1 Introduction

In 2004, the National Renewable Clean Energy Laboratory (NREL) proposed that biomass, similar to petroleum refining, can be converted into value-added chemicals [1]. Levulinic acid, as one of the 12 top platform chemicals identified by NREL, can be further transformed into various downstream products such as γ-valerolactone [2, 3], angelica lactone [4, 5], 2-methyl tetrahydrofuran [6], valeric acid [7], 1,4-butanediol [8], and levulinate esters [9]. Among these value-added products, levulinate esters are identified as the most vital energy chemicals due to their fuel-blending properties which can assist to achieve the target of European Union to secure the future energy demands [10]. Compared with alcohol fuels, levulinate esters have a higher energy density, better hydrophobicity, and lower volatility. These excellent characteristics are more suitable as fuel additives [11, 12]. Currently, the cleaner combustion properties of levulinate esters/diesel compound fuel have been confirmed by many tests [13,14,15]. Upon combustion, the carbon fixed by plant photosynthesis returns to the atmosphere, making bio-based levulinate esters a promising near-zero-emission fuel [16].

The above near-zero-emission conclusions are based on the material flow of biomass conversion, however, it should be noted that the production of these chemicals not only consumes biomass materials but also consumes energy, catalysts, solvents, etc. For comprehensive consideration and comparison, a useful tool is needed to demonstrate the emission reduction potential of bio-based chemicals. Life cycle assessment (LCA) quantitatively examines all the environmental impacts that occur throughout the entire value chain (supply chain plus use and disposal, i.e. “cradle to grave”) of a product [17,18,19,20]. At present, LCA has been used by many researchers to investigate the environmental impacts of levulinic acid chemicals.

Table 1 summarizes the current LCA studies of levulinic acid derivatives. The vast majority of existing research focused on the preparation of levulinic acid, with only a few reports related to the preparation of ethyl levulinate. The scope of their investigation is mainly “cradle to gate”, covering from the planting of the crops to the delivery of the products. Fiorentino et al. performed the first evaluation of a biomass refining model using the LCA method [21]. The raw material of the model comes from non-grain crops (Brassica carinata) in Campania, Italy and hypothetical cases were proposed regarding the local chemical factory and the private investigation of the author. Furthermore, Wang et al. reported the LCA study of ethyl levulinate based on an actual biomass pilot plant in China [22], which was the only case where foreground data were obtained from a real plant. Sensitivity analysis showed that the environmental impact of the factory stage accounts for the vast majority (96.8%) of the entire life cycle. However, the description of the operation of the factory is vague, and there is no additional information on the breakdown of the factory stage, so it cannot identify which process contributes to the main environmental impact. Finally, the remaining literatures in Table 1 obtain foreground datasets mainly through laboratory or simulation. Due to differences in raw materials, production processes, waste disposal, and distribution methods, the greenhouse effect corresponding to the production of 1 kg of levulinic acid varies greatly (from 2.4 kg CO2-eq to 11.6 kg CO2-eq). Furthermore, quantitative data cannot be obtained sometimes. The reason may be the different functional unit, e.g., the functional unit of Gonzalez-Garcia et al. is dealing with 100 kg dried Pinus pinaster chips [23]. Therefore, regardless of any changes in product type or quantity, the functional unit remains constant. As a result, there is no need to set the allocation method and it is impossible to obtain environmental impact results for each product. The same situation applies to Khoo et al. [24] and Chaparro-Garnica et al. [25].

Table 1 Current LCA studies on levulinic acid chemicals
Full size table

In addition, uncertainty analysis (UA) has always been the difficulty of life cycle assessment. Bamber et al. reviewed recent LCA literatures and found that UA was not widespread (less than 20% of LCA studies published in the past 5 years reported any kind of uncertainty analysis), even when it was applied, the focus was often on parameter-related uncertainty [31]. Other sources of uncertainty in LCA such as scenario and model uncertainty are considered equally important but receive less attention [32]. The same situation also exists in the LCA studies of levulinic acid derivatives. From Table 1, it can be found that only two cases have used the Monte Carlo method for uncertainty analysis, while the remaining literatures lack corresponding analysis. Compared with actual factories, the uncertainty of laboratory and simulated data will increase. Since the data in many literatures were obtained from laboratory or simulation, the LCA results obtained from these data need further uncertainty assessment to be of significant reference for practical production. Besides, the location represented by these studies varied widely, including Europe, Asia, and America, but most of them chose Ecoinvent as background database. The environmental impact of the same product in different regions sometimes has considerable differences, therefore, the geographical, time, and technical differences between the background and the foreground data should also be considered [33,34,35].

In this study, we report the life cycle assessment of production of methyl levulinate (ML) in real biomass refineries in China whose feedstock is rice straw from cropland. This study uses the LCA software (eFootprint) combined with the local database (Chinese Life Cycle Database, CLCD) to model, calculate and analyze the whole process. eFootprint, an online platform designed and developed by China IKE Environmental Technology Co., Ltd. [36], has many advantages such as complete record data collection, integrity check, material balance check, and uncertainty analysis. At present, eFootprint and CLCD have been used by many researchers to investigate the environmental impact of Chinese local products [37,38,39,40]. To overcome the unclear objectives and ambiguous process description of life cycle assessment, a detailed and transparent model is presented and elaborated. The target representativeness, technical representativeness, geographical representativeness, temporal representativeness, etc. of each product are classified and interpreted for follow-up research and comparison. Given the lack of sufficient uncertainty analysis in literatures at present, the data quality assessment system in eFootprint is applied to assess the LCA results uncertainty of the ML supply chain. At the same time, we also select the Ecoinvent as the background database and make a corresponding uncertainty analysis for comparison. Finally, to explore the potential for further energy saving and emission reduction, an Aspen process simulation was done based on the sensitivity analysis and the actual production process.

2 Materials and methods

This LCA study was performed following international standards ISO 14040 [41] and ISO 14044 [42]. It mainly includes four steps: goal and scope definition, inventory analysis, impact assessment, and interpretation.

2.1 Goal and scope definition

2.1.1 Goal

The goal of this study is to investigate the environmental impact of producing methyl levulinate (ML) from rice straw, representing the technology and supply chain information of specific companies. The specific goals are as follows:

  1. (1)

    Investigate the production of ML in real biomass refineries in China, and identify key processes with high environmental impact to provide reference for subsequent technical improvement;

  2. (2)

    Comparing ML with diesel to explore whether biomass fuel has an emission reduction effect;

  3. (3)

    Conduct a comprehensive and detailed assessment of data quality to compensate for the lack of effective uncertainty analysis in LCA studies of levulinic acid chemicals. Use non-native background data to explore the impact of background data selection on the uncertainty of LCA results;

  4. (4)

    Explore the potential of energy saving and emission reduction through Aspen simulation;

2.1.2 System boundary

The boundary of this system is “cradle to grave”. As shown in Fig. 1, the whole process starts from the waste straw collection of Pujiwei Farm in Tongling (Process 1). Compared with food, straw is a low-value by-product in agricultural production, so it is recognized as “zero environment burden”. The rice straw is packed and transported to the nearby biomass refinery of Zhongke Wanye Co., Ltd. Here, rice straw undergoes a three-element separation process to obtain high-purity lignin and cellulose (Process 2). The cellulose is extruded and transported to the biomass refinery of Hefei Leaf Biotechnology Co., Ltd. Through catalytic conversion, cellulose is converted into ML and by-products such as methyl formate (Process 3), and the residual humin is extruded and sent to biomass thermal power plant for incineration (System expansion). Finally, the ML is transported and mixed with diesel to be burned in the engine, which is the “gate-to-grave” stage (Process 4).

Fig. 1
figure 1

System boundary diagram

Full size image

2.1.3 Functional unit and allocation method

The functional unit for this life cycle model is set as 1 kg ML (purity > 98 wt%).

In a multifunctional process, the environmental burden must be allocated to different products and co-products obtained during the life cycle. Usually, there are three kinds of allocation methods: mass, energy, and economic value. First, the form and function of the products are different, so the use of energy allocation is not appropriate. Second, obtaining prices for ML is not easy due to the current low capacity and small market size. Furthermore, the price of certain intermediary products (such as cellulose) remains inaccessible due to inter-enterprise collaboration. Therefore, it is difficult to apply economic allocation to this life cycle model. So, to obtain the deterministic values and simplify the model, this study chooses the mass allocation method to assign the environmental burden of the products.

2.2 Inventory data quality evaluation and result uncertainty analysis

eFootprint has a complete and detailed data quality evaluation system, which not only evaluates the parameter uncertainty of the foreground data itself but also gives various uncertainty scores for the comparison between the target representation and actual representation. In addition, there are also corresponding uncertainty scores for the geographical, time, and technical differences between the background data and the foreground data. Finally, the system integrates these uncertainties and synthesizes the total inventory data uncertainty, and uses Taylor series expansion instead of Monte Carlo simulation to transfer the inventory data uncertainties to the final LCA results uncertainty. In this way, the system can not only obtain the uncertainty of LCA results but also identify the main sources of uncertainty according to the order of inventory data uncertainty contribution.

2.3 Process simulation

The research on energy conservation and emission reduction was mainly carried out through the chemical simulation software Aspen Plus V11. The simulation mainly refers to the real process. The physical and chemical properties of biomass are mainly obtained from NREL’s public reports [43].

2.4 Life cycle inventory (LCI) and process description

2.4.1 Process 1: field waste rice straw collection

The biomass raw material comes from the discarded straw at Pujiwei Farm in Tongling. The collection process of rice straw storage and transportation is shown in Fig. 2a. The collection of rice straw is mainly realized by bundling machines, which break and compress the raw straw into square or round bales. Bales of straw are loaded and transported to a nearby factory (about 20 km). The main consumption of this process is diesel and its dosage is mainly obtained through surveys of the activities of biomass collection. The emissions from diesel combustion are calculated regarding IPCC Guidelines for National Greenhouse Gas Inventories [44]. The average data is shown in Table 2, and more details are given in Table S1. These data are derived from reasonable hypotheses and applied to the straw collection process at Pujiwei Farm.

Fig. 2
figure 2

a Schematic diagram of rice straw collection; b Schematic diagram of cellulose production in Tongling Biomass Refinery; c Schematic diagram of methyl levulinate production in Bengbu Biomass Refinery

Full size image
Table 2 Process 1 inventory data (field waste rice straw collection based on reasonable hypothesis)
Full size table

2.4.2 Process 2: cellulose production (Tongling Biomass Refinery)

The technological process of the Tongling Biomass Refinery is shown in Fig. 2b. The rice straws are further broken by a crusher and then sent to the percolating bed reactor through a conveyor belt. About 200 kg of straw is loaded into one reactor and multiple reactors are run in parallel at a time. Then the p-toluenesulfonic acid (PTSA)/glycol solution is heated and injected into the percolation bed. After a long period of flow and circulation, the solution gradually separates and dissolves hemicellulose and lignin in rice straw. When the reaction is over, most of the cellulose remains in the reactor, while hemicellulose and lignin enter the percolate. The cellulose is taken out and sent to the granulator for extruding granulation. High-purity lignin can be obtained from percolating after polar transformation and decanting. This lignin is sold as high value-added material. After filtration, the remaining percolate enters a vacuum tank and undergoes distillation to separate the polar solvent. The residual glycol solution in the tank contains hemicellulose oligomers which can be separated and filtered through the membrane. These hemicellulose oligomers are currently collected and stored in the factory and do not enter subsequent life cycle processes as commodities or downstream raw materials. Therefore, in our setting, the environmental impact of this process is allocated to lignin and cellulose, while hemicellulose oligomers do not bear any environmental impact. The inventory data of this process is obtained from the production ledger of Zhongke Wanye Co., Ltd. See Tables 3 and 4 for specific data.

Table 3 Process 2 inventory data (Straw crushing process based on production ledger of Zhongke Wanye Co., Ltd.)
Full size table
Table 4 Process 2 inventory data (Cellulose production based on production ledger of Zhongke Wanye Co., Ltd.)
Full size table

2.4.3 Process 3: ML production (Bengbu Biomass Refinery)

The process flow of the Bengbu Biomass Refinery is shown in Fig. 2c. The sulfuric acid/methanol solution is first loaded into the enamel reaction kettle, and then cellulose particles transported from Tongling are put in. The reactor is subsequently heated to a specific temperature and maintained at that level for a designated duration. Upon completion of the reaction, the reactor temperature is reduced to approximately 30 °C through the circulation of cooling water. The vacuum pump is then turned. Light components in the kettle such as methyl formate and methanol are distilled out. Methanol and methyl formate are collected respectively after passing through the two-stage condenser and the concentrated solution is removed from the bottom of the reactor into the neutralization reactor for neutralization treatment. In the neutralization kettle, slowly add sodium hydroxide and stir with the motor until the solution is weakly alkaline. Due to the low solubility of the neutralized product (sodium sulfate) and its strong water absorption capacity, it precipitates in the concentrated solution and brings out water. After filtration, the clear liquid is sent to the distillation tank for further distillation. Methanol and ML are obtained by two-stage vacuum distillation. Finally, the humin at the bottom of the kettle is sent to the power plant for incineration after pressure filtration. The inventory data of this process is obtained from the production ledger in Bengbu Biomass Refinery of Hefei Leaf Biotechnology Co., Ltd. See Table 5 for specific data.

Table 5 Process 3 inventory data (Methyl levulinate production based on production ledger of Hefei Leaf Biotechnology Co., Ltd.)
Full size table

2.4.4 System expansion (Biomass Thermal Power Plant)

Humin from the Bengbu Biomass Refinery is considered as solid fuel and sent to the Biomass Thermal Power Plant for incineration. A brief description of the process is as follows: the humin is sent to the bottom of the boiler for combustion, and the desalinated water is heated into steam, which is sent to the steam turbine to generate electricity. The condensed water is circulated and pumped by the heating network to the heat-consuming units. Users in the heating network can play the role of cooling towers. The electricity and heat supplies can offset the corresponding environmental impact. For the convenience of modeling, negative numbers are displayed in the data set. Finally, the gas from the boiler is purified by ammonia denitrification, cyclone dust removal, and gypsum desulfurization, and then discharged through the chimney. The data for this process was obtained by investigating the environmental impact assessment (EIA) reports of several biomass power plants in China. Here, the postprocessing of humin of Bengbu Biomass Refinery is combined with data from other biomass thermal power plants as a hypothetical process for solid waste post-treatment. The average data is shown in Table 6, and more details are given in Table S2.

Table 6 System expansion inventory data (Biomass Thermal Power Plant based on reasonable hypothesis)
Full size table

2.4.5 Process 4: combustion in engine

ML can be used as an additive to replace part of the diesel. Due to the current capacity limit (1000 t/a) of Bengbu Biomass Refinery, it can be fully digested by nearby petrol stations, so the transport and distribution radius of ML is assumed to be 200 km. ML is then mixed with diesel and burned in engines. The CO2 emission data is calculated according to the chemical equation of 100% oxidation: C6H10O3 + 7 O2 = 6 CO2 + 5 H2O. This part is the “gate-to-grave” stage of the entire life cycle. See Table 7 for specific data.

Table 7 Process 4 inventory data (Combustion in engine based on the chemical equation of 100% oxidation)
Full size table

2.5 Life cycle impact assessment (LCIA)

This model selects nine types of common characteristic environmental indexes: Global warming potential (GWP, 100 yr), Primary energy depletion (PED), Abiotic depletion (ADP), Water usage (WU), Acidification (AP), Eutrophication potential (EP), Respiratory inorganics (RI), Ozone layer depletion (ODP), Photochemical ozone formation (POFP). Most of them choose CML 2002 as the basic evaluation method, and other evaluation methods are selected as supplements for the indexes that CML 2002 does not include, such as WU, RI, POFP, etc. See Table S3 for the specific index list and units.

3 Results

3.1 Life cycle assessment results

Table 8 lists the environmental impact results of ML from different stages of this model. It can be seen from the data that all environmental impact indicators for 1 kg of ML are positive across the entire life cycle (raw material acquisition, factory production, and final combustion in the engine, i.e. “cradle-to-grave”). This result shows that the whole life cycle process of biomass-based ML has certain side effects on the environment. Among them, the global warming potential (GWP) is 2.90 kg CO2-eq, the primary energy depletion (PED) is 78.2 MJ, and the water usage (WU) is 34.7 kg, all of which are considerable. This indicates that bio-based products do not always have the desired positive impact on the environment.

Table 8 Life cycle assessment results of products at different stages
Full size table

Furthermore, the GWP of 1 kg ML in “cradle-to-gate” stage is 2.55 kg CO2-eq, that is, the subsequent transportation and combustion process of methyl levulinate only increases the GWP of 0.35 kg CO2-eq. According to the stoichiometric equation of 100% oxidation of ML, the complete combustion of 1 kg of ML will produce about 2.03 kg of CO2 emissions, which is inconsistent with the above LCA results. The reason is that the five carbon atoms in ML come from biomass cellulose, and only the carbon atom on the methoxyl group comes from fossils. The methoxyl in ML is derived from methanol in Process 3, most of which is made from coal due to the Chinese special energy structure. So, 1/6 of CO2 emissions are attributed to fossil sources and the remaining 5/6 of CO2 emissions are counted as non-fossil sources (Table 7). In the calculation method of IPCC, carbon emissions from non-fossil sources will not be included in the final global warming potential.

3.2 Sensitivity analysis and critical process identification

Sensitivity analysis identifies each process of the product life cycle and obtains its proportion in the overall result, which is an essential reference for determining the key process and proposing improvement plans. Figure 3a shows the decomposition of the environmental impact of the ML from “cradle-to-grave”. It can be found that the ML production stage is the main contributor to all types of environmental impact, which is similar to Wang et al. [22]. Therefore, the ML production stage is a key process that causes environmental impacts. Further breakdown of the ML production stage shows that cellulose supply contributes about half of the GWP and PED (Fig. 3b). The contribution of methanol, steam, electricity, and sodium hydroxide follows closely, and the combined environmental impact of these materials or energy exceeds 60% of the GWP and PED.

Fig. 3
figure 3

a Schematic diagram of contribution analysis of ML from “cradle-to-grave”; b Schematic diagram of contribution analysis of ML production; c Schematic diagram of contribution analysis of cellulose production

Full size image

Note that, except for the post-processing of humin, the sum of the contribution of each process in Fig. 3b exceeds 100%. This is because the humin post-processing co-produces electricity and heat (system expansion), and both of them are connected to the background data as environmental impact deduction (Table 6). The results of environmental deduction are represented by grids with negative values, as shown by the purple squares in Fig. 3b, and the contribution rates of all processes are based on the final calculation results (the sum of positive and negative values).

As mentioned above, the supply of cellulose is responsible for more than half of the contribution rate in multiple environmental indicators in ML production stage. Therefore, it is necessary to continue the decomposition analysis of cellulose production. Figure 3c is the contribution rate analysis for the cellulose production stage. At this stage, the use of electricity and steam contributes to the main environmental impact. In particular, the total contribution rate of electricity and steam in GWP and PED exceeds 80%, indicating that a large amount of energy consumption is a key process in the cellulose production stage, and subsequent technical improvements should focus on energy conservation.

In summary, after three decompositions and identifications, the environmental impacts of the life cycle of ML mainly come from electricity, steam, methanol, and sodium hydroxide. To reduce environmental impact, subsequent technological improvements should focus on cleaner electric energy, higher thermal energy conversion rate, cleaner methanol, etc.

3.3 Uncertainty analysis

3.3.1 Uncertainty analysis based on CLCD

We have carried out the corresponding data quality assessment and LCA uncertainty analysis of the life cycle model of ML based on CLCD. First, the quality of the extensive data and intermediate streams encompassed within the entire model is evaluated. For detailed information regarding the specific configurations and uncertainty outcomes of the inventory data, please refer to Table S6. In general, the foreground data obtained from the factories exhibit good quality (uncertainty within 20%), whereas the data quality of system expansion and PTSA production is poor (uncertainty over 30%). These uncertainties of inventory data are ultimately propagated to the final uncertainty of LCA results and presented as upper and lower bounds (Fig. 4a). The uncertainty analysis of LCA results indicates that all environmental impact indicators have satisfactory uncertainty, all of which are controlled within 20%. Six environmental indicators, namely GWP, PED, ADP, WU, AP, and RI exhibit fluctuations within a range of 10%, indicating that two-thirds of the indicators demonstrate high credibility. This is primarily attributed to the utilization of data from actual biomass plants and its integration with the latest localized background database.

Fig. 4
figure 4

a The uncertainty of the LCA results based on CLCD; b The uncertainty of the LCA results based on Ecoinvent

Full size image

As mentioned above, the data quality of humin postprocessing and PTSA is poor. However, the sensitivity analysis conducted in Section 3.2 indicates that their contribution to the environmental impact is relatively minor compared to other stages. Therefore, despite the poor quality of data from humin postprocessing and PTSA, its influence on the uncertainty of LCA results is relatively small.

Different from Monte Carlo simulation, Taylor series analysis can further analyze the contribution of each inventory data to the uncertainty of the LCA results and identify the main sources of uncertainty [45,46,47]. By ranking these contributions, targeted improvements of data quality can be made. The uncertainty contribution of GWP and PED in this model is shown in Tables S7 and S8. As can be seen, the main sources of uncertainty are humin postprocessing. Therefore, to further reduce the uncertainty of the LCA results, a crucial step is to improve the data quality of humin postprocessing, such as investigating real biomass thermal power plants rather than obtaining data from their EIA reports.

3.3.2 Uncertainty analysis based on Ecoinvent

Due to differences in energy structure, raw material sources, and manufacturing techniques, the environmental impact of similar products produced in different regions may sometimes be quite different. To explore the impact of different background database selection on LCA results and uncertainty, we also select Ecoinvent database as the background data and link to this model.

Figure 4b shows the LCA results and uncertainty after connecting the Ecoinvent database. First, the LCA results show varying degrees of change, such as the significant increase in GWP from 2.90 kg CO2 eq to 4.11 kg CO2 eq. Second, the uncertainty of LCA results is significantly higher when using the Ecoinvent compared to CLCD. Differences in the time, sources of raw materials, and product specifications of background data are the main reasons for the above situation. For example, for the use of steam, the CLCD data provides a closer time (2020) and has a clear pressure specification (0.7 MPa), while the data provided by Ecoinvent is a bit aged (2010) and there is no classification of pressure specifications (Tables S4 and S5). For the use of methanol, the data in CLCD mainly comes from the coal-to-methanol process, which is in line with the Chinese energy structure, while the methanol in Ecoinvent mainly comes from natural gas (Tables S4 and S5).

In summary, the use of non-local background data will indeed have an impact on the LCA results and their uncertainty. The use of such data requires careful consideration and corresponding uncertainty analysis to be of reference for actual production and improvements. And for better data quality and LCA results uncertainty, the use of local database should be encouraged.

3.4 Compared with the entire life cycle of fossil diesel

Biomass-based ML can be used as a fuel additive to replace part of diesel. One of the objectives of this LCA study is to explore whether ML has the potential to reduce emissions compared to diesel. The full life cycle results (raw material acquisition, production, transportation, and combustion) of 1 kg diesel are listed in Table 9. For GWP, no matter in CLCD or Ecoinvent, the overall life cycle of diesel is very similar. If the CLCD database is selected, the GWP of ML within the uncertainty range is always lower than that of diesel (the carbon emission reduction range is 17.8% to 30.4%). If the Ecoinvent database is selected, the GWP of ML within the uncertainty range is the opposite conclusion compared with diesel (from 11.4% carbon emission reduction to 34.1% carbon emission increase). Such a contradictory conclusion is not conducive to providing a reliable reference for actual production and technological improvement.

Table 9 LCA results of diesel in different background database
Full size table

3.5 Aspen simulation based on the real production system

In the published LCA studies of levulinic acid chemicals, the Aspen simulations all take place before the actual production activities. The data obtained in this manner lacks endorsement from the actual factories, making it impossible to obtain corresponding feedback and propose effective improvement measures. To address this deficiency, we conduct an Aspen simulation based on the actual factories and derive potential improvement strategies from it. Since this simulation is carried out after the actual production activities and is supported by the actual factories, the feedback between theory and practice can be formed to provide a reliable guarantee for the improvement plans.

Figure 5 is the simulation flow chart of Aspen Plus. For specific materials and energy information, see Tables S10-S12. The main energy in the Aspen simulation is focused on electricity and steam. However, it should be noted that Aspen Plus does not have power consumption models for some devices such as vacuum pumps and reactor stirrers, that is, the power consumption information of these devices cannot be obtained through Aspen simulations. This indicates that the energy consumption information obtained through Aspen is sometimes incomplete. Different from electricity, the amount of steam can be obtained completely. Since the calculation is strictly following the physical and chemical properties, the Aspen simulation can give the optimal solution of thermal energy utilization (that is, the optimal solution of steam consumption). The optimal steam usage of the two simulated factories by Aspen Plus is presented in Table 10. Referring to the actual consumption of steam, the thermal efficiency of the two factories is about 60%, which is a low value, indicating that the thermal efficiency of the actual production process needs to be further improved.

Fig. 5
figure 5

Aspen simulation flow diagram based on actual biomass refineries

Full size image
Table 10 Thermal efficiency result
Full size table

4 Discussion and prospect

4.1 Effects of transport distance on the carbon footprint

The conversion of waste rice stalks into ML involves multiple steps and three scenarios involving the use of trucks to transport intermediate products (Fig. 6). To explore the impact of transportation distance on product carbon footprint, we list the ML carbon footprint as a function of transport distance according to the mass transformation relationship (Eq. 1).

$${GWP\, of\, transportation\, distance}^{*}=7.33\times {10}^{-5}\times {D}_{3}+7.154\times {10}^{-5}\times {D}_{2}+ 1.274\times {10}^{-4}\times {D}_{1}$$
(1)
Fig. 6
figure 6

Supply chain transport process diagram

*The unit basis is 1 kg ML; D1 represents Distance 1, D2 represents Distance 2, and D3 represents Distance 3; The unit of distance is km

Full size image

The impact of the collection and transportation distance of the original straw bale (D1) carries the most significant weight in the carbon footprint calculation. This is because biomass conversion requires a multi-step process (it takes nearly 10 kg of rice straw to produce 1 kg of ML). However, it should also be noted that the proportion of GWP during transportation is not obvious compared to the overall GWP of ML (2.90 kg CO2 eq/kg ML). In this model, the collection and transportation distances are as follows: D1 is 20 km, D2 is 300 km, and D3 is 200 km. The total GWP of these transportation process is 0.039 kg CO2 eq, representing 1.34% of the overall life cycle impact of ML. If the straw collection radius (D1) is expanded to 100 km, the carbon emissions from the transportation process account for only 1.68% of the total carbon footprint. This suggests that the transportation process is not the primary contributor to carbon emissions. Altering the transportation distance has minimal impact on the carbon emissions in this model. This conclusion is consistent with most of the literature in Table 1 [21, 22, 27, 29].

4.2 Technical improvement program and results

Based on the previous sensitivity analysis in Section 3.2 and Aspen simulation results, the following four emission reduction improvement schemes are proposed:

  • Case 1: Improve thermal efficiency to 80%;

  • Case 2: Using clean electricity (biomass power generation);

  • Case 3: Using clean methanol (methanol from natural gas);

  • Case 4: Combination of Case 1, Case 2 and Case 3;

In order to explore the results after the implementation of the above measures, the corresponding background data set is changed and recalculated. The changed background data is shown in Table S13. Table 11 shows the LCA results for Case 1–4. It can be seen that the GWP decreases to varying degrees after different technical improvements. Finally, the GWP of Case 4 is 1.28 kg CO2 eq. Compared with the actual situation (2.90 kg CO2 eq), it is reduced by 56%, and compared with diesel, it can reduce carbon emissions by 65.6%. This indicates that the multi-strategy improvement scheme will have a good effect on the energy saving and emission reduction of ML production, and the subsequent technical improvements can be considered from the above aspects.

Table 11 LCA results of ML of different technical improvements
Full size table

4.3 LCA results of products when the hemicellulose oligomers bear the environmental impact

Section 3.1 mentions that due to the temporary lack of appropriate treatment technology, the hemicellulose oligomers are currently stored in Tongling Biomass Refinery and not incorporated into the next life cycle. Therefore, in our initial setting, the environmental impact of Process 2 was all assigned to cellulose and lignin, while hemicellulose oligomers did not bear the environmental effects. However, given the continuous production, hemicellulose will continue to accumulate. Hemicellulose can be easily converted into furfural and other downstream products, which has great application potential. In summary, if hemicellulose enters the next life cycle and is applied in the future, it should also bear the environmental impact. In this chapter, we discuss the changes in LCA results of various products if hemicellulose is also subject to environmental impacts.

Table 12 shows the updated LCA results. By comparing with the original results, the environmental impact of other products has been reduced to varying degrees. The GWP of cellulose and lignin was reduced by about 33.1% and the GWP of the whole life cycle process of ML was reduced by about 16.9%. Since hemicellulose is produced along with cellulose and lignin, the additional allocation of hemicellulose has a direct impact on the carbon footprint of cellulose and lignin. As cellulose continues to be converted into ML, these reduced carbon footprints are further reflected in the GWP of ML. This shows that the effective utilization of various components of biomass can also share considerable carbon emissions for subsequent biomass products.

Table 12 Life cycle assessment results of products (Hemicellulose oligomers bear the environmental impact)
Full size table

5 Conclusion

In this study, we established and evaluated the full life cycle of methyl levulinate based on real biomass refineries in China. This model showed a transparent and detailed biomass conversion process, and the production process and technology type of the products were clearly explained. Given the lack of effective data quality assessment and uncertainty analysis in the current LCA research on levulinic acid chemicals, we conducted a comprehensive and detailed data quality assessment on the inventory data of this LCA model. The final uncertainty analysis of LCA results showed that the LCA results obtained by this model had good reliability. The fluctuation of all environmental indicators was controlled within 20%, and the fluctuation of two-thirds of environmental indicators was even controlled within 10%. This is mainly due to the combination of actual factory data with up-to-date local background databases. However, if non-local databases were selected, the uncertainty of the LCA results would increase due to the differences in raw materials, time, technology, and other aspects between foreground data and background data.

The final LCA results showed that, when using a local database, biomass methyl levulinate could reduce the GWP by approximately 16% to 29% over the whole life cycle compared to fossil diesel of the same mass within the range of uncertainty. However, if a non-local database is used, conflicting conclusions will be obtained within the range of uncertainty.

To explore the potential for further energy conservation and thermal efficiency, use of biomass electricity, and use of clean methanol were proposed. The prospective results showed that the carbon emissions of methyl levulinate can be further reduced by more than half by the above improved methods. In addition, if hemicellulose is further utilized and transformed, the carbon emissions in the life cycle of methyl levulinate will be further reduced, which shows the advantages of fully utilizing biomass.