Whole process efficiency framework for decomposing general industrial solid waste storage and discard in China

Summary

Reducing general industrial solid waste (ISW) storage and discard is important for advancing zero-waste cities. We integrated slacks-based measure data envelopment analysis, index decomposition analysis, and provincial attribution analysis within a whole-process efficiency-oriented framework to identify seven factors across 30 Chinese provinces from 2011 to 2022. General ISW generation, storage, and discard varied markedly among provinces. Average process control efficiency and the utilization-disposal rate were 0.672 and 0.840, respectively, indicating scope for improvement. A reduction in process control inefficiency, reflecting improved process control efficiency, made the largest downward contribution, whereas an increase in economic inefficiency made the largest upward contribution. End-of-pipe treatment did not offset the overall increase, although utilization and disposal of previously stored waste began after 2018 and faced further obstacles during the 2020 pandemic. The framework links source prevention, process control, and end-of-pipe treatment, supporting province-specific strategies for reducing ISW accumulation.

Graphical abstract

Highlights

• •General industrial solid waste storage and discard show strong provincial variation
• •Lower economic efficiency contributed most to increasing storage and discard
• •Higher process control efficiency contributed most to reducing storage and discard
• •Across 2011–2022, end-of-pipe treatment did not offset storage and discard growth

Industrial process; Industrial engineering; Environmental engineering; Industrial ecology

Introduction

China has benefited from rapid industrialization since its reform and opening-up. As a result, industrial solid waste (ISW) generation has significantly increased, growing from 816 million tons in 2000 to 4.375 billion tons in 2023. This is a 5.36-fold increase over 23 years. Due to underdeveloped technology and management practices, China’s comprehensive ISW utilization rate was only 60.45% in 2023, which is less than that rate for developed countries.¹ China’s general and hazardous ISW storage volumes were 931 million tons and 88 million tons, respectively, in 2023. Over the years, China has accumulated 60 billion tons of ISW storage, occupying a more than 2-million-hectare area.² This highlights the need for better ISW governance.

China’s “Work Plan for Zero Waste City Construction during the 14^th Five-Year Plan Period” (2021) establishes mandatory industrial constraints, including ISW generation intensity, ISW comprehensive utilization rate, and reductions in ISW storage and discard volumes. The “14^th Five-Year Plan for Industrial Green Development” also set specific requirements to reduce ISW generation intensity and promote its large-scale utilization. China’s Central Economic Work Conference, held on December 11–12, 2024, generated a proposal to include the “formulation of a comprehensive solid waste management action plan” in the 2025 work agenda of the Ministry of Ecology and Environment. It is anticipated that this action plan may have the same impact as the “Air Pollution Prevention and Control Action Plan,” the “Water Pollution Prevention and Control Action Plan,” and the “Soil Pollution Prevention and Control Action Plan.” These reflect a surge of solid waste management initiatives in China.

Improving comprehensive ISW utilization and reducing its storage and discard are important research topics. These activities contribute to the development of the circular economy and help avoid resource waste. However, compared to other environmental problems, such as greenhouse gas emissions and air and water pollution, ISW has not received sufficient attention at research and application levels. Previous studies exploring the factors driving industrial pollutants have mainly focused on pollutant emission intensity or pollutant emissions. They generally use two main research methods: decomposition analysis and econometric models.^3,4,5 Decomposition analysis includes structural decomposition,⁶ index decomposition,⁷ and production-theoretical decomposition.⁸ Studies using econometric models to examine the factors driving industrial pollutants have typically focused on the impact of specific factors on industrial pollutants. They explore causal effects and influencing mechanisms, with important economic implications. For example, Liu et al.⁹ applied an econometric model to examine the impact of digital technology innovation on China’s industrial SO₂ emissions. Lin and Xu¹⁰ used a vector autoregression model to analyze the factors driving increased industrial carbon dioxide emissions. However, econometric models are often inadequate for comparing the contributions of different influencing factors. In addition, it is challenging to address endogeneity issues in regression analysis.

The structural decomposition analysis (SDA) method is based on input-output models and adequately characterizes intrinsic relationships among sectors, technologies, and demand. However, studies are generally inter-temporal studies, as several countries do not provide annual input-output tables.¹¹ This hinders the application and development of SDA methods. Pasurka¹² proposed the production-theoretical decomposition analysis (PDA) method, which combines distance functions with data envelopment analysis (DEA) models. It requires a significant amount of data and requires the calculation of DEA models. Among the index decomposition methods, the logarithmic mean divisia index (LMDI) is a mainstream research tool, because it has consistent aggregation^13,14 and provides additive and multiplicative decomposition approaches.¹⁵

Table 1 lists selected studies on solid waste index decomposition. These studies generally focus on the factors driving source prevention, such as energy intensity and industrial structure. They do not consider the impact of process control and end-of-pipe treatment driving factors on solid waste. Moreover, they focus on the factors driving greenhouse gas emissions from municipal solid waste (MSW) during the treatment process, such as solid waste generation intensity and economic development. They do not consider the impact of preventing solid waste sources and factors from impacting process control. Some scholars have investigated the factors driving industrial pollutants across source prevention, process control, and end-of-pipe treatment processes,²⁷ however, ISW was not the research object. Chen et al.²⁸ were the first to incorporate end-of-pipe treatment factors into the decomposition framework. Since then, scholars have applied proposed methods to perform index decompositions of industrial pollutants from both source prevention and end-of-pipe treatment perspectives; they also explore their driving factors. Representative studies include Fujii et al.,²⁹ Zhang,³⁰ and Wang et al.³¹ However, little research considers process control as a driving factor of pollutant prevention and control.

Table 1.

Selected empirical studies on solid waste factor decomposition

Author	Country/Region	Year	Explained variable	Methods	Source prevention	Process control	End-of-pipe treatment
Kuai et al.16	China	2011, 2015	general ISW generation, hazardous ISW generation	LMDI	✓	–	–
Guan et al.17	Guangdong, China	2002–2015	ISW generation	LMDI	✓	–	–
He et al.18	Australia	2007–2014	solid waste generation	SDA	✓	–	–
Xiao et al.19	Shanghai, China	2000–2017	greenhouse gas emissions from treating MSW	LMDI	–	–	✓
Zhang et al.20	China	2011–2015	ISW generation	SDA	✓	–	–
Kang et al.21	294 cities in China	2006–2019	greenhouse gas emissions from treating MSW	LMDI	–	–	✓
Zhang et al.22	China	2010–2019	carbon emissions from managing MSW	LMDI	–	–	✓
Zhao et al.23	China	2004–2018	MSW generation	LMDI	✓	–	–
Lee et al.24	South Korea	2008, 2018	hazardous ISW generation	SDA	✓	–	–
Liu and Wang25	China	2011–2020	carbon emissions from MSW treatment	LMDI PDA	–	–	✓
Boré et al.26	China	2005–2020	nine MSW incineration emissions	LMDI	–	–	✓
This study	China	2011–2022	general ISW storage and discard	LMDI SBM	✓	✓	✓

This study focuses on general ISW, applying an interdisciplinary perspective to examine the factors influencing the general ISW management process. First, the slacks-based measure (SBM) DEA model is an accurate way to assess efficiency.³² As such, the study uses that model to calculate the general ISW process control inefficiency (PCI), economic inefficiency (EI), and the potential generation intensity from China’s provincial industries from 2011 to 2022. Second, the study applies the LMDI method to decompose the general ISW storage and discard volumes into seven influencing factors. These factors include investment quality, investment structure, general ISW PCI, potential generation intensity, EI, end-of-pipe treatment, and investment scale. Of these factors, investment quality and structure are key drivers for source prevention of general ISW. Notably, PCI, potential generation intensity, and EI are key drivers for process control. End-of-pipe treatment is the key driver for end-of-pipe treatment. Third, the attribution analysis (AA) method is applied to calculate the contribution of each Chinese province to a specific driving factor.

The reduction of data about general ISW storage and discard includes identifying driving factors across the source prevention, process control, and end-of-pipe treatment processes. This study offers the following innovations in this area. First, production investments are allocated across industrial types, which significantly impacts general ISW storage and discard. Given the role of general ISW storage and discard as a form of source prevention, this study incorporated driving factors such as investment quality and structure when assessing contributions to general ISW storage and discard. Second, this study proposes a new decomposition analysis method that evaluates efficiency, making it possible to identify the impacts of process control factors, such as PCI and EI, on changes in general ISW storage and discard. This method explores the mechanisms by which general ISW PCI and EI impact general ISW storage and discard. Third, the results of the decomposition and AA provide a reference for formulating targeted policies for the comprehensive use, disposal, storage, and discard of general ISW. This research used changes in general ISW storage and discard across China’s provinces from 2011 to 2022 as a case study to illustrate the proposed efficiency-oriented decomposition method. This method can also be applied to identify the drivers of other pollutant types using a whole-process prevention-and-control perspective.

Results and discussion

The ISW management process has two stages: generation and treatment.³³ Source prevention and process control are combined into the ISW generation stage, as source prevention is not a single process with a series of measures. The study framework diagram in Figure 1 shows that in the generation stage, capital, labor, energy, and other inputs generate economic value and ISW (including common ISW and hazardous ISW). In the treatment stage, funds are invested to govern ISW. Some ISW enter the generation stage after comprehensive utilization to create new economic value.³⁴ Some ISW-related hazards can be eliminated after disposal. Others are stored in environmentally friendly sites for future utilization or disposal. A small amount of ISWs is discharged into the environment. The amount of ISW generated is a common measure across ISW generation treatment stages. The investment in ISW treatment is mainly used to purchase reactors, incinerators, crushing equipment, or to build garbage landfills.

Figure 1.

A framework of ISW process management system

General ISW: Quantities and subsystem efficiency

Figure 2 presents the trajectories of general ISW generation across 30 Chinese provinces, from 2011 to 2022. To facilitate visualization, the provinces were grouped using K-means clustering, with k = 3; the resulting groups are displayed in Figures 2A–2C. As detailed in the quantification and statistical analysis section, the cluster-validation diagnostics did not identify a single common statistical optimum across the general ISW generation and general ISW storage/discard datasets. Therefore, k = 3 was retained as a pragmatic choice for a consistent three-panel presentation. The clustering is descriptive and does not affect the subsequent SBM-DEA, decomposition, or attribution analyses. Most provinces peaked their levels of general ISW generation in 2019; subsequent drops were due to the COVID-19 pandemic and the resulting economic downturn. Figure 2A represents a high-generation, resource-intensive heavy-industry group comprised of Liaoning, Hebei, Shandong, Shanxi, and Inner Mongolia. Inner Mongolia and Shanxi saw significant growth in general ISW generation throughout the study period. Shandong, Liaoning, and Inner Mongolia maintained relatively stable generation levels, with minor fluctuations.

Figure 2.

Trends in ISW generation from 2011 to 2022 in China by provincial groups

(A) Panel A shows the high-generation, resource-intensive heavy-industry group.

(B) Panel B shows the moderate-generation group with mixed manufacturing and resource-based economies.

(C) Panel C shows the lower-generation group with heterogeneous economic structures.

Figure 2B represents a moderate-generation group with mixed manufacturing and resource-based economies, including Jiangsu, Anhui, Jiangxi, Henan, Hubei, Yunnan, Sichuan, Guizhou, Shaanxi, Qinghai, and Xinjiang. These provinces showed similar patterns, with initial increases followed by declines after 2019. Henan experienced a surge in generation during 2018 and 2019, followed by a decline in 2020. Figure 2C represents a lower-generation group with heterogeneous economic structures. Included in this group are Beijing, Tianjin, Shanghai, Zhejiang, Fujian, Guangdong, Guangxi, Hainan, Jilin, Heilongjiang, Hunan, Chongqing, Gansu, and Ningxia. Most of these provinces are either economically developed areas or have a small economic scale. Beijing, Hainan, Tianjin, Shanghai, and Chongqing showed a relatively stable generation performance, with minimal overall variation. Ningxia trended steadily upward, while Heilongjiang, Guangdong, Gansu, Fujian, Zhejiang, Hunan, and Jilin showed fluctuating patterns in their ISW generation levels.

Figure 3 presents the trajectories of general ISW storage and discard across 30 Chinese provinces, from 2011 to 2022. To facilitate visual comparison, the provinces were grouped into three clusters using k-means clustering, with k = 3. Details of cluster selection and validation are provided in the quantification and statistical analysis section. The resulting groups are displayed in Figures 3A–3C. Figure 3A represents a Hebei-specific legacy-stockpile adjustment trajectory. Hebei is historically one of China’s most important centers for heavy industry and steel manufacturing; the province’s general ISW storage and discard volumes exceeded 200 million tonnes annually in both 2011 and 2012. Following the 2012 revision of the Steel Industry Standard Conditions and other capacity regulation policies, Hebei’s storage and discard volumes fell sharply, to approximately 8.84 million tonnes in 2015. This represented a decline of over 95 percent from the 2012 peak. After 2015, the volumes remained below 40 million tonnes for several years but then rose to approximately 100–130 million tonnes in 2021 and 2022. This dramatic peak and decline trajectory were unique among all 30 provinces, reflecting the concentrated impact of regulatory intervention on a single province with an outsized legacy stockpile.

Figure 3.

Trends in ISW storage and discard from 2011 to 2022 in China by provincial groups

(A) Panel A shows the Hebei-specific legacy-stockpile adjustment trajectory, which forms a separate single-province cluster.

(B) Panel B shows the persistent high storage-and-discard group dominated by resource-intensive activities.

(C) Panel C shows the low and stable storage-and-discard group.

Figure 3B represents a persistent high storage-and-discard group dominated by resource-intensive activities. This group is comprised of Inner Mongolia, Liaoning, Jiangxi, Sichuan, Qinghai, and Shanxi. These six provinces maintained consistently high general ISW storage and discard volumes throughout the study period, reflecting the sustained pressure of resource extraction and heavy industrial activity on waste management systems. To this day, Inner Mongolia and Liaoning are major heavy industrial hubs with large steel, coal, and chemical sectors. Jiangxi and Sichuan have extensive manufacturing and mineral resource development industries. Qinghai has a relatively small industrial economy, yet, during the study period, the province’s storage and discard volumes were substantial and fluctuated considerably. This pattern reflects the province’s remote mining areas, where poor transportation infrastructure and limited access to downstream industries constrain the utilization and disposal of general ISW. Shanxi, dominated by coal mining, generated large quantities of general ISW but maintained relatively stable storage and discard volumes through the province’s utilization and disposal capacity.

Figure 3C represents a low and stable storage/discard group—the remaining 23 provinces—with heterogeneous economic structures. These provinces had comparatively low general ISW storage and discard volumes that typically remained below 30 million tonnes annually, with generally stable trends. This cluster encompasses several distinct types of provincial economies and includes economically diversified and developed regions, such as Beijing, Shanghai, Jiangsu, Zhejiang, Guangdong, and Fujian. In this cluster, a service-oriented and advanced manufacturing economic structure generates less general ISW, and stricter environmental enforcement supports effective waste management. The cluster also includes provinces such as Yunnan, Xinjiang, Guangxi, and Guizhou, which have mining and resource extraction activities but the storage and discard volumes of which have been moderated by utilization and disposal capacity.

The general ISW PCE and the general ISW utilization and disposal rate (UDR) represent the efficiencies of the general ISW generation and treatment subsystems. Next, the 30 provinces were classified into four categories based on the provinces’ average general ISW PCE and UDR. The first category, in the upper right quadrant of Figure 4, includes provinces with a high UDR and high PCE. These include Hainan, Guangdong, Ningxia, Zhejiang, Jiangsu, Beijing, and Shanghai. These provinces typically have strict environmental protection regulations and advanced technology levels and equipment. This ensures efficient utilization and disposal of general ISW. Most of these provinces are economically developed areas that can invest resources and funds in general ISW management. The second category includes provinces with high UDR and low PCE. Shown in the lower right quadrant of Figure 4, this category includes Tianjin, Fujian, Hunan, Henan, Chongqing, Shanxi, Shandong, Shaanxi, Anhui, and Hebei. These provinces focus on general ISW utilization and disposal, but process control may benefit from further improvement and enhancement. The third category encompasses provinces with low UDR and high PCE. Shown in the upper left quadrant of Figure 4, this category includes Yunnan, Gansu, Xinjiang, Liaoning, and Inner Mongolia. These provinces have good process control technology and strict and standardized associated procedures. However, their UDRs are low due to resource constraints, such as funding and end-of-pipe technology limitations. The fourth category includes provinces with low UDR and low PCE. Shown in the lower left quadrant of Figure 4, this category includes Heilongjiang, Qinghai, Jiangxi, Sichuan, Guangxi, Guizhou, Hubei, and Jilin. These provinces have deficiencies in both UDR and PCE due to insufficient resources and funding for investments.

Figure 4.

Classification of Chinese provinces based on general ISW PCE and UDR from 2011 to 2022

This study’s control process and end-of-pipe treatment efficiency analysis identified substantial areas for improvement. Targeted interventions could yield significant environmental and economic benefits, particularly in provinces with lower efficiency metrics. Provinces with high UDR and high PCE could be benchmarks for other provinces. Provinces with high UDR and low PCE should strengthen their process control. Provinces with low UDR and high PCE should invest in improving general ISW utilization and disposal. Provinces with both low UDR and low PCE should improve their general ISW management standards throughout the process.

Decomposition analysis

Based on the results of the general ISW PCE and economic efficiency (EE) evaluation, Equations 11 and 12 were used to calculate the seven factors influencing general ISW storage and discard across 30 Chinese provinces. A decomposition analysis was conducted for two-year periods spanning 2011–2022. Table 2 shows the 11 two-year periods and indicates that from a whole-process governance perspective, the end-of-pipe treatment factor made the largest positive cumulative contribution to the increase in general ISW storage and discard. It had a cumulative contribution value of 1.1583, meaning that the growth rate of general ISW storage and discard exceeded the growth rate of general ISW generation. In other words, the increase in general ISW comprehensive utilization and disposal did not keep pace with the increase in general ISW generation. Both the source prevention effect and the process control effect had cumulative impact values below 1 (0.9705 and 0.9708, respectively) on general ISW storage and discard. This means both processes effectively reduced general ISW storage and discard volumes. Of the seven driving factors, EI contributed the most (1.5019), followed by investment scale (1.3307), and end-of-pipe treatment (1.1583).

Table 2.

Changes in general ISW storage and discard and their decomposition for 2011–2022

Period	Dtot	Source prevention			Process control				End-of-pipe treatment	IScal
			IQ	IStr		PCI	PGI	EI
2011–2012	0.9837	0.9079	0.9073	1.0007	0.9114	2.0613	0.3748	1.1798	0.9647	1.2323
2012–2013	0.7119	0.9204	0.9220	0.9982	0.9219	0.5378	1.4380	1.1921	0.7102	1.1817
2013–2014	1.0546	1.5132	1.5189	0.9962	0.9250	0.8153	1.1213	1.0118	1.0627	0.7090
2014–2015	1.2977	0.6828	0.6793	1.0051	1.0407	1.0233	0.9968	1.0203	1.2396	1.4733
2015–2016	1.0721	1.1809	1.1829	0.9983	1.0019	0.5478	1.5505	1.1796	1.1623	0.7796
2016–2017	1.2548	1.0230	1.0228	1.0002	1.1785	1.0227	1.0625	1.0845	1.1666	0.8922
2017–2018	1.0045	1.1531	1.1520	1.0009	1.0164	0.8188	1.3649	0.9095	0.8635	0.9925
2018–2019	1.0045	0.8517	0.8516	1.0001	1.2301	1.3302	0.8600	1.0753	0.9320	1.0288
2019–2020	1.0045	1.0405	1.0389	1.0015	0.8493	1.4034	0.6798	0.8903	1.1660	0.9749
2020–2021	1.1077	0.9970	0.9966	1.0004	0.9939	0.8557	1.1889	0.9769	1.0347	1.0804
2021–2022	1.0028	0.9134	0.9124	1.0011	0.9631	0.9209	1.1003	0.9505	0.9748	1.1694
Geo-mean	1.0345	0.9973	0.9970	1.0002	0.9973	0.9580	1.0032	1.0377	1.0134	1.0263
2011–2022	1.4517	0.9705	0.9679	1.0027	0.9708	0.6240	1.0358	1.5019	1.1583	1.3307

Note: IQ indicates investment quality, IStr indicates investment structure, PCI indicates process control inefficiency, PGI indicates potential generation intensity, EI indicates economic inefficiency, and IScal indicates investment scale.

The year 2012 was particularly significant. In 2012, PCI contributed more to general ISW storage and discard (2.0613) than in 2011. However, in 2013, the contribution of PCI fell below 1, reducing general ISW storage and discard volumes. This was primarily because the steel industry generates significant amounts of general ISW, significantly influencing these results. Before 2012, China’s steel industry developed rapidly, but it also developed serious problems, such as unauthorized construction before approval. This occurs when steel enterprises (especially private steel companies) privately invest in, construct, and operate steel smelting projects without obtaining all necessary documentation, including filing and approval documents from the National Development and Reform Commission and other departments, land use permits, and environmental impact assessment approvals. This unauthorized steel development reached approximately 300 million tons in 2012. To address this specific problem, in 2012, China’s Ministry of Industry and Information Technology released the revised normative conditions of the steel industry, bringing them uniformly under national industry management. This led to a significant increase in general ISW PCI and improved general ISW storage and discard volumes. In 2013, to actively and effectively resolve serious overcapacity conflicts in industries such as steel, the State Council issued the “Guiding Opinions on Resolving the Serious Overcapacity Contradiction”.³⁵ Eliminating underdeveloped capacity in 2013 has significant results, increasing general ISW PCE and lowering general ISW storage and discard volumes.

General ISW storage and discard varied significantly across different provinces. As such, they were classified into six sub-categories based on order of magnitude, to clearly characterize the contribution of prevention effects, process control effects, end-of-pipe treatment effects, and investment scale effects on general ISW storage and discard volumes. Figure 5 shows that the investment scale effect made a positive contribution to the growth of general ISW storage and discard in 23 provinces, excluding Liaoning, Shanxi, Jilin, Hainan, Beijing, Shanghai, and Shandong. This result indicates a widespread contradiction between economic growth and environmental pressure. China still needs to break through the traditional development model, characterized by the inevitable combination of economic development and environmental pressures. For provinces with higher levels of general ISW storage and discard, end-of-pipe treatment positively impacted general ISW storage and discard. These provinces included Inner Mongolia, Liaoning, Jiangxi, Qinghai, Shanxi, Heilongjiang, Shandong, Henan, Sichuan, Yunnan, Hubei, Hunan, Guizhou, and Gansu. This indicates that the general ISW utilization and disposal systems in these provinces were overwhelmed. General ISW management approaches shifted from “utilization and disposal” toward “storage.”

Figure 5.

Cumulative decomposition results of the four factors of source prevention, process control, end-of-pipe treatment, and investment scale for each province, from 2011 to 2022

Based on the contributions of prevention effects, process control effects, end-of-pipe treatment effects, and investment scale effects on general ISW storage and discard volumes, provinces were classified into three categories: provinces with a high level of utilization and disposal; provinces under dual pressure from scale and end-of-pipe treatment; and provinces with advanced green transformation. Utilization and disposal-dominated provinces had large positive investment scale effects that were effectively offset by strong negative end-of-pipe treatment effects. This indicates that general ISW in these provinces was effectively recycled and made harmless. These provinces included Hebei, Anhui, Ningxia, Jiangsu, Zhejiang, and Chongqing.

Provinces under the dual pressure of scale and end-of-pipe treatment showed large positive investment scale effects and positive end-of-pipe treatment effects, leading to rapid growth in general ISW storage and discard. These provinces included Inner Mongolia, Jiangxi, Guangxi, Qinghai, Heilongjiang, Henan, Sichuan, Yunnan, Hubei, Hunan, Guizhou, Gansu, Fujian, Guangdong, and Tianjin. Finally, provinces with an advanced green transformation showed negative effects in investment scale, process control, and end-of-pipe treatment, and positive effects in prevention. The negative effects outweighed the positive ones. These provinces included Hainan, Beijing, and Shanghai.

Figure 6 shows the cumulative decomposition results of the seven factors, where general ISW storage and discard were decomposed across multiple periods. General ISW storage and discard volumes increased by 45.17% from 2011 to 2022, and EI made the largest positive cumulative contribution to this growth. In addition, PCI was the most important factor influencing reductions in general ISW storage and discard. This indicates that adjustments to industrial structure and improvements in technology and management levels played important roles, despite a significant decline in PCI. Both EI and investment scale remained above 1 throughout the study period. This indicates that these two factors played an important role in promoting general ISW storage and discard, and were the first- and second-largest growth drivers. Investment structure consistently remained around 1, with a 0.3% increase by 2022. This prompted an overall increase in general ISW storage and discard volumes.

Figure 6.

Cumulative changes of general ISW storage and discard and their decomposition from 2011 to 2022

Additionally, from 2011 to 2017, the end-of-pipe treatment effect followed a trend similar to changes in general ISW storage and discard, diverging after 2018. Before 2018, the potential generation intensity continued to rise, increasing from below 1 to above 1. This indicated that the gap in general ISW generation intensity narrowed, then widened. In other words, the disparities in China’s industrial structure, process control technology, and management levels increased. After 2018, the potential generation intensity fluctuated. By 2022, the cumulative contribution of the potential generation intensity to general ISW storage and discard was 1.0358, representing small-magnitude growth.

Figure 7 illustrates how different factors influence the storage and discard of general ISW in each province. For investment scale, 23 provinces (76.67%) showed a positive investment scale effect, indicating that an expanded investment scale increased overall ISW storage and discard. Most of these provinces had a relatively large investment scale; only 23.33% of provinces showed a negative effect.

Figure 7.

Cumulative decomposition results of general ISW storage and discard by province

For end-of-pipe treatment, a few provinces performed well. These included Beijing, Zhejiang, Shanghai, Jiangsu, Ningxia, Hainan, Hebei, and Shaanxi. Some provinces, including Shandong, Guangdong, and Sichuan showed positive end-of-pipe treatment effects and significantly contributed to general ISW storage and discard. This indicates that these provinces performed poorly with respect to resource utilization and harmless general ISW disposal.

In addition, 13 provinces (including Xinjiang, Henan, and Jilin) showed that PCI made a positive contribution; increases in PCI were associated with increased general ISW storage and discard. In contrast, 16 provinces (including Guizhou, Yunnan, and Gansu) made negative contributions, as a decreased PCI effectively reduced general ISW storage and discard. Finally, the contribution of PCI to changes in general ISW storage and discard in Beijing was negligible.

For potential generation intensity, 17 provinces (including Chongqing, Jilin, and Hunan) had negative contributions, indicating that a reduction in the gap of general ISW generation intensity decreased general ISW storage and discard. In contrast, 13 provinces (including Hubei, Shaanxi, and Gansu) showed that a potential generation intensity made a positive contribution. This means that the increasing gap in general ISW generation intensity in these provinces led to increased general ISW storage and discard. The results show that 17 provinces narrowed the gap in general ISW generation intensity with frontier provinces.

Concerning investment quality effects, only eight provinces (including Shanxi, Beijing, and Hainan) saw a positive contribution. This indicates that improved investment quality promoted increased general ISW storage and discard. In contrast, 19 provinces (including Inner Mongolia, Guangxi, and Qinghai) showed a negative contribution from investment quality effects. This may be due to decreasing returns from investment, which also reduced general ISW storage and discard. Finally, EI effects and investment structure effects were not dominant contributors to general ISW storage and discard across provinces.

Attribution analysis

Table S1 includes the cumulative attribution results of the effects of investment quality, investment structure, PCI, potential generation intensity, EI, end-of-pipe treatment, and investment scale for 2011–2022 for each province. From 2011 to 2022, the cumulative contribution rate of China’s industrial investment quality to general ISW storage and discard was −3.21%. This indicated a decline in the overall return on industrial investment in China. Only a few provinces, such as Shanxi, Liaoning, Hainan, Sichuan, Guizhou, Gansu, Ningxia, and Xinjiang showed increasing investment returns. All of these show lagging economic development. In developed provinces such as Beijing (0.01%), Tianjin (0.00%), Shanghai (0.01%), Jiangsu (0.00%), Zhejiang (−0.01%), and Guangdong (−0.02%), industrial investment returns remained steady. Figure 8 shows that in 2012, 2013, 2015, and 2022, industrial investment returns declined in most provinces.

Figure 8.

Multi-period contribution of investment quality changes to general ISW storage and discard (base year = 2011)

However, in 2014, industrial investment returns increased in most provinces, primarily due to a significant contraction in industrial investment. In that year, China’s economy was in a stage of “triple overlap”, meaning it was in a period marked by growth rate transition, structural adjustment challenges, and the absorption of previous stimulus policies. The Chinese government proactively reduced its pursuit of fast economic growth. Instead, it focused on quality and sustainability and introduced the concept of a “new normal” for the economy. In 2014, the government also implemented a series of policies that strictly prohibited the approval of new capacity projects in industries with overcapacity, such as steel, cement, and electrolytic aluminum, and banned the provision of land and credit support for these projects. This meant that the investment pathways of traditional industries, which were highly dependent on investment expansion, were blocked, which restricted investment. Additionally, environmental policies, such as the “Air Pollution Prevention and Control Action Plan,” began to be rigorously enforced, with increasingly strict production restrictions and transformation requirements for high energy-consuming and high-polluting enterprises. This reduced new investments in these sectors.

From 2011 to 2022, the cumulative contribution rate of China’s general ISW PCI to general ISW storage and discard was −37.60%. Overall, China’s PCI for general ISW declined, with only a few provinces showing an increase. The provinces with the largest increase in PCI were Hebei (39.60%), Henan (3.36%), Jilin (2.24%), Hunan (1.37%), Guangxi (1.36%), and Guizhou (1.14%).

Figure 9 shows that in 2012, PCI for general ISW increased in almost all provinces compared to other years, with Hebei experiencing a particularly sharp increase (56.297%). This validates the hypothesis that the steep decline in PCE for general ISW in 2012 was due to the Ministry of Industry and Information Technology’s publication of the 2012 version of the “Steel Industry Standard Conditions.” This helped standardize steel capacity management, which had previously operated outside regulatory oversight. This led to significant decreases in PCE for general ISW in most provinces, including Hebei, Jiangxi, and Xinjiang.

Figure 9.

Multi-period contribution of PCI changes to general ISW storage and discard (base year = 2011)

From 2013 to 2016, administrative measures dominated the capacity reduction efforts in steel, coal, and other industries, with strong intensity and rapid short-term results. This increased the PCE for general ISW in most provinces. After 2017, the marginal effects of administrative capacity reduction measures began to diminish, and the effectiveness of capacity reduction in steel, coal, and other industries dropped compared to earlier periods.

Across multiple cycles spanning 2011–2022, EI inhibited general ISW storage and discard in only nine provinces. These inhibitive effects were relatively low (e.g., −1.47% in Xinjiang, −0.80% in Henan, −0.63% in Sichuan, and −0.44% in Hubei). Most provinces saw an increase in EI, including Liaoning (22.78%), Inner Mongolia (8.33%), Hebei (6.57%), Yunnan (4.83%), and Heilongjiang (3.57%). This contributed to an increase in general ISW storage and discard. The most significant period of declining EE occurred from 2012 to 2015, as shown in Figure 10. The launch of supply-side structural reform in 2016 was associated with a recovery in EE. This recovery was not driven by the past; rather, it was a crucial turning point, as China’s industry shifted from “scale and speed-oriented” extensive growth to “quality and efficiency-oriented” intensive growth.

Figure 10.

Multi-period contribution of EI changes to general ISW storage and discard (base year = 2011)

Looking at multiple cycles spanning the period 2011–2022, end-of-pipe treatment inhibited general ISW storage and discard in 12 provinces (accounting for 40% of the total), including Hebei (−16.65%), Anhui (−0.93%), Ningxia (−0.68%), and Shaanxi (−0.65%). Most other provinces did not effectively utilize and dispose of general ISW, and end-of-pipe treatment declined significantly. Examples of such declines included Liaoning (7.44%), Shandong (5.08%), Inner Mongolia (4.24%), Sichuan (3.79%), and Heilongjiang (3.69%).

The year 2018 was a turning point for China’s end-of-pipe treatment of general ISW. In 2018, 2019, 2021, and 2022, end-of-pipe treatment in most provinces inhibited general ISW storage and discard, as shown in Figure 11. In contrast, in 2020, end-of-pipe treatment in most provinces promoted general ISW storage and discard. This was because in 2020, China experienced the effects of the COVID-19 pandemic, and factors such as transportation disruptions and shutdowns of resource recycling enterprises impeded general ISW utilization and disposal. This led to increased storage and discard.

Figure 11.

Multi-period contribution of end-of-pipe treatment changes to general ISW storage and discard (base year = 2011)

There are three reasons why 2018 was a turning point for China’s end-of-pipe treatment of general ISW. First, on January 1, 2018, China’s first Environmental Protection Tax Law came into effect, stipulating the exemption of legally utilized solid waste from the environmental protection tax. Second, in December 2018, the General Office of the State Council issued the “Pilot Work Plan for ‘Zero Waste City’ Construction.” This provided clear direction for solid waste management and opened a new chapter in systematic, city-level solid waste governance. Third, in 2018, China’s import controls on foreign waste peaked, with multiple types of solid waste being added to the prohibited import list. This forced Chinese resource recycling enterprises to focus fully on domestic ISW, and created a large market space for domestic solid waste use.

Conclusions

This study constructed a three-step analytical framework that included PCE measurement, index decomposition analysis, and AA. General ISW storage and discard was decomposed into seven driving factors covering source prevention, process control, and end-of-pipe treatment processes. The study then identified the contribution of each province of China. This method provides a scientific basis and an empirical reference for developing differentiated policies for ISW process management. The main conclusions and implications of this study are as follows.

First, there was significant provincial heterogeneity in general ISW management performance. The average general ISW PCE was 0.672, and the average UDR reached 0.840. Because the PCE estimated in this study is defined relative to a production frontier formed by the 30 Chinese provinces, the PCE is not directly comparable with efficiency scores estimated for other countries under different production frontiers. Similarly, China’s general ISW UDR does not have an exact counterpart in the OECD database, because waste coverage, treatment categories, and statistical definitions vary across countries. Nevertheless, broader international indicators provide useful context. For example, OECD evidence shows that, in 2018, OECD economies generated approximately USD 4,000 of GDP per tonne of non-energy materials used. That figure is more than four times the approximately USD 900 per tonne generated by the OECD Key Partner economies, which include China.³⁶ This contextual comparison, together with the average PCE of 0.672 estimated in this study, suggests that China retains substantial scope to improve material productivity and general ISW prevention. The regional differences in general ISW management performance reflect the uneven distribution of technological capabilities, regulatory enforcement, and economic development levels across different provinces.³⁷ These variations indicate that a one-size-fits-all policy approach may not address China’s complex ISW challenges. Instead, region-specific strategies that account for local industrial structures, technological capacities, and environmental conditions may be better to improve overall system efficiency.³⁸ For example, in cities in coal-rich provinces such as Shanxi, Shaanxi, and Inner Mongolia, a strategy that combines energy recovery and ecological restoration may effectively address the environmental burden of legacy ISW. Specific activities could include developing coal gangue-fired combined heat and power plants to convert higher-calorific-value gangue into electricity and heat. For historically stockpiled low-calorific-value gangue, a backfill reclamation and landscape reconstruction approach could help create engineered landforms in mining subsidence areas and restore vegetation. In provinces with a high concentration of small and medium-sized enterprises, general ISW management lacks economies of scale. Local governments could use fiscal subsidies to incentivize specialty firms to build shared treatment and disposal facilities and provide small and medium-sized enterprises with one-stop services.³⁹

Second, EI made the largest positive contribution to the increase in general ISW storage and discard, whereas PCI made the largest contribution to reducing these volumes. This paradoxical relationship shows the complex interplay between economic growth and environmental sustainability in China’s industrial development trajectory.⁴⁰ The positive correlation between EI and general ISW storage/discard indicates that rapid industrial expansion has outpaced the development of adequate waste management infrastructure and circular economy practices. Temporary storage has become the buffer mechanism.⁴¹ In contrast, improved PCE shows that economic growth can be decoupled from general ISW generation using cleaner production technologies, resource optimization, and industrial symbiosis initiatives.⁴² To reinforce this decoupling, provincial government could provide performance-based subsidies to firms that reduce general ISW generation per unit of industrial value added or use industrial by-products as secondary materials. Furthermore, accelerated depreciation could be permitted for eligible cleaner-production and resource-recovery equipment. Making support conditional on verified reductions in generation intensity and storage and discard volumes would help prevent these incentives from merely financing production-capacity expansion. Digital transformation may have influenced both PCE and EE during the study period. Internet-of-things sensors, AI-enabled process monitoring, digital twins, and predictive-maintenance systems can identify production losses, optimize material inputs, and extend equipment life, potentially reducing general ISW generation relative to desirable output. Digital traceability and industrial symbiosis platforms can facilitate the exchange of industrial by-products. These technologies may also increase desirable output per unit of input and thereby improve EE. However, productivity-induced output expansion and digital infrastructure material requirements may partly offset these efficiency gains. Future research could combine indicators of industrial robot adoption, smart-manufacturing pilot policies, and enterprise digitalization with sector-level waste data to test these mechanisms.

Third, end-of-pipe treatment technologies have limited effectiveness in constraining general ISW storage and discard, illustrating a limitation in current waste management approaches. This study’s analysis indicates that traditional end-of-pipe approaches did not achieve the goal of reducing overall ISW storage and discard levels. There are several reasons for the limited effectiveness of end-of-pipe treatment, including technological constraints, economic considerations, and regulatory gaps.⁴³ Many end-of-pipe technologies are designed to address specific waste characteristics, and may not fit the diverse range of general ISW generated across different sectors.⁴⁴ The high capital and operational costs associated with advanced treatment technologies may discourage their widespread adoption, particularly among small and medium-sized enterprises. These firms make up a large part of China’s industrial landscape.⁴⁵ The inadequacy of end-of-pipe solutions highlights the need for upstream approaches for waste prevention, process optimization, and circular design principles. This could lead to more sustainable waste management outcomes. With the successive rollout of policy documents such as the Guiding Opinions on the Comprehensive Utilization of Bulk Solid Wastes during the 14th Five-Year Plan and the Catalog of Advanced and Applicable Technologies and Equipment for the Comprehensive Utilization of Industrial Resources (2021, 2023, and 2025 editions), and with the emergence of AI-driven sorting systems, it is expected that end-of-pipe treatments for general ISW will become more effective. The pace of comprehensive utilization and disposal should ultimately catch up with the pace of generation, end-of-pipe measures will be increasingly important in general ISW management.

Fourth, there were dynamic patterns in general ISW management over time, with notable policy-driven improvements after 2018 and pandemic-induced disruptions in 2020. These highlight the sensitivity of general ISW management systems to external shocks. Improvements in general ISW utilization and disposal rates after 2018 reflect the effectiveness of policy packages that combined regulatory enforcement, economic incentives, and technological support.⁴⁶ However, the pandemic-related disruptions observed during the 2020 pandemic period revealed the vulnerability of general ISW management systems to external shocks and supply chain disruptions.⁴⁷ The pandemic-induced obstacles to general ISW utilization and disposal highlight the interconnected nature of ISW generation and utilization (disposal) systems, as well as their dependence on logistical networks. This finding highlights the importance of building resilience into general ISW management systems through diversified disposal pathways, robust supply chains, and adaptive regulatory frameworks that can respond effectively to crises.⁴⁸

Limitations of the study

First, the bootstrap assessment showed generally high convergence, but convergence rates did vary across province-year observations (Table S2). Bounded bias-corrected estimates and 95% bounded basic (reverse-percentile) bootstrap intervals are provided as supplementary uncertainty diagnostics for the SBM-DEA estimates. The main LMDI decomposition and provincial attribution calculations use the original SBM-DEA estimates and preserve the exact decomposition identity of the LMDI model. Future research could extend the bootstrap procedure to quantify how uncertainty in the DEA-derived PCI, potential generation intensity, and EI affects the estimated contributions of the seven driving factors.

Second, the scope of the available data are limited in both temporal coverage and sectoral resolution. The study period ends in 2022 and therefore captures only the early implementation of the strengthened “zero-waste city” and industrial green transformation policies introduced between 2021 and 2025. In addition, the use of province-level aggregate data may conceal industry-level heterogeneity in the effects of the seven driving factors covering source prevention, process control, end-of-pipe treatment, and investment scale. Future research could extend the dataset beyond 2022 and construct a province-industry panel to examine whether the effects of the seven driving factors differ across industries. Where consistent data are available, firm-level analyses could further identify heterogeneity in production technologies, waste-management practices, and driver contributions across enterprises. Industry-specific analyses could prioritize high-volume waste-generating sectors like coal mining and coal-fired power generation, which are major sources of coal gangue and fly ash, respectively. These extensions would help assess the generalizability of the provincial-level findings and support more targeted industrial and waste-management policies. Such data could also support causal designs, including a difference-in-differences analysis of the post-2019 implementation of the “zero-waste city” initiative.

Resource availability

Lead contact

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Jiexin Tang (008008@yzu.edu.cn).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •Data: The processed province-year analytical dataset and the formula-preserving Excel workbooks supporting the LMDI decomposition and provincial attribution calculations have been deposited in Mendeley Data: https://doi.org/10.17632/43tn7yz3s5.1. The repository identifier is also listed in the key resources table.
• •Code: The original MATLAB and Python code underlying the reported SBM-DEA estimation, bootstrap uncertainty assessment, alternative DDF sensitivity analysis, and k-means clustering and validation has been deposited in Mendeley Data: https://doi.org/10.17632/43tn7yz3s5.1. The repository identifier is also listed in the key resources table.
• •All other items: Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This study was supported by the National Natural Science Foundation of China (grant no. 72373129), the Natural Science Foundation of Jiangsu Province (grant no. BK20240939), the General Project of Philosophy and Social Science Research in Colleges and Universities of Jiangsu Province (grant no. 2024SJYB1522), and the Humanities and Social Science Research Fund Project of Yangzhou University (grant no. xjj2023-10). We also gratefully acknowledge the anonymous reviewers and editors for their constructive comments and suggestions that significantly improved this manuscript.

Author contributions

Conceptualization, Q.W. and J.T.; methodology, J.T. and Q.W.; investigation, J.T., Q.W., and J.X.; writing – original draft, J.T.; writing – review and editing, Q.W. and J.X.; funding acquisition, J.T. and J.X.; resources, Q.W. and J.X.; supervision, Q.W. and J.X.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Deposited data
Standard coal equivalent	China Energy Statistical Yearbook	https://www.stats.gov.cn/english/
Capital stock, average annual employment	China Industrial Statistical Yearbook	https://www.stats.gov.cn/english/
Industrial value added, total industrial investment scale	China Statistical Yearbook	https://www.stats.gov.cn/english/
General ISW generated, general ISW stored, general ISW discarded, industrial pollution source treatment investment	China Environment Statistical Yearbook	https://www.stats.gov.cn/english/
Province-year analytical dataset	This paper	https://doi.org/10.17632/43tn7yz3s5.1
Formula-preserving Excel workbooks for the LMDI decomposition and attribution calculations	This paper	https://doi.org/10.17632/43tn7yz3s5.1
Software and algorithms
MATLAB R2024a	MathWorks	https://ww2.mathworks.cn/products/new_products/release2024a.html
Original MATLAB code for SBM-DEA estimation and sensitivity analysis	This paper	https://doi.org/10.17632/43tn7yz3s5.1
Python	Python Software Foundation	https://www.python.org/
Original Python code for K-means clustering and validation	This paper	https://doi.org/10.17632/43tn7yz3s5.1
Stata/MP 18	StataCorp	https://www.stata.com/statamp/
Microsoft Office 2021	Microsoft	https://www.microsoft.com/zh-cn/microsoft-365/microsoft-office

Experimental model and study participant details

Omitted because this study uses publicly available aggregate socioeconomic and environmental statistics at the provincial level and does not involve biological models or individual human participants.

Method details

Kaya identity

In the ISW process management system, the quantities of stored and discarded ISW are undesirable outputs, because storage occupies land, and discarded materials pollute the environment. Therefore, this study uses the sum of the quantities of general ISW storage and discard as a proxy indicator for general ISW treatment amounts. Using the Kaya identity, the general ISW storage and discard quantities are decomposed into seven key driving factors using a whole-process perspective, as shown in Equation 1. The first and the second terms of this equation express the factors driving source prevention. Specifically, the first term is the added value per unit of investment in production, representing the investment quality effect. The second term is the proportion of production investment in total industrial investment, representing the investment structure. The third to fifth terms are the driving factors for process control. The third term is the level of general ISW PCI, which represents the general ISW PCE factor. The fourth term is the potential general ISW generation intensity, which represents the gap in the intensity of general ISW generation compared to the frontier provinces. The fifth term is the EI level, which represents the economic efficiency factor. The sixth term is the factor driving end-of-pipe treatment, which is the ratio of general ISW storage and discard quantities to the general ISW generation quantity. This represents the end-of-pipe treatment effect. Finally, the seventh term is the total industrial investment scale. The resulting formula is as follows:

$$\begin{array}{ccc}IS{W}_i& =& \frac{IS{W}_i}{b_i}\ast \frac{b_i}{\left(1-{\alpha }_{b,i}\right)b_i}\ast \frac{\left(1-{\alpha }_{b,i}\right)b_i}{\left(1+{\alpha }_{v,i}\right)V_i}\ast \frac{\left(1+{\alpha }_{v,i}\right)V_i}{V_i}\ast \frac{V_i}{P{I}_i}\ast \frac{P{I}_i}{I_i}\ast I_i\\ & =& \underset{source}{\underbrace{\frac{V_i}{P{I}_i}\ast \frac{P{I}_i}{I_i}}}\ast \underset{process}{\underbrace{\frac{b_i}{\left(1-{\alpha }_{b,i}\right)b_i}\ast \frac{\left(1-{\alpha }_{b,i}\right)b_i}{\left(1+{\alpha }_{v,i}\right)V_i}\ast \frac{\left(1+{\alpha }_{v,i}\right)V_i}{V_i}}}\ast \underset{end}{\underbrace{\frac{IS{W}_i}{b_i}}}\ast I_i\end{array}$$ (Equation 1)

Where, for province i, b_i represents its general ISW generation; ISW_i represents its general ISW storage and discard; PI_i represents its industrial production investment; I_i represents its total industrial investment scale, and V_i represents the industrial added value. The terms α_b,i and α_v,i indicate the ISW generation reduction ratio and industrial added value increase ratio, respectively. The associated calculation is described in the following section addressing the efficiency evaluation model.

Baseline fixed-input output-oriented global VRS-SBM model

Identifying the driving factors of process control required measuring general ISW process control inefficiency (PCI), potential ISW generation intensity, and economic inefficiency (EI). A traditional radial DEA model assumes that all inputs and outputs change in the same proportion to reach the efficiency frontier. The model only measures this proportional distance and overlooks input excesses and output shortfalls. The DEA model also treats all outputs as desirable, which is incompatible with the dual goal of increasing economic benefits while reducing general ISW. Recasting general ISW as an input to force a reduction would contradict the basic principles of production theory. Therefore, this study constructed an SBM-DEA model that allows non-proportional adjustments and explicitly handles undesirable outputs, and this model was used to estimate the three indicators. For an economic system, it is particularly important to achieve the maximum economic value with the minimum input and the minimum generated ISW.^49,50 Assuming there are n provinces, each of which uses m inputs in the generation stage, resulting in l desirable outputs and e ISW. The term x_tij represents the input i of province j in the generation stages in period t. The term y_tij represents the desirable output i of province j in the generation stage in period t. The term b_tij represents the generated ISW i in province j in period t. The term $s_{tio}^{+}$ is the slack variable for the desirable output i of province o in period t. This indicates the amount by which the desirable output needs to increase to reach the production frontier. The term $s_{tio}^{b-}$ is the slack variable for the generated ISW i by province o in period t. This indicates the amount by which the generated ISW needs to decrease to reach the production frontier. Based on Tone et al.,⁵¹ an SBM-DEA model was constructed to estimate the overall efficiency of process control efficiency (PCE) and economic efficiency (EE) in an output-oriented situation, as shown in Equations 2 and 3. This makes it possible to decompose the proportion by which general ISW generation needed to be reduced (α_b,i) and the proportion by which industrial economy needed to expand (α_v,i) for inefficient provinces to achieve efficiency. Using α_b,i and α_v,i makes it possible to calculate general ISW PCI, potential generation intensity, and EI. The general ISW generation reduction ratio for province o in period t is calculated using Equation 4, and the industrial economic expansion ratio for province o in period t is calculated using Equation 5:

$${\beta }_0^t=\mathit{Min}.\frac{1}{1+\frac{1}{2}\left(\frac{s_{tio}^{+}}{y_{tio}}+\frac{s_{tio}^{b-}}{b_{tio}}\right)}$$ (Equation 2)

$$\begin{array}{l}\begin{array}{cc}s.t.& \sum _{t=1}^T\sum _{j=1}^n{\lambda }_{tj}{x}_{tij}\le x_{tio},t=1,...,T;i=1,...,m\end{array}\\ \begin{array}{cc}& \begin{array}{cc}\sum _{t=1}^T\sum _{j=1}^n{\lambda }_{tj}{y}_{tij}-s_{tio}^{+}=y_{tio},t=1,...,T;& i=1,...,l\end{array}\end{array}\\ \begin{array}{cc}& \sum _{t=1}^T\sum _{j=1}^n{\lambda }_{tj}{b}_{tij}\end{array}+s_{tio}^{b-}=b_{tio},t=,...,T;i=1,...,e\\ \begin{array}{cc}& \sum _{t=1}^T\sum _{j=1}^n{\lambda }_{tj}=1\end{array}\\ \begin{array}{cc}& s_{tio}^{+},s_{tio}^{b-}\ge 0\end{array}\end{array}$$ (Equation 3)

$${\alpha }_{tbo}=\frac{s_{tio}^{b-}}{b_{tio}}$$ (Equation 4)

$${\alpha }_{tvo}=\frac{s_{tio}^{+}}{y_{tio}}$$ (Equation 5)

The two equations 1/(1+α_tbo) and 1/(1+α_tvo) denote the general ISW PCE and the EE, respectively, of province o in period t. The term ${\lambda }_j^{\left(1,t\right)}$ represents the intensity vectors associated with the generation stage, with $\sum _{t=1}^T\sum _{j=1}^n{\lambda }_{tj}=1$ indicating variable returns to scale (VRS) in the generation stages. Equations 2 and 3 represent a nonlinear programming problem that needs to be converted into a linear programming form for a solution, as shown in Equations 6 and 7. The variable δ is the scaling factor. The general ISW generation reduction ratio and industrial economic expansion ratio for province o in period t were then calculated using Equations 8 and 9, respectively:

$${\beta }_0^t=\mathit{Min}.\delta \dots$$ (Equation 6)

$$\begin{array}{l}\begin{array}{cc}s.t.& \sum _{t=1}^T\sum _{j=1}^n{\mathrm{\Lambda }}_{tj}{x}_{tij}\le \delta x_{tio},t=1,...,T;i=1,...,m\end{array}\\ \begin{array}{cc}& \begin{array}{cc}\sum _{t=1}^T\sum _{j=1}^n{\mathrm{\Lambda }}_{tj}{y}_{tij}-S_{tio}^{+}=\delta y_{tio}& ,t=1,...,T;i=1,...,l\end{array}\end{array}\\ \begin{array}{cc}& \sum _{t=1}^T\sum _{j=1}^n{\mathrm{\Lambda }}_{tj}{b}_{tij}\end{array}+S_{tio}^{b-}=\delta b_{tio},t=1,...,T;i=1,...,e\\ \begin{array}{cc}& \sum _{t=1}^T\sum _{j=1}^n{\mathrm{\Lambda }}_{tj}=\delta \end{array}\\ \begin{array}{cc}& \delta +\frac{1}{2}\left(\frac{S_{tio}^{+}}{y_{tio}}+\frac{S_{tio}^{b-}}{b_{tio}}\right)=1\end{array}\\ \begin{array}{cc}& S_{tio}^{+},S_{tio}^{b-}\ge 0\end{array}\end{array}$$ (Equation 7)

$${\alpha }_{tbo}=\frac{s_{tio}^{b-}}{{\delta \ast b}_{tio}}$$ (Equation 8)

$${\alpha }_{tvo}=\frac{s_{tio}^{+}}{{\delta \ast y}_{tio}}$$ (Equation 9)

Alternative fixed-input global VRS non-radial DDF specification for sensitivity analysis

In efficiency measurement, both the directional distance function (DDF) model and the SBM model can address undesirable outputs. Equation 10 re-estimates the previously-defined industrial added value increase ratio and general ISW generation reduction ratio using a fixed-input global VRS non-radial directional distance function. All variables, indices, and symbols follow the definitions used in the baseline global VRS-SBM model. The alternative DDF specification uses the same inputs, desirable output, undesirable output, pooled global reference technology, and VRS assumption. The model includes one desirable output and one undesirable output, and equal weights (w_y=w_b=0.5) are assigned to the two outputs.⁵² For the sensitivity analysis, the ratios estimated by the baseline SBM model are replaced by the corresponding ratios estimated by the non-radial DDF model. The definitions and calculation formulas of PCI, potential general ISW generation intensity, and EI remain unchanged. The other four LMDI factors and the decomposition and attribution procedures are also unchanged.

$$\begin{array}{l}\max w_y{\alpha }_{tvo}+w_b{\alpha }_{tbo}\\ \begin{array}{cc}s.t.& \sum _{t=1}^T\sum _{j=1}^n{\lambda }_{tj}{x}_{tij}\le x_{tio},t=1,...,T;i=1,...,m\end{array}\\ \begin{array}{cc}& \begin{array}{cc}\sum _{t=1}^T\sum _{j=1}^n{\lambda }_{tj}{y}_{tij}\ge \left(1+{\alpha }_{tvo}\right)y_{tio}& ,t=1,...,T;i=1,...,l\end{array}\end{array}\\ \begin{array}{cc}& \sum _{t=1}^T\sum _{j=1}^n{\lambda }_{tj}{b}_{tij}\end{array}\le \left(1-{\alpha }_{tbo}\right)b_{tio},t=1,...,T;i=1,...,e\\ \begin{array}{cc}& \sum _{t=1}^T\sum _{j=1}^n{\lambda }_{tj}=1\end{array}\\ \begin{array}{cc}& {\lambda }_{tj}\ge 0\end{array},{\alpha }_{tvo}\ge 0,0\le {\alpha }_{tbo}\le 1\end{array}$$ (Equation 10)

Decomposition model

During [0,T], the relative change and the absolute change in general ISW storage and discard are represented by $D_{ISW}^{0,T}$ and ${\Delta }_{ISW}^{O,T}$, respectively, using the following specific formulas:

$$\begin{array}{lll}{D}_{ISW}^{O,T}& =& \frac{{ISW}^T}{{ISW}^O}\\ & =& \left(D_{IQal}^{O,T}\times D_{IStr}^{O,T}\right)\times \left(D_{GE}^{O,T}\times D_{PGStre}^{O,T}\times D_{VE}^{O,T}\right)\times D_{TQal}^{O,T}\times D_{IScal}^{O,T}\\ & =& D_{Pre}^{O,T}\times D_{Con}^{O,T}\times D_{Tre}^{O,T}\times D_O^{O,T}\end{array}$$ (Equation 11A)

$$\begin{array}{lll}{\Delta }_{ISW}^{O,T}& =& {ISW}^T-{ISW}^0\\ & =& \left({\Delta }_{IQal}^{O,T}+{\Delta }_{IStr}^{O,T}\right)+\left({\Delta }_{GE}^{O,T}+{\Delta }_{PGStre}^{O,T}+{\Delta }_{VE}^{O,T}\right)+{\Delta }_{TQal}^{O,T}+{\Delta }_{IScal}^{O,T}\\ & =& {\Delta }_{Pre}^{O,T}+{\Delta }_{Con}^{O,T}+{\Delta }_{Tre}^{O,T}+{\Delta }_O^{O,T}\end{array}$$ (Equation 11B)

Where, $D_{\mathrm{Pr}e}^{0,T}$ represents the temporal changes in general ISW storage and discard caused by source prevention. This is expressed as the ‘source prevention effect’, including the investment quality effect and investment structure effect. The term $D_{Con}^{0,T}$ represents the process control effect, including the PCI effect, potential generation intensity effect, and EI effect. The term $D_{Tre}^{0,T}$ represents the end-of-pipe treatment effect, containing one driving factor, i.e., end-of-pipe treatment effect. The term $D_O^{0,T}$ represents another driving effect, i.e., investment scale. For symbols not defined in the main text, please see Table S3 in supplemental information. The variation effects were calculated using the LMDI-II method. Using the investment scale effect as an example, during the period [0,T], the relative and absolute changes of this term were calculated using Equations 12A and 12B, respectively. The driving effects of other factors were calculated accordingly:

$$D_{ISca{l}_i}^{0,T}=\exp \left({\sum }_i{w}_i\ln \left(\frac{ISca{l}_i^T}{ISca{l}_i^0}\right)\right),w_i=\frac{L\left(ISW{P}_i^T,ISW{P}_i^0\right)}{{\sum }_{k=1}L\left(ISW{P}_k^T,ISW{P}_k^0\right)},ISW{P}_i^T=\frac{IS{W}_i^T}{IS{W}^T}$$ (Equation 12A)

$${\Delta }_{ISca{l}_i}^{0,T}=\frac{IS{W}_i^T-IS{W}_i^0}{\ln IS{W}_i^T-\ln IS{W}_i^0}\left(\ln ISca{l}_i^T-\ln ISca{l}_i^0\right)$$ (Equation 12B)

Attribution analysis

This study’s decomposition model identifies seven factors driving general ISW storage and discard. The in-depth analysis includes exploring the contributions of different provinces to each driving factor. This provides a reference for formulating provincial policies to reduce general ISW storage and discard. This involved adopting the AA method proposed by Choi and Ang.⁵³ Taking investment scale (IScal) as an example, the single-period attribution analysis equation was expressed as follows. The other six driving factors were derived accordingly:

$$D_{IScal}^{\left(t-1,t\right)}-1=\sum _{i=1}^N{c}_{IScal,i}^{\left(t-1,t\right)}=\sum _{i=1}^N{s}_i\left(\frac{ISca{l}_{i,t}}{ISca{l}_{i,t-1}}-1\right)$$ (Equation 13a)

$$s_i=\frac{{\pi }_i{I}_{i,t-1}}{\sum _{k=1}^N{\pi }_k{I}_{k,t-1}}=\frac{\frac{W_i}{L\left(I_{i,t},I_{i,t-1}\ast D_{IScal}^{\left(t-1,t\right)}\right)}{I}_{i,t-1}}{\sum _{k=1}^N\frac{W_k}{L\left(I_{k,t},I_{k,t-1}\ast D_{IScal}^{\left(t-1,t\right)}\right)}{I}_{k,t-1}}$$ (Equation 13b)

$${\pi }_i=\frac{w_i}{L\left(I_{i,t},I_{i,t-1}\ast D_{IScal}^{\left(t-1,t\right)}\right)}$$ (Equation 13c)

Where, $D_{IScal}^{\left(t-1,t\right)}-1$ indicates the single-period percentage change in investment scale from time t-1 to t; and $c_{IScal,i}^{\left(t-1,t\right)}$ represents the contribution coefficient of province i to the percentage change in investment scale from time t-1 to t. The multi-period attribution results were directly derived from single-period attribution results, as follows:

$$D_{IScal}^{0,T}-1=\sum _{i=1}^I{c}_{IScal,i}^{0,T}=\sum _{i=1}^I\sum _{t=0}^T{D}_{IScal}^{0,t}{c}_{IScal,i}^{t,t+1}$$ (Equation 14)

Where, $D_{IScal}^{0,T}-1$ indicates the multi-period percentage change in investment scale from time 0 to T; and $c_{IScal,i}^{0,T}$ indicates the contribution coefficient of province i to the percentage change in investment scale from time 0 to T. This reflects the degree of contribution of that province to the change in a specific driving factor.

Data collection sources

This study uses industrial data from 30 provinces in Mainland China from 2011 to 2022, excluding Tibet due to data unavailability. All data were sourced from the China Statistical Yearbook, the China Environment Statistical Yearbook, the China Industrial Statistical Yearbook, and the China Energy Statistical Yearbook. These publications are compiled with the participation of the National Bureau of Statistics of China, and are considered to have significant authority and reference value. The China Environmental Statistical Yearbook, however, reports only aggregate figures for general ISW, without disaggregating by specific types such as fly ash, slag, coal gangue, or sludge. This data constraint therefore limits the present study to an exploratory analysis of the drivers underlying general ISW storage and discard. Using 2011 as the base year, price factors were removed from all monetary indicators. The specific indicator statistics for efficiency evaluation and drivers are in Table S4. A description of the selection and explanation of the specific indicators is as follows:

• (1)
  Input indicators for efficiency evaluation. This study used the perpetual inventory method to estimate the industrial capital stock of each province,⁵⁴ using Equation 15. The terms K_i(t-1) and K_it represent the capital stock of province i in year t-1 and year t respectively; σ_it represents the capital depreciation rate of province i in year t; and I_it represents the total industrial investment of province i in year t.⁵⁵ The total industrial investment was deflated using the provincial fixed asset investment price index. Industrial energy consumption was sourced from provincial energy balance sheets. All energy types were converted to standard coal equivalent.⁵⁶ Labor was represented by the average annual employment in above-scale industrial enterprises.⁵⁷

  $$K_{it}=K_{i\left(t-1\right)}\left(1-{\sigma }_{it}\right)+I_{it}$$ (Equation 15)
• (2)Desirable output indicators for efficiency evaluation. This study selected industrial value added as a proxy indicator for economic benefits. The value was deflated using the provincial industrial producer price index.
• (3)Undesirable output indicators for efficiency evaluation. General ISW generation was considered an undesirable output indicator for process control.
• (4)Indicators for explained variables and drivers. General ISW storage and discard were calculated as the sum of general ISW storage and discard volumes. Industrial production investment was obtained by subtracting industrial pollution source treatment investment from the total industrial investment scale. Seven categories of factors driving general ISW storage and discard were constructed based on indicators, including industrial production investment, total industrial investment, industrial value added, general ISW storage and discard volume, and general ISW generation.

Quantification and statistical analysis

K-means clustering and cluster validation

For both general ISW generation and general ISW storage/discard, each province was treated as the clustering unit, and the 12 annual observations from 2011 to 2022 were treated as clustering features. Before clustering, each annual feature was standardized across the 30 provinces to have a mean of 0 and a population standard deviation of 1. No missing values remained in either 30 × 12 feature matrix. The K-means solutions for k = 1–8 were estimated using k-means++ initialization, 25 initializations, the Lloyd algorithm, and a fixed random seed of 42. Cluster-number diagnostics included the within-cluster sum of squares (WCSS), a silhouette coefficient, the Davies–Bouldin index, and a gap statistic. The gap statistic was calculated using 500 uniformly-generated reference datasets within the observed feature-wise ranges, with a fixed reference-data seed of 20260710. For each reference dataset, K-means was fitted using 10 initializations and a deterministic random seed equal to the reference-dataset index. Gap-statistic standard errors were calculated and used to evaluate the one-standard-error stopping rule. The complete results are reported in Table S5.

For general ISW generation, the silhouette coefficient was highest at k = 2 (0.693). The Davies–Bouldin index at k = 2 (0.500) was lower than that at k = 3 (0.605), although the index’s global minimum within k = 1–8 occurred at k = 8 (0.471). For general ISW storage and discard, the silhouette coefficient was highest at k = 3 (0.631), and the Davies–Bouldin index at k = 3 (0.493) was substantially lower than that at k = 2 (0.732), although the global minimum again occurred at k = 8 (0.436). The gap statistic increased monotonically for both datasets, and the one-standard-error rule did not identify an interior stopping point within k = 1–8. Thus, the diagnostics did not establish a single common statistical optimum for general ISW generation and general ISW storage and discard.

This study retained k = 3 for both figures as a pragmatic visualization choice that provides a parallel three-panel presentation of the provincial trajectories. The two datasets were clustered separately, and the cluster memberships of the two datasets should not be interpreted as directly equivalent. The clustering was used only to organize Figures 2 and 3 and did not enter the subsequent SBM-DEA estimation, LMDI decomposition, or provincial attribution analysis.

This study further assessed the stability of the retained k = 3 solutions. Agreement between K-means and Ward hierarchical clustering, measured using the adjusted Rand index (ARI), was 0.596 for general ISW generation and 0.746 for general ISW storage and discard. Across 500 random 80% provincial subsamples, the mean ARI relative to the corresponding full-sample assignments was 0.768 (SD = 0.182; median = 0.842) for general ISW generation and 0.821 (SD = 0.209; median = 1.000) for general ISW storage and discard. The corresponding mean subsample silhouette coefficients were 0.506 (SD = 0.030) and 0.606 (SD = 0.045), respectively. These results indicate a reasonable degree of assignment stability for the descriptive purpose of the clustering, with stronger stability for general ISW storage and discard. Complete stability results are reported in Table S6.

Bootstrap uncertainty assessment of the SBM-DEA estimates

Uncertainty in the SBM-DEA estimates was assessed using 2,000 bootstrap replications in MATLAB R2024a. In each replication, a reference set of the same size was sampled with replacement from the pooled global reference observations; the global VRS SBM model was re-estimated for all 360 province-year observations, using MATLAB’s linprog function with the dual-simplex algorithm. A fixed random seed of 12345 was used. Replications without a valid optimal solution were treated as missing. Bias-corrected estimates were calculated as twice the original estimate minus the mean of the valid bootstrap estimates and were bounded by the theoretical parameter space of the corresponding measure. Two-sided 95% basic (reverse-percentile) bootstrap intervals were constructed as $\left[2{\hat{\theta }}_i-Q_{0.975}\left({\hat{\theta }}_i^{\ast }\right),2{\hat{\theta }}_i-Q_{0.025}\left({\hat{\theta }}_i^{\ast }\right)\right]$. Each interval was intersected with the corresponding theoretical parameter space: [0,1] for the overall SBM efficiency score and the general ISW generation reduction ratio, and [0,∞)] for the desirable-output expansion ratio. Intervals were reported only when at least 50% of the 2,000 replications were valid.

The overall convergence rate was 95.95%. All 360 province-year observations exceeded the 50% reporting threshold. Specifically, 320 achieved convergence rates of at least 85%, while the remaining 40 ranged from 60.95% to 84.65%. The mean original and bias-corrected overall SBM efficiency scores were 0.6742 and 0.6570, respectively. The mean absolute differences between the original and bias-corrected estimates were 0.0172 for the overall SBM efficiency score, 0.1004 for the desirable-output expansion ratio, and 0.0540 for the general ISW generation reduction ratio. The overall SBM efficiency score is distinct from PCE, which is derived from the observation-specific general ISW generation reduction ratio. Bias correction was applied separately to each reported measure, and the resulting bias-corrected values were used only as measure-specific uncertainty diagnostics. Complete province-year convergence statistics, original estimates, bounded bias-corrected estimates, and 95% bounded basic (reverse-percentile) bootstrap intervals are reported in Table S2.

Sensitivity analysis using the alternative DDF specification

To assess sensitivity to the DEA efficiency-measure formulation, the three DEA-derived factors were re-estimated using the fixed-input global VRS non-radial DDF specified in Equation 10. To isolate the effect of replacing the baseline slack-based SBM with the non-radial DDF, the fixed-input global VRS reference technology was held constant. The absolute differences in the national cumulative effects were below 2.5×10^-7, and the province-year rank correlations for PCI, potential general ISW generation intensity, and EI were 0.9989, 1.0000, and 0.9949, respectively. The seven-factor contribution ranking remained unchanged. These results indicate that the principal factor ordering is stable under this alternative DEA formulation (Table S7).