simulation method yields study results that can be applied to regions with similar environments [20] and are highly generalizable. The commonly used models are mainly the Carnegie–Ames–Stanford Approach (CASA) [21,22], the biome–biogeochemical cycles (BIOME-BGC) model [23], the integrated valuation of ecosystem services and tradeoffs (INVEST) [24], etc. Net ecosystem productivity (NEP) is an essential scientific indicator for the quantitative analysis of ecosystem carbon sinks, which can be used to measure the carbon sequestration level of ecosystems and is widely used in carbon cycle research [25,26,27,28]. Therefore, this study adopts the carbon sequestration mechanism model to assess carbon uptake in the terrestrial ecosystem, which is clear in the calculation principle, easy to obtain data, and convenient to calculate.

Despite the extensive research on ecosystem carbon sinks, there is still a research gap when exploring the indirect contribution of ecosystem services to carbon neutralization beyond CO₂ uptake. Ecosystem services support the functioning of the Earth’s life-support system [29,30,31], which also plays an intangible but powerful role in carbon neutrality. For example, through the temperature regulation services of ecosystems, vegetation in cities can effectively reduce near-surface temperatures in summer, creating a locally cooler environment and mitigating the heat island effect [32]. As a result, ecosystems can reduce anthropogenic CO₂ emissions by reducing the energy consumption of air conditioning. Roxon et al. discussed the cost of cooling and heating in relation to UHI in the United States and pointed out that it might equivalate to 550 million tons of CO₂ emissions per year [33]. Similarly, ecosystem air purification services can remove some pollutants (e.g., SO₂, NO₂, and total suspended particulate) through the dry deposition process and are considered an alternative solution to air pollution through technical means [34]. Liu et al. also pointed out that in the source control of atmospheric pollutants, equipment consumes a large amount of energy and indirectly emits a large amount of CO₂ [35]. The air purification services of the ecosystem can reduce the pressure of source control, thus reducing the indirect emissions of CO₂. These previous studies have shown the enormous potential for CO₂ emissions to be reduced via ecosystem services. However, this part of the CO₂ emission reduction contributed by the aforementioned ecosystem services is not considered in the current carbon accounting system, leading to a certain underestimation of the ecosystem carbon benefit (C benefit). It has been shown that the capacity of ecosystems to supply ecological services decreases with proximity to urban centers and that ecosystems surrounded by urban buildings play a weaker role in regulating and purifying the environment [36,37]. However, the drivers that influence the indirect reduction in CO₂ emissions via ecosystem services are not clear yet. Further exploration of the influence mechanisms affecting the C benefits of terrestrial ecosystems is needed in order to understand what factors are critical to the enhancement of ecosystem C benefits so that more scientific carbon management decisions can be made.

Based on research gaps, we quantified the C benefits of temperature regulation and air purification services in ecosystems. Based on the terrestrial ecosystem carbon accounting model, an ecosystem C benefit accounting framework was constructed from the perspectives of CO₂ uptake and emissions reduction, and the value of C benefits was measured with the CO₂ shadow price, which has been widely used to guide environmental policy [38]. Chengdu was selected as a reference sample to estimate C benefits over the past 20 years and explore variation trends in C benefits from the perspectives of time and space. Through a machine learning approach, ten influencing factors were selected to analyze the influence of natural and social factors on the C benefit of ecosystems, thus providing a scientific basis for optimizing urban ecosystem management.

2. Materials and Methods

2.1. Study Area

Chengdu is located in Southwest China of the Sichuan Basin and covers an area of 14,335 km². It is one of the most fertile and water-rich areas in China, known as the ‘Land of Heaven’ [39], and is a breeding base for giant pandas. Chengdu is rich in ecological resources, with 40.2% forest coverage, 8 nature reserves, and 25 forest parks. The altitudes of its multiple districts vary considerably (ranging from 359 to 5364 m), leading to different climatic conditions. Chengdu is also a typical and representative mega city in Western China. Over the past 20 years, Chengdu has been optimizing urban ecological construction and exploring the harmonious coexistence between humans and nature. Therefore, we selected Chengdu as the study area to analyze its C benefits. Figure 1 shows the geographical location and land use of the study area.

2.2. C Benefits Assessment Framework

This study integrated three ecosystem service models based on which an ecosystem C benefits accounting framework was established (Figure 2); the framework comprises three parts, i.e., CO₂ sequestration services (${\mathrm{Q}}_{\mathrm{C}\mathrm{S}}$), the CO₂ emission reductions contributed by temperature regulation services (${\mathrm{Q}}_{\mathrm{T}\mathrm{R}}$), and CO₂ emission reductions contributed to by air purification services (${\mathrm{Q}}_{\mathrm{A}\mathrm{P}}$). The C value of an ecosystem is the C benefit of the ecosystem multiplied by the shadow price of CO₂ (P_CO2) [38]. The calculation formulas are as follows:

$$\mathrm{C} \mathrm{b}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{f}\mathrm{i}\mathrm{t}={\mathrm{Q}}_{\mathrm{C}\mathrm{S}}+{\mathrm{Q}}_{\mathrm{T}\mathrm{R}}+{\mathrm{Q}}_{\mathrm{A}\mathrm{P}},$$

(1)

$$\mathrm{C} \mathrm{v}\mathrm{a}\mathrm{l}\mathrm{u}\mathrm{e}=\mathrm{C}\mathrm{a}\mathrm{r}\mathrm{b}\mathrm{o}\mathrm{n} \mathrm{b}\mathrm{e}\mathrm{n}\mathrm{e}\mathrm{f}\mathrm{i}\mathrm{t}\times {\mathrm{P}}_{{\mathrm{C}\mathrm{O}}_2},$$

(2)

2.2.1. Carbon Uptake Model

The carbon sequestration mechanism model [40] was used to calculate ${\mathrm{Q}}_{\mathrm{C}\mathrm{S}}$:

$${\mathrm{Q}}_{\mathrm{C}\mathrm{S}}={\mathrm{M}}_{{\mathrm{C}\mathrm{O}}_2}/{\mathrm{M}}_{\mathrm{C}}\times \mathrm{N}\mathrm{E}\mathrm{P},$$

(3)

$$\text{NEP}=\text{α}\times \text{NPP}\times \frac{{\text{M}}_{{\text{C}}_6}}{{\text{M}}_{{\text{C}}_6{\text{H}}_{10}{\text{O}}_5}},$$

(4)

where NEP represents the net ecosystem productivity (t C/a), $\mathsf{\alpha }$ represents the conversion coefficient of NEP and NPP, and NPP represents net primary productivity (t C/a).

2.2.2. Carbon Emission Reduction Model

Carbon emission reductions contributed by ecosystem services can be calculated by power consumption reduction and grid emission factors. The total energy consumed via ecosystem evapotranspiration was considered as the power consumption reduction contributed by temperature regulation services [40]:

$${\mathrm{Q}}_{\mathrm{T}\mathrm{R}}={\mathrm{E}}_{\mathrm{t}\mathrm{t}}\times \mathrm{f},$$

(5)

$${\mathrm{E}}_{\mathrm{t}\mathrm{t}}={\mathrm{E}}_{\mathrm{p}\mathrm{t}}+{\mathrm{E}}_{\mathrm{W}\mathrm{e}},$$

(6)

$${\mathrm{E}}_{\mathrm{p}\mathrm{t}}={\sum }_{\mathrm{i}}^3{\mathrm{E}\mathrm{P}\mathrm{P}}_{\mathrm{i}}\times {\mathrm{S}}_{\mathrm{i}}\times \mathrm{D}\times {10}^6/(3600\times \mathrm{r}),$$

(7)

$${\mathrm{E}}_{\mathrm{W}\mathrm{e}}={\mathrm{E}}_{\mathrm{W}}\times \mathrm{q}\times {10}^3/\left(3600\right),$$

(8)

where ${\mathrm{E}}_{\mathrm{t}\mathrm{t}}$ represents the total energy consumed via ecosystem evapotranspiration (kWH/a), $\mathrm{f}$ represents the emissions factor of the China Power Grid (0.5810 tCO₂/MWh), ${\mathrm{E}}_{\mathrm{p}\mathrm{t}}$ represents the energy consumed by ecosystem vegetation transpiration (kWH/a), ${\mathrm{E}}_{\mathrm{W}\mathrm{e}}$ represents the energy consumed via ecosystem water surface evaporation (kWH/a), ${\mathrm{E}\mathrm{P}\mathrm{P}}_{\mathrm{i}}$ represents the heat consumed via transpiration per unit area of the type i ecosystem (kJ/m²d), ${\mathrm{S}}_{\mathrm{i}}$ represents the area of the type i ecosystem (km²), r represents the air conditioning efficiency ratio (3.0), D represents the number of days the air conditioning is on for, ${\mathrm{E}}_{\mathrm{W}}$ represents the amount of evaporation from the water surface (m³), and q represents the latent heat of volatilization (J/g).

The air purification services of the ecosystem mainly include the uptake of pollutants such as particulate matter (PM), SO₂, and NO_X [41]. Among them, the treatment effect of PM is much higher than that of other pollutants. Therefore, in this study, we only assessed the reduction in CO₂ emissions when contributed indirectly by the purification of PM in ecosystems [42]:

$${\mathrm{Q}}_{\mathrm{A}\mathrm{P}}={\mathrm{R}}_{\mathrm{a}\mathrm{p}}\times \mathrm{e}\times \mathrm{f},$$

(9)

$${\mathrm{R}}_{\mathrm{a}\mathrm{p}}={\sum }_{\mathrm{i}=1}^{\mathrm{m}}{\mathrm{R}}_{\mathrm{i}}\times {\mathrm{S}}_{\mathrm{i}},$$

(10)

where ${\mathrm{R}}_{\mathrm{a}\mathrm{p}}$ represents the ecosystem air purification capacity (kg/a), $\mathrm{e}$ represents the power consumption factor of the electrostatic precipitator (kWH/kg), and ${\mathrm{R}}_{\mathrm{i}}$ represents the dust retention capacity per unit area of the type i ecosystem (kg/km²∙a).

Details on the C benefits assessment model parameters can be found in Table S1.

2.3. Drivers Analysis of C Benefits

Since ecosystem C benefits include natural carbon sinks and carbon emission reductions from ecosystem regulation services, factors that might affect carbon sinks and ecological services must be considered simultaneously when selecting drivers of C benefits. Summarizing the results of several researchers, it was found that temperature [43,44], precipitation [43], elevation [45], vegetation structure [46], and soil properties [44] are important factors affecting ecosystem carbon sinks. Additionally, precipitation [47], economic factors [48], the characteristics of the soil matrix [49], and vegetation structure [47,48] also affect the capacity of ecosystem regulation services. Therefore, we selected 10 factors to analyze their driving effects on carbon benefits, including precipitation (PRE), temperature (TEM), elevation (DEM), the normalized differential vegetation index (NDVI), topsoil (0–30 cm) organic carbon content (TOC), topsoil pH (TPH), subsoil (30–100 cm) organic carbon content (SOC), subsoil PH (SPH), gross domestic product (GDP), and population (POP). All of the aforementioned data were converted to data with a uniform spatial resolution of 1 km × 1 km via data preprocessing. A grid method was used to achieve the spatial matching of C benefits and influencing factors, and a total of 11,778 samples were obtained.

To determine the overall relationship among the 10 influencing factors, the principal component analysis (PCA) method was used to downscale and extract features from the sample dataset in this study. The PCA revealed the simple structure behind complex data by mapping high-dimensional data to the number axis with the largest variance and excluding the number axis with a smaller variance [50]. The algorithm is as follows:

$$\mathrm{P}\mathrm{C}\mathrm{X}={\sum }_{\mathrm{k}=1}^{\mathrm{n}}\mathsf{\alpha }1\times \mathrm{X}1,$$

(11)

where PC represents the principal component, X represents the variable and $\mathsf{\alpha }$ represents the weight.

The extreme gradient boosting (XGBoost) algorithm was used to predict the C benefits of four types of ecosystems using 10 influencing factors. XGBoost is a tree model for regression and classification [51]. It generates a series of models by series iteration and adds these models linearly weighted to obtain the final integrated learner. Moreover, the XGBoost optimizes the structured loss function and adds the complexity of the tree model to the regular term to reduce the risk of overfitting. Consequently, XGBoost generally has a higher inferential capability and higher prediction accuracy. The algorithm is as follows:

$$\mathrm{F}({\mathrm{x}}_{\mathrm{i}})={\sum }_{\mathrm{k}=1}^{\mathrm{K}}{\mathrm{f}}_{\mathrm{k}}({\mathrm{x}}_{\mathrm{i}}),$$

(12)

where $\mathrm{F}\left({\mathrm{x}}_{\mathrm{i}}\right)$ represents the output of the model, ${\mathrm{f}}_{\mathrm{k}}\left({\mathrm{x}}_{\mathrm{i}}\right)$ represents the $\mathrm{k}$th weak learner, and ${\mathrm{x}}_{\mathrm{i}}$ represents the feature of the training sample set.

To solve the problem of the low interpretability of the XGBoost model, we used the Shapley Additive Explanation Approach (SHAP) to explain the influence of input factors on the model’s output results. The SHAP approach was based on the coalitional game theory [52], which treats the model predictions as a result of the joint action of the feature values. The Shapley value can be defined as the average marginal contribution of the eigenvalues to all possible coalitions. Because the Shapley value takes into account all possible predictions for all possible combinations of input instances, it can ensure both consistency and local accuracy.

$$\mathrm{g}({\mathrm{z}}^{\prime })={\mathrm{Ø}}_{\mathrm{o}}+{\sum }_{\mathrm{j}=1}^{\mathrm{M}}{\mathrm{Ø}}_{\mathrm{j}}{\mathrm{z}}_{\mathrm{j}}^{\prime }$$

(13)

where $\mathrm{g}\left({\mathrm{z}}^{\prime }\right)$ represents the interpreter function, M represents the number of features, and ${\mathrm{z}}_{\mathrm{j}}^{\prime }\in {\left\{0,1\right\}}^{\mathrm{M}}$ and ${\mathrm{Ø}}_{\mathrm{j}}$ represent the $\mathrm{j}$th features of the Shapley value.

2.4. Data

The meteorological observation data used herein were obtained from the China Meteorological Data Network (http://data.cma.cn/data/weatherBk.html) (accessed on 10 August 2022), land-use and land-cover change (LUCC) data were obtained from the GlobeLand30 dataset (http://www.globeland30.org/) (accessed on 20 June 2022), and net primary productivity (NPP) data were obtained from the MODIS17A3 dataset (https://search.earthdata.nasa.gov) (accessed on 20 June 2022). The precipitation (PRE), temperature (TEM), normalized difference vegetation index (NDVI), gross domestic product (GDP), and population (POP) datasets were obtained from the Resource and Environment Science Data Center (https://www.resdc.cn/) (accessed on 10 August 2022). The digital elevation model (DEM) data were obtained from the Geospatial Data Cloud (http://www.gscloud.cn/) (accessed on 10 August 2022), and the soil property data were acquired from the Harmonized World Soil Database Version 1.1 (http://data.tpdc.ac.cn/zh-hans/) (accessed on 20 June 2022).

3. Results and Discussion

3.1. Analysis of the Temporal Trend Regarding C Benefits

Figure 3 shows the changes in terrestrial ecosystem C benefits in Chengdu from 2000 to 2020. The C benefit of the Chengdu terrestrial ecosystem in 2020 was 66.1 Tg, with the C value as high as 15.8 billion dollars: an increase of 39.4% compared with that in 2000. The average annual growth rate of C benefits was 0.91 Tg/a, showing a substantial growth trend and also proving that China’s ecological conservation efforts have achieved good results. According to the CO₂ emission inventory of the Emissions Database for Global Atmospheric Research (EDGAR) version 6.0 [53], the total CO₂ emissions in Chengdu in 2020 was 82.3 Tg. It can be concluded that the terrestrial ecosystem offsets about 45.41% of CO₂ emissions in Chengdu, which is much more than the results of previous studies [19,54]. This finding extends our understanding of the role ecosystems play in carbon neutralization goals.

Figure 3b shows that from 2000 to 2020, C benefits from temperature regulation services (Q_TR), carbon sequestration services (Q_CS), and air purification services (Q_AP) increased by 18.4 (41.43%), 0.218 (8.42%), and 0.031 Tg (7.38%), respectively. The largest increase was observed for Q_TR, which was the main reason for the increase in ecosystem C benefits. We speculated that this might be due to the increased role of ecosystem temperature regulation services as global warming and the heat island effect intensified during recent years [55], which could help reduce more energy consumption and carbon dioxide emissions. Figure 3b shows that, among various ecosystems, forests and wetlands had the most pronounced upward trend of C benefits, with increases of 47.17% and 42.81% from 2000 to 2020, respectively. This indicates that Chengdu’s efforts in forest conservation and park city construction over the years have achieved good results, and the area and quality of forests and wetlands have improved [39]. From 2000 to 2020, the C benefits of farmland remained stable, rising slightly by 8.41%, while that of grassland decreased by 21.51%. We found that this was caused by the significant decrease in the grassland area in Chengdu city during 20 years [39], which had 901 km² of grassland in 2000, but only 518 km² remained in 2020.

3.2. C Benefits Components and Spatial Distribution Trends

Figure 4d shows the components of C benefits for different ecosystems. This figure shows that the CO₂ emissions reduction from the temperature regulation service (Q_TR) and air purification service (Q_AP) accounted for 95.07% and 0.67% of the C benefits, respectively, while the CO₂ uptake by the carbon sequestration service only accounted for 4.25% of the C benefit. Surprisingly, the CO₂ emissions reduction due to ecosystem services was 22.47 times that of carbon sinks. Although previous studies have focused on the management of natural carbon sinks [56,57,58], these results suggest that ecosystems can also cut a large amount of potential CO₂ emissions at their source through regulating services, providing a new insight to facilitate the achievement of carbon neutrality. In addition, Q_TR is mainly contributed by forests (60.28%) and wetlands (35.46%), Q_CS is mainly contributed by farmlands (58.12%) and forests (36.94%), and Q_AP is mainly contributed by forests (94.90%). The contribution of different ecosystems to C benefits is ranked as forests (59.51%) > wetlands (33.75%) > grasslands (4.23%) > farmlands (2.50%). However, due to the areas of different ecosystems varying considerably, space matching was performed in this study using LUCC and C benefit data to determine the per unit area of C benefits for each type of ecosystem. The results showed that the average C benefit per unit area for the wetland, forest, grassland, and farmland were 940.55, 99.17, 47.04, and 2.11 t ha⁻¹, respectively. One unanticipated finding was that wetlands exhibit a higher carbon-neutral capacity than the same area in other ecosystems. Previously, numerous scholars have regarded forests as key to helping mitigate climate change [59,60,61,62]. However, our study shows that although the direct CO₂ sequestration of wetlands is considerably less than that of forests, they contribute significantly to mitigating the heat island effect and reducing energy consumption due to their powerful thermoregulatory capacity, indirectly decreasing substantial CO₂ emissions in cities.

Table S2 and Figure 4 show the C benefits of each county in Chengdu, and its spatial distribution presents characteristics that are high in the west and low in the east while appearing lowest in the middle. The high C benefits area of Chengdu was distributed primarily in the western mountainous and the Minjiang River Basin, including Dayi, Dujiangyan, Qionglai, Chongzhou, and Xinjin. They covered 36.84% of the whole area of Chengdu, but their C benefits accounted for 57.09% of the overall C benefits. Their C benefits per unit area were 7892, 8363, 5602, 6633, and 7330 t CO₂ km⁻² a⁻¹, which exceeded the average level of Chengdu by 72.04%, 82.30%, 22.12%, 44.59%, and 59.78%, respectively. We speculated that this might be due to these regions having large areas of nature reserves or dense river networks. For example, Dujiangyan and Dayi rank in the top two in terms of nature reserve areas, and Xinjin ranks first in terms of river network density. This suggests that China’s nature reserve policy also plays an important role in the carbon–neutral strategy. Pidu and the central urban area cover 6.3% of the whole area of Chengdu, while their C benefit is only 1.04% of the overall C benefits. Their C benefits per unit area are only 14.13% and 18.81% of the average level of Chengdu, respectively. This indicates that ecological resources in economically developed urban centers are relatively poor, and constructed ecosystems have less capacity to provide ecosystem services that match those observed in earlier studies [63]. Meanwhile, the spatial distribution of Q_TR, Q_CS, and Q_AP in Chengdu are roughly similar, with slight differences in local areas (Figure 4a–c). Both Q_TR and Q_AP have their highest values in the northwest area since their main contributors are forests, which are more abundant in the northwestern mountains. The highest QCS is observed in southwestern regions such as Qionglai and Dayi, which are rich in forest resources and have vast agricultural land.

3.3. Analysis of C Benefit Drivers

Figure 5 shows the experimental dataset of terrestrial ecosystem C benefits modeling. There were 11,778 sets of 1 km resolution grid samples in the dataset, and the sample numbers of farmland, forest, grassland, and wetland were 7997, 3051, 538, and 192, respectively. Each sample contained terrestrial ecosystem C benefits (Figure 5k), which were calculated by the method described in Section 2.2 and 10 drivers (Figure 5a–j). Figure 5 shows that the mountainous area in the west of Chengdu had a high terrain, while the eastern part was low and flat. The temperature in the southwest was higher than that in the north, and PRE decreased with increasing latitude. The central district was where the population and GDP were mainly concentrated due to a high level of economic development. The soil pH showed a distribution pattern of acidic to alkaline soil from the west to the east.

Dimensionality reduction was performed for the experimental dataset using principal component analysis. The number of PCs was determined to be three according to the parallel analysis scree plot (Figure 6a). Figure 6b–d shows the load values of 10 drivers and the scores of 11,778 samples on these three PCs. The three PCs explained 67% of the variance for the original dataset. The vectors presented in Figure 6 indicate the correlation between the drivers and the PCs. The included angle between the vectors represents the relationship between the drivers. The vectors with a smaller angle had a more substantial positive correlation. Conversely, if the two vectors were in opposite directions, they showed an evidently negative correlation. For example, GDP was positively correlated with POP and negatively correlated with NDVI. This is because densely populated areas are usually economically and socially developed, where vegetation growth tends to be suppressed [64]. The opposite direction of the DEM and TEM vectors indicated that these two variables were negatively correlated, and the reason for this was the decrease in temperature with increasing elevation. Similarly, there was a strong positive relationship between PRE and pH. This occurred because the study area was located in southwestern China, where the soil was acidic, and as precipitation increased, the dilution effect made the soil less acidic [65]. Figure 6c shows that the included angle between TOC and SOC was large, i.e., no substantial correlation existed between the organic carbon content in the soil layers at different depths. This occurred because the soil’s surface was more susceptible to influences such as climate. Therefore, TOC was more varied [66], while SOC tended to be more stable. In addition, the representation of PCs could be determined based on the projection of the drivers on the three axes. POP and GDP have large loading coefficients on the PC2 axis. Therefore, PC2 could be considered a vector representing the level of development. Similarly, PC1 and PC3 could be considered vectors describing geographical conditions and soil properties, respectively.

The four-colored scatters presented in Figure 6 represent the samples of the four types of ecosystems. The scatters of different colors cluster with each other, indicating good repeatability in the dataset. The farmland samples (purple points) show a high degree of dispersion along the PC1 and PC3 axes. This occurs because farmlands in Chengdu are widely scattered, with large differences in soil properties and economic development levels among different regions. However, due to the high demand for flat terrain for agricultural activities [67], farmlands are mainly distributed in the plains. Therefore, the sample points are very concentrated on the PC2 axis. By contrast, the forest samples (blue points) are more discrete along the PC2 axis. This occurs because forests are mainly distributed in the western mountainous areas with large altitude differences.

Figure 7 shows the performance of the XGBoost model in predicting the C benefits of the four types of ecosystems. To prevent overfitting, we used 80% of the samples (green points) to train the model and 20% of the samples (purple points) to validate the model. Here, the XGBoost model showed excellent prediction performance in forests, grasslands, and farmlands, with R² values of 0.82, 0.78, and 0.86, respectively. This indicates that the model could reliably simulate their C benefits. However, the XGBoost model of wetlands did not show good performance as well as that of other terrestrial ecosystems, with an R² value of 0.56. A possible explanation for this might be that wetlands are not abundant in Chengdu, resulting in an insufficient training sample size of wetlands.

Moreover, NDVI had a positive effect on the ecosystem’s C benefit, while GDP and POP weakened the ecosystem’s C benefit (Figure 8). This was due to the fact that vegetation in areas with high NDVI had more biomass, and the capacity of vegetation transpiration and photosynthesis was stronger [68]. Correspondingly, it provided more carbon sequestration and temperature regulation services. Therefore, the benefits of C were higher. In the same way, when the temperature increased within a certain range, the transpiration of vegetation was also strengthened [69], and the carbon benefits brought about by temperature regulation services increased. Therefore, temperature is also an important factor affecting C benefits. In contrast, GDP affects the growth of vegetation through anthropogenic interference [70], which, in turn, affects carbon benefits. DEM seems to exhibit different effects on different ecosystems for forests and farmlands, where samples with higher DEM eigenvalues are distributed in areas with the highest SHAP values, i.e., the C benefits of forestland increase with an increasing elevation in Chengdu. For grasslands, medium-altitude areas (784–2156 m) exhibited higher C benefits, while alpine meadows tended to have lower C benefits. We speculated that this might be due to the fact that, as altitude increases within a certain range, human activity decreases, and vegetation grows more naturally. Therefore, the C benefits of all ecosystems increase with elevation. However, when the altitude is higher than a certain value, the cold environment is no longer suitable for vegetation growth, and carbon benefits should start to decrease [45]. However, since the trees are no longer viable [71] and there are no samples of forests at very high altitudes, the C benefits of forests cannot be observed to “increase-decrease” with increasing altitude. One unanticipated finding was that although both POP and GDP are factors reflecting the level of social development, the negative impact of POP on ecosystem C benefits is much smaller than that of GDP, which indicates that developed economic levels are more damaging to ecosystem C benefits than extensive traces of human life. We speculated that this was due to the fact that affluent areas tended to invest more in the management of constructed ecosystems for landscaping needs with regular pruning, irrigation, and fertilization. However, studies have shown that finely managed vegetation tends to provide fewer ecosystem services [63].

3.4. Policy Implications

To further enhance the C benefits of terrestrial ecosystems, we put forward the following three recommendations for decision-makers based on the above findings:

• When constructing green infrastructure in urban centers with scarce land resources, small wetlands should be given priority over urban forests. Achieving the goal of carbon neutrality requires a rational approach to optimizing the allocation of ecological resources. Our findings suggest that, although wetlands might not absorb as much CO₂ as forests, they play a substantial role in regulating temperature and mitigating heat island effects [72], reducing huge amounts of potential CO₂ emissions. Compared with other types of ecosystems, wetlands have higher C benefits for the same area. Their urban center is not only short of land resources but also the area with the most severe heat island effect. Therefore, building blue spaces can maximize C’s benefits.
• The restoration of pre-existing ecosystems needs to be prioritized over afforestation. The C benefits of the terrestrial ecosystem are the result of the joint effort of multiple ecosystem services. Based on the estimation result of ecosystem C benefits in Chengdu, we know that the direct sequestration of CO₂ by ecosystems only accounts for a small part of C benefits. It is an end-of-pipe CO₂-emissions uptake measure that is the focus of current carbon sink management. However, an indirect reduction in CO₂ emissions from ecosystem temperature regulation services is even more significant and is usually overlooked. Merely focusing on changing existing land use to increase carbon reserves may not be conducive to achieving carbon neutrality [73,74]. Therefore, ecological restoration measures should be adopted to enhance the overall capacity of ecosystem services and give full play to the huge potential of ecosystems to reduce carbon emissions, thereby contributing to the realization of the carbon neutrality goal.
• The over-elaborate management of constructed ecosystems in cities should be avoided, and vegetation should be left in its original state as much as possible. Our study found that the most negative impact on C benefits was the fine management of constructed ecosystems. Therefore, more consideration should be given to the ecological service function of urban green infrastructure beyond aesthetics, and less manual intervention, such as pruning and fertilization, should be conducted so as to enhance C benefits.

3.5. Contributions and Limitations

Our paper is unique in that it explores the indirect role of ecosystems in carbon neutrality beyond CO₂ uptake and renews knowledge about the contributions of ecosystems to carbon neutrality. In this study, we developed a comprehensive carbon benefit accounting framework that incorporated indirect reductions in CO₂ emissions from ecosystem services (e.g., temperature regulation and air purification), typically overlooked in conventional carbon accounting methods. By applying this integrated model to a case study in Chengdu, China, our research provides a unique perspective on the important yet undervalued contribution of ecosystem services to carbon neutrality. In addition, this study also used machine learning to simulate the carbon benefits of four different types of ecosystems and analyzed their impact factors. Additionally, this will help inform strategic decisions for sustainable urban development and climate change mitigation.

However, it must be recognized that there are some limitations in this study. Firstly, there are a large number of ecosystem services, and the terrestrial ecosystem C benefit accounting framework constructed in this study only covers three key regulating services (carbon sequestration, temperature regulation, and air purification). In the future, more ecosystem services (e.g., soil conservation and water purification, etc.) should be included in the C benefit assessment framework, and a more comprehensive measure of the contribution of terrestrial ecosystems to achieving carbon neutrality should be provided. Secondly, this study pre-selected some social and natural factors for driver analysis based on the principle of common and easily available factors. However, social and economic activities are complex, and GDP and population density cannot fully reflect the impact of human activities on ecosystem C benefits. Therefore, subsequent studies should expand on the selection of driver indicators to fully understand the key drivers affecting ecosystem carbon efficiency.

4. Conclusions

In this study, we constructed a C benefits assessment framework that integrated multiple ecosystem services. The quantity and value of terrestrial ecosystem C benefits in Chengdu from 2000 to 2020 were estimated as a reference, revealing the relative importance of different ecosystems and ecosystem services for achieving carbon neutrality. Meanwhile, based on XGBoost and SHAP algorithms, we constructed models to explain the driving mechanisms affecting C benefits and proposed urban ecosystem construction and conservation strategies. The results of the study are as follows:

• The contribution of terrestrial ecosystems to carbon neutrality is not limited to CO₂ uptake, but ecosystem regulation services can also indirectly reduce a large amount of CO₂ emissions from this source. In Chengdu City, the CO₂ emissions reduction brought by temperature regulation and air purification services was 62.79 Tg and 0.45 Tg, respectively, far exceeding the 2.87 Tg of CO₂ uptake by carbon sequestration services. Therefore, in addition to traditional carbon sequestration projects, such as afforestation, overall ecosystem services can be enhanced through ecological conservation and restoration projects. The enormous potential of ecosystems in CO₂ emissions reduction should be fully exploited to accelerate the achievement of carbon neutrality.
• The C benefits of Chengdu have improved considerably over the past 20 years, with an average annual growth rate of 0.91 Tg/a. Terrestrial ecosystems have offset 45.41% of anthropogenic CO₂ emissions for Chengdu. In addition, wetlands considerably reduced energy consumption and CO₂ emissions due to their efficient temperature regulation service. The C benefits of wetland ecosystems are much higher than those of other ecosystems. Therefore, in urban centers with a shortage of available land, building micro wetlands is a better way of maximizing C benefits than planting wood or building parks.
• NDVI, GDP, and DEM are the main factors affecting ecosystem C benefits. Among the social development factors, GDP has a considerably higher negative impact on ecosystem C benefits than the population. This is due to the fact that the fine management of green infrastructure in economically developed areas greatly reduces C benefits. Therefore, constructed ecosystems should be allowed to grow naturally and avoid excessive artificial intervention for aesthetic purposes.