Nationwide Decline of Wet Sulfur Deposition in China from 2013 to 2023

Abstract

Atmospheric sulfur (S) deposition, a key component of acid deposition, poses risks to ecosystems, human health, and sustainable development. In China, decades of coal-dominated energy use caused severe S pollution, but recent emission-control policies and energy restructuring have sought to reverse this trend. However, the effectiveness and regional differences in these measures remain insufficiently quantified. Here, we combined continuous observations from 43 monitoring sites (2013–2023), satellite-derived SO₂ vertical column density, and multi-source environmental datasets to construct a high-resolution record of wet S deposition. A random forest model, validated with R² = 0.52 and RMSE = 1.2 kg ha⁻¹ yr⁻¹, was used to estimate fluxes and spatial patterns, while ridge regression and SHAP analysis quantified the relative contributions of emissions, precipitation, and socioeconomic factors. This framework allows us to assess both the environmental and health-related sustainability implications of sulfur deposition. Results show a nationwide decline of more than 50% in wet S deposition during 2013–2023, with two-thirds of sites and 95% of grids showing significant decreases. Historical hotspots such as the North China Plain and Sichuan Basin improved markedly, while some southern provinces (e.g., Guizhou, Hunan, Jiangxi) still exhibited high deposition (>20 kg ha⁻¹ yr⁻¹). Over 90% of the reduction was attributable to emission declines, confirming the dominant effect of sustained policy-driven measures. This study extends sulfur deposition records to 2023, demonstrates the value of integrating ground monitoring with remote sensing and machine learning, and provides robust evidence that China’s emission reduction policies have delivered significant environmental and sustainability benefits. The findings offer insights for region-specific governance and for developing countries balancing economic growth with ecological protection.

1. Introduction

Atmospheric sulfur (S) deposition is a major component of acid deposition and a significant atmospheric pollutant and has profound impacts on ecosystem structure and function [1], and further threatens the sustainable development of ecosystems and human societies. When the level of S deposition exceeds the critical load—the maximum input an ecosystem can tolerate without adverse effects—it can trigger irreversible ecological damage. For example, excessive S deposition leads to the leaching of nutrient ions and disruption of soil acid-base balance, resulting in soil acidification [2,3]. It can also damage plant root systems, impair stomatal structure, and reduce enzymatic activity [4,5], thereby further compromising biodiversity, ecosystem productivity, and stability [6]. On the other hand, S deposition also exerts profound impacts on human society, such as corroding buildings and materials, and endangering respiratory health. In conclusion, these adverse effects pose significant threats to sustainable development by impacting ecosystem stability, public health, atmospheric chemistry, and energy structure transformation. They are closely linked to key Sustainable Development Goals (SDGs), especially SDG 15 (life on land), SDG 3 (good health and well-being), SDG 13 (climate action), and SDG 7 (affordable and clean energy) [7].

Excessive sulfur deposition was primarily driven by the widespread burning of high-sulfur fossil fuels, industrial emissions, and metal smelting activities, which released large amounts of sulfur dioxide (SO₂) into the atmosphere. These anthropogenic sources accounted for over 70% of global SO₂ emissions, with coal combustion alone contributing nearly 50% [8,9]. With the implementation of long-term emission reduction strategies, such as cleaner fuel use, flue-gas desulfurization, and stricter industrial regulations, S deposition levels have gradually declined in many industrialized regions [10,11]. Nevertheless, in rapidly developing and ecologically sensitive areas, sulfur deposition remains a pressing environmental and health concern [12,13].

Building on these global trends, we next focus on China, where rapid industrialization and policy responses provide a critical testbed for assessing deposition dynamics. In China, reliance on a coal-dominated energy structure and rapid industrialization resulted in sustained high SO₂ emissions over the past decades, placing substantial S deposition loads and associated ecological pressures on many regions [14,15]. To achieve sustainable development goals, since the mid-2000s, China has implemented a series of pollution control and emission reduction policies to balance socioeconomic development with environmental pollution control, including the adoption of flue gas desulfurization (FGD), adjustments in energy and industrial structures, and the enactment of the Air Pollution Prevention and Control Action Plan (2013) [16]. These measures have effectively alleviated the pressure of pollutant emissions. Notably, the widespread implementation of FGD in the energy sector after 2005 contributed to a significant decline in national SO₂ emissions from 2006 [17,18]. Existing studies utilizing observations, remote sensing, and modeling have demonstrated a concomitant decreasing trend in S deposition across China [19,20], corresponding to the reduction in SO₂ emissions [21], with some ecosystems showing partial recovery and improvement [22]. However, the spatiotemporal patterns and dynamics of S deposition in recent years—particularly since 2020—remain inadequately quantified. It is still uncertain whether this declining trend has persisted during the most recent years in response to ongoing emission changes, such an assessment is crucial for evaluating the socio-environmental benefits of these policies and informing future sustainable governance strategies.

In this study, we focus on wet S deposition, the dominant component of total S deposition. We integrate multi-source data, including site observations collected from 2013 to 2023 by the China Wet Deposition Observation Network (ChinaWD) [23] and satellite remote sensing. We apply a machine learning model to construct a gridded S deposition dataset for China covering 2013–2023. Based on this, we comprehensively assess the spatial patterns and temporal trends of S deposition over the past decade and evaluate the relative contributions of emission reductions and climate variability. Furthermore, ridge regression and SHAP feature importance analysis are employed to quantitatively examine the impact of these factors on deposition changes, providing insights into maximizing the environmental and health benefits of sulfur emission control policies. This study aims to provide updated evidence for assessing the effectiveness of sulfur emission reduction policies and offers a scientific basis for environmental governance and sustainable development.

2. Materials and Methods

2.1. Data Sources

This study develops a gridded sulfur deposition dataset for China (2013–2023) by integrating site observations, satellite remote sensing, meteorological, and socioeconomic data. Spatiotemporal patterns and driving factors are systematically analyzed. Details of the data sources and characteristics are provided below.

The wet S deposition observations were obtained from the China Wet Deposition Observation Network (ChinaWD). Established in 2013 under the China Ecosystem Research Network (CERN) and additional monitoring stations, ChinaWD comprises 43 sites distributed across 22 provinces and 8 ecological regions, covering the major ecosystem types in China, including forests, grasslands, deserts, wetlands, and croplands. The stations were selected to provide a spatially representative coverage of China rather than focusing solely on areas with high sulfur deposition. For detailed station information and spatial distribution, refer to our previously published article [24]. During 2013–2023, precipitation was collected at each site using three plastic buckets placed 1.5 m above the ground, which were deployed only during precipitation events and removed afterwards. Sampling was conducted 3–5 times per month, and equal volumes were combined to produce monthly samples, which were stored in polyethylene bottles at −20 °C to prevent chemical degradation and microbial transformation. Precipitation data were obtained from manual observations at each station. In the laboratory in Beijing, insoluble particles were first removed by passing the samples through 0.45 μm filters, and sulfate (SO₄²⁻) concentrations were then measured using inductively coupled plasma optical emission spectrometry (ICP-OES). ICP-OES enables the simultaneous determination of more than seventy metallic and non-metallic elements, and offers distinct advantages for large-scale sulfate analysis, including high throughput, wide linear range, minimal chemical interference, and excellent accuracy [25,26]. Quality control was ensured through the use of parallel samples, blanks, and certified reference materials to guarantee analytical accuracy and precision [27]. Wet deposition flux was calculated by multiplying the ion concentration in samples by the precipitation amount, with all values uniformly expressed on an S basis, i.e., as SO₄²⁻-S (kg ha⁻¹ yr⁻¹).

Meteorological data were obtained from the CRU TS v4.09 dataset [28]. The dataset is produced by interpolating global meteorological station observations and provides monthly gridded data since 1901 with a spatial resolution of 0.5°. The variables used in this study include precipitation (P), temperature (T), wind speed (WS), vapor pressure (VAP), and radiation (R). The land surface variables used in this study included land use/land cover (LULC), enhanced vegetation index (EVI), and digital elevation model (DEM). LULC data were obtained from the ESA CCI product (https://cds.climate.copernicus.eu/, accessed on 4 May 2025), which provides annual global land cover data since 1992 with a spatial resolution of 300 m. EVI data were obtained from the MOD13A3 product (https://ladsweb.modaps.eosdis.nasa.gov/, accessed on 3 May 2025), which provides monthly global vegetation index data since 2000 with a spatial resolution of 1 km. DEM data were obtained from the SRTM digital elevation product (https://www.earthdata.nasa.gov/, accessed on 4 May 2025), with a spatial resolution of 90 m. Socioeconomic data included population (POP), gross domestic product (GDP), waste collection and transport (TRS), harmless disposal of municipal waste (HAR), the human footprint index (HFP), energy consumption (E), and coal consumption (C). POP and GDP data were obtained from the World Bank database. TRS and HAR data were obtained from the China Urban-Rural Construction Statistical Yearbook. HFP data were obtained from the global human footprint dataset [29]. E and C data were obtained from the China Statistical Yearbook.

Pollutant emission and atmospheric concentration data included SO₂ emissions (SO2-E), SO₂ VCD (SO2-C), PM_2.5 emissions (PM25-E), PM_2.5 concentration (PM25-C), NH₃ emissions (NH3), NO₂ VCD (NO2), NO_x emissions (NOX), and emissions of volatile organic compounds (VOC). Several emission datasets were obtained from the Multi-resolution Emission Inventory for China (MEIC) [21,30], a high-resolution emission inventory developed by Tsinghua University. SO2-C and NO2 data were obtained from the Aura/OMI OMSO2e L3 product (https://disc.gsfc.nasa.gov/datasets/, accessed on 4 May 2025) [31], which provides daily global gridded data since October 2004 with a spatial resolution of 0.25°. PM25-C data were obtained from the NASA MERRA-2 reanalysis (https://gmao.gsfc.nasa.gov/, accessed on 4 May 2025), which integrates satellite observations, surface measurements, and chemical transport modeling. PM_2.5 and PM₁₀ concentrations are partly related to SO_x and NO_x due to common emission sources and atmospheric chemical processes.

For datasets with daily or monthly temporal resolution, the data were aggregated into annual means to obtain yearly averages. For datasets with missing years in the time series, linear temporal interpolation was applied to complete the full sequence for 2013–2023. This method was selected for its simplicity and suitability for short gaps where no abrupt changes are expected. We note that linear interpolation assumes a steady change between adjacent years and may not capture sudden fluctuations, which could introduce minor biases. Subsequently, all data were resampled to a spatial resolution of 0.1° using bilinear interpolation, clipped to the land area of China, and compiled into a unified dataset on an annual timescale for 2013–2023. Due to insufficient data, Taiwan was not included in our assessment and analysis.

2.2. Random Forest Model

In this study, the random forest (RF) model was applied to construct the gridded S deposition dataset. RF is a non-parametric ensemble learning method that integrates multiple decision trees to improve prediction accuracy and reduce the risk of overfitting [32]. It was selected for its robustness, ability to handle non-linear relationships, and strong performance with high-dimensional datasets. Additionally, RF provides interpretable outputs, such as feature importance, which are valuable for assessing the relative contributions of environmental and anthropogenic factors. These aspects make RF particularly suitable for capturing the complex interactions between environmental factors and atmospheric pollutants.

The observed S deposition values from ChinaWD sites were used as the dependent variable, combined with multiple explanatory variables. By annually matching the site observations with the explanatory variables, a training dataset for 2013–2023 was constructed. To ensure model accuracy and generalization, random seeds were applied for sample partitioning, and the data were divided into training and validation sets. In addition, k-fold cross-validation was performed to assess model stability and robustness across different subsets of the data. The main parameter settings were as follows: (1) the ratio of training to validation sets was 8:2; (2) the number of trees (n_estimators) was set to 1000; (3) the maximum number of features (max_features) was 5; and (4) out-of-bag error was enabled to evaluate model performance during training. In practice, 3000 non-repeating random seeds were drawn within the range of 0–10,000 to train 3000 RF models. Subsequently, the 10 best-performing models were selected based on the coefficient of determination (R²) and root mean square error (RMSE) evaluated on both the training and validation sets. The cross-validation results confirmed that these models consistently maintained good performance, indicating reliable predictive capability across different data splits. The formulas are as follows.

$$R^2=1-{\displaystyle \frac{{\sum }_{a=1}^{a=n}{\left(y_a-\widehat{y_a}\right)}^2}{{\sum }_{a=1}^{a=n}{\left(y_a-\overline{y}\right)}^2}}$$

(1)

$$RMSE=\sqrt{{\displaystyle \frac{1}{n}}{\sum }_{a=1}^{a=n}{\left(y_a-\widehat{y_a}\right)}^2}$$

(2)

Here, $y_a$ represents the observed values, $\widehat{y_a}$ the simulated values, $\overline{y}$ the mean of the observations, and $n$ the sample size. $R^2$ ranges from (−∞, 1], with values closer to 1 indicating better model performance. $RMSE$ reflects the average deviation between the simulated and observed values, with smaller values representing higher model accuracy.

The selected 10 optimal models achieved R² values greater than 0.5 (averaging 0.52) in the validation phases, and the RMSE values were all below 1.5 kg ha⁻¹ yr⁻¹ (averaging 1.2 kg ha⁻¹ yr⁻¹), which is less than 10% of the mean simulated value. Moreover, the distribution of residuals is centered around zero, indicating no systematic bias toward either overestimation or underestimation. These results indicate good model performance and robustness. Subsequently, the explanatory variables for each year from 2013 to 2023 were input into the 10 optimal models to generate 10 sets of outputs, and the multi-model mean was calculated as the final result. In this way, an annual 0.1°-resolution S deposition dataset for China during 2013–2023 was constructed.

2.3. Temporal Trend Analysis

To detect the interannual trends of S deposition, we employed the Mann–Kendall (MK) trend test and Sen’s slope estimator. The MK test is a non-parametric method that determines the strength and significance of a trend by evaluating the rank correlation of data pairs in a time series [33]. It has been widely used as an important tool for analyzing the dynamics of variables in meteorology, hydrology, and ecology. The relevant formulas are as follows.

$$S=\sum _{i=1}^{n-1}\sum _{j=i+1}^{n}sgn\left(x_j-x_i\right)$$

(3)

$$tau={\displaystyle \frac{2·S}{n·(n-1)}}$$

(4)

$$slope=median\left\{{\displaystyle \frac{x_j-x_i}{j-i}}\right\}$$

(5)

Here, $n$ is the length of the time series; $S$ is the test statistic; $x_i$ and $x_j$ are the sulfur deposition values in year $i$ and year $j$, respectively; and $sgn$ is the sign function, which returns 1, 0, or −1 when the input is greater than, equal to, or less than zero, respectively. $tau$ is the statistic used to measure the strength of the trend in the time series, ranging from −1 to 1, with positive values indicating an increasing trend and negative values indicating a decreasing trend, and larger absolute values representing stronger trends. $slope$ denotes Sen’s slope of the time series.

After the MK test, we further evaluated the significance level of the trend, as expressed by the following formulas.

$$t={\displaystyle \frac{S-sgn\left(S\right)}{\sqrt{Var\left(S\right)}}}$$

(6)

$$Var\left(S\right)={\displaystyle \frac{n·\left(n-1\right)·\left(2n+5\right)}{18}}$$

(7)

$$p=2·\left(1-\theta \left(\left|t\right|\right)\right)$$

(8)

Here, $Var\left(S\right)$ is the expected variance of the statistic $S$ under the null hypothesis of no trend; $t$ is the standardized test statistic; $\theta$ is the cumulative distribution function of the standard normal distribution; and $p$ is the significance indicator, ranging from [0, 1], with smaller values indicating stronger trend significance.

To comprehensively characterize the temporal variation in S deposition, the MK test and Sen’s slope estimator were applied at five scales: (1) grid scale: trends were tested for each 0.1° grid cell of the S deposition dataset to obtain the nationwide spatial distribution; (2) site scale: individual trends were analyzed based on the time series at each ChinaWD site; (3) site-mean scale: the annual mean series of all sites was tested to reflect the overall changes observed by the monitoring network; (4) national average scale: the nationwide mean series derived from the S deposition dataset was analyzed to capture the macroscopic trend characteristics; (5) atmospheric concentration scale: the nationwide mean SO₂ VCD series was tested for comparison with S deposition trends.

2.4. Driving Factors Analysis

To evaluate the influence of different factors on S deposition changes, this study employed a combined approach of ridge regression and SHAP (SHapley Additive exPlanations). Ridge regression is an improved method developed within the framework of multiple linear regression, which maintains model robustness in the presence of multicollinearity among explanatory variables [34]. It has been widely applied in attribution analyses in the field of environmental science. Compared with Lasso or Elastic Net, ridge regression retains all explanatory variables, which is important for interpreting the contributions of multiple environmental and anthropogenic factors. In this study, ridge regression provides a stable estimation framework to disentangle the relative roles of emission-related versus climate-related variables, ensuring that all potential drivers are considered simultaneously. The formulas are as follows.

$$Y={\sum }_{i=1}^{i=p}{a}_i·X_i+\epsilon$$

(9)

$$a=\arg min\left\{{\epsilon }^2+\gamma {\sum }_{i=1}^{i=p}{X_i}^2\right\}$$

(10)

Here, $Y$ denotes the observed sulfur deposition; $X_i$ is the $i$ explanatory variable; $a_i$ represents the regression coefficient of $X_i$; and $\gamma$ is the ridge regression parameter, which constrains the regression coefficients to preserve explanatory power while avoiding overfitting. In this study, $\gamma$ was set to 1 to maintain consistency across all regions.

Based on the estimated regression coefficients, we quantified the contributions of different factors to S deposition changes. Furthermore, the explanatory variables were grouped into three categories, and the contributions of each category were calculated: (1) climatic and land-surface factors, including P, T, WS, VAP, R, LULC, EVI, and DEM; (2) atmospheric pollutant emissions and concentrations, including SO2-E, SO2-C, PM25-E, PM25-C, NOX, NO2, NH3, and VOC; (3) socioeconomic factors, including POP, GDP, HFP, E, C, TRS, and HAR. The corresponding formulas are as follows.

$$C_n=\sum _{i=1}^q{a}_i·{∆X}_i$$

(11)

Here, $C_n$ denotes the contribution of the $n$-th category of factors; $q$ is the number of explanatory variables within the $n$-th category. ${∆X}_i$ represents the change in the $i$-th factor within the $n$-th category, calculated as the mean for 2018–2023 minus that for 2013–2017.

SHAP is an interpretation method based on cooperative game theory that establishes an additive attribution relationship between model predictions and the input explanatory variables [35]. Its core idea is to compute the marginal contribution of each feature across all possible feature combinations, thereby achieving a fair allocation of contributions to the prediction outcome. Unlike traditional regression coefficients, SHAP values not only reflect the relative importance of variables but also reveal the direction of their effects (positive or negative) on the outcome. This approach allows readers to understand how each variable contributes to the model prediction transparently and intuitively, even when the underlying model is complex. This makes SHAP particularly appropriate for our purpose, as it allows for a transparent quantification of how emission factors and climate factors individually contribute to S deposition changes. By visualizing SHAP values, one can see both the magnitude and direction of each feature’s effect, improving interpretability for readers less familiar with machine learning techniques. This method has been widely applied in the interpretation of complex environmental systems and machine learning models and is well-suited for attributing the drivers of S deposition in this study. Building upon the training and selection of the top 10 random forest models, we further employed SHAP values from these models for attribution analysis.

We acknowledge that uncertainties exist in measurements, satellite retrievals, and model predictions. ICP-OES provides high-precision measurements, but sampling and chemical analysis may introduce minor errors. Satellite data, such as SO₂ VCD and PM_2.5, contain retrieval uncertainties, while model outputs may be affected by interpolation and incomplete representation of explanatory variables. Cross-validation and selection of multiple optimal models help mitigate these uncertainties, but results should still be interpreted with these limitations in mind.

3. Results

3.1. Spatial Patterns of Wet S Deposition in China

The mean flux of wet S deposition across 43 monitoring sites from 2013 to 2023 was 11.30 kg ha⁻¹ yr⁻¹. Among these, the sites at HTF (located in Hunan), YTA (Jiangxi), and FQA (Henan) exhibited relatively high deposition levels, each exceeding 20 kg ha⁻¹ yr⁻¹, with HTF recording the maximum value of 24.11 kg ha⁻¹ yr⁻¹. These higher values likely reflect the greater influence of local and regional anthropogenic emissions, such as coal combustion and industrial activities. In contrast, remote stations such as NMD (Inner Mongolia), SJM (Heilongjiang), and DXA (Tibet) showed comparatively low deposition, all below 4 kg ha⁻¹ yr⁻¹, consistent with their distance from major emission sources and lower local industrial activity. This spatial variation highlights the role of emission intensity, atmospheric transport, and regional environmental conditions in shaping the distribution of sulfur deposition.

Random forest simulation results (Figure 1) indicate that the average wet S deposition flux over China from 2013 to 2023 was 9.64 kg ha⁻¹ yr⁻¹. Spatially, the deposition exhibited a decreasing gradient from the southeastern coastal toward the northwestern inland. Higher deposition was widespread across North and Central China, with notable hotspots including the North China Plain, the middle and lower reaches of the Yangtze River, southern China, and the Sichuan Basin (local deposition > 20 kg ha⁻¹ yr⁻¹). In contrast, deposition is relatively low in Northwest China and Qinghai–Tibet, with most areas receiving less than 8 kg ha⁻¹ yr⁻¹. Areal analysis indicated that approximately 55% of China experienced deposition ranging from 0 to 8 kg ha⁻¹ yr⁻¹, while less than 10% of the area had high deposition exceeding 15 kg ha⁻¹ yr⁻¹.

3.2. Trends in Wet Sulfur Deposition over the Past Decade

Among the 43 monitoring sites, approximately two-thirds exhibited a statistically significant decreasing trend (

Table 1. Wet S deposition trends at observation sites from 2013 to 2023.

Site	Trend (kg ha−1 yr−1)	Significance (p-Value)	Site	Trend (kg ha−1 yr−1)	Significance (p-Value)
ASA	−0.001	0.251	HLA	−0.831	0.022
BJF	−0.878	0.336	HSF	−0.018	0.241
BJU_A	−2.218	0.000	HTF_A	−0.409	0.335
BJU_B	−1.880	0.004	HTF_B	−0.188	0.152
BNF_A	−0.375	0.156	LCA	−1.351	0.037
CBF_A	−0.230	0.298	LSA	−0.538	0.366
CBF_B	−0.195	0.091	LZD	−0.457	0.113
CSA	−2.534	0.040	MXF	−0.023	0.237
CWA	−0.728	0.037	NMD	−0.594	0.262
DHF	−2.201	0.000	NMG	−0.363	0.015
DXA	−0.165	0.159	QYA	−0.919	0.087
ESD	−0.126	0.077	QYF	−0.961	0.095
FKD	0.046	0.118	SJM	−0.002	0.103
FQA	−2.481	0.000	SNF	−1.306	0.009
GGF_A	−1.091	0.044	SPD	−0.666	0.012
GGF_B	−0.894	0.026	SYA	−2.497	0.224
HBG_A	−0.127	0.067	TYA	−1.577	0.000
HBG_B	−0.028	0.095	YCA_A	−0.251	0.189
HBG_C	0.110	0.132	YCA_B	−0.681	0.094
HBG_D	0.169	0.075	YGA	−2.230	0.002
HJA	0.245	0.108	YTA	−2.391	0.000

Note: Only 42 sites with more than six years of observations are included.

). Notable declines were observed at sites including BJU (located in Beijing), CSA (Jiangsu), FQA (Henan), and YTA (Jiangxi), where reduction rates exceeded −2 kg ha⁻¹ yr⁻¹ (p < 0.05). In contrast, a slight but non-significant increasing trend was detected at a few stations, such as HJA (Guangxi) and HBG (Qinghai). The overall annual mean trend across all 43 sites (Figure 2B) indicates a significant decline in wet S deposition at a rate of approximately −0.895 kg ha⁻¹ yr⁻¹ (p < 0.05) from 2013 to 2023, with a cumulative reduction of over 50% over the past decade.

The random forest simulations further confirmed a consistent declining trend (Figure 2C), with a decline rate of −0.244 kg ha⁻¹ yr⁻¹, amounting to an approximately 28% reduction. Grid-level analysis (Figure 2A) revealed that the majority of regions exhibit significant decreases in wet S deposition, with North and Central China, as traditional high-value areas, demonstrating the most pronounced reductions, thereby contributing significantly to the overall national decline. Only a few localized areas exhibit slight upward trends. Area statistics indicated that over 95% of the country experienced a decrease in S deposition, reflecting a widespread alleviation and mitigation on a national scale. The weaker declining trend in the simulations compared with observations can largely be attributed to the smoothing effect of the random forest model and the spatial averaging over large areas, which may dilute the sharp decreases detected at individual observation sites. Furthermore, satellite-retrieved national SO₂ VCD showed a synergistic decline alongside deposition (p < 0.05; Figure 2D), with a nearly consistent reduction over the past decade (59%) closely aligning with the decrease observed at the scale of deposition monitoring stations (58%).

3.3. Attribution of S Deposition Changes in China over the Past Decade

The grid-level attribution analysis based on ridge regression (Figure 3A) indicates that approximately 90% of the variations in wet S deposition across the nation from 2013 to 2023 are predominantly driven by atmospheric emissions and concentrations, particularly in northern regions such as North China, Inner Mongolia, and the Northwest. Localized deposition changes in Central China and northern Northeast China were mainly influenced by socioeconomic and activity-intensity factors. In contrast, climate and surface factors played a relatively limited role, with their effects concentrated mainly in certain areas of Qinghai–Tibet.

Further, the random forest model combined with the SHAP method was employed to quantify the contributions of various factors to S deposition (Figure 3B,C). The results indicated that SO₂ emissions were the primary driving factor, exhibiting the highest contribution ratio. This is followed by energy consumption, coal consumption, and SO₂ VCD, all of which are closely related to precursor emissions. Among surface and climatic factors, precipitation demonstrates the strongest explanatory power, ranking immediately after emission-related factors. Overall, the reduction in wet S deposition in China over the past decade is primarily attributed to the decline in atmospheric precursor emissions and concentrations.

4. Discussion

Based on observational data from national wet S deposition monitoring sites between 2013 and 2023, this study revealed a significant decline in wet S deposition over the past decade, primarily driven by reductions in atmospheric emissions and concentrations. This finding was consistent with existing research [18,20,36]. Multiple observational and modeling studies have indicated that since the 2000s, both SO₂ emissions and atmospheric S deposition in China decreased markedly, with the decline accelerating since the 2010s, reflecting a strong emission-deposition coupling [15,21]. Our study further revealed that the national wet S deposition had decreased to 8.5 kg ha⁻¹ yr⁻¹ by 2023, lower than that recorded at the beginning of China’s reform and opening-up period (21.8 kg ha⁻¹ yr⁻¹ in 1984) [37]. Significant reductions were observed in historical deposition hotspots, including the North China Plain, the middle and lower reaches of the Yangtze River, and the Sichuan Basin (Figure 1D). These spatial differences can be explained by variations in local energy consumption patterns, the intensity of industrial and residential emissions, topographic and microclimatic constraints, and differences in the implementation and enforcement of emission control policies. For instance, most areas of the North China Plain and the Sichuan Basin, which previously had high deposition due to dense industrial activity and limited pollutant dispersion, exhibited deposition values below 15 kg ha⁻¹ yr⁻¹ by 2023, reflecting both effective policy interventions and improvements in emission management.

Overall, the sustained decline in S emissions and deposition has been driven by a multifaceted sustainability transition and emission reduction policies, including continuous socioeconomic and industrial restructuring, energy structure adjustment, promotion of low-sulfur coal, and implementation of FGD technologies in industry [10,11,38]. Compared with Europe, where the UNECE Sulphur Protocol led to sharp declines in S deposition since the 1990s, and the US, where the Clean Air Act achieved substantial reductions, China’s coordinated policies over the past decade have similarly resulted in marked national decreases in wet S deposition, highlighting the effectiveness of combined regulatory and technological measures. Despite the general alleviation of S deposition across the country, certain areas—including parts of the middle and lower Yangtze River region and southern China (such as Guizhou, Hunan, and Jiangxi)—still exhibited elevated deposition levels, with localized values exceeding 20 kg ha⁻¹ yr⁻¹. Attribution analysis indicates that these regional differences are partly influenced by variations in socioeconomic and industrial factors, such as local population density, residential coal use, and industrial activity intensity [39]. This interpretation is supported by the MEIC emission inventory, which shows that Guizhou and Hunan rank first and third, respectively, in residential coal-related SO₂ emissions. These findings suggest that while national trends are clear, the rates of decrease differ among regions due to local drivers.

When S deposition exceeds the critical load of ecosystems, it can induce a range of ecological risks and adverse effects. Integrating the findings of this study with critical load research indicates that in remote regions such as Northwest China, Qinghai–Tibet, and Inner Mongolia, deposition levels were generally below critical load due to naturally higher baseline loads and limited anthropogenic disturbance [40,41]. Moreover, rapid decreases in S deposition at the North China Plain and the Sichuan Basin have brought most areas below the critical load. Nevertheless, in certain high-deposition zones within the South of the Yangtze River (e.g., Guizhou, Hunan, and Jiangxi), deposition remains > 20 kg ha⁻¹ yr⁻¹, surpassing the local critical load (generally < 16 kg ha⁻¹ yr⁻¹). It should be noted that climate change, through altered precipitation patterns, increased frequency of extreme events, and shifts in atmospheric circulation, may modulate sulfur deposition dynamics by affecting wet scavenging and pollutant transport. While this study primarily focuses on observed trends under current climatic conditions, future variability in precipitation and extreme weather could influence the spatial and temporal distribution of S deposition, particularly in sensitive regions. To mitigate ecological risks in these areas, targeted control measures are recommended, including restricting residential coal use, promoting clean energy sources, and providing energy subsidies to further reduce S deposition and further realize the sustainable development goals.

This study provides new empirical evidence for understanding the changes in S deposition in China, demonstrating that sustained emission reduction measures and coordinated sustainable development strategies have contributed to a significant national decline in wet S deposition, though the magnitude of decrease varies among regions due to differences in local emissions, industrial structure, and socioeconomic factors. We also acknowledge that uncertainties in measurement, satellite retrievals, and modeling may influence the precise quantification of deposition trends, and readers should interpret the results with this consideration in mind. Furthermore, the findings are directly relevant to several Sustainable Development Goals (SDGs), including SDG 3 (good health and well-being) by reducing exposure to atmospheric pollutants, SDG 7 (affordable and clean energy) through promotion of low-sulfur fuels and industrial emission control, SDG 13 (climate action) by mitigating co-emissions, and SDG 15 (life on land) by alleviating ecological risks associated with sulfur deposition. Given these co-benefits, we suggest that developing countries could prioritize mitigation in the energy and industrial sectors by implementing large-scale emission reduction technologies (e.g., SCR–FGD combinations), phasing out highly polluting industries, and optimizing energy structures for coordinated control of SO₂, NO_x, and PM_2.5. Such measures are likely to enhance regional and national reductions in sulfur deposition and support sustainable development objectives.

5. Conclusions

Based on observational data from 43 sites across China between 2013 and 2023, combined with multi-source data including satellite remote sensing, this study systematically reveals the spatiotemporal dynamics of wet S deposition and its driving factors. By integrating ground observations, satellite retrievals, and machine learning modeling, we constructed a high-resolution, gridded dataset of wet S deposition, providing a unique methodological contribution. The results indicate a clear declining trend nationwide, consistent with the decrease in SO₂ VCD. Traditional high-deposition regions such as the North China Plain and the Sichuan Basin also showed notable improvement. Emission reductions driven by environmental policies were identified as the dominant factor, accounting for the majority of the decrease across most regions. Nevertheless, some areas in southern China, including Guizhou, Hunan, and Jiangxi, still exhibit relatively high deposition due to local sources such as residential coal combustion, highlighting the need for targeted mitigation. This study provides robust, up-to-date evidence supporting the effectiveness of China’s sustainability-oriented emission reduction policies, offering guidance for other developing countries aiming to balance economic growth with environmental protection.

Looking forward, future research could link S deposition trends with ecosystem recovery, refine uncertainty quantification, and apply this framework to other developing countries to support sustainable development planning. Moreover, our findings are directly relevant to several Sustainable Development Goals, including SDG 3 (good health and well-being), SDG 7 (affordable and clean energy), SDG 13 (climate action), and SDG 15 (life on land), underscoring the broader international significance of effective sulfur emission reductions.