Future Reductions in Suitable Habitat for Key Tree Species Result in Declining Boreal Forest Aboveground Biomass Carbon in China

Abstract

China’s forest ecosystem plays a crucial role in carbon sequestration, serving as a cornerstone in China’s journey toward achieving carbon neutrality by 2060. Yet, previous research primarily emphasized climate change’s influence on forest carbon sequestration, neglecting tree species’ suitable area changes. This study combinates the Lund–Potsdam–Jena model (LPJ) and the maximum entropy model (MaxENT) to reveal the coupling impacts of climate and tree species’ suitable area changes on forest aboveground biomass carbon (ABC) in China. Key findings include the following: (1) China’s forests are distributed unevenly, with the northeastern (temperate coniferous broad-leaved mixed forest, TCBMF), southwestern, and southeastern regions (subtropical evergreen broad-leaved forest, SEBF) as primary hubs. Notably, forest ABC rates in TCBMF exhibited a worrisome decline, whereas those in SEBF showed an increasing trend from 1993 to 2012 based on satellite observation and LPJ simulation. (2) Under different future scenarios, the forest ABC in TCBMF is projected to decline steadily from 2015 to 2060, with the SSP5-8.5 scenario recording the greatest decline (−4.6 Mg/ha/10a). Conversely, the forest ABC in SEBF is expected to increase under all scenarios (2015–2060), peaking at 1.3 Mg/ha/10a in SSP5-8.5. (3) Changes in forest ABC are highly attributed to climate and changes in tree species’ highly suitable area. By 2060, the suitable area for Larix gmelinii in TCBMF will significantly reduce to a peak of 65.71 × 10⁴ km² under SSP5-8.5, while Schima superba Gardner & Champ and Camphora officinarum in SEBF will expand to peaks of 94.07 × 10⁴ km² and 104.22 × 10⁴ km², respectively. The geographic detector’s results indicated that the climate and tree species’ suitable area changes showed bi-variate and nonlinear enhanced effects on forest ABC change. These findings would offer effective strategies for achieving carbon neutrality.

1. Introduction

Forests play a pivotal role in global carbon sequestration, significantly contributing to climate change mitigation and sustainability efforts [1,2]. It is noteworthy that forest biomass carbon stock alone constitutes more than 90% of the global vegetation biomass carbon reservoir [3,4]. However, the accurate estimation of forest carbon sequestration remains a challenging scientific issue. Considerable debate persists concerning whether forthcoming forest ecosystems will enhance carbon sequestration within the context of climate variation [5]. Consequently, precise evaluations of forest carbon sequestration hold great importance in appraising strategies for ecosystem management and mitigation policies [6,7].

Estimating China’s forest carbon sequestration potential has remained a focal point of research, given that forest coverage in China reaches 22.96% [8,9]. However, preceding studies have focused on forecasting carbon sink trends and examining the correlation with climate alterations, often lacking in-depth mechanistic investigations [8]. For example, Tong et al. (2018) [10] explored the trend of vegetation aboveground biomass carbon (ABC) and its response to natural drivers in China’s karst region from 1982 to 2015 through integrating satellite data with the Lund–Potsdam–Jena model (LPJ). Meanwhile, leveraging the LPJ model and datasets of global climate models (GCMs), Kong et al. (2022) [2] demonstrated that China’s vegetation ABC is projected to experience an upward trajectory from 2041 to 2060 under the SSP1-2.6 and SSP2-4.5 scenarios, with temperature and precipitation identified as primary influencing factors. While the LPJ model comprehensively considers the dynamic characteristics of the carbon cycle using the theory of vegetation physiology and ecology [2,11], its applicability is limited in explaining key ecological functional outcomes at the species level due to its typical focus on temporal and spatial variations of general vegetation functional types [12].

Climate change drives changes in forest vegetation growth and suitable zones, thereby influencing the spatial pattern of forest carbon sequestration capacity. Nevertheless, current research generally lacks an in-depth understanding of the connection between shifts in tree species’ suitable areas and forest carbon sequestration under changing conditions. For instance, under the background of climate warming, the distribution patterns of the dominant conifer tree species may exhibit a shift towards higher latitudes and altitudes [13,14,15]. Li et al. (2011) and Yu et al. (2011) [16,17] demonstrated that rising temperatures would lead to a decline in spruce growth within the Changbai Mountain Natural Reserve, while other prevalent tree species would experience benefits from the warming climate. Although recent studies have started to consider the impacts of forest type and age on carbon sequestration capacity [18], such investigations are often constrained by experimental scales and data collection limitations. Presently, statistical models rooted in the maximum entropy theory (MaxENT) have gained prominence due to their small sample requirements. These models accurately project species point data to suitable distribution regions on a regional scale through establishing relationships between species data and environmental variables. For instance, Wang and Guan (2023) [19] employed the MaxENT model to investigate the impact of climate change on the suitability of conformation trees. Nevertheless, these studies fall short of delving into the underlying variations in forest carbon sequestration resulting from changes in tree species’ suitable areas. Consequently, the mechanism of climate variation’s effects on species migration and forest carbon sequestration potential remains unclear.

This study aims to bridge the gap in understanding the combined impacts of climate variation and changes in species’ suitable areas on forest carbon sequestration capacity. Through integrating the LPJ model and the MaxEnt model, our research seeks to (1) investigate the temporal and spatial variations in forest ABC across China and (2) explore the coupling effects of climate and tree species’ suitable area changes on forest carbon sequestration potential in China. Through conducting case studies in different forest regions of China, this research aims to provide empirical evidence and enhance our understanding of the complex interplay between climate variation, tree species’ suitable area changes, and the carbon sequestration potential of forests. Ultimately, this study will contribute to the development of targeted strategies for sustainable forest management and carbon neutrality in China.

2. Materials and Methods

2.1. Study Area and Data Sources

China’s complex topography, coupled with variations in temperature, precipitation, and vegetation types, creates a rich tapestry of ecosystems that respond differently to climate drivers. Figure 1 illustrates that China is divided into seven distinct geographical zones. The response of terrestrial ecosystems to climate change varies across these regions. China’s diverse ecology, ranging from the tropical forests in the south to the alpine meadows in the north, offer an ideal setting for investigating the impacts of climate change on ABC [2]. Forests in China occupy approximately 23.57% of the total land area, mainly distributed across the eastern, central, southern, southwestern, and northeastern regions. The typical natural vegetation zone in the northeastern region is temperate coniferous broad-leaved mixed forest (TCBMF), while the southern region is characterized as subtropical evergreen broad-leaved forest (SEBF).

The forest ABC dataset was obtained from a VOD dataset collected by global satellites monthly, with a resolution of 0.25°, covering the years 1993–2012 (http://www.wenfo.org/wald/globalbiomass, accessed on 22 August 2019). The meteorological observation data were collected from 2472 stations with a spatial resolution of 0.5° × 0.5° in China from 1982 to 2014 (https://data.cma.cn/site/index.html, accessed on 31 January 2020). The data on wet days and cloud cover were obtained from the Climate Research Unit (CRU) version TS4.04 datasets spanning the period between 1982 and 2014. These datasets were all provided at a resolution of 0.5° × 0.5° and can be accessed at (https://sites.uea.ac.uk/cru/, accessed on 2 April 2022). The carbon dioxide (CO₂) data were sourced from the CO₂ Program of the Scripps Institution of Oceanography (SIO), accessible through (https://scrippsco2.ucsd.edu/, accessed on 2 April 2022).

Larix gmelinii (L. gmelinii) is one of the commonly encountered species in the boreal forests of TCBMF in Northeast China, playing a significant role in the high carbon storage capacity of the forest [20,21]. In addition, L. gmelinii habitats provide a diverse range of support for various organisms, including endemic species, and any alterations to the habitat would impact their distribution [22]. Similarly, Schima superba Gardner & Champ (S. superba) and Camphora officinarum (C. officinarum) are dominant evergreen broad-leaved tree species widely distributed in China’s SEBF, with a high capacity for CO₂ absorption [23,24,25]. Therefore, L. gmelinii serve as the typical tree species of TCBMF, while S. superba and C. officinarum represent typical tree species for SEBF in this study. The initial geographic points of L. gmelinii, S. superba, and C. officinarum within China were acquired from the Global Biodiversity Information Facility (GBIF, https//www.gbif.org, accessed on 12 May 2023). These data primarily originated from field observations and preserved specimens. Duplicate entries and original records lacking geographic coordinates (longitude or latitude) were excluded. Additionally, only one occurrence point per 1 km × 1 km grid cell was retained. This selection was primarily due to the spatial resolution of bioclimatic variables being 30 arcseconds (equivalent to approximately 1 km), which aimed to mitigate potential clustering effects predicted using the model [19]. Adhering to the input data format of the MaxEnt model, the distribution data for the target species were organized into CSV files, ordered by species name, longitude, and latitude (refer to Figure 1).

The environmental information considered in this research encompassed 19 bioclimatic datasets, elevation data, and 34 soil-related variables. The bioclimatic data, spanning the years 1970–2000, were computed as averages and possessed a spatial resolution of 30 arcseconds. These bioclimatic data and elevation information were sourced from the dataset provided by WorldClim (http://www.worldclim.org, accessed on 28 May 2021). This study took into account the impact of soil attributes on species distribution modeling (SDM) predictions, where a comprehensive set of 34 soil variables was procured from the Harmonized World Soil Database (HWSD v. 1.21). All data were clipped to China for model simulation.

In this study, forthcoming climate datasets (precipitation and temperature) of China were extracted from the BCC-CSM2-MR, MRI-ESM2-0, and CanESM5 global climate models, considering three shared socio-economic pathways (SSP1-2.6, SSP2-4.5, and SSP5-8.5). These three sets of GCMs data, bias-corrected through the Equidistant Cumulative Distribution Functions (EDCDF) [2,26], accurately captured the spatial-temporal variation characteristics of temperature and precipitation in China (Figures S1 and S2). All the data points were averaged over 20-year intervals (from 2041 to 2060). The anticipated future distribution for each species was determined based on the average results generated by the MaxEnt model, utilizing inputs from the three specified GCMs.

To establish an efficient model using a reduced set of variables, all the variables were initially fed into the model for a preliminary run. Variables that collectively contributed to over 90% of the model’s simulation were then retained. Subsequently, a Pearson correlation analysis was conducted on the retained variables. If the correlation between the two variables exceeded 0.8, the variable with a greater contribution rate during the preliminary model run was preserved [27]. The variable selection results of L. gmelinii, S. superba, and C. officinarum are presented in

Table 1. Variable selection results of L. gmelinii, S. superba, and C. officinarum.

Variable	Variable Description	L. gmelinii	S. superba	C. officinarum
Bio2	Mean diurnal range (mean of monthly (max temp–min temp))	√	√
Bio3	Isothermality (bio2/bio7) (×100)	√
Bio4	Temperature seasonality (standard deviation × 100)	√		√
Bio5	Max temperature of the warmest month		√	√
Bio6	Temperature annual range (bio5–bio6)		√
Bio8	Mean temperature of wettest quarter	√
Bio12	Annual precipitation		√
Bio13	Precipitation of the wettest month	√
Bio14	Precipitation of the driest month		√	√
Bio15	Precipitation seasonality (coefficient of variation)		√
Bio17	Precipitation of the driest quarter	√
elev	Elevation	√
t_ph_h2o	Topsoil pH (H2O)		√
t_bs	Topsoil Base Saturation	√
s_ece	Subsoil Salinity (Elco)	√
s_cec_soil	Subsoil CEC (soil)	√
s_teb	Subsoil TEB (Total exchangeable bases)	√

.

2.2. Methods

2.2.1. LPJ-DGVM

The Lund–Potsdam–Jena Dynamic Global Vegetation Model (LPJ-DGVM) serves as a widely utilized computational tool devised for simulating the intricate dynamics of global vegetation under the influence of environmental factors and climate shifts. This model, at the forefront of technology, amalgamates insights from ecology, biogeography, and climatology to replicate the intricate interplays among the atmosphere, land surface, and terrestrial ecosystems [11]. The LPJ-DGVM relied on a process-driven methodology, embodying essential ecological mechanisms like photosynthesis, respiration, carbon allocation, and vegetation evolution. The model effectively captures the multifaceted feedback loops between vegetation and the natural environment through encompassing meticulous depictions of plant physiology, the carbon cycle, and land surface procedures [12]. The embedded physical mechanisms and causalities allow LPJ to provide valuable insights into the potential future trajectories of global vegetation and the associated feedback with climate change.

In the LPJ model, ten distinct plant functional types (PFTs) are featured, each governed by constraining factors that regulate the effective influences of light, temperature, and moisture [28]. Every simulated PFT represents the collective characteristics of the entire population (e.g., tree height and vegetation carbon pools; population-based method) [29]. The simulation process for the LPJ model involves employing climate data spanning 1981 to 2010, allowing for 1000 model years spin-up to establish ecosystem equilibrium. Following the attainment of equilibrium, a dynamic simulation was executed, employing climate data from both 1981 to 2014 and 1981 to 2060 [2]. To support the LPJ model simulations from 1981 to 2014, observed meteorological data such as temperature and precipitation, along with the CRUTS 4.04 climate dataset containing wet days and cloud cover information, were incorporated. Furthermore, the annual average atmospheric CO₂ concentration was integrated.

2.2.2. MaxEnt Model

Using MaxEnt (version 3.4), the training dataset encompassed 75% of the occurrence data, while the remaining 25% was allocated for testing purposes [19,30]. The algorithm underwent 1000 iterations of these steps. This process was iterated 10 times to yield the mean simulation outcomes of tree species’ geographical distributions [31]. Subsequently, these results were consolidated within ArcGIS 10.5 and transformed into a raster format for subsequent analysis.

Model performance was assessed through the utilization of the receiver operating characteristic (ROC) curve and the corresponding area under the ROC curve (AUC), a metric that varied from 0 to 1 [27,31]. Evaluation of the model’s performance was typically categorized into five grades as follows: failure (0.5–0.6), moderately accurate (0.6–0.7), reasonably accurate (0.7–0.8), highly accurate (0.8–0.9), and exceptionally accurate (0.9–1) [32]. According to the AUC results generated from the MaxEnt model, the average AUC values of 10 repetitions for L. gmelinii, S. superba, and C. officinarum were 0.819, 0.928, and 0.914, respectively, indicating that the model yielded highly reliable results in predicting the potential distribution (Figure S3).

The MaxEnt model’s output for species distribution was presented in ASCII format, with values ranging from 0 to 1 [27]. Employing the natural breaks (Jenks) methodology within ArcGIS, the suitability levels for L. gmelinii, S. superba, and C. officinarum were divided into five categories: unsuitable habitat (<0.10; <0.08; <0.10), poorly suitable habitat (0.10–0.28; 0.08–0.27; 0.10–0.27), moderately suitable habitat (0.28–0.50; 0.27–0.48; 0.27–0.48), and highly suitable habitat (0.50–1.00; 0.48–1.00; 0.48–1.00), respectively.

2.2.3. Statistical Methods

The Theil–Sen trend analysis approach was applied to compute long-term trends on a pixel-by-pixel basis. The Theil–Sen trend analysis method is a robust non-parametric estimation algorithm that effectively mitigates the impact of outliers in extended time-series analyses, representing an enhancement over the least squares linear regression method [33]. The calculation formula is provided below:

$$\mathrm{s}\mathrm{l}\mathrm{o}\mathrm{p}\mathrm{e}=\mathrm{m}\mathrm{e}\mathrm{d}\mathrm{i}\mathrm{a}\mathrm{n}\left(\frac{x_j-x_i}{j-i}\right),i<j\le n$$

(1)

where “slope” signifies the rate of change for the particular factor; $i$ and j denote the respective years; $n$ refers to the length of the time series; and $x_i$ and $x_j$ represent the values associated with the corresponding years. Ordinarily, a slope of 0 indicates the absence of change in the variable. However, in practice, instances of a slope precisely equaling 0 may be rare. Consequently, to account for this, a very small slope value is considered, implying negligible change. Thus, we establish a threshold interval of [−0.0005–0.0005] to signify no change. An affirmative slope value (>0.0005) signifies an ascending trend, while a negative value (<−0.0005) signifies a descending trend.

The Theil–Sen trend analysis is frequently paired with the Mann–Kendall test, a nonparametric statistical method, to evaluate long-term data trends. This approach was initially introduced by Mann in 1945 and subsequently improved by Kendall and Sneyers. This method does not require normal distribution or a linear trend and is not affected by missing values or outliers. This approach had found broad application in assessing trends in extended time-series data [33,34]. The test statistic $\mathrm{S}$ was computed using the following formula:

$$\mathrm{S}={\sum }_{i=1}^{n-1}{\sum }_{j=i+1}^{n}sign(x_j-x_i)$$

(2)

$$\mathrm{s}\mathrm{i}\mathrm{g}\mathrm{n}(v)=\left\{\begin{array}{c}1 for v>0\\ 0 for v=0\\ -1 for v<0\end{array}\right.$$

(3)

The variance of $\mathrm{S}$ was:

$$\mathrm{v}\mathrm{a}\mathrm{r}\left(\mathrm{S}\right)=\frac{n(n-1)(2n+5)}{18}$$

(4)

Statistic $\mathrm{Z}$ was defined as:

$$\mathrm{Z}=\left\{\begin{array}{c}\frac{S-1}{\sqrt{var}} for S>0\\ 0 for S=0\\ \frac{S+1}{\sqrt{var}} for S<0\end{array}\right.$$

(5)

where |$\mathrm{Z}$| ≥ 1.96 signifies that the trend has achieved statistical significance at the 95% confidence level.

Geodetector has proven to be an effective technique for the comprehensive analysis of the interplay between independent and cooperative variables concerning dependent variables [35,36]. The Geodetector model encompasses components such as the factor detector, interaction detector, risk detector, and ecological detector. In this study, we employed the factor detector and interaction detector to investigate the impact of climate factors and tree species’ suitable area changes on forest ABC changes in China.

(1) The factor detector systematically measures the degree to which driving factor “X” elucidates the spatial disparities in forest ABC via the q-statistic value:

$$q=1-\frac{{\sum }_{h=1}^L{N}_h{\delta }_h^2}{{N\delta }^2}$$

(6)

where q symbolizes the explanatory potency of the influencing factor concerning the temporal and spatial shifts in ABC, $h$ signifies the categorization levels of a driving factor, $L$ represents the sample size of the impact factor, $N_h$ and $N$ stand for the counts of units within layer $h$ and the entire region, respectively, and ${\delta }_h^2$ and ${\delta }^2$ denote the variances of $h$ and the overall region, respectively.

(2) The interaction detector was designed to identify the way two driving factors interact and affect the ABC. Firstly, we calculated the q values of two driving factors that affect ABC, namely q(X1) and q(X2). Subsequently, we calculated the q values of the interactive effect (q(X1∩X2)) and compared them with q(X1) and q(X2) to identify the interaction type between the two driving factors (

Table 2. Definition of the different interaction types in the Geodetector model.

Interaction Relationship	Interaction Types
q(Xi∩Xj) < Min(q(Xi), q(Xj))	Nonlinear-weaken
Min(q(Xi), q(Xj)) < q(Xi∩Xj) < Max(q(Xi), q(Xj))	Uni-variable weaken
q(Xi∩Xj) = q(Xi) + q(Xj)	Independent
Max(q(Xi), q(Xj)) < q(Xi∩Xj) < q(Xi) + q(Xj)	Bi-variable enhanced
q(Xi∩Xj) > q(Xi) + q(Xj)	Nonlinear-enhanced

).

3. Results

3.1. Satellite-Observed and Simulated Forest ABC in China

Satellite observations of the spatial distribution of aboveground biomass carbon in Chinese forests reveal that from 1993 to 2012, regions with high carbon density were located in the SEBF and TCBMF. Among these regions, North, Northeast, and the southern part of Southwest China exhibited carbon densities exceeding 70 Mg/ha. The LPJ model driven by observed data, including temperature and precipitation, successfully captured the spatial distribution of aboveground carbon stocks in forests. However, overestimation was observed in the Northeast region, as well as in Central, South, East, and West China. The forest carbon stocks simulated using CMIP6-driven models showed good consistency with those driven by observed data, but underestimated carbon stocks were observed in North China and the southeastern coastal areas (Figure 2).

The LPJ model simulation results showed that the areas with high forest ABC in China from 1982 to 2014 were mainly distributed in the SEBFs and TCBMFs of China (Figure 3a). However, these forest areas with high ABC experienced a decreasing trend, and the area with a significant decreasing trend accounted for 11.56%, mainly distributed in Northeast China and Southwest China. Approximately 40.15% of the forest ABC showed a significant increasing trend, mainly in southwest, central, east, and northern China (Figure 3b). The results of the average annual forest ABC in China clearly show that the overall ABC in the study area increased at a growth rate of 0.19 Mg/ha/a before 1992 and then changed to a decreasing trend (−0.03 Mg/ha/a). The overall trend in TCBMF was similar to that in the study area, and the decline rate is more obvious (−0.17 Mg/ha/a) after 1992. Meanwhile, the forest ABC in SEBF showed a slower growth rate (before 2000: 0.12 Mg/ha/a; after 2000: 0.01 Mg/ha/a).

The results of simulations using three different models and a multiple-model ensemble (MME) to drive the LPJ model to calculate the ABC trends in China from 2015 to 2060 under different SSP scenarios are shown in Figure 4. These simulations consistently indicate an overall increasing trend in forest ABC from 2015 to 2060. Especially in the SSP5-8.5 scenario, all models and the MME predicted a higher proportion of pixels with significantly increased forest ABC compared to other scenarios. The forest ABC of the TCBMF will significantly decrease in all scenarios, while forest ABC of the SEBF generally exhibit a significant increasing trend, except in the SSP1-2.6 scenario. The MME results show that under the SSP1-2.6, SSP2-4.5, and SSP5-8.5 scenarios, the proportion of pixels with a significantly increasing trend in forest ABC will be 46.16%, 67.18%, and 75.53%, respectively. Conversely, the proportion of pixels with a significantly decreasing trend will be 10.49%, 12.93%, and 16.09%, respectively. Under the SSP5-8.5 scenario, the BCC model results exhibit the highest proportion of pixels with a significantly increasing trend in forest ABC at 80.34%, while the proportion with a significantly decreasing trend will be lowest at 5.90%. The Can model results show that under the SSP1-2.6 scenario, the smallest proportion of forest ABC significantly increasing will be 29.53%, whereas, under the SSP5-8.5 scenario, the largest proportion of significantly decreasing forest ABC will be 19.63%. Overall, from 2015 to 2060, the forest ABC trajectory in TCBMF is expected to experience a continuous decrease, in stark contrast to SEBF, which is projected to undergo a significant increase.

Under different scenarios, the statistical results of annual forest ABC in different regions of China from 2015 to 2060 are shown in Figure 5. Overall, the rates of change in forest ABC in China are expected to remain stable. However, there will be notable differences in various regions across the country. In the TCBMF area, the forest ABC change rate will be consistently declining across various scenarios. The most significant rate of decreae is projected to be −0.46 Mg/ha/a under the SSP5-8.5 scenario. In contrast, the rate of change in forest ABC in the SEBF region is expected to increase in all scenarios. The maximum change rate will be 0.13 Mg/ha/year in the SSP5-8.5 scenario (Figure 5).

3.2. Potential Suitable Area Shifts of Typical Tree Species in Response to Climate Change

Figure 6 illustrates the current potential geographic distribution of the three typical species. The statistical results showed that the high-, moderate-, and low-suitability areas for L. gmelinii are currently estimated to cover an area of 133.47 × 10⁴ km², 145.31 × 10⁴ km², and 223.58 × 10⁴ km², respectively. The high-suitability areas for L. gmelinii are mainly distributed in the northern part of North China and the Northeast region. Under different scenarios, the highly suitable area of L. gmelinii is projected to decrease, with a reduction to 50.56 × 10⁴ km² by 2060 under the SSP5-8.5 scenario. The total suitable area is expected to decrease from 475.25 × 10⁴ km² in the current period to 375.29 × 10⁴ km² in 2060 under the SSP5-8.5 scenario (Figure 6a–d,

Table 3. Statistical results of potentially suitable area (unit: 10⁴ km²).

Species	Periods	Scenarios	Low	Moderate	High	Total
L. gmelinii	Current		223.58	145.31	133.47	475.25
	2041–2060	SSP1-2.6	246.86	102.60	67.81	417.27
		SSP2-4.5	241.41	96.62	58.45	396.48
		SSP5-8.5	234.24	90.49	50.56	375.29
S. superba	Current		98.20	63.55	49.96	211.71
	2041–2060	SSP1-2.6	81.85	62.68	114.47	259.00
		SSP2-4.5	88.57	66.48	113.25	268.30
		SSP5-8.5	79.09	59.57	136.97	275.63
C. officinarum	Current		93.37	76.79	60.41	230.57
	2041–2060	SSP1-2.6	84.85	55.75	141.78	282.38
		SSP2-4.5	91.86	56.63	140.98	289.47
		SSP5-8.5	92.86	52.13	157.90	302.89

).

In the current period, it was estimated that the high-, moderate-, and low-suitability areas for S. superba are 49.96 × 10⁴ km², 63.55 × 10⁴ km², and 98.20 × 10⁴ km², respectively. The high-suitability area is primarily located in the southern region, with expansions expected in the future to cover all of Central China, East China, and the eastern part of Southwest China. Under the scenarios of SSP1-2.6, SSP2-4.5, and SSP5-8.5, the high-suitability area for S. superba will expand in 2060. Under the scenario of SSP5-8.5, the expansion of the high-suitability area will be the largest, with an expected expansion of 136.97 × 10⁴ km² in 2060 (Figure 6e–h, Table 3).

For C. officinarum, the currently suitable areas were estimated to be 60.41 × 10⁴ km², 76.79 × 10⁴ km², and 93.37 × 10⁴ km² for the high-, moderate-, and low-suitability areas, respectively. The high- and medium-potential suitable zones for C. officinarum are mainly distributed in Northeast China and North China. The high- and moderate-suitability areas for C. officinarum are primarily distributed in Central China, East China, and South China. By 2060, the area of highly suitable areas for C. officinarum in the SSP1-2.6 and SSP2-4.5 scenarios will increase to 141.78 × 10⁴ km² and 140.98 × 10⁴ km², respectively, with a maximum of 157.90 × 10⁴ km² in the SSP5-8.5 scenario (Figure 6i–l, Table 3).

We further examined the potential changes in the distribution of suitable areas for different tree species based on the distribution of high-suitability areas in both current and future scenarios. (Figure 7). The results indicate that the future distribution of the high-suitability zone for L. gmelinii is projected to shrink primarily in North China and the Northeast region, with the greatest reduction in area under the SSP5-8.5 scenario in 2060, reaching up to 65.71 × 10⁴ km². As for S. superba and C. officinarum, they will both tend to expand towards the central regions of Central China and East China, as well as the eastern part of Southwest China, under all future scenarios. The expansion is most significant under the SSP5-8.5 scenario, with an increase in area of 94.07 × 10⁴ km² for S. superba and 104.22 × 10⁴ km² for C. officinarum (Figure 7,

Table 4. Statistical results of area change of potential highly suitable areas (unit: 10⁴ km²).

Species	Scenarios	Range Expansion	No Occupancy	No Change	Range Contraction
L. gmelinii	SSP1-2.6	3.97	823.86	68.86	49.05
	SSP2-4.5	2.80	825.03	60.29	57.61
	SSP5-8.5	2.28	825.54	52.20	65.71
S. superba	SSP1-2.6	70.12	819.28	55.42	0.18
	SSP2-4.5	68.67	820.73	55.44	0.16
	SSP5-8.5	94.07	795.33	55.46	0.14
C. officinarum	SSP1-2.6	87.00	793.95	66.65	0.03
	SSP2-4.5	86.04	794.91	66.60	0.09
	SSP5-8.5	104.22	776.73	66.66	0.02

).

3.3. Coupling Impacts of Climate and Species’ Suitable Area Change on Forest ABC

The results of the superposition of the ABC changes simulated using the LPJ model and the changes in the suitable areas of dominant tree species simulated using the Maxent model from 2015 to 2060 are shown in Figure 8. Under SSP1-2.6 scenario, high-suitability areas of L. gmelinii will decrease in some TCBMF areas in Northeast China, but forest ABC will still increase, accounting for 20.46% of the total area. Meanwhile, the increase in forest ABC in most SEBF areas will be consistent with the single or co-increase in the high-suitability areas of S. superba and C. officinarum, which account for 12.66%, 13.54%, and 24.55%, respectively (Figure 8a,d). Under the scenarios of SSP2-4.5 and SSP5-8.5, the main theme in the TCBMF region will be more evident in the reduction in forest ABC and the shrinking of highly suitable areas for dominant tree species, with an area proportion of 18.23% and 21.96%, respectively. The synergistic trend between the increase in forest ABC and the expansion of highly suitable habitat for dominant tree species in the southern SEBF region will be even more pronounced, with the proportion of this trending area accounting for 63.24% and 65.10% under the SSP2-4.5 and SSP5-8.5 scenarios, respectively (Figure 8b–d).

The individual effects of the driving factors behind forest ABC variations were analyzed using factor detectors (Figure 9). For Northeast China, climate factors have a relatively strong ability to explain forest ABC, especially temperature, followed by precipitation and the distribution of L. gmelinii. The q value of temperature interpretation will reach 0.224 in the scenario of SSP1-2.6. For the southern parts of China, the temperature will still be the driving factor with the highest explanatory power (0.094) for forest ABC in the SSP1-2.6 scenario, while in the SSP2-4.5 scenario, the q value of the explanatory power for forest ABC shows the order of C. officinarum, precipitation, S. superba, and temperature (Figure 9b).

The results of synergistic detection among the factors showed that the effects of the interaction between temperature and precipitation and L. gmelinii on forest ABC in northeast China will be bi-enhanced and nonlinear enhanced in the SSP1-2.6 and SSP2-4.5 scenarios, respectively (Figure 10a,b). In the SSP5-8.5 scenario, precipitation has an effect on the ABC of northeastern forests through the interaction of bi-enhanced, with temperature, and nonlinear enhanced interaction with L. gmelinii, respectively. Temperature and L. gmelinii also showed a bi-enhanced interaction affecting the northeastern forest ABC (Figure 10c).

In the SSP1-2.6 and SSP5-8.5 scenarios, temperature, precipitation, S. superba, and C. officinarum will affect the forest ABC in southern China through the synergistic effect of nonlinear enhancement between each two (Figure 10d,f). In the SSP2-4.5 scenario, the bi-enhanced synergistic effect between pairwise factors will have a stronger effect on forest ABC (Figure 10e).

4. Discussion

The satellite dataset showed that high values of forest ABC in China from 1993 to 2012 were mainly distributed in North China, Northeast China, and Southwest China, especially in northeastern and southwestern China (>70 Mg/ha), and forest ABC increased at a rate of approximately 0.37%yr⁻¹, which is aligned with the results of Chen et al. [6] and Tang et al. [1], who integrated various types of remote sensing observations and field measurements to analyze the spatial distribution and temporal trends of forest ABC using regression analysis and machine learning methods. The mean value of satellite observation of forest ABC (37.58 Mg/ha) was lower than Masson pine with 18-year-old stands (86.0 Mg/ha of total biomass carbon) [37]. Meanwhile, simulated forest ABC (62.87 Mg/ha) was greater than Masson pine with 28-year-old stands (112.70 Mg/ha of total biomass carbon). The main reasons for the differences in results are primarily attributed to variations in carbon sequestration capabilities among different species, disparities within the same species at different ages, and human-induced factors of interference [37]. To enhance the accuracy of remote sensing retrieval or model simulation of forest biomass carbon, it is essential to incorporate considerations for vegetation species diversity, stand age, and human-induced disturbances.

The forest ABC of TCBMF in Northeast China showed a significant downward trend from 1993 to 2012, while the ABC of SEBF in South China showed a continuous upward trend. Combining the LPJ model and GCM data, the future forest ABC indicated a significant overall increase in ABC in SEBF regions under all SSP1-2.6, SSP2-4.5, and SSP5-8.5 conditions from 2015 to 2060, which was consistent with the results obtained by Kong et al. (2022) [2]. This phenomenon would be attributed to the higher precipitation and temperature conditions in SEBF regions, which can meet the requirements for vegetation growth [2]. Liu et al. (2015) [38] believed that tropical deforestation declines in China had also contributed to the increase in forest ABC in the region. Furthermore, the forest ABC in TCBMF of China will experience a continuing significant downward trend from 2015 to 2060. Sun et al. (2022) [39] also obtained similar results regarding the changes in forest carbon stocks in the Northeast region of China using the INVEST model based on land use and land cover changes. The main reasons for these changes are associated with the conversion of forest land to agricultural land and fluctuations in the economic timber forests of the Lesser Khingan Mountains. In addition, common forest pests and diseases in temperate forests, wildfires, and the degradation of old-growth forests may also lead to a decline in the forest’s carbon sequestration potential in this region [38].

Understanding the impact of climate change on forests’ carbon sequestration potential needs to be linked to changes in areas suitable for tree species. Our results showed that the future distribution of the highly suitable area for coniferous species (L. gmelinii) is projected to shrink primarily in Northeast China, while evergreen broad-leaved tree species (S. superba and C. officinarum) will tend to expand towards the central regions of China from 2015 to 2060 based on the MaxEnt model. Similar conclusions have also been reported, for example, by Bai et al. (2016) and Chen et al. (2022) [40,41], who found, through research on the relationship between tree rings and temperature changes, that rapid warming will lead to the gradual decline of L. gmelinii in Northeast China. Du et al. (2022) [42] found that the future suitable habitat area of eight species of tree species in the Northeast region, including L. gmelinii, will decrease by 10%–30%. Wu et al. (2012) [43] found that climate warming caused a significant decrease in the growth of L. gmelinii in semi-northern coniferous forests during the last two decades. These findings are consistent with ours, but such studies rarely explore the impact of changes in species’ suitable areas on forest ABC. Combining climate change, species’ suitable area changes and forest ABC changes can enhance our understanding of the carbon sequestration potential of forests.

Previous research has primarily revealed changes in the carbon sequestration potential of China, and emphasized climate change’s influence on them, but ignored the effects of changes in tree species’ suitable area [43,44,45]. Overlay analysis of the LPJ and MaxENT models showed that the future trends of forest ABC reduction in TCBMF and increase in the SEBF region of China are spatially consistent with the shrinking of highly suitable areas for L. gmelinii in TCBMF and the expansion of highly suitable areas for S. superba and C. officinarum in the SEBF region. The outcomes of the Geodetector model found that climate and tree species’ highly suitable area changes showed a bi-variable and nonlinear synergistic enhancement of forest ABC change [46]. Generally, warming and humidification of the climate are conducive to the growth of vegetation, but this has species and space differences [47]. Larch pollen is negatively correlated with temperature [48]. For the Northeast region, which is dominated by coniferous species such as Larch, a reduction in Larch pollen leads to an insufficient supply of young species under warming conditions, coupled with the degradation of old-aged coniferous forests in the region, resulting in a decline in forest ABC. In addition, the thawing of permafrost in Northeast China due to climate warming may provide more water and space for plants in the short term [40,49], but as warming intensifies in the upcoming decades (0.36 °C per decade), the thawing of permafrost will lead to a long-term reduction in the coniferous forest ecosystem [50], leading to forest fragmentation and forest loss and reducing the potential of forests to sequester carbon [51]. Notably, the reduction in the carbon sequestration potential of forests in the Northeast may be short-term. Temperate broad-leaved species are expanding into northern forest patches, causing the Northeastern coniferous forest to retreat and be replaced by deciduous broad-leaved forests [40,48,52,53]. The carbon sequestration potential of forests in the Northeast remains uncertain in the long term, as some relatively uncommon species will become more common under climate change. Therefore, the long-term carbon sequestration potential of forests in the Northeast would depend on newly established tree species.

Despite this study innovatively combines embedded physics-based models with statistical models, revealing changes in China’s future forest carbon sequestration potential and its relationship with climate change and tree species’ suitable area shifts, limitations still exist within this study. Firstly, while this study considered the impact of climate and vegetation on carbon sequestration potential, it did not account for factors like forest fires, pests, diseases, stand age, and the effects of carbon dioxide fertilization [54]. Secondly, due to model limitations, this research did not include forest belowground biomass carbon, a crucial component of the forest ecosystem’s carbon composition [4,55]. Additionally, apart from the inherent uncertainties of the LPJ and MaxEnt models, the accuracy of future meteorological data, which served as input for both models, as well as the correction of biases, requires further improvement. Despite these limitations, this study introduces a novel approach to assessing the relationship between forest carbon sequestration potential, climate change, and shifts in tree species’ suitable distribution areas.

5. Conclusions

Based on GCM model data coupled with the LPJ model and MaxENT model, this study predicted the distribution of forest ABC and suitable areas for typical tree species in China and explored the impact of climate variations and changes in the suitable area for tree species on forest ABC in China from 2015 to 2060. The main results are as follows:

(1) The geographical distribution of forests in China has spatial differences, with the northeast (TCBMF), southwest, and southeast (SEBF) being the predominant hubs of distribution. However, the forest ABC generated from the LPJ model clearly showed that the overall ABC in the study area increased at a growth rate of 0.19 Mg/ha/a before 1992, and then changed to a decreasing trend (−0.03 Mg/ha/a). Meanwhile, a clear and concerning downward trajectory was noticeable in the forest ABC in TCBMF (before 2000: 0.38 Mg/ha/a, after 2000: −0.17 Mg/ha/a), while that of SEBF showed an increasing trend (before 2000: 0.12 Mg/ha/a, after 2000: 0.01 Mg/ha/a) after 2000 based on satellite observation and LPJ simulation.

(2) The trajectory of forest ABC in TCBMF is set to undergo a sustained decline with a maximum decline rate of −0.46 Mg/ha/a under the SSP5-8.5 scenario. In stark contrast, the forest ABC in SEBF is anticipated to experience a notable upswing at a rate of 0.13 Mg/ha/a under SSP5-8.5. Under the SSP5-8.5 scenario, the proportion of pixels with significantly increased or decreased forest ABC is higher compared to other scenarios. Among them, the BCC model results will have the highest proportion of pixels with a significantly increasing trend in forest ABC at 80.34%, whereas the Can will have the largest proportion of significantly decreasing forest ABC at 19.63%.

(3) The variations in forest ABC are highly attributed to climate and changes in tree species’ highly suitable area. By 2060, the suitable area for L. gmelinii in TCBMF will be significantly reduced to a maximum of 65.71 × 10⁴ km² under SSP5-8.5, while S. superba and C. officinarum in SEBF will expand to peaks of 94.07 × 10⁴ km² and 104.22 × 10⁴ km², respectively. Spatially, the decrease in forest ABC in TCBMF will be consistent with the reduction in highly suitable areas of L. gmelinii, while the increase in forest ABC in SEBF regions will be consistent with the expansion of S. superba and C. officinarum. Climate and tree species’ suitable area changes showed bi-variable and nonlinear synergistic enhancement of forest ABC change based on the geographic detector’s results. The findings of this work can provide valuable insights for developing effective sustainability management strategies under a carbon neutrality context.