Development of the Food Systems–Related Greenhouse Gas Emissions Factor Database Using the Korea National Health and Nutrition Examination Survey (2016–2018)

Abstract

Background:

The increase in the frequency and scale of climate-related disasters is closely linked to greenhouse gas emissions (GHGEs) from food systems. Due to a lack of a comprehensive emission database that covers entire food systems, data on per capita dietary GHGEs are limited.

Objective:

We created the Food Systems–related GHGE Factor Database (FS-GHGEF-D) to cover an entire food system and estimate per capita GHGEs for Korea.

Methods:

We include GHGE factors for 3,894 food items derived from 24-h recall data of the seventh Korea National Health and Nutrition Examination Survey (2016–2018). We extracted these factors from 920 articles, excluding studies that focused only on specific GHG types or single-system boundaries and used a Monte Carlo Markov chain simulation to assess uncertainty of estimates.

Results:

The FS-GHGEF-D covered 96.6% of food items. A total of 265 food items, primarily alcoholic and nonalcoholic beverages, were characterized by high degrees of uncertainty. However, removing these foods did not significantly alter the average GHGE factor across all food groups or affect coverage significantly, with the exception of the beverage category (both alcoholic and nonalcoholic). The average daily diet–related GHGEs per capita in Korea, as calculated using FS-GHGEF-D, were $5.08{\text{\hspace{0.17em}kgCO}}_2\text{eq}$. Among food groups, meats contributed the most to the total variation in dietary GHGEs in the Korea population (75.7%). Men generally emitted more GHGEs than did women, with men in their 30s being the highest emitters.

Discussion:

This study highlights the utility of a full-system GHGE database that addresses prior limitations in global estimates. Korean dietary patterns exceed climate-compatible thresholds, with substantial variation by demographic groups. These findings support the need for equity-focused strategies and integration of environmental considerations into national dietary guidelines for sustainable and climate-resilient food systems. https://doi.org/10.1289/EHP15534

Introduction

In recent years, the world has witnessed a significant surge in climate-related catastrophes, such as floods, droughts, and heatwaves. This phenomenon is closely linked to human activities, particularly greenhouse gas emissions (GHGEs).^1,2 Addressing these environmental challenges necessitates a comprehensive understanding of key contributors to GHGEs. Among these, food systems play a pivotal role, as they account for a substantial portion of total emissions. According to previous study,³ food systems–related GHGEs (FS-GHGEs) constitute $\sim 34\%$ of global emissions, underscoring the critical need for sustainable transformations within food systems.

As defined previously,⁴ “food systems encompass the entire range of actors and their interlinked activities involved in the production, aggregation, processing, distribution, consumption, and disposal of food products.” However, despite the well-documented significance of FS-GHGEs, there is limited data quantifying national food systems–related per capita GHGEs from daily food consumption. This gap is largely due to the absence of comprehensive databases containing detailed food systems–related GHGE factors for commonly consumed foods. Understanding FS-GHGEs at the national level is essential for assessing real-world environmental exposures and their potential implications for human health, as dietary habits are intricately linked to both individual health outcomes and environmental sustainability.

The farming stage, encompassing agricultural production and land-use change, is the primary contributor to total FS-GHGEs, responsible for $\sim 71\%$ of emissions.³ However, to evaluate the environmental impact of individual food items comprehensively, a life cycle assessment (LCA) is required. An LCA examines the full life history of a food item, typically considering five system boundaries: farming (e.g., land use, water consumption, energy inputs, fertilizer and pesticide application, fishing, or livestock production), processing (e.g., food processing and packaging), retail (e.g., transportation, distribution, and storage), cooking (e.g., preparation and consumption), and waste management.^5–7 Definitions of key terms used throughout this manuscript are provided in the Supplemental Material, “Glossary of terms.”

Globally, several efforts have been made to establish FS-GHGE factor databases, including the following: the SHARP Indicators Database (SHARP-ID) in Europe,⁸ the Database of Food Impacts on the Environment for Linking to Diets (dataFIELD) in the United States,⁹ the Chinese Food Life Cycle Assessment Database (CFLCAD),¹⁰ and the Japanese database of GHGEs.¹¹ However, these databases often underestimate FS-GHGEs due to limited coverage of food items or system stages. Additionally, none of these databases evaluate the uncertainty of their data,¹² despite their reliance on comprehensive literature reviews. This results in variability in the variables considered at each LCA stage across studies.¹³

Korea faces similar limitations. The only existing GHGE factor database for commonly consumed foods in Korea¹⁴ has critical drawbacks. First, it includes only 32 food items, limiting its comprehensiveness. Second, it relies solely on one database for the farming stage. Third, it considers only ground transport, excluding other modes of transportation. Fourth, the GHGE coefficient for the cooking stage is based exclusively on the use of liquefied natural gas. Last, the database does not account for GHGEs from packaging and waste stages.

To address these gaps and advance previous work, we aimed to develop a comprehensive food system–related GHGE factor database (FS-GHGEF-D) for foods consumed in Korea. This database was constructed using food items identified from the 24-h dietary recall data collected during the seventh Korea National Health and Nutrition Examination Survey (KNHANES).¹⁵ Additionally, we aimed to estimate per capita daily diet-related GHGEs (dGHGEs), thereby contributing to a more detailed understanding of the environmental impacts associated with dietary patterns in Korea and providing a foundation for future health and environmental policy initiatives.

Methods

Study Data Sources

A list of foods appropriate for the development of a Korean FS-GHGEF-D and participants for estimation of daily diet–related per-capita GHGEs were drawn from the seventh KNHANES between 2016 and 2018. A total of 21,271 participants aged 1 year or older completed a single 24-h dietary recall (24HDR) (from midnight to midnight), reporting on 3,894 food items categorized into 20 food groups. For children under 12 years of age, dietary recall data were collected through proxy reporting by a parent or primary caregiver, typically the mother. To account for seasonal and weekly variations and ensure accurate estimates of the population’s per capita daily intake, survey days were distributed across weekdays and weekends throughout all four seasons. They were given standard measuring tools and two-dimensional measuring guides to reduce errors resulting from variations in how they perceived the same volumes (e.g., dish, cup, or bowl). For home-cooked meals, a detailed recipe was provided by the person in charge of cooking. Detailed information about the nutritional assessment can be obtained from the KNHANES website (https://knhanes.kdca.go.kr/knhanes).

Development of FS-GHGEF-D

Search strategy of food-related GHGE factors.

The Korean Food Systems–related Greenhouse Gas Emission Factor Database was developed based on a literature review followed the PRISMA (Preferred Reporting Items for Systematic Review and Meta-Analyses) guidelines provided by the PRISMA Group.¹⁶ We used the PubMed (https://www.ncbi.nlm.nih.gov/pubmed/), Ovid-EMBASE (http://ovidsp.tx.ovid.com), Web of Science (https://www.webofscience.com), CINAHL (Cumulative Index to Nursing and Allied Health Literature) (https://web.p.ebscohost.com/ehost/search), and Google Scholar search engines for journals regardless of language. Articles written in languages other than English were translated prior to screening and data extraction to ensure inclusion where relevant information was available. To enhance the precision of the database, the term “GHG” was used to denote and focus on the quantity of the GHGEs across all five stages of LCAs. For a comprehensive search, we developed a search strategy using Medical Subject Headings (MeSH) and all fields (e.g., title, abstract, original title, name of substance word, subject heading word, protocol supplementary concept word, rare disease supplementary concept word, unique identifier, manuscript, etc.) to capture studies relevant to our research objectives. The search terms were “diet,” “food,” “greenhouse gas,” “greenhouse effect,” “GHG,” and “greenhouse gas emission” (see Table S1 for exact search strings).

Inclusion/exclusion criteria and study selection.

Articles were included if they reported the GHGE factor of one or more food items in kilograms or grams of ${\mathrm{CO}}_2$ equivalent per $1\mathrm{kg}$ of food (${\text{kgCO}}_2\text{eq}/\mathrm{kg}$ or ${\text{gCO}}_2\text{eq}/\mathrm{kg}$), by LCA stages such as farming, processing, retail, cooking, and waste. Articles were excluded if they were not related to food, estimated only single GHG components [rather than ${\mathrm{CO}}_2$, nitrous oxide (${\mathrm{N}}_2\mathrm{O}$), and methane (${\text{CH}}_4$)], focused on a single LCA stage (not cumulative), presented simulation results only, or were reviews or nonoriginal articles. Using Rayyan, a web-based automated screening tool developed by the Qatar Computing Research Institute (https://www.rayyan.ai/), duplicate articles were automatically identified and excluded. Although the Rayyan tool did not fully support non-English articles, it was partially effective in identifying both English and non-English duplicates by utilizing metadata such as author names, abstracts, and DOIs. Titles and abstracts were then screened as part of the initial exclusion process, with specific inclusion and exclusion keywords applied to systematically filter and categorize relevant studies. During the title and abstract screening phase, articles were included if they contained at least one of the following dietary terms (“diet” or “food”) and at least one of the following environmental terms (“greenhouse gas,” “greenhouse effect,” “GHG,” or “greenhouse gas emission”). Articles were excluded if they contained the keywords “literature review,” “systematic review,” and “meta-analysis.” The authors manually reviewed the remaining articles to determine eligibility. Articles that met the inclusion criteria were transferred to EndNote, where the software was used to retrieve full texts as much as possible. For articles not accessible through the program, we searched them on the internet. During the full-text article assessment for eligibility, we initially evaluated whether the articles satisfied the inclusion criteria based on their methods section. For eligible articles, emission factors were manually extracted from the results sections. Although this process was time-intensive, automated extraction software was not employed.

Data extraction.

During the full-text review step, we first verified the types of GHG assessed, identified the LCA stages considered, confirmed the functional unit used, and examined the specific food items for which emission factors were calculated. Based on this systemic process, we then extracted the following data from each study: title, author(s), country, year of publication, publication journal, food items or group, system boundary, unit of ${\mathrm{CO}}_2$ equivalence, and GHGE factor values. The GHGE factor values were imported from the literature. To obtain a better approximation to truth than a value of 0 for food items with missing GHGE factors,^17,18 we considered the following steps: First, for 3,894 listed food items, we matched foods that have the same processing and consumption types and then imported the GHGE factors (cumulative values from LCA stages) for these matched foods. Second, if data were not available for a specific food, we imputed the GHGE factors of foods with the most similar cultivation process, as cultivation accounts for the largest share of emissions.¹⁹ Similarity was determined based on key agronomic attributes, including primary crop type (e.g., solanaceous, root, or leafy vegetables), climate requirements, cultivation methods, and input intensity (e.g., irrigation and fertilizer use), drawing on information from agricultural databases and prior LCA studies. In the case of plants, the specific parts consumed (e.g., leaves, roots, and stems) were not considered separately. Instead, we used the values of foods sharing identical cultivation conditions, even if the parts differed (e.g., the raw value of red pepper leaves is taken from red pepper). In addition, if the processing method was different, the value for the most similar processing method (e.g., stew, soup, additives, fortified food, dilution, drying, powdering, or concentration) was selected. For imported foods, such as some meats or sauces, when the origin was unclear or labeled as imported, we applied the GHGE factors from all countries. However, when the origin was clearly specified, we only imported values for the relevant country. To account for the GHGEs associated with transporting foods and ingredients overseas, we added additional emissions by selecting and applying emissions factors for bulk ($0.00211{\text{\hspace{0.17em}kgCO}}_2\text{eq}/\text{ton}\times \mathrm{km}$) and container ($0.00902{\text{\hspace{0.17em}kgCO}}_2\text{eq}/\text{ton}\times \mathrm{km}$)¹⁴ sea freight as follows. The GHGEs during food transportation were calculated by multiplying the amount of food transported (ton) by the distance traveled (km) and the corresponding GHGE factor (${\text{\hspace{0.17em}kgCO}}_2\text{eq}/\text{ton}\times \mathrm{km}$) [greenhouse gas emissions at the food transportation stage = food material weight × transport distance × carbon labeling (sea transport) emission factor]. The cumulative GHGE factors were extracted for 3,894 food items. In the finalized FS-GHGEF-D, we used the average value of the GHGE factor of each food item as the definitive GHGE factor. The database was built in Excel (version 16.0.5413.1000; Microsoft).

Uncertainty assessment of FS-GHGEF-D.

In the present study, we considered parameter uncertainty.²⁰ The primary aim of the uncertainty assessment procedure was to comprehensively and precisely identify and record the origins of uncertainty in individual estimates as well as in the cumulative dataset.^12,21 We applied two methods to assess the uncertainty of the developed database (FS-GHGEF-D). First, we collected source data as comprehensively as possible, with no restrictions by country, year, or language of the literature, to minimize the uncertainty associated with random sampling errors and biases.²² This approach aimed to mitigate potential errors and biases by filling in data gaps. Second, we evaluated the uncertainty of food GHGE factors using a Monte Carlo Markov chain (MCMC) simulation,¹² which is a Bayesian approach to statistical analysis. This nonparametric approach was used to confirm whether the uncertain distribution derived from GHGEFs of various discontinuous values in several articles is appropriate.²³ In the process of FS-GHGEF-D development, we calculated the mean and variance for individual foods, and the mean values were used as the final values. Instead of assuming an exact shape for the distribution of many food items, we assumed a Cauchy distribution, for which expected values do not exist.²⁴ While there are no clear-cut criteria for appropriateness, suitability is usually based on three points, and the assumed distribution is considered appropriate if all three criteria are met^25–27: a) the trace plot of mean values from countless distributions, which the MCMC algorithm creates, shows fluctuations, but there is convergence to one value (mean); b) autocorrelations rapidly converge; and c) the kernel density estimation graph also shows a unimodal Cauchy distribution without multiple inflection points.

Estimation of Daily Diet–Related GHGEs per Capita

The method for calculating daily diet–related GHGEs among 21,271 participants 1 year of age or older who completed 24HDR involved the following steps. First, to estimate dGHGE, we assigned a value of 0 to a total of 131 food items without available GHGE factors. The diet-related emissions of each specific food were determined by multiplying the quantity consumed (in grams) by an emissions factor (expressed in ${\text{kgCO}}_2\text{eq}/\mathrm{kg}$) divided by 1,000. Next, the calculated values for all foods were summed to obtain the total daily GHGs. Daily diet–related GHGEs per capita were calculated after assigning a value of 0 to foods for which a GHGEF was not available. The process can be represented by the following equation: daily amount of GHGEs $\left({\text{kgCO}}_2\text{eq}/\text{day}\right)=\sum$ [GHGE factor ${(\text{kgCO}}_2\text{eq}/\mathrm{kg})/\mathrm{1,000}$ for each food × corresponding food intake (g)]/1,000.

Statistical Analysis

All analyses were conducted using SAS version 9.4 (SAS Institute Inc.). Coverage of the database was quantified as the percentage of food items out of the total foods in the list ($n=\mathrm{3,894}$) that were assigned a GHGE factor. Five-number summaries (minimum, Q1, median, Q3, maximum) of the GHGEF were presented by food group. Correlation coefficient ($r^2$) values were calculated using a linear regression model that related the GHGEs from each food item (y) to daily diet–related GHGEs (x) to assess the extent to which each food item explains the variability in diet-related GHGEs. The appropriateness of this database was assessed by PROC MCMC.²⁸ The differences in daily diet–related GHGEs before and after excluding food items with relatively low certainty were calculated.

As described in the “Study Data Sources” section, demographic information was sourced from the KNHANES, which includes nationally representative data on participant characteristics: sex (male, female), age groups (seven groups: 1–18, 19–29, 30–39, 40–49, 50–59, 60–69, and $70+$), region (three groups: large city, medium-sized city, and rural area), household income quintiles [low (first quintile), middle-low (second), middle (third), middle-high (fourth), and high (fifth)], and education level (four groups: elementary school and under, middle school, high school, and college and above). In terms of household income level, 54 people were missing, and 1,553 people were missing education level. For education level, only adults who could specify their own education levels were included. Weights were applied to account for the complex sampling design, using PROC SURVEY. Means and standard errors of GHGEs according to groups of those characteristics were obtained using a general linear model. Differences between groups were assessed by Tukey’s post hoc comparison test and the significance level was $p<0.05$.

This study was conducted in accordance with the guidelines of the Declaration of Helsinki and approved by the institutional review board of Hanyang University (IRB number HYUIRB-202303-004). All participants provided written informed consent before participating in the survey.

Results

Construction of the FS-GHGEF-D

Database coverage and structure.

A total of 26,498 articles were initially identified (4,068 from PubMed; 3,650 from EMBASE; 17,192 from the Web of Science; and 1,588 from CINAHL) (Figure 1). After removing 5,212 duplicate records, 19,890 articles were excluded by screening the titles and abstracts, and 12 studies were added by searching Google Scholar; 1,408 remained for the data extraction from full-text evaluation. Among them, 488 articles (102 not related to food, 49 single GHG components, 247 single LCA stage, 9 simulation studies, and 81 nonoriginal) were excluded. From 920 original sources, which were produced in 91 countries (by stages: 40,651 farming, 2,069 processing, 151,461 retail, 52,681 cooking, and 53,558 waste), in total, 300,420 values for 3,894 food items were identified. The coverage of the FS-GHGEF-D for food items is shown in Table 1. For 131 food items, such as “pudding,” “seaweeds,” “insects,” “breast milk,” and “sauce,” we could not find matching foods to extract GHGEFs. Total coverage was therefore $\sim 96.6\%$, and out of a total of 3,894 food items, “cereals” ($n=681$ items) accounted for the largest number of items (17.49%), followed by “milk and milk products” ($n=460$ items; 11.81%), “seasonings” ($n=452$ items; 11.6%), and “beverages” ($n=398$ items; 10.22%). “Eggs” ($n=13$ items) and “oil and fats (animal)” ($n=13$ items) accounted for the smallest number of items (0.33%) and “others (animal)” ($n=15$ items) accounted for 0.39%. Most food groups presented with coverage $>90\%$, with the exception of seaweeds (15.6%) and others (animal) (66.7%). The final GHGE factor was calculated as the average value of the collected items,^8–11 with “seaweeds” being the lowest food group (0.59) and “meats” the highest (16.34). Among 20 food groups, meats showed the highest average GHGE factor ($16.34{\text{kgCO}}_2\text{eq}/\mathrm{kg}$), followed by fishes ($7.92{\text{kgCO}}_2\text{eq}/\mathrm{kg}$), with seaweeds presenting with the lowest value ($0.59{\text{\hspace{0.17em}kgCO}}_2\text{eq}/\mathrm{kg}$).

Figure 1.

PRISMA flow chart of selecting data sources. Note: CINAHL, Cumulative Index to Nursing and Allied Health Literature; GHG, greenhouse gas emissions; LCA, life cycle assessment; PRISMA, Preferred Reporting Items for Systematic Review and Meta-Analyses.

Table 1.

The coverage and the range of the food systems–related greenhouse gas emission factor (GHGEF) (${\text{kgCO}}_2\text{eq}/\mathrm{kg}$) database (FS-GHGEF-D) by food groups of the seventh Korea National Health and Nutrition Examination Survey (2016–2018).

Food groups (20 groups)	GHGEF database
	Total number of food items	Number of food items the GHGEF assigned	Coverage (%)	GHG emission factor (${\text{kgCO}}_2\text{eq}/\mathrm{kg}$)
				Mean	Minimum	25th	50th	75th	Maximum
Total food items	3,894	3,763	96.6	4.21	0.04	1.83	2.34	4.14	36.07
Plant foods (11 groups)
Cereals	681	678	99.6	2.59	0.78	1.84	1.96	2.34	33.37
Potatoes and starches	68	67	98.5	1.45	0.35	0.62	0.90	2.46	3.12
Sugars and sweeteners	263	263	100	2.53	0.19	2.06	2.27	3.17	4.17
Pulses	137	137	100	2.06	1.14	1.41	1.41	3.00	3.35
Nuts and seeds	61	61	100	1.46	0.93	1.18	1.35	1.66	2.33
Vegetables	311	306	98.4	2.15	0.35	1.68	1.92	2.89	5.39
Mushrooms	33	32	97	2.81	1.92	2.04	2.62	3.65	3.65
Fruits	100	100	100	2.11	0.19	1.47	1.65	2.54	5.09
Seaweeds	128	20	15.6	0.59	0.04	0.04	0.39	0.39	3.12
Oils and fats (plants)	64	64	100	3.73	1.30	2.39	2.74	3.74	7.51
Others (plants)	12	12	100	2.55	1.55	1.87	2.7	3.22	3.27
Animal foods (6 groups)
Meats	234	234	100	16.34	2.09	6.20	15.23	30.38	36.07
Eggs	13	13	100	4.08	4.02	4.08	4.08	4.09	4.09
Fishes	349	349	100	7.92	0.99	4.03	5.22	11.31	18.7
Milk and milk products	460	459	99.8	3.56	1.73	1.73	2.24	5.09	9.48
Oils and fats (animal)	13	13	100	11.1	7.8	11.02	11.02	11.02	15.39
Others (animal)	15	10	66.7	4.23	1.31	4.59	4.59	4.59	4.59
Seasonings	452	445	98.5	1.87	0.91	1.52	1.89	1.89	10.73
Beverages	398	398	100	5.71	0.21	2.64	4.17	8.83	11.73
Alcoholic beverages	102	102	100	2.06	0.80	0.80	2.40	2.80	4.17

Note: The 131 food items with no data were “pudding,” “seaweeds,” “insects,” “breast milk,” and “sauce,” which were not the main food items in greenhouse gas estimation studies. GHG, greenhouse gas.

Uncertainty assessment of FS-GHGEF-D.

To construct an FS-GHGEF-D incorporating 3,894 foods, average values of GHGE factors from all data-source articles were used. However, limited data sources may lead to biased average values. The trace plot, autocorrelation, and kernel density estimation graph by MCMC are shown in Figure S1. For 3,629 food items, trace plots of most MCMC results converged to a common value, autocorrelations rapidly converged, and unimodal Cauchy distributions without multiple inflection points in the kernel density estimation graph were found ($n=\mathrm{3,629}$; 93.2%) as shown in Figure S1A. In total, 265 food items demonstrated uncertainty as shown in Figure S1B. Among them, data were not available for 131 food items, including 108 seaweeds, pudding, insects, breast milk, and sauces. We found another 134 foods, such as alcoholic and nonalcoholic beverages (cocktails, traditional Korean liquors, instant soups, ginseng, and instant coffee), which had available GHGEFs but with low certainty, as depicted in Figure S1B. Although there are no commonly agreed criteria for the extent of uncertainty, the database appeared to be acceptable as it considered 6.81% of all food items. Table 2 presents the coverage and mean values of GHGEFs after excluding food items with high uncertainty. Compared with before excluding uncertain food items, these 265 food items did not significantly affect the average GHGEFs of each food group in the database and the daily diet–related GHGEs per capita (1.18%). However, the GHGEs from alcohol consumption showed a 12% reduction after excluding foods with high uncertainty (Table S2). Unfortunately, no data were available to compare with the uncertainty found in the present study.

Table 2.

Coverage and mean values of the food systems–related greenhouse gas emission factor (GHGEF) (${\text{kgCO}}_2\text{eq}/\mathrm{kg}$) database (FS-GHGEF-D) by food groups after excluding food items with high uncertainty in the seventh Korea National Health and Nutrition Examination Survey (2016–2018).

Food groups (20 groups)	Number of food items with high uncertainty by the MCMCa,b	After excluding food items with high uncertainty			Difference from before the exclusion
		Number of food items	Coverage (%)	Mean value of the GHGEF	Number of food items	Coverage (%)	GHGEF
Total food items	265 (131)	3,629	93.2	—	$-134$	$-3.4$	—
Plant foods (11 groups)
Cereals	15 (3)	666	97.8	2.43	$-12$	$-1.8$	$-0.16$
Potatoes and starches	2 (1)	66	97.1	1.45	$-1$	$-1.4$	0
Sugars and sweeteners	0	263	100	2.53	0	0	0
Pulses	0	137	100	2.06	0	0	0
Nuts and seeds	0	61	100	1.46	0	0	0
Vegetables	5 (5)	306	98.4	2.15	0	0	0
Mushrooms	1 (1)	32	97.0	2.81	0	0	0
Fruits	0	100	100	2.11	0	0	0
Seaweeds	108 (108)	20	15.6	0.59	—	0	0
Oil and fats (plants)	0	64	100	3.73	0	0	0
Others (plants)	4 (0)	8	66.7	2.21	$-4$	$-33.3$	$-0.34$
Animal foods (6 groups)
Meats	1 (0)	233	99.6	16.21	0	$-0.4$	$-0.13$
Eggs	0	13	100	4.08	$-1$	0	0
Fishes	0	349	100	7.76	0	0	$-0.16$
Milk and milk products	1 (1)	459	99.8	3.56	0	0	0
Oil and fats (animal)	0	13	100	11.10	0	0	0
Others (animal)	5 (5)	10	66.7	4.23	0	0	0
Seasoning	17 (7)	435	96.2	1.85	$-10$	$-2.3$	$-0.02$
Beverages	70 (0)	328	82.4	5.20	$-70$	$-17.6$	$-0.51$
Alcoholic beverages	36 (0)	66	64.7	1.87	$-36$	$-35.3$	$-0.19$

Note: —, no data; MCMC, Monte Carlo Markov chain.

^a Total food items with high uncertainty determined by the MCMC are shown, with the number of food items not assigned the GHGEF indicated in parentheses.

^b The value in the parenthesis is the number of food items which did not have GHGEF.

Estimation of Daily dGHGEs per Capita in Korea

Table 3 presents the estimated daily diet–related greenhouse gas emissions (dGHGEs) (${\text{kgCO}}_2\text{eq}/\text{day}$) from 20 food groups across age categories, based on dietary intake data from the Korea National Health and Nutrition Examination Survey (KNHANES) for the years 2016–2018. Daily per capita dGHGEs were $5.08{\text{\hspace{0.17em}kgCO}}_2\text{eq}$, and they were the highest for participants in their 30s. Among 20 food groups, the meats group was the largest contributor to dGHGEs. Within the plant food groups, cereals, including cooked rice (a Korean staple) also contributed statistically significantly. Meats had the highest contribution to the variation in dGHGEs ($r^2=75.69\%$), followed by beverages ($r^2=9.13\%$). After these, alcoholic beverages ($0.528{\text{\hspace{0.17em}kgCO}}_2\text{eq}/\text{day}$) exhibited significant emissions, while fish ($r^2=6.73\%$) and cereals ($r^2=2.21\%$) also made substantial contributions. Table 4 provides further insights into dGHGEs by sex, age, regional, income level, and education variations. We found that men were responsible for more dGHGEs compared with women, and individuals in their 20s, 30s, and 40s were the leading emitters of GHGs due to food intake (Figure S2).

Table 3.

The amount of daily diet–related greenhouse gas emission (dGHGE) (${\text{kgCO}}_2\text{eq}/\text{day}$) by food groups and age groups in the seventh Korea National Health and Nutrition Examination Survey (2016–2018) ($n=\mathrm{21,271}$).

Food groups (20 groups)	1–18	19–29	30–39	40–49	50–59	60–69	$70+$	Total	$r^2$ in daily diet–related GHGEs (%)
Total dGHGE	4.27	5.87	6.17	5.91	5.05	4.35	3.41	5.08	—
Number	4,417	1,958	2,673	3,089	3,108	2,952	3,074	21,271	—
Plant foods (11 groups)
Cereals	0.511	0.535	0.530	0.494	0.410	0.374	0.323	0.453	2.21
Potatoes and starches	0.029	0.034	0.030	0.031	0.033	0.042	0.031	0.032	0.03
Sugars and sweeteners	0.029	0.024	0.020	0.018	0.015	0.013	0.012	0.019	0.02
Pulses	0.077	0.086	0.105	0.103	0.102	0.097	0.083	0.092	0.25
Nuts and seeds	0.004	0.006	0.008	0.009	0.009	0.011	0.007	0.007	0.01
Vegetables	0.238	0.339	0.450	0.551	0.545	0.501	0.390	0.423	2.00
Mushrooms	0.017	0.020	0.023	0.026	0.023	0.020	0.017	0.021	0.01
Fruits	0.216	0.152	0.237	0.266	0.317	0.285	0.221	0.242	1.13
Seaweeds	0.041	0.034	0.044	0.052	0.054	0.061	0.051	0.047	0.12
Oils and fats (plants)	0.014	0.017	0.018	0.015	0.012	0.009	0.007	0.013	0.00
Others (plants)	0.010	0.024	0.038	0.050	0.046	0.017	0.016	0.029	0.00
Animal foods (6 groups)
Meats	1.281	1.708	1.846	1.662	1.237	1.113	1.040	1.422	75.69
Eggs	0.140	0.141	0.143	0.156	0.145	0.133	0.110	0.138	0.21
Fishes	0.284	0.408	0.527	0.512	0.428	0.332	0.244	0.392	6.73
Milk and milk products	0.502	0.400	0.344	0.347	0.313	0.272	0.231	0.363	0.80
Oils and fats (animal)	0.015	0.024	0.020	0.016	0.012	0.011	0.005	0.017	0.00
Others (animal)	0.000	0.260	0.128	0.018	0.146	0.157	0.006	0.113	0.00
Seasonings	0.043	0.062	0.069	0.063	0.052	0.040	0.030	0.052	0.04
Beverages	0.792	1.042	0.933	0.792	0.502	0.355	0.237	0.680	9.13
Alcoholic beverages	0.031	0.551	0.660	0.726	0.650	0.510	0.352	0.528	1.60

Note: The 131 food items with no data were assigned as zero value when calculating the dGHGE per day per capita. —, no data.

Table 4.

Daily diet-related greenhouse gas emission (dGHGE) by some characteristics of the seventh Korea National Health and Nutrition Examination Survey (2016–2018) ($n=\mathrm{21,271}$).

	Number	Percentage	Daily dGHGE (${\text{kgCO}}_2\text{eq}/\text{day}$)
			Meana	(SE)	Median (minimum, maximum)
Total population	21,271	—	5.08	0.04	4.11 (0.02, 53.92)
Sex
Men	9,440	44.38	5.97 A	0.07	4.89 (0.11, 59.92)
Women	11,831	55.62	4.19 B	0.04	3.50 (0.02, 40.17)
Age groups
1–18 years	4,417	20.77	4.27 A	0.07	3.49 (0.14, 40.96)
19–29	1,958	9.21	5.88 BC	0.13	4.74 (0.10, 49.24)
30–39	2,673	12.57	6.26 B	0.13	5.09 (0.11, 53.92)
40–49	3,089	14.52	5.78 C	0.11	4.73 (0.08, 46.98)
50–59	3,108	14.61	4.94 D	0.07	4.18 (0.11, 39.54)
60–69	2,952	13.88	4.31 A	0.08	3.60 (0.05, 45.70)
70 or older	3,074	14.45	3.41 E	0.07	2.66 (0.02, 49.37)
Regionb
Big city	9,193	43.22	5.09 A	0.07	4.14 (0.05, 49.24)
Medium-sized city	8,178	38.45	5.23 A	0.09	4.20 (0.10, 53.92)
Rural area	3,900	18.33	4.66 B	0.12	3.74 (0.02, 43.97)
Household Income levelc
Low	3,373	15.86	3.86 A	0.11	2.98 (0.02, 53.92)
Middle-low	4,021	18.90	4.55 B	0.09	3.69 (0.05, 43.97)
Middle	4,529	21.29	5.13 C	0.10	4.15 (0.08, 49.37)
Middle-high	4,690	22.05	5.39 D	0.11	4.37 (0.11, 50.64)
High	4,604	21.64	5.81 E	0.11	4.77 (0.10, 45.70)
Education leveld
Elementary school or under	3,300	15.51	3.42 A	0.06	2.75 (0.05, 39.95)
Middle school	1,547	7.27	4.33 B	0.11	3.52 (0.16, 31.19)
High school	4,808	22.60	5.38 C	0.08	4.38 (0.07, 49.24)
College or above	5,646	26.54	5.96 D	0.08	4.95 (0.11, 50.64)

Note: 54 missing in income level; 1,553 missing in education level. —, no data.

^a $p$-difference was marked as letters, which were determined using a general linear model (Tukey’s multiple comparison).

^b Region was classified as big cities (urban neighborhoods in metropolitan municipalities with $\ge 1\text{\hspace{0.17em}million}$ population), medium-sized cities (urban areas in nonmetropolitan provinces), and rural areas (towns and townships in counties or rural parts of mixed cities).

^c Household income categories are based on equalized household income, calculated as the monthly household income divided by the square root of the number of household members. Participants are then classified into income quintiles.

^d Reanalyzed only for adults who could use an individual’s final educational background.

Discussion

Through this study, we developed FS-GHGEF-D, a comprehensive and reliable database of food-related GHG emission factors using 24HDR from KNHANES. To the best of our knowledge, this database represents the initial attempt in Korea to create a comprehensive GHG database. Notably, it accounts for emissions up to the grave stage, addressing a previous limitation observed in overseas databases,^8–11,14 which tended to underestimate emissions. With the aid of this database, we were able to calculate the GHG emissions originating from the dietary habits of Koreans, revealing that, on average, each individual emits $5.08\mathrm{kg}$ of ${\mathrm{CO}}_2$ per day.

When building a food database, employing imputed or well-documented values based on similar food values is almost always a closer approximation compared with assigning a value of 0.^17,18 We therefore incorporated data from a wide array of sources, broadening the scope of the variables and creating a versatile database that accommodated various research needs regardless of year, country, or language, which may reduce random errors in the database.²² However, articles may vary in terms of sources of GHGE, such as energy production, transportation, and industrial processes, and data from those articles may differ in measurement methodologies, standards, or units, making it necessary to standardize and harmonize the data for consistency. Most GHGEF databases were established similarly, particularly when estimating at least three types of GHGs (${\mathrm{CO}}_2$, ${\text{CH}}_4$, and ${\mathrm{N}}_2\mathrm{O}$) and converting units into a single functional unit²⁹ in accordance with the present study. However, regarding the coverage of the five stages on a LCA, most databases do not encompass GHGE factors from all of these stages. The SHARP-ID⁸ and the CFLCAD¹⁰ collected cumulative GHGE factors from farm gate to cooking stage, the dataFIELD⁹ gathered cumulative GHGEFs from farm gate to processor gate, and the Japanese GHGE database¹¹ collected emission factors from farm to regional distribution center or retail stage. In the present study, we tried to address this issue by selecting only articles that presented a cumulative amount of GHGEs of all five system boundaries (from farm gate to waste gate), although specific sources within each boundary might vary between articles.

According to data from the Ministry of Environment, Korea is the 10th-largest carbon emitter in the world, releasing 680 million tons of carbon annually, which accounts for 1.8% of global emissions.³⁰ By using the database developed in this study, we estimated that Korea’s 51.74 million people emit $\sim 5.08\mathrm{kg}$ of carbon daily due to food intake, resulting in an annual total of 96 million tons of carbon. Therefore, Korea contributes 15.99% of global GHGEs due to food consumption. A report from the World Wildlife Fund (WWF) on carbon emissions related to food in the United Kingdom revealed that $\sim 20\%$ of the country’s GHGEs were associated with food, with a daily per person emission of $5.17{\text{\hspace{0.17em}kgCO}}_2\text{eq}$.³¹ However, as WWF focused solely on calculating GHGEs from four iconic UK dishes, the estimated amount of GHGEs from foods may be underestimated. Another global report suggests that, on average, each person worldwide emits $\sim 6.03{\text{\hspace{0.17em}kgCO}}_2\text{eq}/\text{day}$ due to dietary choices.³² The estimated daily diet–related GHGEs per capita in Korea, $5.08{\text{\hspace{0.17em}kgCO}}_2\text{eq}$, may be relatively low. However, given the global necessity for reducing emissions to $<4.76{\text{\hspace{0.17em}kgCO}}_2\text{eq}$,³¹ the need for improvement in Korea remains.

We also found that men emitted more than women, and those in their 20s, 30s, and 40s were the highest emitters. Some participants were excluded due to missing income and education data. This variation can be ascribed to multiple factors, including elevated meat consumption among men and sex-specific distinctions in dietary habits.²⁹ Participants who lived in small- and medium-sized cities emitted the most GHGs from dietary intake, followed by those in large urban and rural areas. No statistical difference was identified between large urban areas and small- and medium-sized cities, but a significant difference was observed between cities and rural areas. These variations can be ascribed to regional difference in dietary preferences, agricultural methodologies, and food distribution systems.³³ According to income level, the higher the income, the higher the GHGEs from diet. Similarly, when the education level was high, the amount of GHGs emitted due to diet was also relatively high. In terms of income levels, a clear correlation emerged: higher income levels corresponded with elevated GHGEs from diet. This may be partially explained by increased consumption of meat, dairy products, and processed foods (which typically have higher carbon footprints) by individuals with higher incomes and education levels and who often opt for resource-intensive diets that are rich in meat and dairy.^34–36 Despite the increasing environmental awareness among highly educated individuals, demand for convenience and personal preferences may deter such individuals from choosing a sustainable diet.³⁷ Creating an environment that makes it easier to choose a sustainable diet is therefore a critical challenge.

This database has several limitations. Creating an FS-GHGEF-D without imposing any consideration of publication year or geographical region may have expanded the coverage. However, this approach can overlook the impact of temporal changes and regional characteristics on elements within system boundaries. Second, ignoring the publication year may neglect technological advancements or shifts in environmental policies over time, and ignoring regional variations may fail to account for newer technologies and environmental impacts that are influenced by geography and climate. While diverse countries employ distinct agriculture production methods and cover varied distribution distances for various food items, distinguishing among these factors is not feasible.

Despite these limitations, the present study’s FS-GHGEF-D is a comprehensive and reliable database of food systems–related GHGE factors, accounting for emissions up to the waste stage. To the best of our knowledge, this database represents the first attempt to create a comprehensive GHG database for Korea. Given that the present FS-GHGEF-D encompasses a variety of countries and measurement methods, it is possible that this new database could be applied to food data (with the exception of imported foods) not only in Korea but internationally. This database is an essential tool for researchers exploring the connections between dietary patterns, GHGEs, and health outcomes. Both the database and subsequent research findings should help inform dietary guidelines that prioritize environmentally conscious food choices.

Conclusions and opinions are those of the individual authors and do not necessarily reflect the policies or views of EHP Publishing or the National Institute of Environmental Health Sciences.