S. Hu, H. Qi, Z. Wang et al.
Environmental Science and Ecotechnology 30 (2026) 100682
However, energy-based accounting approaches neglect direct carbon emissions from pulping and wastewater treatment pro- cesses, resulting in the omission of approximately 20% of total emissions in energy-based emission estimates [10] and potentially introducing substantial biases into total emissions estimations. In addition to methods based on energy consumption, plant- level carbon emissions can be estimated by integrating publicly available production-capacity data with product-specific carbon- emission factors. This method has been applied to construct large- scale, plant-level carbon emission inventories for the refining [6], steel [11], and primary aluminum smelting [12] industries. It has also been extended to the PPI through life cycle assessment (LCA), which is used to estimate carbon emission factors for major paper products in China [13]. In 2015, by integrating these factors with plant-level production-capacity data, researchers developed a comprehensive carbon-emission inventory covering 814 PPPs across China [14]. In this inventory, production capacity is treated as an indirect indicator of emissions and used to estimate plant- level carbon emissions using product-specific carbon-intensity parameters. It has certain advantages, such as broad applicability and ease of data collection [15]; however, variations in process technology, raw material compositions, and scale effects across plants lead to pronounced differences in product-level carbon emission intensities [16]. These differences undermine the accu- racy of carbon emission estimates, even within the same industrial sector. Moreover, this approach is further complicated by incon- sistent system boundaries and ambiguous emission attribution to individual plants [17]. Remote-sensing imagery — a high-resolution observational tool — allows the spatial layouts of PPPs, including production buildings, raw material storage zones, and wastewater treatment areas [18], to be identified. Such spatial information supports differentiation among raw material structures and scale-related characteristics across plants, providing spatial data for carbon accounting [19]. However, such imagery, when used alone, often cannot distinguish among different types of PPPs, because plants with similar spatial layouts may differ substantially in product types and emission characteristics. In contrast, textual information captures knowledge that cannot be directly obtained from images, including plant product names and structures. Multimodal data fusion refers to the integration of heterogeneous data from mul- tiple sources or modalities to provide a more comprehensive and accurate representation of a system or process. Multimodal data fusion approaches have been applied to address the limitations of single-source remote-sensing imagery, including urban waste pile detection [20], urban village detection [21], and urban area func- tional identification [22]. Therefore, building on these advances, we integrated plant-level remote-sensing imagery with product semantic information and used a multimodal fusion framework to improve the accuracy of PPP classification and plant-level carbon emission estimation. Estimating carbon emissions at the plant level improves the compilation of carbon emission inventories and provides a necessary foundation for differentiated CER management. At the plant scale, currently implementable decarbonization options for PPPs include improving energy efficiency, adjusting the energy mix, increasing waste paper utilization, and deploying carbon capture and storage (CCS) technologies [23]. Optimizing production processes, improving equipment retro- fitting, and enhancing thermal and electrical energy efficiency can reduce carbon emissions by 6.8 – 192.1 kg of carbon dioxide (CO 2 ) per ton of paper [24]. However, substantial heterogeneity in pro- duction equipment lifetimes and process configurations leads to pronounced differences in energy consumption patterns across PPPs. Consequently, establishing a unified retrofitting strategy that
does not disrupt normal production is challenging [25]. Substituting recycled fibers for virgin wood pulp can reduce car- bon emissions by approximately 25% [26], but China’s ban on imported waste paper has greatly reduced recycled fiber re- sources, resulting in an estimated supply gap of around 30 million tons [13]. CCS technology can mitigate the high-concentration direct emissions from thermal power units in PPPs. However, in practice, most Chinese PPPs do not use thermal power units, and their emissions mainly arise indirectly from electricity consump- tion and steam supply [27]. Furthermore, the absence of central- ized emissions sources, together with high capital investment requirements, elevated operational costs, and insufficient CO 2 transport and storage infrastructure, severely restricts the appli- cation of CCS at the plant level [28]. At present, optimizing the energy structure at the plant level seems to be the most feasible pathway for reducing carbon emissions in China’s PPI [29]. Increased use of renewable energy, such as from green electricity and photovoltaic (PV) systems, can reduce the indirect emissions associated with electricity con- sumption [30]. However, the emissions reduction benefits of green electricity depend on the cleanliness of the regional power grid and the external energy structure [31]. In regions dominated by fossil fuel power generation, the mitigation effect of green elec- tricity substitution is limited. In contrast, PV power generation offers great potential and flexibility for industrial decarbonization. Previous studies have demonstrated that PV systems installed on plant rooftops or within industrial parks can reduce a portion of purchased electricity consumption [32,33]. The roofs of PPPs are typically spatially distributed, underutilized, and centrally managed, making them particularly suitable for deploying distributed PV systems. When combined with energy storage systems, PV systems facilitate tiered energy utilization, thereby accelerating the PPI decarbonization process [34]. Recent studies on PPI decarbonization strategies based on PV power have pri- marily focused on national-scale assessments [35]. Few systematic assessments of the potential of PV power generation for plant- level decarbonization have been conducted. To address unclear boundaries and substantial estimation bia- ses in existing carbon accounting methods for PPPs, we developed a multimodal carbon accounting framework by integrating infor- mation from remote-sensing images and textual data. First, we developed a multimodal classification strategy based on the DeepLabv3 + semantic segmentation model and bidirectional encoder representations from the transformers (BERT) text-based model. This facilitated the accurate identification and categoriza- tion of PPPs based on their spatial and semantic features. Second, we developed a data-driven, differentiated carbon-accounting model to estimate 2022 carbon emissions for 720 PPPs and to provide a plant-level emissions inventory for China’s PPI. Finally, we incorporated solar radiation and geographic data to assess the potential of deploying PV systems within existing PPPs to reduce carbon emissions and support actionable decarbonization solu- tions at the plant level.
2. Methodology
This study aimed to address key carbon-accounting issues in China’s PPI, including low spatial resolution of region-level statis- tical data, limited data dimensions, and challenges in accounting for plant heterogeneity. By integrating image recognition, natural lan- guage processing, and data-driven modeling, we developed a framework for estimating plant-level carbon emissions. The overall framework of this study consists of two core components — carbon accounting and an assessment of CER potential (Fig. 1). These two components are described in detail below:
2
Made with FlippingBook interactive PDF creator