Mitigating GHG emissions is critical to prevent climate change and assessing GHG emissions can effectively support industries to make sustainable decisions. The cold chain uses refrigeration technologies to prevent spoilage and preserve the value of perishable food from production to consumption (Global Cold Chain Allian). The cold chain industry is growing rapidly in developing countries such as China. The Chinese cold chain market size is expected to grow from RMB 276.37 billion in 2019 to over RMB 500 billion in 2026 (ResearchMarkets, 2020). The rapid expansion of cold chain will inevitably generate greenhouse gas (GHG) emissions (Zhao et al., 2018), (Hu et al., 2019). However, the potential GHG emissions of using the cold chain to deliver perishable food products are not well-studied. Thus, the purpose of this paper is to evaluate the lifetime GHG emissions of non-processed high-value perishable food products (i.e. meat, aquatic, fruit, and vegetables) that are delivered by the cold chain in China.
GHG emissions associated with food products' cold chain come from refrigerant leakage and energy consumption (Dong et al., 2020), (Hoang et al., 2016). Firstly, traditional refrigerants including chlorofluorocarbon (CFCs), hydrochlorofluorocarbon (HCFCs), and hydrofluorocarbon (HFCs) can leak directly into the environment and cause severe environmental impacts (Xue et al., 2019;Heard and. Miller, 2016;United Nations Environment Programme, 2018). CFCs (e.g., R12) and HCFCs (e.g., R22) are being phased out under the Montreal Protocol due to the ozone depletion effect (United Nations Treaty Collection, 1987). Developed countries and developing countries will completely phase out HCFCs by 2030 and 2040, respectively (Shaik and Babu, 2017; US Environmental Protection Agency, 2020; Zhang et al., 2019a). HFCs (e.g., R404a and R134a) are widely used in refrigeration systems to substitute R22. However, most HFCs have high global warming potentials (GWP) and are restricted by the Kigali Amendment (United Nations Treaty Col, 2016;Zemel et al., 2013;Heredia-Aricapa et al., 2020). Developed countries need to phase down HFCs by 85% of the baseline year 1 by 2036 (The European FluoroCarbons Technical Committee), (UNEP, 2016). Article5 2 Group 1 and Article 5 Group 2 developing countriesfoot_0 need to phase down HFCs by 80% and 85% of the baseline yearfoot_1 by 2040 and 2045, respectively (The European FluoroCarbons Technical Committee), (UNEP, 2016). Secondly, the energy consumption of refrigeration also leads to significant GHG emissions. In refrigerated storage facilities, energy consumption is mainly from electricity consumption. According to the Green Cooling Initiative, refrigeration is responsible for around 5% of the worlds' electricity consumption corresponding to 2.5% of global GHG emissions in 2018 (Green Cooling Initiative, 2020), (International Energy Agency (IEA)a). In refrigerated vehicles, energy consumption is mainly from fuel consumption, and vehicle engines drive the refrigeration system (Tassou et al., 2009). In refrigerated vehicles, energy consumption can be up to 86% of total emissions (Wu et al., 2013). Overall, the combined emissions from leakage and energy consumption of major cold chain activities are estimated to account for 1e3.5% of GHG emissions in the world, 70e80% of which is due to energy consumption (Heard and. Miller, 2016), (Green Cooling Initiative, 2020), (Garnett, 2007), (James and James, 2010).
In addition to refrigerant leakage and energy consumption, GHG emissions of the food cold chain are also associated with food losses and wastes (FLW) (Hu et al., 2019), (Dong et al., 2020). Food losses refer to the lost edible food quantity during the supply chain, whereas food wastes occur at the end of the supply chain related to retailers and consumers (Hu et al., 2019), (Blakeney, 2019). According to James (James and James, 2010), over 200 million tonnes of perishable products could be preserved in developing countries if they were properly stored. FLW leads to the waste of invested energy and carbon in agriculture activities (Hu et al., 2019). The cold chain can prolong the shelf life of perishable products to reduce food and embodied carbon losses. From a system perspective, (Hu et al., 2019), (Dong et al., 2020), (Heard and. Miller, 2018) pointed out a tradeoff between using the cold chain to prevent food losses and the emissions generated from using refrigeration facilities.
Hence, it is critical to examine all three emission causes of the food cold chain: refrigerant leakage, energy consumption, and FLW related emissions.
As a standard method to evaluate GHG emissions, life cycle assessment (LCA) tracks the lifetime of a product (or a service) from production to disposal, which can systematically evaluate the total environmental impacts (Finnvedenet al., 2009;Rebitzeret al., 2004;Penningtonet al., 2004). In evaluating the cold chain industry, Hoang et al. (Hoang et al., 2016) conducted an LCA study to compare the chilling and super-chilling technologies for the salmon cold chain. The study is based on 1 kg of consumed salmon at the end of the cold chain with a European geographical focus. Wu et al. (Wu et al., 2019) investigated the environmental impacts of the orange cold chain and used computational fluid dynamics (CFD) to trace the historical temperature of fresh oranges produced in South Africa and exported to Switzerland. To investigate the GHG emissions of introducing cold chain into developing countries, Heard and Miller (Heard and Miller, 2019) simulated the GHG emissions of the agricultural cold chain in sub-Saharan Africa assuming development pathways similar to North America and Europe. They found that the GHG emissions at the system level (from agricultural production to the pre-retail stage) may increase by 10% or decrease by 15% depending on the diet structure (proportion of meat consumption) in different scenarios.
Regarding the cold chain in China, Zhao et al. (Zhao et al., 2018) and Dong et al. (Dong et al., 2020) reviewed the current status and discussed the environmental impacts. Both studies found that cold chain facilities are insufficient in China. The refrigerated warehouse capacity per capita in China was 0.132 m 3 in 2018, while that data for the US and UK is 0.49 and 0.44 (Salin, 2018). The distribution of refrigerated warehouses is uneven and more developed regions tend to have more cold chain facilities (Dong et al., 2020). Although the current cold chain facilities are still inadequate in China, the growing economy and improved living standards drive its rapid expansion. The refrigerated warehouse capacity increased by 38% from 2014 to 2018 in China, while the increase rate is 14% in the US (Salin, 2018). Simultaneously, the concerns of environmental impacts resulted from using cold chain in food logistics arise. Hu et al. (Hu et al., 2019) took China as an example and investigated the tradeoff between using a cold chain for meat, milk, and aquatic products and environmental impacts. They found that increasing the electricity input to the refrigeration systems can reduce the overall cold chain emissions when considering food loss reduction. However, without considering the refrigerant leakage and emissions from refrigerated transportation, the study of Hu can only partially represent the emissions of the food cold chain in China.
Overall, the attention on the environmental impacts of the cold chain for food products is rising in China. However, most cold chain studies still focus on cold chain management or optimization (Hanet al., 2020;Dai et al., 2019;Zhang et al., 2019b;Zhang et al., 2019c), while few studies (Schmidt Rivera et al., 2014) analyze the lifecycle emissions of food products specifically delivered by the cold chain in China. Therefore, this study aims to fill this knowledge gap and evaluates the lifetime GHG emissions of four types of unprocessed food products (i.e. fruit, vegetables, meat, and aquatic) from production to consumption with the cold chain being responsible for the delivery. One major challenge of conducting an LCA study is data availability. Existing literature contains assumptions based on known data, which may be from a different country or a different application context (Hu et al., 2019), (Heard and Miller, 2019) (Xueet al., 2017). In our study, we have a finer resolution of lifecycle inventory data (Supplementary Materials) to compare with existing studies (Hoang et al., 2016), (Heard et al., 2019). For instance, we model the quantity of food losses along the cold chain as a function of storage time and temperature
We consider the entire lifetime for food products from agricultural production to final consumption, focusing mostly on the postagriculture supply chain where the cold chain is present. The functional unit is defined as 1 kg of edible food consumed in households and the GHG emission results are measured in kg of CO 2 equivalent per kg of food consumed [kg CO 2 eq/kg consumption]. We assume agricultural activities include food production and processing (e.g., bone removing, peeling). As shown in Fig. 1, the post-agriculture cold chain includes refrigerated warehouse at the origins of production, 1st stage refrigerated transportation (long-distance), distribution center refrigerated storage at the places of sale, 2nd stage refrigerated transportation (short-distance), and retail stores. The household activities include household refrigeration, consumption, and food waste. In the system, we include the GHG emissions from refrigerant leakage, energy consumption associated emissions, and emissions due to food losses and wastes.
We defined eight cold chain scenarios in Table 1 to estimate GHG emissions and each scenario represents one food category at one storage condition. We considered four food categories, vegetables, fruit, meat, and aquatic products. Each food category is represented by typical food types (ST2, ST11 in the supplementary materials). We admit this generalization may introduce errors in the modeling. However, considering the scope of this paper is to present a robust emission estimation of using cold chain in China, it is reasonable to use the broad food categories. Similar generalized food categories are also used in the cold chain modeling by Hu et al. (Hu et al., 2019) and Heard et al. (Heard and. Miller, 2018).
Additionally, we defined appropriate temperature levels in each scenario for the specific category: low-temperature (LT), mediumtemperature (MT), and high-temperature (HT) according to the optimal storage temperature (Mercier et al., 2017). For instance, apple and orange belong to the MT fruit scenario, and banana belongs to the HT fruit scenario (Mercier et al., 2017). The detailed definition, value, and distribution of each parameter in all scenarios can be found in the Supplementary Materials.
Food losses refer to the decrease of food quantity or quality throughout the food supply chain which is essentially due to the growth of organisms and biochemical reactions or mishandling (Hu et al., 2019), (Hammondet al., 2015), (Food and Agriculture Organization of the United Nation). In this paper, we consider food processing (e.g., bone removal, peeling) as a part of agricultural activities and we assume the food products entering the cold chain are edible quantities. We assume the edible proportion is 60% for meat and aquatic products (Hamerschlag and Venkat, 2011) and 80% for vegetables and fruits (Hamerschlag and Venkat, 2011). Hence, the food quantity after harvesting is multiplied by the edible proportion to calculate the quantity entering the post-agriculture cold chain. During the post-agriculture cold chain (from the refrigerated warehouse to household refrigerator stages in Fig. 1), we referred to (Hoang et al., 2016), (Van Boekel, 2008) and regarded the food losses are merely due to the generic food quality (e.g., color, moisture, nutrition) degradation without other weighted losses.
The food quality degradation is described by the kinetic model and shown in Eq. (1) (Wu et al., 2019), (Van Boekel, 2008), (Rong et al., 2011). Q refers to food quality, k is the reaction rate, and n represents the reaction order. Using the kinetic modeling for food losses, we can emphasize the impacts of storage temperature. It has been found that the linear zero-order reaction (n ¼ 0) and exponential first-order reaction (n ¼ 1) are good fits for food quality evolution, which are showing in Eqs. (2) and (3), respectively (Rong et al., 2011;Li et al., 2020;Ling et al., 2015) e (Rong et al., 2011;Li et al., 2020;Ling et al., 2015). In Eqs. (2) and (3), Q 0 represents the initial food quality and t refers to the reaction time.
The reaction rate k is mostly influenced by temperature and it is widely modeled by the Arrhenius equation as a function of temperature (Van Boekel, 2008), (Rong et al., 2011), (Ling et al., 2015), (Mizrahi, 2011). As shown in Eq. ( 4), kðTÞ is the reaction rate at temperature T (in K), k 0 is a constant,E a is the activation energy, and R is the universal gas constant. As k 0 and E a are temperature independent, the kðTÞ can be estimated from a known reaction rate at the reference temperature (Van Boekel, 2008), (Mizrahi, 2011).
Additionally, existing studies further simplified the method where the constant k 0 is avoided and the variable Q 10 is introduced. As shown in Eq. ( 5), Q 10 essentially means the ratio between the reaction rate at temperate T þ 10 and that at temperature T (Wu et al., 2019), (Van Boekel, 2008), (Mizrahi, 2011). According to the Van't Hoff's rule, Q 10 is between 2 and 3 at 0 to 10 C temperature range (Wu et al., 2019), (United States Department of Agriculture, 2016). In this study, we assume the Q 10 is 2 for vegetables and fruit, while the Q 10 is 2.5 for meat and aquatic.
Thus, the food quality evolution can be modeled for the targeted food product at a specific temperature. In this study, we refer to Rong et al. (Rong et al., 2011) and assume the zero-order reaction for vegetables and fruit, while we consider a mixture of zero and first-order reaction for meat and aquatic according to Ling et al. (Ling et al., 2015). The reduction is considered as the first-order in the initial 3 h and zero-order otherwise because the reaction rate is faster during the cooling down process but will be slower afterward. According to the temperature level for each scenario defined in Table 1, we assume the product lifetime in each scenario (ST5) (Wu et al., 2019), (Vasavada, 1996), (Cantwell). We then computed the quality evolution based on the kinetic model and present the food losses curve for the post-agriculture stages of all scenarios in Fig. 2. Thus, the food losses in each can be estimated given the residence time in the cold chain.
As we define the functional unit as 1 kg of edible food consumed, the food losses and food quantity entering each cold chain stage are calculated back based on the food loss rate and storage time. The use of the kinetic model likely results in an underestimate of overall food losses, since it does not account for consumer preferences or behavior. The food quantity entering each previous stage can be calculated by the sum of food quantity entering the current stage and food losses at the previous stage. For instance, if the current cold chain stage is the 1st refrigerated transportation, the food quantity entering the pervious stage, refrigerated warehouse, can be calculated by Eq. ( 6) and the food losses at the refrigerated warehouse are calculated from Eq. ( 7).
The paper estimates GHG emissions associated with the cold chain for vegetables, fruit, meat, and aquatic product in China. To take into account various sources of variability and uncertainty, Monte Carlo simulation is conducted of variables to enhance the robustness of results. We use the programming language Python3 to perform the Monte Carlo simulation and we run the model 10,000 times with randomly generated parameters every time. The lifecycle emissions are the summation of emissions at each stage defined in Section 2.5. Table 2 lists the definition of each parameter and the details regarding the specific parameter distribution and data sources are in ST1 in Supplementary Materials. Additionally, a sensitivity analysis is conducted to determine the degree of influence of each parameter (in Table 2) on the final results, using oneat-a-time perturbation. All parameters are fixed at their median value except the targeted parameter which is increased by 20% from its median.
The cradle-to-gate emission factors for agricultural activities of vegetables, fruit, meat, and aquatic products are obtained from existing studies. Porter et al. (Porter et al., 2016) studied the emission factors for industrialized Asia and Hamerschlag (Hamerschlag and Venkat, 2011) calculated the cradle-to-gate emissions factors. To characterize food emission factors (C food ) for China, we primarily use the industrialized Asia data from Table 1 Cold chain scenarios considered in this study. The category is according to the optimal storage temperature of each product.
Storage type Optimal storage temperature Temperature in this study
Fig. 2. Perishable product quality evolution. Fruit and vegetables follow zero order reaction. Aquatic and meat follow the mixed zero-first order reaction. Based on the reference quality evolution curve, the food losses for each product at any given temperature can be estimated. (Hamerschlag and Venkat, 2011) and supplement not present data in (Porter et al., 2016). The original data source is summarized in the ST2 in the Supplementary Materials. The C food taken from (Porter et al., 2016) are primarily US-based and we converted it to model China by the difference of agriculture production emission intensity between the US and China (ST3). Notice that the emission factors are also used to calculate the food loss emissions at each cold chain stage, as we follow the definition of Hu et al. (Hu et al., 2019) considering that the food loss emissions result from the waste of invested carbon in agricultural activities.
The refrigerated warehouse mainly involves precooling and refrigerated storage. In our paper, we assume all food products use warehouse precooling because it is the most used approach in China (Zhao et al., 2018). Hence, the energy consumption of the precooling process is included in the refrigerated storage energy consumption. Besides, we add the packaging emissions in the refrigerated warehouse stage. Note that packaging emissions are not refrigeration related, and we simply consider it at the beginning of the cold chain. We refer to the study by Heller (Heller et al., 2019) and consider the packaging emission factor (C pack ) is proportional to C food by the coefficient food to packaging ratio (FTP). In other words, C pack equals C food divided by FTP. The original data for FTP is shown in ST6 in the Supplementary Materials.
The GHG emissions of refrigerated storage are allocated to refrigerant leakage, energy consumption, and food losses. Regarding refrigerant leakage, we consider a 20-year equipment lifetime with 8% annual operational leakage, 1% of installation leakage, and 5% disposal leakage (Hoang et al., 2016), (United Nations Environment Programme, 2018). Firstly, the annual leakage rate is multiplied with the initial refrigeration charge of a commercial refrigerated warehouse (2000e10,000 kg) and then divided by the annual throughput product quantity (approximately six times of warehouse capacity) (United Nations Environment Programme, 2018), (China Federation of Logistics & Purchasing (CFLP), 2018). Secondly, the installation and disposal leakage are divided by the lifetime throughput product quantity. The refrigerant leakage allocated to one kg food product in 1 h (leak RW ) is the sum of both leakages over each associated time spent. We then calculate the direct GHG emissions (Em leak; RW ) by Eq. ( 9) where t RW is the storage time and GWP is the mean of the most used refrigerant in China (R22, R404a, and R134a) (Zhao et al., 2018), (Zhang et al., 2019a). See ST7 in Supplementary materials for detailed data refrigerant leakage allocation.
The cclcn.com reports the daily electricity consumption for different capacities of refrigerated warehouse in China (cclcn.com, 2012). We assume a 10-h compressor operating time and food products load 75% of refrigerated warehouse capacity on average. Thus, the daily electricity consumption data is divided by 75% of warehouse capacity and then divided by 10 to reach the electricity consumption for one ton product in 1 h (E RW ). Eq. ( 10) calculates the GHG emissions from electricity consumption of refrigerated warehouse (Em ele; RW ) where Em ele [gCO 2 e/kWh] refers to the weighted lifecycle power generation emissions in China. Em ele is calculated from the Chinese electricity generation structures and emissions of each power source (Li et al., 2017), (International Energy Agency (IEA)b). See ST8 and ST9 in Supplementary Materials for daily refrigerated warehouse electricity consumption and power source in China.
Food quantity entering each stage Calculated back based on one kg food consumedQ loss , and Qwaste.
Calculated at each cold chain stage and for each food category. Specified for emission sources. For instance, Em leak; RW is the calculated GHG emissions from refrigerant leakage at the refrigerated warehouse
Eq. ( 11) calculates the food losses emissions (Em loss; RW ) which is multiplied from food losses quantity at refrigerated warehouse (Q loss; RW ) and losses emission factors (C loss )
Em loss; RW ¼ Q loss; RW Â C loss (11)
The 1st stage transportation is the long-distance truck refrigerated transportation from the origins of production to distribution centers. We assume an 8% annual refrigerant leakage rate and multiply it with the refrigerant charge quantity (3e8 kg); then divided the value by the annual truck travel distance (assume 60,000 km in China) and truck capacity (3e11 ton) to allocate the refrigerant leakage to one kg food product per km travelled (leak Trans ) (United Nations Environment Programme, 2018), (Tassou et al., 2009). The food miles of big cities in China are studied by Huang et al. (Huang et al., 2014), which provides data for the 1st stage travel distance (D 1 ) of each food category. Thus, the GHG emissions associated with leakage (Em leak; Trans ) can be calculated by Eq. ( 12) where Q Trans1 is the food quantity entering this stage and GWP is the mean value of the most used refrigerant (R134a/R404a). Original data source is in ST7 in Supplementary Materials.
We refer to a UK based study (Tassou et al., 2009) for the fuel efficiency [mL/ton-km] of refrigerated vehicles. The fuel efficiency is assumed to be the same in the UK and China. Then, we use an emissions factor of refrigerated vehicles in China (2.63 kgCO 2 /L) (Liu et al., 2020) to compute the specific emissions of refrigerated transportation (C Trans ) measured in gram of CO 2 equivalent per ton product per km travelled. The compiling of C Trans can be found in ST 4 in the supplementary materials. Eq. ( 13) calculates the energy usage related GHG emissions with specific refrigeration temperature conditions.
As we define the food losses is a function of time (Section 2.3), we use D 1 divided by the truck speed (v 1 ) to reach the time spent during the 1st transportation stage where the speed limit for trucks on expressways in China is 80e100 km/h (Road Traffic Safety Law o, 2011). Thus, the food losses in the 1st refrigerated transportation stage (Q loss; Trans1 ) are calculated and food loss emissions can be calculated by Eq. ( 14).
Em loss; Trans ¼ Q loss; Trans1 Â C loss (14)
The calculation method for the distribution center stage is the same as the refrigerated storage process in the refrigerated warehouse stage. Eq.15e17 calculate the GHG emissions from refrigerant leakage, electricity consumption, and food losses where t DC differentiates the refrigerated warehouse stage and the distribution center stage.
Em loss; DC ¼ Q loss; DC Â C loss (17)
The 2nd stage, short-distance refrigerated transportation from distribution centers to local markets. We use the same approach in Section 2.5.3 to calculate the GHG emissions from the 2nd refrigerated transportation with travel distance (D 2 ) and truck speed (v 2 ) differentiate both steps.
2.5.6. Retail Refrigerant leakage at retailing stores (leak Retail ) is assumed to be the same as leak RW (leak DC ). The leakage emissions (Em leak; Retail ) can therefore be calculated by Eq. ( 21) where Q Retail is the food quantity entering the retail stage and t Retail refers to the time duration in this stage. See ST7 in Supplementary Materials for details refrigerant leakage allocation and refrigerant assumptions.
The electricity consumption in the retail stage are divided into refrigeration electricity consumptions (E Retail; ref ) and all other energy consumptions (E Retail;other ) (Heller et al., 2019). We calculate the E Retail; ref from the energy usage of display cases where we assume closed cases are used for LT applications while open cases are used for HT/MT products. According to Fricke and Becker (Fricke and Becker, 2011), the energy consumption of open and closed display cases is 56 and 40 kWh/day per case and we convert it to kWh/day per cubic meter based on the case size (considering 75% full of total capacity). Then, the data is divided by the food bulk density and 24 to reach kWh/ton-h. See the ST10 for display case parameters and ST11 for the food bulk density in Supplementary Materials (Charrondiere et al., 2012).
We use the economic data of the retail sector in China (ST12 in Supplementary Materials) to allocate the E Retail;other . The annual sales revenue in China in 2018 of vegetables, fruit, meat, and aquatic are divided by the total revenue of the retail sector (National Bureau of Statistics of China). Afterward, the ratio is multiplied with the total energy consumption of the retail sector in 2018, and then divided by the consumed food quantity of each product to calculate E Retail;other (National Bureau of Statistics of China). Hence, the energy consumption related emissions (Em ele; Retail ) can be calculated by Eq. ( 22) where Q Retail is the food quantity entering retailing stores and t Retail refers to the storage time in the retail stage.
Eq. ( 23) calculates the food losses emissions (Em loss; Retail ) by the multiplication of the quantity of food losses at retailing stores (Q loss; Retail ) and losses emission factors (C loss ) Em loss; Retail ¼ Q loss; Retail  C loss (23)
The calculation of the domestic fridge storage is the same as previous cold storage stages as shown in Eq. ( 24) and Eq. ( 25). Note that the annual leakage rate is merely 0.3% of household refrigerators, so it is neglected (United Nations Environment Programme, 2018).
Em ele; House ¼ E House  Em ele  Q House  t House (24)
Em loss; House ¼ Q loss; House  C loss (25)
Food wastes are generated during household storage and the food waste quantity is calculated by Eq. ( 26) where R Waste refers to the food waste rate. Blakeney from the Food and Agriculture Organization of the United Nations (FAO) recorded the food waste rate for industrialized Asia, which is used to model the food waste in China (Blakeney, 2019). The emissions associated with food losses (Em Waste ) can be calculated by Eq. ( 27).
Thus, the lifecycle GHG emissions of each scenario defined in Table 1 can be calculated by accumulating the emissions of each stage from Section 2.5.1 to Section 2.5.7. The definition of all parameters in our modeling is listed in Table 2 and the details of data descriptions can be found in ST1 in Supplementary materials. It is worth noting that it is challenging to use China-specific data for all parameters due to data availability. In Table 2, D 1 , D 2 , v 1 , v 2 , E RW , E DC , E Retail;other , Em ele , GWP, and R waste are specified for China. Meanwhile, C food combines the food losses emissions in the US and industrialized Asia; however, the data from the US is converted by a coefficient to model China (ST3). Besides, to compile C Trans (ST4), we assume the fuel efficiency of refrigerated vehicles are the same in the UK and China (Tassou et al., 2009); however, the fuel emission factor is from a China-based study (Liu et al., 2020). Overall, most parameters are China-specific and non-China data are also corrected; hence, the results of our study can still robustly indicate the cold chain emissions in China.
We first calculate the lifetime GHG emissions of each scenario defined in Table 1. The lifetime GHG emissions include the cradleto-gate agriculture production emissions, post-agriculture cold chain, and household emissions. We report the median emissions of 10,000 times of Monte Carlo simulation and divide them into three main stages (agriculture, cold chain, and household). The results are shown in Fig. 3, with Fig. 3a illustrating the absolute values and Fig. 3b showing the contributions in percentage.
The most apparent character is the difference in GHG emissions between meat/aquatic and vegetable/fruit scenarios. As expected (Clune et al., 2017), the lifecycle GHG emissions of meat and aquatic products are much higher than fruit and vegetable products in China. On average, four meat/aquatic scenarios generate 15.6 kg CO 2 eq/kg consumption more than vegetable/fruit scenarios. In meat/aquatic scenarios, cradle-to-gate agriculture activities contribute to a significant portion. 74% (8.2 kg CO 2 eq/kg consumption) and 83% (21.0 kg CO 2 eq/kg consumption) of GHG emissions in LT aquatic and LT meat scenarios are from agriculture activities. Similarly, agriculture emissions contribute to 84% (7.7 kg CO 2 eq/kg consumption) and 89% (19.8 kg CO 2 eq/kg consumption) in the MT meat and MT aquatic scenarios. In the vegetable/fruit scenarios, agriculture activities and cold chain activities contribute to an approximately equal amount of GHG emissions (~47%), while the emissions from household activities are around 5%.
It is not surprising that the cradle-to-gate agriculture activities constitute for a large portions as it includes raw materials production (e.g., fertilizer), waste management (e.g., on-farm manures), on-farm energy use, and raw materials transportation at the agriculture stage (Hamerschlag and Venkat, 2011), (Porter et al., 2016). Besides, we consider the processing (e.g., bone removing, peeling) as a part of agriculture activities in this paper, which also generate GHG emissions. Several studies (Hoang et al., 2016), (Heard et al., 2019), (Hamerschlag and Venkat, 2011) also showed that food production contributes to a significant proportion of total GHG emission. For instance (Hamerschlag and Venkat, 2011), found nearly 90% of lamb lifetime GHG emissions are in the production phase.
In China, the cradle-to-grave food lifetime emissions are rarely studied, especially when using cold chain logistics. Our following results present an early estimation of the entire lifecycle GHG emissions for food products in China and the higher resolution calculations of post-agriculture emissions allow greater analysis of the environmental impacts of the cold chain.
This paper focuses on the post-agriculture GHG emissions to better analyze the contributions of the cold chain. Results of the Monte Carlo simulation are given in Fig. 4. Unless otherwise indicated, all values reported are the median value of the 10,000 simulations. The first observation from Fig. 4 is that the eight scenarios can be divided into two groups (vegetable/fruit and meat/aquatic) based on the emission level. Even when agricultural production is not taken into account, the post-agriculture emissions from meat/ aquatic products are still significantly higher than those of vegetables and fruit. On average, the median emissions of LT/MT meat scenarios are 0.9 kg CO 2 eq/kg consumption higher than that of LT/ MT aquatic scenarios and 2.4 kg CO 2 eq/kg consumption higher than that of four vegetable/fruit scenarios. The significant emission differences across eight scenarios mainly result from food losses, which are a function of (C food ) of different food categories (Hu et al., 2019). Meat products have the most embodied carbon associated with their production; therefore, even when the loss quantity of meat is the same as vegetable or fruit, its associated GHG emissions are amplified by C food .
The median emissions of LT/MT meat scenarios are 4.0 and 2.2 kg CO 2 eq/kg consumption, and ranging from 0.8 to 8.3 kg CO 2 eq/kg consumption. The C food for producing one kg poultry, pork, and beef are 3, 6, and 18 kg CO 2 eq, which makes significant emissions variations in meat (Hamerschlag and Venkat, 2011), (Porter et al., 2016). Moreover, a median of 2.8 and 1.4 kg CO 2 eq emissions are generated from LT/MT aquatic scenarios. The range of lifecycle emissions of aquatic products is 0.8e4.0 kg CO 2 eq/kg consumption which is much smaller than that of meat products. Regarding vegetables and fruit, the median emissions of HT and MT fruit (0.73 and 0.77 kg CO 2 eq/kg consumption) are slightly higher than that of HT/MT vegetable scenarios (0.60 and 0.64 kg CO 2 eq/kg consumption). Over 80% of the Monte Carlo simulation results of HT fruit (and MT fruit) are higher than the median emissions of HT vegetable (and MT vegetable). The main reason is that the fruit loss rate is larger than the vegetable loss rate during the cold chain. As a result, emissions associated with fruit losses are larger than that of vegetable losses leading to more lifecycle emissions.
Fig. 4 also indicates that lower temperature conditions tend to have higher total GHG emissions. Comparing LT and MT meat scenarios, the LT meat scenario has 1.7 kg CO 2 eq/kg consumption more emissions than the MT meat scenario. The deciding reason is again food losses related emissions. Although the meat and aquatic loss rate in LT conditions (À22 C) is merely 20% of that in MT conditions (À2 C), the frozen meat and aquatic products spent around 900 h (~37 days) more than chilled meat and aquatic products in refrigerated storage facilities. Our modeling results show that the quantity of FLW in the LT meat scenario is larger than that of the MT meat scenario by 0.05 kg due to the long residence time. Besides, the energy usage related emissions of the LT meat scenario at the refrigerated warehouse, distribution center, and two distribution stages are higher than that of the MT meat scenario, while the energy emission at retailing stores of LT meat is lower than the MT meat because closed display cases used for LT products are more energy-efficient than open cases for MT cases. Hence, combining the impacts of residence time and energy consumption, the LT meat scenario generates 1.7 kg CO 2 eq/kg consumption more than the MT meat scenario. Due to the same reason, the total GHG emissions of the LT aquatic scenario is 1.5 kg CO 2 eq/kg consumption more than the MT aquatic scenario. However, the emission difference between HT/MT vegetable scenarios (also HT/MT fruit scenarios) is not significant with merely around 0.04 kg CO 2 eq more in MT conditions on average. It is because the cold chain residence time of vegetable/fruit is merely 12 h longer than HT vegetable/ fruit.
We also divide the post-agriculture GHG emissions into six stages (Fig. 5) with Fig. 5a showing the absolute emissions and Fig. 5b illustrating the emissions contribution in percentage. As we discussed before, the food emissions factor (C food ) amplified the GHG emissions due to FLW in meat/aquatic scenarios. Comparing emissions from each cold chain stage, Fig. 5 shows that the aquatic/meat and vegetable/fruit scenarios have different features. In the LT aquatic scenario, refrigerated warehouse, retail, and household are the largest three emission contributors accounting for 36% (1.1 kg CO 2 eq/kg consumption), 32% (0.9 kg CO 2 eq/kg consumption), and 22% (0.6 kg CO 2 eq/kg consumption), respectively. Those three stages also produce the largest emission in the LT meat scenario with 33% (1.3 kg CO 2 eq/kg consumption), 29% (1.19 kg CO 2 eq/kg consumption), and 31% (1.25 kg CO 2 eq/kg consumption) for each. In the MT meat and MT aquatic scenarios, the contribution of GHG emissions from household activities increases to 41% (0.9 kg CO 2 eq/kg consumption) and 31% (0.5 kg CO 2 eq/kg consumption), respectively. The absolute GHG emissions from the refrigerated warehouse in the MT meat (0.4 kg CO 2 eq/kg consumption) and MT aquatic (0.3 kg CO 2 eq/kg consumption) scenarios have decreased compared with that in LT conditions due to shorter residence time. However, the wastes of meat and aquatic embodied emissions do not change significantly leading the household stage to be the most emissions contributor. In comparison, the emissions from the refrigerated warehouse and 1st refrigerated transportation constitute the largest proportion in vegetable/fruit scenarios. On average, refrigerated warehouse and 1st refrigerated transportation account for 37% (0.26 kg CO 2 eq) and 28% (0.19 kg CO 2 eq) in MT/HT vegetable scenarios. Comparing vegetables and fruit, the emissions of both fruit scenarios are slightly larger than that of vegetable scenarios. It is because more embodied emissions are generated in agriculture activities for fruit; hence the FLW associated emissions of fruit scenarios will be higher than that of vegetables.
To analyze the source of GHG emissions, we break down the post-agriculture GHG emissions into refrigerant leakage, energy consumption, and FLW associated emissions. As Fig. 6 displays, the majority of emissions in fruit/vegetable scenarios are from energy usage related emissions, while FLW associated emissions represent a considerable number of emission in meat/aquatic scenarios.
Firstly, 61% (0.28 kg CO 2 eq/kg consumption) of emissions on average are energy consumption related emissions in the four vegetable/fruit scenarios. The emissions from energy usage of vegetables and fruit in MT conditions are slightly higher than that in HT conditions (by around 0.05 kg CO 2 eq/kg consumption), which is mainly because that more energy is consumed to reach the MT conditions (2 C) than the HT conditions (8 C). The refrigerant leakage and FLW emissions in vegetable/fruit scenarios share a similar proportion (~19%). When it comes to aquatic/meat scenarios, the largest emissions contributor becomes FLW associate emissions. In the MT meat scenario, 86% of emissions (1.7 kg CO 2 eq/ kg consumption) are made up of FLW emissions, and this number decreases to 62% (2.3 kg CO 2 eq/kg consumption) in the LT meat scenario. Similarly, the proportion of FLW emissions in the MT aquatic scenario is 69% (0.8 kg CO 2 eq/kg consumption) and it decreases to 40% (1.1 kg CO 2 eq/kg consumption) in the LT aquatic scenario. Noted that the absolute emissions from FLW increase from MT to LT conditions in aquatic/meat scenarios. The percentage decrease comes from the gained refrigerant leakage and energy usage related emissions due to the much longer storage time and slightly increased energy consumption in LT conditions. The FLW associated emissions are calculated by multiplying C food and FLW quantity. Considering the high C food value for meat and aquatic products, the FLW emissions in meat/aquatic scenarios could be augmented small quantity of FLW.
We conduct sensitivity analysis to determine how variability and uncertainty of the model parameters impact the results of GHG emissions. Additionally, the results of sensitivity analysis also potentially imply approaches to reduce GHG emissions. It again should be noticed that the sensitivity analysis is conducted from post-agriculture stages, due to the overall focus on cold chain related activities.
In this study, we reduce the value of one parameter by 20% at each simulation run while keeping all other parameters constant at their median. The results of sensitivity analysis are shown in the heat map (Fig. 7), with Fig. 7a displaying the emission changes in absolute value and Fig. 7b illustrating the emission changes in percentage. Overall, the visible variation in magnitude in Fig. 7 suggests that C food and t (time spent at each stage) are the most influential parameters in this study. Generally, reducing C food leads to considerable changes in all scenarios. At least 0.22 kg CO 2 eq/kg consumption (7.4%) and 0.015 kg CO 2 eq/kg consumption (2.2%) emission reduction can be achieved in meat/aquatic and vegetable/ fruit scenarios by decreasing the C food . The largest emission reduction is 0.47 kg CO 2 eq/kg consumption in the LT meat scenario. Moreover, time spent in cold chain facilities (t RW , t DC , t Retail , and t House ) also have strong influences on the emissions. The longer time the product is stored, the more embodied emissions the product would have. For instance, reducing 20% of storage time in retailing stores would lead to 0.25 kg CO 2 eq/kg consumption (6.3%) and 0.2 kg CO 2 eq/kg consumption (7.1%) decrease in LT meat and LT aquatic scenarios, and the emissions of the other six scenarios also decrease by 4.2% on average. Besides, reducing C pack could also decrease emissions especially in vegetable/fruit scenarios.
Significant emission reductions in absolute values are normally found in meat/aquatic scenarios, while the notable emission reductions in percentage are generally in vegetable/fruit scenarios. Taking the LT meat scenario and an example, leak, GWP, R waste , and Em ele are also crucial variables other than C food and t. Parameters leak and GWP are associated with direct emissions from refrigerant leakage. Reducing refrigerant leakage at the refrigerated warehouse by 20% can decrease 0.089 kg CO 2 eq/kg consumption (2.2%), and more significantly, 0.18 kg CO 2 eq/kg consumption (4.4%) can be decreased if lower GWP refrigerant is used. Reducing the GHG emissions from the power grid (Em ele ) can reduce 0.079 kg CO 2 eq GHG emissions (2%). Additionally, food waste is a substantial emission source; hence decreasing R waste is helpful to reduce the emissions by 0.13 kg CO 2 eq/kg consumption (3.2%). In the MT vegetable scenario, the parameters associated with the 1st refrigerated transportation (C Trans and D 1 ) greatly contribute to GHG emissions besides of critical variables analyzed before. If C Trans and D 1 are decreased, 0.04 kg CO 2 eq/kg consumption (6.3%) and 6. GHG emissions breakdown into refrigerant leakage, energy consumption, and FLW associate emissions for refrigeration related stages (post-agriculture emissions). 0.042 kg CO 2 eq/kg consumption (6.6%) emission reduction can be achieved in the MT vegetable scenario.
Based on our results of 1 kg consumed food product, we also approximate the aggregated lifecycle emissions at the national level in China. We multiply the consumption quantity of vegetables, fruit, meat, and aquatic in China in 2018 (ST12 in Supplementary Materials) with the 1 kg based lifecycle emissions of each food product. The estimation shows that if the cold chain were fully developed in China, the aggregated post-agriculture GHG emissions of the studied food category would be roughly in the range of 280e400 Mt CO 2 eq. As a reference, the total CO 2 emissions of China is 10,064 Mt in 2018, and thus, the post-agriculture emissions of vegetables, fruit, meat, and aquatic account for approximately 3% of total emission in China based on the 2018's level (International Energy Agency (IEA)a). The 3% of total emission in China is slightly higher than James's estimation that the food cold chain accounts for about 1% of carbon emission in the world (James and James, 2010). It should be noted that our result is based on the cold chain fully developed assumption. With the total GHG emissions in China being expected to grow in the coming years, it is reasonable to expect 1e3% of total GHG emissions to come from the food cold chain in China in the future. Li (Li et al., 2016) reported that over 10% of GHG emissions are from the food system in China; however, one should note that a dominant proportion is from agriculture activities. Hence, this paper conveys a key message in that, even though agricultural emissions dominate the environmental impacts of the food system, post-agriculture cold chain emissions are also significant on an absolute basis, and should not be obscured by a focus on agricultural emissions.
The results of the sensitivity analysis for post-agriculture emissions illuminate important measures that might be Fig. 7. The GHG emission changes (post-agriculture emissions) of sensitivity analysis. Keeping all other parameters at their median value while reducing the current targeted parameter by 20% from its median.
undertaken for potential emission reduction. Above all, it is well understood that reducing food embodied emissions from agriculture activities is the most critical step to reduce the lifecycle emissions of food products. In the post-agriculture stages, it is important to reduce cold chain energy usage related emissions because it constitutes approximately 60% of total post-agriculture GHG emissions for vegetables and fruit (see Fig. 6). On one hand, emission reduction from electricity usage can be achieved either by improving the energy efficiency of refrigeration equipment or using renewable electricity sources. For instance, Wu concluded that if solar power is used, the emissions can be reduced by 8.5% (Wu et al., 2019). On the other hand, emissions from non-electricity energy usage, mainly from refrigerated vehicles, can be reduced by replacing diesel/gasoline with electric vehicles. In addition, reducing FLW is essential to decrease lifecycle emissions of meat and aquatic products (see Fig. 6). Due to the considerable embodied carbon from production, even a small quantity of FLW will lead to significant emissions. Beyond using refrigeration technologies to prevent food spoilage, it is more important for consumers to reduce food waste which is approximately 5%e10% of the initial production quantity in China (Blakeney, 2019). In this paper, we also highlight the impacts of food storage time of frozen meat/aquatic in cold chain facilities, especially in the refrigerated warehouse, 1st refrigerated transportation, and retailing stores. The longer the products are stored, the more emissions will be embodied in food products. Additionally, diminishing emissions from packaging and using cleaner refrigerants are effective approaches to reduce food lifecycle emissions.
We admit our study is not without flaws. Firstly, data quality is always a challenge in LCA studies, especially for an emerging industry (Miller and Keoleian, 2015). In this paper, 51 parameters are used in the model (ST1 in the supplementary materials). Although 40 parameters are China-specific or universal, the compiling of C food , and C Trans partially referred to studies based on the US (Hamerschlag and Venkat, 2011) or the UK (Tassou et al., 2009). We made corrections for C food , and C Trans to model China (ST3 and ST4), but undoubtedly, higher data quality can improve our study. Since most parameters are China-based, our results can still reveal a robust estimation of cold chain emissions in China. Secondly, our study reflects the future developed scenario but does not capture the transitions of the supply chain as it develops over time. One should note that the food supply chain structure is changing with the penetration of the cold chain in China. For instance, Garnett concluded that the food supply chain is lengthening in China (Garnett and Wilkes, 2014).
Additionally, the present paper, similar to other food LCA studies (Hoang et al., 2016), (Heard et al., 2019), represents a food productoriented, cross-sector perspective of GHG emissions. However, one should properly understand that those studies estimate the coupled effects between perishable food products embedded emissions and cold chain facilities operation emissions. It does not reflect the lifetime emissions of the cold chain facilities themselves (i.e. the construction, operation, and retirement of refrigerated warehouses and refrigerated vehicles), which are interdependent with the food system and are not easily measured in terms of a standalone functional unit. Nevertheless, the impacts of cold chain infrastructure are not well understood and future studies are needed on the GHG emissions of the cold chain infrastructure to reveal the environmental impacts of the cold chain industry.
In this paper, we investigate the lifecycle GHG emissions of consuming 1 kg of unprocessed vegetables, fruit, meat, and aquatic products delivered by cold chain logistics in China, based on the assumptions of a fully developed cold chain to approximate the total potential contribution of the cold chain industry to nationwide GHG emissions. In total, eight cold chain scenarios with each scenario representing one food category in one temperature condition (HT, MT, or LT) are considered in our study. We analyzed the total lifecycle GHG emissions in each cold chain scenario, breaking down emissions into different stages and sources, and conducting a one-at-a-time parameter sensitivity analysis. The main conclusions are drawn as follows:
(i) Although agriculture activities dominate the environmental impacts along the food lifecycle, the GHG emissions of postagriculture cold chain and household activities are still significant on an absolute basis and should not be overshadowed. The post-agriculture activities of food supply chains could potentially contribute to 1e3% of total emissions in China. (ii) In the post-agriculture stages, it is found that household, retail, and refrigerated warehouses are the largest three emission stages for meat/aquatic scenarios, while most of the emissions of fruit/vegetable scenarios are from the refrigerated warehouse and 1st refrigerated transportation stages. In the four vegetable/fruit scenarios, most of the emissions are from cold chain activities with an average of 54%. (iii) The energy usage of the cold chain (e.g., refrigeration, transportation) results in around 61% of post-agricultural emissions in fruit/vegetable scenarios, which suggests that using lower-carbon energy sources is an effective approach to reduce post-agriculture cold chain emissions of fruits and vegetables. In contrast, FLW emissions are the highest contributor to meat/aquatic post-agricultural emissions. It is because the FLW emissions are amplified by significant carbon invested in producing meat/aquatic products. Even a small quantity of food loss or waste could lead to considerable GHG emissions. Hence, using the cold chain to reduce food losses and changing behaviors to reduce food wastes are critical approaches. (iv) According to the observations of sensitivity analysis, decreasing the food losses and wastes and time duration spent in cold chain activities would have significant influences on the post-agriculture GHG emissions. Reduction of those factors, together with reducing refrigerant leakage, using lower-GWP refrigerants, and reducing emissions associated with transportation and electricity production can be effective to reduce the total lifecycle emissions of the food system. Such changes require improvement from multiple aspects across the supply chain including agriculture technologies, refrigeration technologies, and behaviors of managing food products.
The perishable food cold chain will inevitably develop significantly in the coming years in China due to economic growth and improvements in the standard of living. This paper presents an estimation of what lifecycle GHG emissions of perishable food cold chain would be expected after China has a fully developed food cold chain if significant interventions are not enacted. Essentially, by understanding the GHG emissions at each stage of the food lifetime, decisions can be made to support the sustainable development of the cold chain in China.
Article 5 Group 1: parties not in Group 2; Article 5 Group 2: Bahrain, India, the Islamic Republic of Iran, Iraq, Kuwait, Oman, Pakistan, Qatar, Saudi Arabia and the United Arab Emirates (The European FluoroCarbons Technical Committee),(UNEP, 2016).
Article
Group 1 baseline: Average production/consumption of HFCs in 2020e2022 þ
65% of HCFCs baseline production/consumption; Article 5 Group 2 baseline: Average production/consumption of HFCs in 2024e2026 þ 65% of HCFCs baseline production/consumption (The European FluoroCarbons Technical Committee), (UNEP, 2016).