Assessment of carbon footprint of milk production and identification of its major determinants in smallholder dairy farms in Karnataka, India

Indian dairy enterprise is dominated by smallholder dairy farms that contribute 72% of the country's total milk production. These smallholder dairy farms are often considered to emit substantial greenhouse gases (GHG) but are poor in productive performances. Therefore, it is crucial to estimate the carbon footprint (CF) of milk production of the smallholder Indian dairy farms. The primary objectives of the study were: 1) Assessing the CF of milk production of smallholder dairy farms through life cycle analysis in south-interior Karnataka, India; 2) Identifying the hotspots of GHG emissions and significant factors influencing the CF of milk production in smallholder dairy production system. The study accounted GHG emissions from different sources and considered multiple functions of the smallholder production system. Estimations were made based on primary data collected from 47 farms and associated secondary data. For estimating the CF of milk production, the emissions of carbon dioxide (CO 2 ), methane (CH 4 ) and nitrous oxide (N 2 O) on CO 2 -equivalent (CO 2 - eq) basis from feed production, enteric fermentation, manure management, transport and energy usage were allocated to fat-protein-corrected milk (FPCM) based on mass balance, price (crop by-products and residues) and feed digestibility. Principal component analysis and stepwise linear regression analysis were performed to identify the major factors influencing the CF. The average total GHG emissions (kg CO 2 -eq yr-1 farm-1) attributable to milk production based on mass, economic and digestibility allocations were 8936, 8641 and 8759, respectively. The contributions of CH 4 , N 2 O and CO 2 to the total farm GHG emission were 70.6, 20.5 and 7.69%, respectively. The major emission hotspots were CH 4 emission from enteric fermentation (66.8%) and GHG emission from feed production (23.0%). The average CF of cradle-to-dairy cooperative milk production varied from 1.45 to 1.81 kg CO 2 -eq kg FPCM −1 . The CF of milk production was more than 2-fold greater, when milk yield was below 3500 kg lactating cow −1 yr −1 . The FPCM yield 100 kg body weight −1 , dry matter intake and CH 4 emission from manure management were the strongest determinants of the CF and explained 83.4% of the observed variation. The study emphasized the importance of considering multiple functions of a mixed crop-livestock-based dairy production system for estimating CF per unit of product. The results suggest that maintaining high-yielding dairy animals and adopting appropriate feeding strategies for better feed utilization are the possible effective interventions for reducing the CF of milk production.


INTRODUCTION
Livestock is considered as one of the primary sources of anthropogenic greenhouse gas (GHG) emission, and dairy farms are major contributors in this regard (Rotz et al., 2018).The significant source of GHG emission in dairy farms is methane (CH 4 ) from enteric fermentation (Verge et al., 2007).Additionally, there is the emission of nitrous oxide (N 2 O) directly from animal manure and fodder or pasture land, and indirectly from ammonia emissions and nitrate leaching that would transform into N 2 O in the ecosystem (Rotz, 2018).These sources are treated as independent sources, and interactions among them affect the overall GHG emission.The emission of carbon dioxide (CO 2 ) from land use depends on the demand for food grains, grazing land, and energy usage for farm operations.In addition, the decomposition of lime applied to crop and pasture land also contributes to the emission of CO 2 (Herron et al., 2022).
Assessment of carbon footprint of milk production and identification of its major determinants in smallholder dairy farms in Karnataka, India A Mech, 1 * G Letha Devi, 1 M Sivaram, 2 S Sirohi, 3 A Dhali, 1 AP Kolte, 1 PK Malik, 1 RK Veeranna, 1 L Niketha, 1 and R Bhatta 1 India is the largest producer of milk in the world, contributing 21% of global milk production that employs more than 80 million rural households with the majority being small and marginal farmers as well as the landless.(Ministry of Fisheries, Animal Husbandry and Dairying, 2022).This sector benefits the farm households in terms of nutritive food, supplementary income, and employment of family laborers (Belhekar and Dash, 2016).According to the latest livestock census, the total cattle population of India is 192 million that includes 81.3 million adult female cattle (DAHD, 2019).The total milk production of India is 188 million MT for the year 2018-2019(NDDB, 2022)).Among the 28 Indian States, Uttar Pradesh is the highest milk producing State contributing 16.2% of the total milk and possesses 11.7% of India's total adult female cattle population.The State Karnataka is ranked at 11th contributing 4.20% of the total milk, and the State possesses 5.20% of the country's total adult female cattle.The average annual milk production per adult female cattle is 2312 kg in India, and the respective figures are 3315 kg in Uttar Pradesh and 1946 kg in Karnataka (DAHD, 2019;NDDB, 2022).
Determining GHG emissions from important sources and identifying strategies to reduce the overall GHG emission from dairy farms are important.Efforts have been made previously to assess enteric and manure CH 4 emissions in India (Singhal et al., 2005;Patra, 2012;Kumari et al., 2019) and other regions and countries such as North America, Europe, Australia and New Zealand (Boadi et al., 2004;Beauchemin et al., 2008;Islam and Lee, 2019).These studies suggest that for reducing the environmental impact of dairying, CH 4 mitigation strategies must not come at the expense of greater CO 2 and N 2 O emissions.Therefore, to accurately assess the net impact of dairying on GHG emissions, it is necessary to quantify all emissions and removals using whole-system analysis by life cycle assessment (LCA; Geough et al., 2012).The LCA is a holistic tool used across industries and agriculture to quantify the carbon footprint (CF) of specific products.According to the International Organization for Standardization (ISO, 2006a;ISO, 2006b), LCA is defined as the compilation and evaluation of a product system's inputs, outputs, and potential environmental impact throughout its life cycle.
Indian dairy enterprise is dominated by smallholder dairy farms, which maintain 85% of the country's total cattle population.Nearly 97% of these farms (73 million) have only 2 cows and contribute 72% of the milk produced in India (Moran and Chamberlain, 2017).The smallholder livestock production system is multifunctional in nature.Under such a system, in addition to milk, animals produce fertilizer and provide draft power, and they are considered as the source of capital asset.Therefore, the LCA of smallholder livestock production system should essentially account for multifunctionality, as CF and mitigation strategies depend on the functions included in the assessment (Weiler et al., 2014).All products and services have an environmental impact during their production, use or disposal.The exact nature of this impact is complex and difficult to quantify.The heterogeneity, landscapes and diversity of methods used in the multi-functionality of small-scale livestock production systems in developing countries present a major challenge for assessing the level of GHG emissions (Garg et al., 2016).
Many LCA studies have been conducted previously to estimate the CF of dairy and related co-products under specialized production systems in several developed countries (Geough et al., 2012;Marton et al., 2016;Thoma et al., 2013).In contrast, limited efforts have been made to estimate CF of milk production in the smallholder dairy farms in India, and other developed and developing countries (Bartl et al., 2011;Garg et al., 2016;Salvador et al., 2016;Wilkes et al., 2020).In addition, to the best of our knowledge, no published report is available on this aspect from the southern part of India, which is one of the major contributors to the country's total milk production.The current study aimed to determine the CF of liquid milk production of smallholder dairy farms in south-interior Karnataka, India.Further, the hotspots of GHG emissions and significant factors influencing the CF of milk production were identified in these farms.

Study area and data collection
The study was conducted in the south-interior region of Karnataka, India.The state Karnataka forms the west-central part of peninsular India between North Latitudes 11°35′30" and 18°25′30" and East Longitudes 74°06′00" and 78°35′30."Metrologically, the state is divided into coastal, north-interior, and south-interior regions.The south-interior region receives an annual average rainfall of 1286 mm and experiences a semi-arid type of climate (KSNDMC, 2018).
The primary data (farm size, animal management, body measurements, milk production etc.) were collected from dairy farmers, and secondary data (emission factors for fertilizer, transport, energy etc.) were obtained from the reports published by government agencies and peer-reviewed literature.The field data were collected (June 2017 to December 2018) from 47 smallholder dairy farms located in the 8 districts (Bengaluru rural, n = 9; Tumkur, n = 4; Mysore, n = 5; Ko- The selection of villages within a district was done considering the information and suggestions received from the local veterinary doctor.Only those farms that sold milk to a dairy cooperative were selected.Within a district, depending on the accessibility and availability, the dairy farms located within 2 to 5 km distance were selected for the study.Primary data related to culled cows were also collected to assess their contributions to GHG emissions.The farms were representative of the mixed crop and dairy production system of the region.The main crop grown in the region was finger millet, and dairy farming was the secondary occupation.The produced milk was used for household consumption, as well as sold to local dairy cooperatives.The cooperatives transported the collected milk to district-level processing plants. A questionnaire was developed for survey and data collection related to feeding practices, performances of animals, manure disposal, feed and fodder production, and other associated farm management practices for the dairy cows.A preliminary version of the questionnaire was used for initial data collections from the 5 randomly selected smallholder dairy farms.Subsequently, the questionnaire was checked, reformed and finalized based on the initial study (Supplementary Table S1).It was used for final data collection from the selected 47 farms employing semi-structured interviews.The key person as well as other persons assisting in daily farm operations were questioned for gathering the information.The gathered information was further validated by direct observation at each farm.

Goal
The study employed a partial LCA approach to determine and allocate the whole farm GHG emission among different sources up to the dairy cooperative.The LCA conducted in the current study was based on a product approach.The goal was to assess the impact of milk production on the environment, wherein analyzing the CF of milk production in smallholder dairy production system was the focus.The economic functions considered for the assessment were the amount of milk produced and the role of cattle as finance and insurance.For better comparison among the farms with different breeds and or different feeding regimens, milk yield was corrected to a standardized level based on fat and protein contents in milk (IDF, 2015).Milk produced in a dairy farm was corrected to the milk yield with 3.5% fat and 3.2% protein using the formula given below (Opio et al., 2013).Fat-protein-corrected milk (FPCM, kg) = Milk yield (kg) × (0.337+0.116 × milk fat % +0.06 × milk protein %)

Scope and system boundary
The system boundary of the current study is depicted in Figure 1.It was up to the milk selling point at the dairy cooperative society located in each village.Upstream processes included the inputs (fertilizer, fuel and energy) for the production of feed produced onfarm or purchased from market.Core milk production processes included dairy farm operations and animal management.Downstream processes included the sale of calves, culling of cattle and transportation of liquid milk to milk cooperatives.The GHG emission related to milk processing, packaging, retail consumption and product waste that usually constitute10 to 20% of the total emissions (IDF, 2015) were not included in the current study.An attributional LCA approach was applied to assess GHG emissions under the studied production scenario.The production system was considered multifunctional, wherein milk, finance and insurance were handled as multiple products.In Karnataka, according to the prevention of cow slaughter and cattle preservation act, the sale of unproductive cattle is prohibited, and they are reared at local gaushalas (protective shelters for cows).Therefore, meat was not a co-product of the studied farming system.Primary data on feeding practices, body measurements, health management and other associated farm management practices pertaining to culled cows (n = 75) were collected from 2 gaushalas located in South Karnataka.Dairy farmers from near and far areas would leave their culled cattle in these shelters.The culled animals are maintained in these shelters until their natural death.The male calves were sold at a nominal price (₹3000 to 5000) within 2 mo of birth and were considered in the current assessment.Manure produced was used exclusively at own farm.The contribution of culled dairy cows to CF of milk production was also included in the system boundary.
The ISO 14044 (2006a) and ISO 14040 (2006b) provide guidelines for allocating the environmental impact of a process for multiple outputs from a production system.ISO recommends avoiding allocation by dividing the primary process into 2 or more sub-processes and collecting the input and output data related to these sub-processes, and expanding the product system to include the additional functions related to coproducts.Under the situations, where allocation cannot be avoided, GHG emissions can be allocated based on casual and physical relationships (Opio et al., 2013).In the latter case, the most commonly used approach is economic allocation, which allocates emissions to each product according to its share of the products' Mech et al.: Carbon footprint of milk production combined economic value (IDF, 2015).Other indexes, such as products' weight or protein content, can also be used (Cederberg and Stadig, 2003).In compliance with the aforesaid guidelines, in the current study, mass, economic and digestibility allocations were applied to the feed (crop by-products and residues) inputs (Supplementary Table S2 and S3).Mass allocation was calculated for the proportion of each feed after considering the harvesting loss.Economic allocation was computed based on the proportion of crop by-products and residues for the total economic value of the respective crop.The proportion of digestible dry matter (DM) of each feed ingredient was calculated for the computation of biological allocation.The digestibility value of each ingredient was recorded from Feedipedia (2018).Total GHG emissions were allocated among the sale of milk and animal (male calf) and other functions (finance and insurance) based on their economic value (Supplementary Table S2 and S3).Further, the CF was also calculated by assigning GHG emission to milk yield (FPCM) and live weight of the sold calves and culled cows as co-products of the system according to the physical allocation method (IDF, 2015).A small amount of the produced milk was consumed at household in all the farms, which was considered in the assessment and included in the total milk yield.

Inventory analysis
In the LCA model, animals were divided into lactating cows, dry cows, heifers, calves and culled cows.However, the calves below 3 mo of age were not included in the estimation of enteric CH 4 emission (O'Brien et al., 2012).The herd composition of each dairy farm was recorded.The animals' heart girth and body length were measured to calculate their body weight (BW) using the Shaeffer's formula given below.BW (kg) = ((heart girth (cm)/2.54) 2 × length of the body (cm)/2.54/300)× 0.4536 The average daily milk yield and the quantities of milk consumed at household and sold to cooperatives were recorded for each farm.Fat and protein contents in milk were obtained from the receipts of dairy cooperatives given to the farmers at the time of milk sale.
Information was also collected regarding the area under cultivation, the application rate of synthetic fertilizers, manure, crop yield, edible crop and its by-products, details of purchased feeds and transportation of feeds.
The farmers purchased concentrate feed manufactured by the "Karnataka Cooperative Milk Producers' Federation Limited" from the nearest market or cooperative society.Details regarding off-farm feed products such as purchased cattle feed, crop by-products, concentrates and oilseed meals, and information on their sources, mode of transportation and average annual price were obtained from the farmers and local markets.The quantity of feed used in each farm was estimated based on information provided by the farmers.
The GHGs related to dairy farms considered were CO 2 , CH 4 and N 2 O.However, soil carbon fluxes for onfarm and off-farm crops were not included in this study, and soil carbon sinks were assumed to be in equilibrium (IPCC, 2006;O'Brien et al., 2012).The IPCC (2019) Tier 2 approach was followed to quantify GHG emissions from dairy animals.Feed digestibility for a particular farm was calculated based on the collected data on feed ingredients included in animals' ration and intake of feed dry matter (DMI).The digestible energy of different feed ingredients was expressed as the percentage of gross energy (Feedipedia, 2018).Subsequently, the average daily feed intake was expressed in gross energy intake using the IPCC Tier 2 methodology.Finally, the emission of enteric CH 4 was computed using the default conversion factor (Ym = 6.50% of gross energy intake, Table 10A.1 New, IPCC 2019) for different categories of animals (Equation 10.21; IPCC, 2019).The retention and excretion of nitrogen (N) were calculated using the Equations 10.33 and 10.31A (New) of IPCC ( 2019).The retention of N for dairy cows was determined based on the calculated milk protein content (1.9 + 0.4 × fat%).The excretion rate of volatile solid (VS) was calculated based on gross energy intake, digestibility and ash content of feed, using the updated Equation 10.24 (IPCC, 2019).The maximum CH 4 producing capacity (B0) of manure considered was 0.13 for dairy cows and non-dairy cattle (Table 10.16 Updated; IPCC, 2019).The average CH 4 conversion factor (MCF) was considered 5.00% for the solid manure management system (Table10.17Updated; IPCC, 2019) considering the average ambient temperature ≥25.0°C in the studied region.The MCF was then multiplied by the excretion rate of VS and B0 for different categories of animals.The default value taken for the solid manure management system was 0.005 kg N 2 O-N kg N excreted −1 (Table 10.21; IPCC, 2019).To determine indirect N 2 O emission from manure management, the percentage of managed manure N for a specific livestock category that volatilized as NH 3 and NO x in solid manure management system was considered as Frac GasMS loss of 30.0% for lactating and 45.0% for other animals (Table 10.22 Updated; IPCC, 2019).The default value for Frac leach for N loss for leaching and run off was considered as 0.24 (Table 11.3;IPCC, 2019).The content of crude protein in feeds was calculated based on the data collected from Feedipedia (2018).The emission of N 2 O from manure was derived after deducting the excreted N and retained N for growth and milk production by the animals from the intake of total N through feeds.Subsequently, the total managed manure N available for application in soil (kg N yr −1 farm −1 ) was calculated.Default values (IPCC, 2019) were used for the fraction of N lost through volatilization (NH 3 and NO x ) and N loss by leaching.The total fraction of managed manure N loss for different categories of animal in solid manure management system was calculated by using the Equation 10.34a (IPCC, 2019).
The emission of GHG from transportation was calculated by using the emission factors for Indian vehicles as specified by Ramachandra and Shwetmala (2009).The distance traveled for transportation of concentrate feeds (farm and gaushala) and fodders (gaushala) varied between 0.50 to 25.0 km.The mode of transportation used for the purpose was either 2-or 3-wheeler (farm), or truck (gaushala).The quantity of concentrate feeds transported at each time varied (50 to 3000 kg) depending on the time interval between 2 subsequent transportations.The total GHG emission for milk production was calculated based on the emissions of enteric CH 4 , manure CH 4 , manure N 2 O, GHGs from feed production, processing and transportation, and energy usage for dairy farm operations.
The factors used for calculating GHG emission from energy usage were taken from a previously published report of the Government of India (GOI, 2015).

Estimation of carbon footprint
The GHG emission was expressed in terms of CO 2 equivalent (CO 2 -eq).The CO 2 -eq units were calculated based on global warming potentials for CO 2 , CH 4 (27.9) and N 2 O (273) from sixth IPCC assessment reports (IPCC, 2021).The calculation of CF was carried out employing a linear additive model programmed in Microsoft Excel (Supplementary Table S3).

Statistical analyses and generation of chord diagram
All data sets were subjected to the Kolmogorov-Smirnov and the Shapiro-Wilk tests to test the normality using SPSS version 16.0 (SPSS Inc., Chicago, IL, USA).Most of the data sets did not follow normal distribution, and therefore, the data are presented as median along with the 1st and 3rd quartile values wherever applicable.
The identified major variables from PCA were subjected to multiple linear regression analysis using SPSS version 16.0 (SPSS Inc., Chicago, IL, USA), and a regression equation was developed.The stepwise linear regression method (combination of forward and backward procedures) was used that regressed multiple variables and simultaneously removed those that were not important.The Variance Inflation Factor (VIF) was less than < 2, and thus there was no multicollinearity among the variables included in the regression equation.As the variables had large variations and not followed normal distribution, log (natural) transformation was employed for both the dependent and independent variables.The analysis following the log-transformation improved the adjusted R 2 value of the regression equation.The normal probability plot and a scatter plot of standardized residuals against the predicted values were drawn to check the key assumptions of linear regression analysis.
A chord diagram was generated to visualize the contributions of different categories of animals (lactating cows, dry cows, heifers, calves and culled cows) toward the major sources of GHG emission (CH 4 from enteric emission; GHG (CO 2 , CH 4 and N 2 O) from feed production; N 2 O from manure management; and CH 4 from manure management), using median values, and Circos table viewer available online (http: / / mkweb .bcgsc.ca/tableviewer/ docs/ ).

Farm characteristics and production system
The information on household and farm characteristics of the smallholder dairy farms are presented in the Table S4 (Supplementary).The dairy herds primarily consisted of Holstein-Friesian crossbred cows, and Jersey cows were also maintained in a few farms.The farms culled adult (8-10 years of age) unproductive cows, and the annual culling rate varied from 6.07 to 11.8% among the farms.The cows were artificially inseminated in all the farms.Average daily milk production was between 11.5 kg cow −1 and 18.5 kg farm −1 .A small quantity of the produced milk was consumed at home (0.50 kg d −1 ), and the remaining milk was sold.The average yield of FPCM and feed conversion efficiency (FCE) were 2763 kg yr −1 cow −1 and 0.78 (kg FPCM kg DM −1 ) respectively.
The details of feed offered, intake and digestibility of feed, and water usage are given in the Table S5 and S6 (Supplementary).The animals were stall-fed in all the farms.They were offered concentrates, crop byproducts, green forages, dry roughages and feed supplements, which varied among the farms.The average intake (kg d −1 animal −1 ) and digestibility (%) of feed DM in adult cows were 9.83 and 59.0, respectively.
Manure was collected every day and stored in open heaps near the farm for 3 to 5 mo.Subsequently, it was used by the farmers for growing crops and fodders at their own farms.The average manure and fertilizer application rates (kg N ha −1 yr −1 ) were 17.4 and 34.1, respectively.The practice of converting manure to dung cakes for fuel, a common practice in India, was not followed in the studied farms.

Emission of greenhouse gases
The emission of GHGs from different sources and hotspots are presented in Table 1.On the basis of mass, economic and digestibility allocations, the average total GHG emissions (kg CO 2 -eq yr −1 farm −1 ) attributable to milk production were 8936, 8641 and 8759, respectively.The major emission hotspots were CH 4 emission from enteric fermentation (66.8%) and GHG emission from feed production (23.0%), when no allocation was considered.The allocation ratios used for liquid milk production, sold calves and culled cows were 0.80, 0.02 and 0.01, respectively.The contributions of CH 4 , N 2 O and CO 2 to the total GHG emission of the farms were 70.6, 20.5 and 7.69%, respectively.
The emissions of GHGs (no allocation) attributed to different categories of animals (lactating, dry and culled cows, heifers and calves) are depicted in the chord diagram (Figure 2).The emission of CH 4 from enteric fermentation (369 to 2112 kg CO 2 -eq yr −1 ) and GHGs from feed production (91.4 to 943 kg CO 2 -eq yr −1 ) were the 2 major emission hotspots in all categories of animals.In contrast, GHG emission (N 2 O and CH 4 ) from manure management contributed the least (25.7 to 111 kg CO 2 -eq yr −1 ) in all the categories.Among the different animal categories, the highest total GHG emission was attributed to the lactating cows (42.9%), followed by dry cows (21.6%), heifers (16.1%), culled cows (12.2%) and calves (7.23%).

Carbon footprint of milk production
The CF of milk production (kg CO 2 -eq kg FPCM −1 ) in the farms varied depending on the method of allocation used for estimation (Table 2 and Figure 3).

Mech et al.: Carbon footprint of milk production
The CF of milk production was 1.81, when no allocation was considered.Based on the mass allocation that was estimated considering live weight and FPCM, and feed GHG emission, the CF values were 1.45 and 1.81, respectively.When GHG emission was allocated only to milk without any co-product, the CF was 1.76 and 1.74, respectively for the digestibility and economic allocations.Further, the CF based on economic allocation with co-products was 1.55.The CF of milk production (no allocation) was < 3.0 kg CO 2 -eq kg FPCM −1 in 80.9% of the farms (Figure 3).
The Pearson correlation analysis (Table 3) indicated a significant (P < 0.01) negative association between the CF and FPCM yields or FCE.In contrast, the CF was positively correlated with CH 4 emission from manure management (P < 0.01) and GHG emission from feed production (P < 0.05).A strong positive correlation between BW and DMI (r = 0.94; P < 0.01), and between FPCM yield and FPCM yield 100 kg BW −1 (r = 0.52; P < 0.01) was observed.Therefore, DMI and FPCM yield 100 kg BW −1 were included in the subsequent PCA, and BW and FPCM yield were considered as redundant factors.
The results of PCA indicated that the first 2 principal components (PC1 and PC2) explained 54.2% of the total variances (Supplementary Figure S1A).It was evident that out of the 10 factors included in the PCA, 5 factors (FPCM yield 100 kg BW −1 , DMI, CH 4 and N 2 O emissions from manure management and FCE) contributed significantly (>10.0%cut off level) to the PC1 and PC2 (Supplementary Figure S1B).The resultant biplot of PCA revealed a distinct partitioning of the low and high levels of CF of milk production across PC1 and PC2 (Figure 4).The DM digestibility was least associated with the PC1 and PC2.In contrast, DMI and CH 4 and N 2 O emissions from manure management exhibited a positive association with the CF of milk production.On the other hand, FPCM yield 100 kg BW −1 and FCE exhibited a negative association with the CF.
The identified 5 major variables from PCA were subjected to multiple linear regression analysis.Among the variables entered in the stepwise regression analysis, only 3 variables were found significant in explaining variation in the CF of milk production.The adjusted R 2 values indicated that FPCM yield 100 kg BW −1 alone explained 69.2% variation in the CF followed by 11.6% and 2.60% variations by DMI and CH 4 emission from manure management, respectively.Hence, a total of 83.4% of the observed variation in the CF was explained by these 3 variables (Supplementary Figure S2A).The normal probability plot showed that the points were closer to the straight line (diagonal line).The result indicated that the residuals followed normal distribution (Supplementary Figure S2B).The scatter plot of standardized residuals also showed that the points were scattered randomly around the 0 line (Supplementary Figure S2C).Any specific trend in both the plots could not be discerned that satisfied the assumptions of regression analysis (normality and homoscedasticity).The final regression model developed was log (CF) = 9.546 -0.860 log (FPCM yield 100 kg BW −1 ) -0.568 log (DMI) + 0.308 log (CH 4 emission from manure management) + e (Adjusted R 2 = 0.834; F-value = 76.11;P < 0.0001).

DISCUSSION
The current study aimed to estimate the CF of liquid milk production in the smallholder dairy farms in southinterior Karnataka, India.The environmental impacts of dairy production can vary substantially depending on farm characteristics and management practices (Poore and Nemecek, 2018).In developing countries, the livestock production system is a multi-functional venture and includes tangible (marketable) and intangible (non-marketable) functions (ISO, 2006a;ISO, 2006b).The smallholder dairy production system in the current study was multi-functional that resembled the small-scale dairy farms in other tropical regions around the world (Mazzetto et al., 2020;González-Quintero et al., 2021;de Léis et al., 2015, Apdini et al., 2021).Milk was the major marketable product that gave direct economic benefit to the farmers.The remaining economic functions of the production system were attributed to finance, insurance and sale of male calves.
A recent LCA study on dual purpose farms in Columbia reports significantly greater environmental burden to beef as compared with milk production (González-Quintero et al., 2021).Lesser CF of milk production in the specialized as compared with dual purpose farms is reported from Costa Rica based on an attributional LCA study that has considered milk production and sale of animals for the assessment (Mazzetto et al., 2020).Very low CF of energy corrected milk (ECM) at farm gate (0.53 to 0.78 kg CO 2 -eq kg ECM −1 ) is reported for different Southern Brazilian dairy production systems (de Léis et al., 2015).In the current assessment, the sale of meat and fertilizer was not co-products of the milk production system.Therefore, the maximum environmental burden was allocated only to milk.A previous report on LCA of smallholder dairy farms in Kenya indicates the inclusion of economic quantification of non-marketable functions such as farmers' motivation, cultural identity, social status and practice of cattle as dowry (Weiler et al., 2014).Nevertheless, inclusion of the non-marketable functions in LCA is challenging and debatable due to their abstract and intangible socio-cultural characteristics (Weiler et al., 2014;Garg et al., 2016).
Therefore, in the current assessment, non-marketable functions (cultural and social values of animal ownership) were not included for allocating GHG due to the associated complexity.On the other hand, the quantity of milk consumed at household was considered for calculating the total economic value of milk, which is usually not considered for the economic allocation approach in LCA (Behnke and Muthami, 2011).In the studied farms, manure was exclusively used for crop production, and its economic value was considered for computing economic allocation to co-products.Manure application has a positive impact on soil health, and it is crucial for the sustainable nutrient cycle in a mixed crop-livestock system (Weiler et al., 2014).Nevertheless, accounting economics of these positive effects in a multi-functional smallholder production system is highly complex and challenging (Garg et al., 2016).It is suggested that the economic sustainability of manurebased cropping systems and opportunities to improve their profitability must be explored (Rayne and Aula, 2020).
In the current study, the IPCC Tier 2 methodology was used to estimate GHG emission from enteric fermentation, manure management and agricultural soils (N 2 O emission from synthetic and organic fertilizer, ammonia re-deposition and nitrate leaching).Similar approach was used previously for estimating GHG emission from smallholder dairy farms in Gujarat, In-dia (Garg et al., 2016).The total GHG emission of the farms varied substantially based on different methods of allocation.The emission of CH 4 contributed most to the total GHG emission as compared with N 2 O or CO 2 , and CH 4 emission from enteric fermentation was a major contributor.The CH 4 emission factor for enteric fermentation and feed digestibility were 94.3 kg yr −1 and 59.0% respectively.A previous LCA study on Indian dairy cows reports lesser CH 4 emission factor (72.3 kg yr −1 ) for lactating cows that fed similar quality feed (59.2% digestibility) (Garg et al., 2016).Greater CH 4 emission factor in this study was likely due to more consumption of DM (9.83 kg d −1 ) by the animals than the previous report (9.04 kg d −1 ).The CH 4 emission from enteric fermentation depends on the nature of ruminant digestion, influenced by the quality of feed (Cabezas-Garcia et al., 2017).In the current study, nearly half of the DM fed to the animals was from crop residues and human inedible crop by-products, which likely contributed to greater enteric CH 4 emission as well.Further, it may be noted that the CH 4 emission factor in this study was calculated based on the new guidelines of IPCC ( 2019) that consider greater GWP (global warming potential) factor as compared with the previous guidelines.
The contribution of CH 4 or N 2 O emissions from manure management to the total farm GHG emission was negligible (<4.50%).In contrast, the contribution of GHG (CO 2 , CH 4 and N 2 O) from feed production to total farm GHG was 2-to 8-fold greater than the con-  tribution of CH 4 or N 2 O emissions from manure.The magnitude of GHG emission from manure management and feed production in the current study was comparable with that reported previously for Indian cattle (Garg et al., 2016).
Nonlactating animals were also kept along with the lactating cows in most of the farms, and calves were maintained in a limited number of farms.The contribution of culled cows to the CF of milk production was included in the current estimation.The average contributions of lactating cows, dry cows, heifers, culled cows and calves to the total GHG emission were 42.9, 21.6, 16.1, 12.2 and 7.23%, respectively.A previous report indicates that in India, the contributions of dairy and non-dairy cattle to total livestock GHG emission are 30.1 and 20.8%, respectively (Chhabra et al., 2013).The Indian farmers usually maintain nonlactating retired cows and oxen due to the religious and sociocultural values, and other functions such as manure, draft power, finance and insurance.However, in the studied farms, nonlactating animals were mostly the pregnant dry cows and replacement heifers.The emission of GHG of a farm is positively associated with the number of animals kept.Total farm GHG emission in the current study was lower than that previously reported for the Indian smallholder dairy farms (Garg et al., 2016), which was likely due to the smaller farm size in the current study.
The results of CF of cradle-to-dairy cooperative milk production in this study are comparable with the previous LCA studies of smallholder dairy farms reported from Kenya, Ethiopia, South Africa and India (Modupeore, 2011;Woldegebriel, 2013;Weiler et al., 2014;Garg et al., 2016;Wilkes et al., 2020).In contrast, as compared with the current study, nearly 6-fold greater milk CF from Peru with low milk productivity (Bartl et al., 2011) and 2-fold lesser milk CF from high producing large dairy farms of Netherlands, Ireland, Sweden and OECD countries (Thomassen et al., 2008;de Vries and de Boer, 2010;Henriksson et al., 2011;O'Brien et al., 2014) have been reported.The CF (kg CO 2 -eq kg FPCM −1 ) of milk production (no allocation) was less than 3.00 in 80.9% of the studied farms, and the average CF value of these farms was only 1.81.The CF of milk production in the current study was much lesser than that reported by Gerber et al., (2013) in the global studies for southern Asia (3.20 to 4.80).Lesser CF level in this study was likely attributed to the difference in sources of GHG emissions and less intensive management system with low energy usage.Further, the average feed digestibility considered by Gerber et al. (2013) for CF estimation was 52.6% as compared with 59.0% in the current study.Moreover, the annual average milk yield of 1000 kg cow −1 was considered by Gerber et al., (2013) for southern Asia for estimating CF, which was much lesser than that of the present study (2637 kg cow −1 ).These differences also explain the variation in CF levels between the present study and the previous report by FAO (Gerber et al., 2013).The CF values were 4.02 to 24.8% lesser, when estimated based on different allocations as compared with the CF that was estimated considering no allocation.This clearly indicates the role of other functions and co-products in sharing the environmental burden of milk production in a multi-functional smallholder dairy system (Garg et al., 2016).Evidently, the CF was increased by more than 2-fold for the farms with milk (FPCM) yield less than 3500 kg lactating cow −1 yr −1 as compared with the farms with greater milk yield.
The CF of milk production was negatively correlated with FPCM yield and FCE, but was positively correlated with the manure CH 4 and feed GHG emissions.Previous reports indicate that the CF of milk production is negatively associated with milk yield and feed digestibility (Garg et al., 2016), and FCE is a critical indicator affecting the CF (Henriksson et al., 2011).In dairy farms with varied management tactics, milk output per cow is the most influential factor on milk CF (Yan et al., 2013).It is established that with increasing milk yield, GHG emission per lactating animal increases, but the emission decreases per unit of milk output (Gerber et al., 2011;Tubiello et al., 2012).Based on the PCA and regression analysis, FPCM yield 100 kg BW −1 , DMI and CH 4 emission from manure management were identified as the strongest determinants of the CF.Therefore, appropriate managemental interventions targeting these parameters can reduce the environmental burden of the milk production system of smallholder dairy farms.

CONCLUSIONS AND IMPLICATIONS
In conclusion, the CF of cradle-to-dairy cooperative milk production varied based on the different estimation methods.The CF was reduced when multiple func- tions of the production system were considered as economic functions.The CF was increased by more than 2-fold when annual FPCM yield was less than 3500 kg lactating cow −1 .The FPCM yield 100 kg BW −1 , DMI and CH 4 emission from manure management were the strongest determinants of CF.The results indicate that the most effective interventions for reducing the CF in smallholder dairy farms are to maintain high-yielding animals and to adopt appropriate feeding strategies for better feed utilization.Therefore, the future breeding strategy should target milk production traits and feed utilization efficiency as critical phenotypes for improving the genetic merit of dairy cattle.This approach will help to reduce the CF of milk production.However, it may be noted that the current study was performed on a limited number of farms at a specific geographical location.Hence, the results are not true reflection of the national scenario.Therefore, large scale future studies considering a greater number of farms at different agro ecological zones are required to establish the CF of milk production of the smallholder dairy farms in India.Further, a scenario analysis based on the practically applicable GHG mitigation interventions is crucial to suggest paths for reducing the environmental burden of the Indian smallholder dairy production system.

Figure 1 .
Figure 1.System boundary for assessing the emission of greenhouse gases in the smallholder dairy farms in south-interior Karnataka, India.
Mech et al.: Carbon footprint of milk production

Figure 2 .
Figure 2. Chord diagram depicting the contributions of different categories of animals (lactating cows, dry cows, heifers, culled cows and calves; shown on the right side) toward the emissions of greenhouse gases (GHG) from different sources (Enteric CH 4 : CH 4 from enteric fermentation; Feed GHG: CO 2 , CH 4 and N 2 O from feed production; Manure N 2 O: N 2 O from manure management; Manure CH 4 : CH 4 from manure management; shown on the left side) in smallholder dairy farms (n = 47) in south-interior Karnataka, India.The inner scale indicates the numerical emission values (kg CO 2 -eq yr −1 ), and the outer scale indicates the contribution (%) of the chords within a segment (emission category).The width of the chords indicates the magnitude of contribution.CH 4 : methane; N 2 O: nitrous oxide; CO 2 : carbon dioxide.
Mech et al.: Carbon footprint of milk production

Table 2 .
Mech et al.:Carbon footprint of milk production Carbon footprint (CF) of milk production (kg CO 2 -eq kg FPCM −1 ) in smallholder dairy farms (n = 47) in south-interior Karnataka, India.The data are presented as median, and the values for 1st and 3rd quartiles (Q1, Q3) are given in parenthesis.CO 2 : carbon dioxide; FPCM: fat-protein-corrected milk; GHG: greenhouse gases