The system under consideration

Modern pig farming systems can be considered to have three distinct production phases: (1) gestation and farrowing where piglets are produced by breeding sows, (2) the nursery or weaning phase when pigs are separated from their mothers and (3) the grower/finisher (G/F) phase where pigs are fattened from about 30 kg to slaughter weight^{( 15 )}. Fig. 1 shows the major components of this system when considered in an LCA model: from the production of feed ingredients to animals shipped from the farm gate for slaughter. There were three main compartments of material flow considered in the LCA model: (1) the production of feed ingredients, (2) the consumption of feed, energy and other materials for on-farm pig production and (3) the storage and land application of manure. Benchmark data from 2012 on Canadian pig farms showed that 78 % of feed consumed per pig produced at least 75 % of the environmental impacts that occurred during the G/F phase⁽ Reference Mackenzie, Leinonen and Ferguson ^{5 )}. Attention therefore was given to formulating diets only for the G/F phase of production. Diets were formulated in two scenarios for pig production systems in Eastern and Western Canada because the main ingredients used in their typical diets are not the same. Pig diets in Eastern Canada are typically based on maize similar to US pig diets⁽ Reference Thoma, Nutter and Ulrich ^{16 )}, whereas pig diets in Western Canada use wheat and barley as the main cereal component/s^{( 17 )}.

Fig. 1 The structure and main components of the pig production systems as considered by the Life Cycle Assessment model. Feed production in the model included the manufacture of fertilisers and pesticides, etc. as inputs to growing crops.

The life cycle assessment model

The environmental impacts resulting from all diets formulated in this study were calculated using an LCA model of pig systems in Eastern and Western Canada⁽ Reference Mackenzie, Leinonen and Ferguson ^{5 )}. Some aspects of this model were also included as part of the diet formulation process (see Diet formulation rules). The details regarding the main components of the LCA model of Canadian pig farming systems are provided below. The system boundaries of the LCA were cradle to farm gate and the functional unit was 1 kg expected carcass weight (ECW)⁽ Reference Mackenzie, Leinonen and Ferguson ^{5 )}. The breeding and nursery production stages were treated as independent to the G/F phase in this study and remained constant for all comparisons made.

Feed production

The average environmental impacts per kg of ingredient for all ingredients used in the G/F diets can be found in Table 1. Important causes of environmental impact in the feed supply chain for pigs include the following: fossil fuel inputs for fertiliser production, emissions resulting from the spreading of fertilisers, fossil fuel use for field operations, energy inputs to processing (drying, grinding, etc.) and transport⁽ Reference Van Der Werf, Petit and Sanders ^{18 )}. When modelling a complex supply chain, as is the case for animal feed, the inputs to the process (wheat, water, energy, etc.) are shared between the different co-products resulting from these processes, and the environmental impacts associated with them must be allocated. Economic allocation was used as the methodology for co-product allocation throughout the feed supply chain, as advised in the FAO Livestock Environmental Assessment and Performance Partnership recommendations^{( 19 )}. The price ratios found in the online Supplementary Material S1 were used for the purposes of economic allocation. Life cycle inventory (LCI) data for the production of major crops were adapted from a previous LCA on Canadian crop production⁽ Reference Pelletier, Arsenault and Tyedmers ^{20 )}. LCI data for amino acids lysine, methionine, threonine and tryptophan were obtained from an LCA study on the impact of amino acids in pig diets⁽ Reference Garcia-Launay, van der Werf and Nguyen ^{21 )}. LCI data for the production of minerals dicalcium phosphate, salt and limestone came from the Ecoinvent databases^{( 22 )}. LCI data for maize dried distillers grains with solubles (DDGS) from Canadian sources were not available, and therefore were adapted from data representative of ethanol production in the USA^{( 22 )}, assuming the use of Canadian maize and typical electricity mix. The LCI for bakery meal was based on data provided by a large retailer of bakery meal (Sugarich Ltd, unpublished results) and adapted for a Canadian scenario⁽ Reference Mackenzie, Leinonen and Ferguson ^{23 )}. Surplus material from bread production is a large proportion of the material used for bakery meal sold for use in monogastric diets (Sugarich Ltd, unpublished results) and was used as a representative input to bakery meal in this study. The LCI for the production of 1 kg bread was adapted from the LCA food database⁽ Reference Nielsen, Nielsen and Weidema ^{24 )} with the input of Canadian wheat and energy sources. A price ratio of 10:1 was assumed for bread and surplus material, with an average 10 % of material collected as surplus from the bread supply chain either during the production process or discarded at the supermarket (Sugarich Ltd, unpublished results). Processing inputs for packaging removal, drying and grinding were estimated to be 20 kWh electricity and 62 kWh natural gas/tonne of material processed (Sugarich Ltd, unpublished results). LCI data for meat meal were adapted from a previous LCA study on rendering, and the yields by mass from rendering were assumed to be 57·7 % for fat and 42·3 % for meat meal⁽ Reference Ramirez, Humphries and Woodgate ^{25 )}. The price ratio of rendered fat: meat meal was assumed to be 1·22⁽ Reference Mackenzie, Leinonen and Ferguson ^{23 )}. The LCI data for wheat milling were adapted from Ecoinvent^{( 22 )} in order to represent Canadian energy inputs. Bread flour yields were estimated to be 73 % on average, with remaining material flows of 2 % wheat germ, 12·5 % wheat shorts and 12 % wheat bran⁽ Reference Blasi, Kuhl and Drouillard ^{26 )}. A price ratio of 1:0·11:0·22:0·44 was assumed for wheat flour–wheat germ–wheat shorts–wheat bran. This was based on the expectation that flour would provide about 90 % of the gross margin for a typical milling operation^{( 27 )} and Canadian price data for the co-products from wheat milling as animal feed⁽ Reference Mackenzie, Leinonen and Ferguson ^{23 )}.

Table 1 Average environmental impacts per kg for all feed ingredients included in the grower/finisher diets tested

eq, Equivalent; NRRU, non-renewable resource use; AP, acidification potential; EP, eutrophication potential; GWP, global warming potential; DDGS, dried distillers grains with solubles.

* Inventory data for these ingredients was compiled as part of a previous life cycle assessment studies of Canadian pig farming systems⁽ Reference Mackenzie, Leinonen and Ferguson ^{5 ,} Reference Mackenzie, Leinonen and Ferguson ^{23 )}.

† Calculated by combining the total normalised NRRU, AP, EP and GWP using the Institute of Environmental Sciences at Leiden University methodology^{( 39 )} with equal weighting.

Manure model

The manure model estimated the emissions of CH₄, NH₃, N₂O, N₂ and NO _x , which occurred during housing, storage and application as well as the leaching of NO₃ and PO₄. Indirect N₂O formation resulting from NH₃ and NO _x emissions and NO₃ leaching was also modelled in accordance with the Intergovernmental Panel for Climate Change (IPCC) principles^{( 28 )}. Manure was assumed to remain in the barn for up to 7 d; it was then transferred to outside storage (except in cases where storage was a pit beneath the barn). Manure was expected to be applied to land twice annually in spring and autumn. The model of NH₃ emissions for housing and storage was based on a previous model of NH₃ emissions from pig production in Canada⁽ Reference Sheppard, Bittman and Swift ^{29 )}. A tier 2 IPCC methodology was adopted for emissions of CH₄, N₂O, NO _x and NO₃, but adapted to reflect small N losses from housing. As average ambient temperatures were considered to be <0°C during winter^{( 30 )}, emissions during this period were considered negligible for outside storage methods. The proportional mix of floor types in pig housing, storage and application techniques in each region was based on information from the Livestock Farm Practice Survey⁽ Reference Sheppard, Bittman and Swift ^{29 )}, as well as Statistics Canada records regarding the storage and application of swine manure⁽ Reference Beaulieu ^{31 , 32 )}. All N, P and K excreted via faeces or urine were assumed to be applied to land as fertiliser, once losses during housing and storage were accounted for. Manure applied to land was assumed to replace the need to apply equivalent synthetic fertilisers at a rate of 0·75, 0·97 and 1 for N, P and K, respectively⁽ Reference Nguyen, Hermansen and Mogensen ^{33 )}. The proportional mixture of the types of synthetic fertilisers replaced by the NPK content of the manure in each region was derived from sales figures for Eastern and Western Canada to assume a regional average fertiliser mix⁽ Reference Korol ^{34 )}.

Farm performance

With the exception of feed intake during the G/F stage and carcass yield, the baseline herd performance characteristics (litter size, mortality, etc.) used in this study were as those modelled for pig systems in Eastern and Western Canada in a previous regional LCA study⁽ Reference Mackenzie, Leinonen and Ferguson ^{5 )}. All characteristics of herd performance other than average feed intake and carcass yield were assumed to be independent of feed composition in the G/F production stage⁽ Reference Mackenzie, Leinonen and Ferguson ^{23 )}. Although this represents a simplification made for the purposes of a modelling exercise, it is valid for the scenarios modelled in this study. All diets formulated were nutritionally balanced and were not be expected to have implications on herd health status or mortality during the G/F phase. It is reasonable to expect that other model inputs such as on-farm energy use are independent of feed composition. The on-farm energy consumption data were adapted from a detailed study of energy consumption in conventional pig housing systems in Iowa⁽ Reference Lammers, Honeyman and Harmon ^{35 )}, as there were no equivalent data for Canadian systems available. To reflect longer and colder Canadian winters in comparison with Mason City, Iowa (which was used in the Lammers et al. ⁽ Reference Lammers, Honeyman and Harmon ^{35 )} calculations), larger loads of liquid petroleum gas (LPG) for heating were assumed to be required to maintain adequate barn temperatures. On the basis of average temperature data for Mason City^{( 36 )}, and regional data for Eastern and Western Canada^{( 30 )}, the LPG inputs for heating barns in Eastern Canada were estimated to be 25 % higher compared with the Iowa case study. LPG input for heating in Western Canada was assumed to be 25 % larger than that for Eastern Canada. These represent approximations as a previous sensitivity analysis showed that the model was not very sensitive to the assumptions made regarding LPG use for any of the impact categories tested here⁽ Reference Mackenzie, Leinonen and Ferguson ^{5 )}. The mix of electricity generation in the LCA was the national mix for the Canadian grid^{( 37 )}; this was assumed for all Canadian unit processes in the LCA.

Quantifying environmental impacts

The environmental impacts of the system were quantified by the LCA using four environmental impact categories. Three of these categories quantified negative impacts resulting from emissions caused by the system – AP, EP and GWP. We included GWP as it has received the most attention in efforts to quantify the impact of livestock systems. The impact categories AP and EP were considered as they quantify the main environmental impacts, which result from the storage and spreading of animal manure. The fourth impact category quantified the systems use of NRRU and was included because of the relatively high usage of cereals and oil seed meals in pig diets, which have a significant input of resources such as fertilisers⁽ Reference Steinfeld, Gerber and Wassenaar ^{3 )}.

System GWP was quantified in CO₂ equivalents (eq) on a 100-year timescale using the IPPC methodology^{( 28 )}. The methodology of accounting for GWP caused by land use change in this study followed the PAS 2050 guidelines^{( 38 )}. The methodologies for calculating AP (SO₂ eq), EP (PO₄ eq) and NRRU (Sb eq) were established by researchers at the Institute of Environmental Sciences at Leiden University (CML)^{( 39 )}. This methodology was chosen as it is designed to quantify these impact categories on a global scale, importantly accounting for the long-term impacts of airborne emissions on global levels of substances, which contribute to AP and EP. The CML methodology for normalising different types of environmental impact⁽ Reference Huijbregts, Breedveld and Huppes ^{40 )} was also utilised to formulate diets to minimise the combined environmental impact score of the system. The impacts that result from a process are normalised against a reference, which is an estimate of the total annual level of global emissions and resource use caused by human activity⁽ Reference Huijbregts, Breedveld and Huppes ^{40 )}. The normalised scores for AP, EP, GWP and NRRU were then combined additively, with equal weighting to generate a combined environmental impact score in the diet formulation tool. Equal weighting was adopted in this example to ensure that large increases in an individual environmental impact category did not occur when optimising to minimise the combined environmental impact score. The cradle-to-grave environmental impact calculations were performed using the software package SimaPro 7.3.3.

Diet formulation rules

A diet formulation tool was developed that predicted the environmental impacts for each category resulting from G/F diets for the feed supply chain and manure management. The tool also quantified the feed cost per kg LW gain for each solution. The tool formulated diets using linear programming in Microsoft Excel^® with the software plug-in open solver⁽ Reference Mason ^{41 )}. Nutritional values for all ingredients in the diets were primarily taken from the Stein Monogastric Nutrition Laboratory ingredient matrix^{( 42 )}. In cases where certain values were missing (or ingredients themselves were missing from the matrix), values from the NRC 2012 feed ingredient tables^{( 43 )} and the Premier Nutrition Atlas^{( 44 )} were used. All the G/F diets were formulated with four feeding phases (starter, grower, finisher and late finisher); this reflected typical feeding programmes adopted by commercial pig operations in Canada⁽ Reference Mackenzie, Leinonen and Ferguson ^{5 )}.

The predicted start weight of the pigs in the diet formulation tool was fixed at 27·4 kg with a finish weight of 124 kg for the G/F phase, based on benchmark data collected for a previous LCA study of Canadian pig farming⁽ Reference Mackenzie, Leinonen and Ferguson ^{5 )}. Diets were not formulated for a fixed nutritional density, but rather this was an outcome of the solution for a specific objective. The average feed intake per pig for each diet within a feeding phase was predicted based on meeting the animal’s requirements for growth. The net energy (NE) requirement for each feeding phase was defined in compliance with the NRC 2012 animal requirement tables^{( 45 )}. Minimum nutrient levels in g/MJ of NE were then defined for each feeding phase, so that the digestible protein and macronutrient contents of the feed would not be limiting for animal growth^{( 45 )}. It was thus assumed that feed intake was driven by the animal’s need to meet its daily energy requirements; as such feed intake increased when diets of reduced energy density were fed⁽ Reference Kyriazakis and Emmans ^{46 ,} Reference Patience ^{47 )}. The average predicted NE intake was constant for all diets. As all diets were nutritionally balanced, the animals were expected to spend the same average number of days in the barn over the course of the G/F phase. When diets were formulated at reduced energy density, daily feed intake was expected to compensate for this. Any effects the increased daily intake may have had on gut fill were taken into account⁽ Reference Mackenzie, Leinonen and Ferguson ^{23 )}.

Average ingredient prices and availability in Ontario and Manitoba for 2015 were provided by Trouw Nutrition (derived from Statistics Canada data^{( 48 )} – see online Supplementary Material S2 for the list of available ingredients and price ratios in each region). These were used to represent typical diet formulation scenarios for Eastern and Western Canada. Ontario and Manitoba produced about 24 and 23 % of the total pigs marketed in Canada in 2011, respectively⁽ Reference Brisson ^{49 )}. Importantly, maize was not considered as an available ingredient in the Western diets as is typical in many scenarios in this region; similarly, barley was not considered as an available ingredient in the Eastern diets (A Pharazyn, unpublished results).

The average gain:feed ratio over the G/F phase in the benchmark data for Canadian pigs was 0·365⁽ Reference Mackenzie, Leinonen and Ferguson ^{5 )} with feed intake of 264 kg/pig based on the mean start and finish weights. This was used as a starting point for the assumptions on average feed intake in this study. A dietary specification was defined, which represented an industry standard to ensure that the feed:LW gain ratio was minimised within reasonable commercial constraints. The specifications for this ‘typical’ diet are given in Table 2, and it was assumed that this diet ensured an average gain:feed ratio of 0·365. Lower limits were defined for the nutritional density of the diets for each feeding phase. These were set at 95 % of the energy content of the typical industry diet in the first two feeding phases and at 92·5 % for the latter two feeding phases. These restrictions were to ensure that feed intake would not be restricted by gut fill, which can be caused by diets of lower nutrient density, which contain a larger proportion of bulky feed⁽ Reference Kyriazakis and Emmans ^{46 )}. These minimum specifications of the G/F diet for each phase can also be found in Table 2. For each ingredient, a maximum inclusion rate was defined for each feeding phase in order to account for any anti-nutritional properties or other negative impacts on animal performance due to variability. These limits were based on guidance for pig farmers provided by the Ontario Ministry of Agriculture, Food and Rural Affairs^{( 50 )} as well as peer-reviewed studies in the case of some important co-products⁽ Reference Mackenzie, Leinonen and Ferguson ^{23 )} (see online Supplementary Material S3 for further details on ingredient inclusion limits).

Table 2 The nutritional specifications of the ‘typical’ grower/finisher diet for Canadian pig systemsFootnote *

* The lower limits permitted in the diet formulation rules used in this study are also shown.

The retention of N in finished pigs was calculated using the principles of Wellock et al. ⁽ Reference Wellock, Emmans and Kyriazakis ^{51 )} and was assumed to be 0·0256 (sd 0·00128) body weight (BW). Retentions of P and K were calculated using an isometric relationship of body composition with BW⁽ Reference Symeou, Leinonen and Kyriazakis ^{52 )} and were assumed to be approximately 0·005 (sd 0·00025) and 0·002 (sd 0·0001) BW, respectively. For K, this assumption represents a linear approximation around slaughter weight of a curvilinear relationship⁽ Reference Rigolot, Espagnol and Pomar ^{53 )}. All N, P and K not retained by the finished pigs were assumed to be excreted via faeces or urine. The predicted levels of nutrient excretion were required as inputs to the manure model.

Diets formulated

The process followed to formulate G/F diets for environmental impact objectives is shown as part of Fig. 2. The average values of NRRU, AP, EP and GWP/kg of each ingredient as seen in Table 1 were added to the list of ingredient properties in the diet formulation tool. In addition to this, equations that predicted the average environmental impact per kg of N, P and K excretion assuming an average mix of manure management practices were extracted from the manure sub-model of the LCA⁽ Reference Mackenzie, Leinonen and Ferguson ^{5 )}. This enabled the tool to account for the environmental impact resulting from predicted levels of nutrient excretion when formulating the diets. Thus, for any diet formulated, the average values of NRRU, AP, EP and GWP resulting from the feed supply chain and manure storage and application were predicted.

Fig. 2 Schematic of the methodology followed in this study to formulate diets for environmental impact objectives.

The tool was used to formulate G/F diets for both economic and environmental impact objectives. Two diets were formulated for economic objectives: (1) to minimise feed cost per kg LW gain (least cost) and (2) to minimise feed cost per kg LW gain with a requirement to maintain a certain level of feed efficiency (least-cost EFF). The NE content of the latter diet was fixed, so that the feed:LW gain ratio was minimised within reasonable commercial constraints. The minimum specifications of this diet were the ‘industry standard’ energy and nutrient levels shown in Table 2. This is a common commercial scenario, whereby diets are formulated for least cost without compromising feed efficiency⁽ Reference Ferguson ^{1 )}. This diet was included to quantify whether this strategy has any benefit for the environmental impact of the system compared with considering feed cost alone.

Four diets were formulated to minimise the individual environmental impact categories NRRU, AP, EP and GWP. A further diet was formulated to minimise the combined environmental impact (least EI) of the G/F phase, as measured using the combined normalised levels of NRRU, AP, EP and GWP under the CML methodology^{( 39 )} with equal weighting. All diets formulated for environmental impact objectives were restricted to a 30 % maximum cost increase in comparison with the least cost diet. Diets were formulated for these objectives in both regional scenarios for ingredient prices and ingredient availability for Eastern and Western Canada. Diets were formulated using linear programming. The resulting diets were optimal solutions based on the mean nutritional and environmental impact properties of the ingredients, as well as the mean impact levels associated with nutrient excretion calculated by the LCA.

Dietary comparisons in the life cycle assessment model

Accounting for the uncertainty in LCA is important to produce credible and reliable results⁽ Reference Lloyd and Ries ^{54 )}. In this study, an uncertainty analysis was used for statistical comparison of the diet formulations. The cradle to farm gate LCA model was hosted in the specialist software SimaPro. All input parameters had a mean, associated distribution (e.g. normal, lognormal, etc.) and standard deviation. The uncertainty in the environmental impact calculations was quantified using Monte Carlo simulations⁽ Reference Mackenzie, Leinonen and Ferguson ^{5 )}. Variability in all characteristics of herd performance other than feed intake was assumed to be independent of feed composition in the G/F production stage. Feed intake for each simulation was a function of the energy density of the diet in relation to the average energy requirement of the herd over the G/F production stage. This requirement had a distribution to represent variation in feed intake due to genetic and environmental factors, which were assumed to be independent of the feed composition.

As shown in Fig. 2, each diet was tested in the cradle to farm gate LCA of pig farming systems in Eastern and Western Canada. In each case, 1000 simulations of the model were run in order to calculate the NRRU, AP, EP and GWP of the system when adopting these diets. This number of simulations ensured that the SEM of the results for each impact category were low enough for good repeatability⁽ Reference Mackenzie, Leinonen and Ferguson ^{5 )}. Parallel Monte Carlo simulations were used to compare all other diets with the least-cost diet. The parallel simulations enabled the model to determine whether the diets had resulted in any significant changes to the environmental impact levels of the system compared with the least-cost scenario. This method of uncertainty analysis to distinguish between two scenarios in an LCA model was described in detail by Mackenzie et al. ⁽ Reference Mackenzie, Leinonen and Ferguson ^{5 )}. In brief, uncertainties were categorised as either specific to the system (α) or shared between the systems being compared (β)⁽ Reference Leinonen, Williams and Wiseman ^{55 ,} Reference Leinonen, Williams and Waller ^{56 )}. For each simulation, a value for each parameter was randomly selected from the specified distribution input for this variable. Where parameters are shared between two scenarios being tested (e.g. maize yield (kg/ha) when feeding two different diets containing maize), for each individual comparison the same point on the distribution is selected. In this case, variations in all parameters except the G/F diet composition, the resulting feed intake and nutrient excretion during the G/F phase and carcass yield were considered as shared uncertainty in the comparisons. Although the average energy requirement was variable to account for differences caused by animal and environmental factors, in each comparison, the NE intake was the same for both diets. The key output of the simulations was the frequency in which the environmental impact of one scenario was greater or smaller than the second scenario for each impact category tested. Environmental impact levels were reported as significantly different in cases where P<0·05 over 1000 parallel simulations of the LCA model. This allowed the model to account for shared uncertainty between two systems (in this case diets) modelled in the LCA, but provide a useful answer as to which is likely to have greater environmental impact.