Role of Corn Silage in the Sustainability of Dairy Buffalo Systems and New Perspective of Allocation Criterion

This paper aims to compare the cradle-to-farm gate sustainability of two dairy buffalo systems, according to life cycle assessment guidelines (LCA). Primary data were obtained by five intensive farms with feeding plans based on non-corn silage (NCS) and five with corn silage (CS) based rations. Both systems were characterized by the presence of two farms with wheat grain yields, sold for human consumption. All the farms were in Southern Italy and seven were included in the Protected Designation of Origin (PDO) area of “Mozzarella di bufala campana”. The functional unit (FU) adopted was 1 kg of normalized buffalo milk (NBM); impact categories investigated were: global warming potential (GWP), acidification potential (AP), eutrophication potential (EP), agricultural land occupation (ALO), water depletion (WD). Two different economic allocation procedures were tested: a first step aimed to mitigate the environmental impacts sharing among wheat grain, where present, and milk. The second stage involved culled buffalo cows. Neither the allocation nor the combination of allocation and feeding system showed significant effects (p > 0.05). Corn silage-based system (CS) showed lower impacts than non-corn silage (NCS) one for AP and EP (p = 0.002 and p = 0.051 respectively). High average dry matter yield per hectare of corn silage probably had a positive effect on SO2 and PO4 equivalents.


Introduction
In Italy, water buffalo (Bubalus bubalis, Mediterranean Buffalo breed) dairy farming is a traditional activity commonly aimed to produce mozzarella cheese [1]. Mozzarella is a typical PDO product that is exported in other European and extra-European countries. Growing exports (mainly to the United States of America, Great Britain and other European Countries) have led to an increase in the production of buffalo mozzarella cheese (more than 25% from 2013 to 2017). Recently in Germany and Great Britain, a noticeable introduction of dairy buffalo farms confirms the interest in this product [2,3]. Berlese et al. [2] showed as in Italian areas, out of the PDO district zone, some farms started transforming buffalo milk in other dairy products, such as ripened cheeses, ricotta, and stracchino cheese. So, the supply chain is introducing new dairy products from buffalo milk, both for satisfying wider market requirements and for avoiding risks related to a system based almost exclusively on a unique product. Moreover, several new dairy products are sold during the entire year, reducing the effect of the seasonal market demand variability that affects mozzarella [4].
The highest number of dairy buffaloes are reared under intensive conditions in the Protected Designation of Origin (PDO) area, comprising the Campania, Lazio, Puglia, and Molise regions with the three main livestock categories (i.e., lactating cows, dry cows, and heifers) feed with different diets [5]. Despite these authors indicated corn silage-based feeding systems as widespread, nevertheless this fodder is not available in the whole PDO area. strategy allowed the comparison between with and without wheat farms in the buffalo rearing scenario. In agreement with several authors, allocation involved culled cows in cattle [18][19][20] and in buffaloes [6,8] but, as suggested by Romano et al. [9], at a second step, we used the principle to identify the amounts of undesirable products, highlighted by environmental mitigation.

Materials and Methods
LCA is the assessment of inputs, outputs, and environmental impacts in agreement with ISO 14,040 and ISO 14,044 methodology [21,22]. Four distinct phases characterize LCA approach: (1) Definition of the goal and scope (including functional unit and limits of the system); (2) life cycle inventory (LCI) analysis (including input and output data collection for all processes); (3) life cycle impact assessment (LCIA); (4) life cycle interpretation.

Functional Unit (FU)
The functional unit (FU) chosen was 1 kg of normalized buffalo milk (NBM), with a reference about milk fat and protein content of 8.3 and 4.73%, respectively. In accordance with Pirlo et al. [6] the NBM acronym was adopted; moreover raw milk was transformed into NBM with the following equation provided by Di Palo [23]:

System Boundary Definition
The system boundaries were set as a cradle-to-farm gate analysis ( Figure 1). This step involved as primary data all on-farm activities (i.e., agricultural management, crop harvesting, feeding, animal care, milking routine) and sources pertaining to fodder production (i.e., seeds or diesel), then processed with the software SimaPro 8.03. The assessment involved also off-farm feeds, fossil fuels, bedding materials. Consumption of energy and off-farm emissions were retrieved from databases of SimaPro 8.03. silage (CS) based feeding system. Inputs and outputs provided by male calves were excluded, and consequently, their allocation. Allocation of cereals (Triticum durum Desf.) byproducts [7,8,17] involved two farms in NCS and CS system respectively. This strategy allowed the comparison between with and without wheat farms in the buffalo rearing scenario. In agreement with several authors, allocation involved culled cows in cattle [18][19][20] and in buffaloes [6,8] but, as suggested by Romano et al. [9], at a second step, we used the principle to identify the amounts of undesirable products, highlighted by environmental mitigation.

Materials and Methods
LCA is the assessment of inputs, outputs, and environmental impacts in agreement with ISO 14,040 and ISO 14,044 methodology [21,22]. Four distinct phases characterize LCA approach: (1) Definition of the goal and scope (including functional unit and limits of the system); (2) life cycle inventory (LCI) analysis (including input and output data collection for all processes); (3) life cycle impact assessment (LCIA); (4) life cycle interpretation.

Functional Unit (FU)
The functional unit (FU) chosen was 1 kg of normalized buffalo milk (NBM), with a reference about milk fat and protein content of 8.3 and 4.73%, respectively. In accordance with Pirlo et al. [6] the NBM acronym was adopted; moreover raw milk was transformed into NBM with the following equation provided by Di Palo [23]:

System Boundary Definition
The system boundaries were set as a cradle-to-farm gate analysis ( Figure 1). This step involved as primary data all on-farm activities (i.e., agricultural management, crop harvesting, feeding, animal care, milking routine) and sources pertaining to fodder production (i.e., seeds or diesel), then processed with the software SimaPro 8.03. The assessment involved also off-farm feeds, fossil fuels, bedding materials. Consumption of energy and off-farm emissions were retrieved from databases of SimaPro 8.03. Figure 1. Cradle-to-farm-gate system boundaries of the production of 1 kg of normalized buffalo milk (NBM). The milk, functional unit (FU) is carried out from non-corn silage (NCS) and corn silage (CS) based systems. Corn silage must be included as on-farm feed in the CS system, wheat crop in two farms for each system. The solid line represents the system boundaries of the study.

Inventory Analysis
Primary data were obtained from ten farms (Table 1): five farms characterized by a corn silage-based feeding system (CS) and five farms with non-corn silage-based rations (NCS). All the farms, located in the Puglia and Basilicata Regions (Figure 2), were specialized in buffalo dairy farming, with animals (Italian Mediterranean Buffalo breed) kept in barns and without pasture. As recorded in the interviews, sexed semen (aimed to increase genetic selective pressure), was adopted mainly for the heifers. Farm characteristics and productive patterns, obtained by interviews, herd book, and test day record reports, are described in Tables 1  and 2, respectively. These data have been recorded in the year 2021, referred to January-December 2020. As indicated by several authors [7][8][9]17], dairy systems often produce crop commodities (cereals), thus both groups (NCS, CS) were balanced with a 3:2 ratio = no:yes wheat crop production. In farms producing wheat, the inputs useful to wheat production, such as synthetic fertilizers and fuels were considered. We set the feeding plan as corn silage-based if animals received at least 3 kg dry matter/day of corn silage in lactating cows ration. The farms included in the CS group administrated to lactating buffalo cows a range from 3 to 7 kg dry matter/day of corn silage. Only the farms of the Foggia Province (Puglia region) are located in the Protected Designation of Origin (PDO) area. According to current knowledge, Mediterranean Italian Buffalo breed animals show a noticeable heterogeneity in size, and this characteristic would affect management, feeding, and productions, also conditioning dry matter intake of feed and milk yields [8].
Inventory data covering livestock production, crop cultivations, straw for litter, inputs of purchased feed, electricity, diesel consumption and farm extension (divided in different crop per Ha year −1 ) were processed as primary data. The consumption and emission factor of natural gases adopted in the farms (water heating) were excluded due to the low impact and because it was not possible to have suitable information on it. Feed consumption, as well as water use and origin, was retrieved by interviews with the farmers, on the contrary water consumption was estimated [5,8,9]. Water consumption was distinguished in specific categories, provided by the software Simapro 8.03: the water needed for extra European crops, the Italian well water mainly adopted for maize, the water As recorded in the interviews, sexed semen (aimed to increase genetic selective pressure), was adopted mainly for the heifers. Farm characteristics and productive patterns, obtained by interviews, herd book, and test day record reports, are described in Tables 1 and 2, respectively. These data have been recorded in the year 2021, referred to January-December 2020. As indicated by several authors [7][8][9]17], dairy systems often produce crop commodities (cereals), thus both groups (NCS, CS) were balanced with a 3:2 ratio = no:yes wheat crop production. In farms producing wheat, the inputs useful to wheat production, such as synthetic fertilizers and fuels were considered. We set the feeding plan as corn silage-based if animals received at least 3 kg dry matter/day of corn silage in lactating cows ration.
The farms included in the CS group administrated to lactating buffalo cows a range from 3 to 7 kg dry matter/day of corn silage. Only the farms of the Foggia Province (Puglia region) are located in the Protected Designation of Origin (PDO) area. According to current knowledge, Mediterranean Italian Buffalo breed animals show a noticeable heterogeneity in size, and this characteristic would affect management, feeding, and productions, also conditioning dry matter intake of feed and milk yields [8].
Inventory data covering livestock production, crop cultivations, straw for litter, inputs of purchased feed, electricity, diesel consumption and farm extension (divided in different crop per Ha year −1 ) were processed as primary data. The consumption and emission factor of natural gases adopted in the farms (water heating) were excluded due to the low impact and because it was not possible to have suitable information on it. Feed consumption, as well as water use and origin, was retrieved by interviews with the farmers, on the contrary water consumption was estimated [5,8,9]. Water consumption was distinguished in specific categories, provided by the software Simapro 8.03: the water needed for extra European crops, the Italian well water mainly adopted for maize, the water adopted for the industrial processes (for example, wheat flour shorts), and the one used to clean the milking facilities.  Inventory analysis involved buildings, facilities as milking parlor, shed, and concrete areas. On the contrary, buildings and bunkers (i.e., horizontal silos) to stock fodders were excluded. In agreement with Romano et al. [8,9], a productive life of 50 years was assumed for the constructions, as suggested by the Ecoinvent 3 allocation database. A four-wheel drive tractor (Transport, tractor and trailer, agricultural {GLO}, market for, Alloc, Def, S) was presumed as adopted to transport on-farm fodders, whereas the purchased feed, carried by truck (Transport, freight lorry >32 metric ton, EURO 5, RER), was computed based on the distance from the farm, as suggested by Bragaglio et al. [24]. The sunflower meal feed, in most cases, is imported [25], thus a 2700-km travel distance by truck from Ukraine was assessed. According to other authors [26], a 10,000 km journey by cargo (Transport, freight, sea, transoceanic ship {GLO}, market for, Alloc, Def, S) for soybean and cotton seeds from South America and a 12,000 km journey for palm oil from Malaysia were assumed; a travel with truck from the Italian harbor to farm gate was added to this computation [24]. For fossil fuel provision, we also considered transport by truck: transport freight lorry of 3.5-7.5 metric ton, EURO 5, RER. Given the significance of feeding in the LCA criterion, a description of the formulation and composition of the diets is shown in Tables 3 and 4, for cows buffalo (lactating and dry, LC and DC), heifers (HF), and young animals (91-365 days old, YA). Diets of the calves up to the 90th day are not indicated in the table, they were comparable in the ten farms as these animals were fed with milk replacers, calf starter, and small amounts of hay. All the rations were managed with the software SimaPro 8.03 and processed.  -480  400  ----200  75  100  300  ----1800  400  400  240  Sunflower meal  480  -440  360  ----1100  400 600 Wheat flour shorts  280  -480  400  ----100  40  60 400 Chemical composition (%)

Allocation Criterion
Allocation is a criterion, commonly used in LCA, to assess the weight of the coproducts, mainly adopted to avoid the expansion of system boundaries. Allocation strategy is also aimed to manage a functional unit (FU) only. In the framework of multifunctional processes, dairy systems are largely investigated, and known for providing meat from culled animals and male calves, in addition to milk [8,9,19,20]; other authors highlighted as significant by-product the cereals and suggested, as example case, the dairy buffalo system [6,9].
In agreement with the criteria suggested by ISO 14040 [21] for partitioning the input and output flows, milk and meat cannot be produced separately, thus dairy systems cannot be split. Ardente and Cellura [27] argued that, despite its limitations, economic allocation has certain qualities that make it flexible and potentially suitable for different contexts. The authors suggested the following equation: where: P i is the partitioning factor of the ith coproduct, n i is the quantity of the ith product, and x i is the price of the ith coproduct.
The option of economic allocation should be considered, for example, whenever the prices of coproducts and coservices differ widely. Therefore, in this study, as happened in others [8,28,29], economic allocation criterion was applied, as more preferable than mass and energetic allocation, as suggested by Pirlo et al. [6].
In the current research, a doble role was ensured by the economic allocation: at the first step the farms without wheat crop were not affected by allocation, while this criterion is adopted to subtract the pollutants in those wheat crop-based (NCS and CS systems both). After that, all on-farm pollutants are related to livestock except male calves, sold on the 15th day of life; thus, the inputs and outputs provided by these animals were not considered. At the second step, in agreement with Romano et al. [9], allocation criterion was adopted to assess the rate of culling cows, an undesirable product. The model of Equation (2) described above, was adopted, even though improved in accordance with the specifics suggested by Mahath et al. [30].

Emissions
All the emissions were estimated, considering differences due to the corn silage (noncorn silage-NCS and corn silage-CS system) crop and different diets (Tables 3 and 4). The fuel combustion, electricity consumption, enteric emissions, crop residue emissions, manure management, and the emissions due to chemical fertilization were considered. In agreement with Holly et al., and IPCC [31,32], CO 2 emissions from livestock respiration and manure were not accounted.
It was assumed that they were balanced by the carbon previously absorbed and metabolized by crops composing the dairy diet, thus, being part of the carbon cycle, they do not constitute an additional source of CO 2 . Table 5 summarizes the characteristics of equations adopted to estimate total on-farm emissions (from animals, crops, energy sources), mainly in accordance with Refinement of the Intergovernmental Panel on Climate Change [32,33]. .21 was adopted to assess enteric emissions on the basis of gross energy (GE) intake. According to the updated guidelines for the specific emission factors (EFs), with the 10.21 formula, MJ head −1 provided by different ingredients, expressed as dry matter where processed (Tier 2 method). In the assessment was also considered the mean period, expressed in days, such as for other similar estimates, of 90, 270, and 365 days, for dry, lactating cows, and heifers respectively.
The data on the GE supplied by different feed sources and concentrates (e.g., hay, straw and soybean meal) were provided and managed on the basis of previous studies [5,6,14]. The different percentages of feed, indicated in Tables 3 and 4, were also considered for the assessment of the GE.

Methane Emissions from Manure Management
Although enteric emissions showing higher methane amount than manure ones [6,14], in agreement with Tier 2 method [32], their accurate assessment was done. As indicated for enteric emissions, the period of productive life was distinguished between different livestock categories. Two strategic inputs that affect the calculation of emissions, (i) MCF (S,k) and (ii) AWMS (T,S,k) are indicated in Table 5, as characterizing with the same default values, all the farms. Volatile solid excretion per day (VS), also involved in the estimation of these emissions, is reported in the same Table 5; the formula and their specifics are the following: where: VS = volatile solid excretion per day on a dry-organic matter basis, kg day −1 ; GE = gross energy intake, MJ day −1 ; DE% = digestibility rate of the feed. Different amounts of feed were also considered for the assessment of the DE, also considering the livestock categories. Different feeding periods were considered: from birth to the 90th day (weaning ration) and up to a year for the calves; 365, 95, and 270 days for the heifers, dry, and lactating cows. Similarly for the GE, the data provided by INRAE were adopted for digestibility; (UE × GE) = urinary energy expressed as fraction of GE. Typically, 0.04 GE can be considered urinary energy excretion by most ruminants, and this value was adopted in the current study; ASH = the ash content of manure calculated as a fraction of the dry matter feed intake, specifying the different diets, as indicated in Tables 3 and 4; 18.45 = conversion factor for dietary GE per kg of dry matter (MJ kg −1 ).

N 2 O Emissions from Manure Management
Several statistical models had been developed from animal variables, nutrient intake, and chemical composition of diets [35] or milk urea nitrogen [36,37] for prediction of nitrogen excretion from dairy and beef cattle. Despite the substantial share of buffaloes in global nitrogen excretion from livestock, nevertheless few models for predicting fecal and urinary nitrogen excretion in buffaloes have been developed, in comparison with cattle.
Therefore, different approaches such as nitrogen balance and default factors have been attempted to estimate nitrogen excretion. Although IPCC [32] shows criteria useful to assess nitrogen excretion, as indicated, the approach suggested by Patra et al. [38] to estimate the nitrogen excretion (urinary and fecal) in buffaloes was adopted, as did Romano et al. [8].
The N intake was counted by knowing the crude protein amount distinguished by livestock categories and farms, as indicated in the Tables 3 and 4, given the conversion factor 6.25 from kg of dietary protein to kg of dietary N, kg feed protein (kg N) −1 . The following models, indicated by the authors as those with high accuracy were adopted for fecal and urinary nitrogen excretion respectively for no-growing (4 and 6) and growing where: NE= is nitrogen excretion, g/day; BW = is body weight, kg; CP = is crude protein content of diet (g/kg); NI = is nitrogen intake, g/day; ADG = is the average daily gain, kg, assumed as 720 and 500 g/day respectively for young animals (YA) and heifers (HF).
Patra et al. [38] stated that, in lactating animals, no models containing milk yield as a predictor were reliable for predicting NE, probably due to poor number of studies included in the dataset. In agreement with these authors, we adopted the same model for both lactating and dry cows.
Afterwards, IPCC guidelines [32,33] were adopted to assess direct and indirect N 2 O emissions.
Direct and indirect N 2 O emissions. A unique equation (10.25), useful to estimate direct N 2 O emissions, is indicated by 2019 IPCC refinement [32,33], without distinction between "tiers", i.e., Tiers 1, 2, and 3. In agreement with IPCC guidelines, being viable the use of country-specific nitrogen excretion rates for livestock categories, also distinguished by diets, the equation adopted can be identified as a Tier 2 methodology. Similar to methane emissions provided by manure, the kind of management (solid storage, in the NCS and CS systems) leads to adopt the same default value of AWMS (T,S,k) indicated in the Table 5, also  [32] used to quantify the nitrogen lost with volatilization and leaching, respectively. In these equations are adopted default values, named Frac GAS and Frac LEACH aimed to assess the losses at solid storage, i.e., 0.12 and 0.02. The IPCC refinement [32] also introduce as additional source the annual nitrogen input via co-digestate, not included in this research.
As indicated in Table 5, IPCC [33] suggests emission factor (EF) values available to assess N 2 O emissions, 0.05 and 0.011 for volatilization and leaching respectively.

NH 3 and NOx Emissions from Manure Management
All the animals (bulls, dry, and lactating cows, heifers, young animals between 91 and 365 days) involved in the study were kept at barn, with straw-based litter. There are two sources of ammonia provided by nitrogen in the housing [11], i.e., straw and the manure. The mineralization (from nitrogen to ammonia) provided by straw is negligible compared to that of manure, with volatilization and leaching losses during storage of bedding assumed as zero. Moreover, compaction and covering show significant effects on reducing ammonia emissions [32].
NEMA guidelines [39] were adopted to calculate ammonia emissions from the barn to the land, taking in account the different shares between housing, "outdoor", storage and field application. Although focusing on dairy cattle, several authors [11] stated as the share of estimated NH 3 emissions at housing and storage was very similar, mainly in lactating cows.
Berlese et al. [2] assumed that the share of nitrogen lost (TAN, total ammoniacal nitrogen) [39], through volatilization is 29% out of the total amount excreted, in housing, and solid storage. These authors also showed as volatilization of ammonia originated from N mineralization, at field level, due to the application of manure, was estimated in agreement with equations proposed by European Environment Agency (EEA 2009, 2016) [40,41]. The emission factor adopted for NH 3 was 0.05.
NEMA guidelines [39] also provide equations for estimating NOx emissions; however, in our research they have been neglected. Baldini et al. [2] investigated dairy cattle and showed a very low amount of NOx at farm, for each livestock category, i.e., lactating and dairy cows, heifers, and calves. Confirming the negligible impact of nitrates, in the research [2] a Monte Carlo Simulation was performed, involving only CH 4 , N 2 O, NH 3 emissions.

Emissions from Diesel Fuel and Electricity
As suggested by ENAMA [42], a standard value of 0.85 kg per liter as diesel density and a 3.13 eq. emission factor to estimate CO 2 release from the combustion of 1 kg of diesel were adopted. For the electricity mix, the Italian emission factor (0.47 eq.) was adopted, in agreement with other studies [2,8,9].

Emissions from Crop, Soil Residues, and Synthetic Fertilizers
Emissions from soils and crop residues are included in each category of materials provided by SimaPro database, some inputs were modified, loaded, and finally managed with the software [9]. Emissions from synthetic fertilizers were assessed in accordance with IPCC and NEMA guidelines [33,39], distinguishing N 2 O, NH 3 , NOx, and phosphorus loss as phosphate. Similar to manure management, N 2 O share (Table 5) shows different patterns between direct and indirect emissions. Main inputs and emission factors are also indicated as suggested by IPCC [33], in agreement with Tier 2 method. Total amount of N provided by nitrogen fertilizers was processed, considering the share of each fertilizer, i.e., 46%, 27%, and 18% for urea, ammonium nitrate, and ammonium phosphate, respectively.
Ammonia and NOx emissions. These pollutants were calculated in accordance with formulas provided by NEMA [39] and the same guidelines indicate as default emission factors (EFs), some specific values distinguished between fertilizers [33]. The phosphorus loss in the form of phosphate, provided with ammonium phosphate as mineral fertilizer, was estimated as proposed by Nemecek and Kägi [43].

Software and Impact Assessment
The software SimaPro 8.03 (PhD, Pré Consultants, 2015) was adopted to estimate the environmental impacts following the LCA criterion. Two methods were used to assess the impact categories: (i) EPD 2013 for global warming potential (GWP), computed according to the CO 2 equivalent factors in a 100-year time horizon, acidification potential (AP, g SO 2eq) and eutrophication potential (EP, g PO 4 3− -eq); (ii) ReCiPe Midpoint (H) for agricultural land occupation (ALO, m 2 y) and water depletion (WD, m 3 ). GWP was computed as index aiming to assess the climate change contribution due to the greenhouse gases (GHGs), according to the CO 2 -equivalent factors defined by the IPCC (2019a): CO 2 , 1; CH 4 , 25; and N 2 O, 298. To calculate the AP of the different trace gases the SO 2 -equivalent factors were adopted: SO 2 , 1; NOx, 0,56 NH 3 , 2.45 [44].
To calculate the EP of different pollutants the PO 4 3− equivalents factors were adopted: NO 3 , 0.1 and P 2 O 3 3.06 [45].
Although land use (LU) and land use change (LUC) could be recognized as sources of GHGs emissions [46,47], we adopted the agricultural land occupation, the area of land needed to produce the functional unit (FU) chosen, as distinct impact category. When the land occupation is not directly available with the inventory analysis, Gerßen-Gondelach et al. [48] suggested an equation aimed to estimate this index. Some methods given by SimaPro are exclusively developed to assess the water footprint "sensu lato" and they are named with the publication of the researchers (e.g., Mekonnen and Hoekstra, 2012, 2014) [49,50]. ReCiPe Midpoint method, in agreement with the standards and principles developed by these authors [49,50] in water-rich areas, allows to investigate the water depletion (WD).

Statistical Analysis
The R software (R Core Team, 2021) [51] was adopted for the statistical analysis. Milk traits as NBM yield kg lactation −1 , fat and protein percentages, were analyzed. A comparison between non-corn silage and corn silage-based feeding systems was done, through the Student's t, with the following model: where: t is Student's t; µ 1,2 are the means of values observed; s 2 is the standard error; n 1,2 are the observation numbers, n 1 = n 2 = 5 Afterwards, the t-value will be compared with the critical t table (t-test). The datasets were processed to evaluate the homogeneity of the variances, through the Bartlett test with the bartlett.test function of the stats package. The normal distribution of the data was evaluated through the Anderson-Darling test, through the ad.test function of the nortest package. For the analysis of the significance of the difference between the means through the t-test Student, the t.test function of the stats package was used.
Analysis of variance (ANOVA) was developed by applying the anova function of the stats package. The data were analyzed with the following model: Y (a,b,c,d,e) = µ + CS i + A j + (CS × A) ij + ε ijk (9) where: Ya sub > a,b,c,d,e are dependents variables, i.e., LCA impact categories; 1. µ is the overall mean;

2.
CS is the effect of the ith inclusion of the corn silage in the diets (system); 3.
is the effect of the jth inclusion of the allocation criterion; 4. CS * A is the effect of the interaction of the ith inclusion in the diets with jth inclusion of the allocation; 5.
ε ijk is the error term

Milk Production Traits
As shown in Table 2, no significant differences (p > 0.05) of NBM yields and crude protein between systems were found. These results can be explained because the diets are formulated with different fodders sources. Regardless of the experimental clustering, in the farms of both systems (NCS and CS), feed sources ensure adequate milk performances and qualitative traits, as the diets satisfied macronutrient needs according to live weight and production level.
On the other hand, NCS system showed a lower (p < 0.05) fat percentage than CS one. This parameter is strongly influenced by amount and quality of digestible fiber. The fat share is also affected by speed of rumen transit and by the rumen microbiota [52].
As indicated by Krause et al. [53,54], the fiber of corn silage, although responsible of a faster rumen transit, in comparison with hay or roughages, increases the volatile fatty acids (VFA) production, ensured by cellulolytic bacteria.
Focusing on cattle, Couvreur et al. [55] compared milk traits of Holstein cows feed with corn silage-based diets and pasture-based diets. The diets, balanced in terms of energy provided to the animals, showed a different fat profile in the milk. The study highlighted significant higher percentages of milk fat in cows fed with corn silage. The corn silage-based diets also matched larger milk fat globules. More recently several authors [56] tested a substitution of corn silage with low-inputs silages such as sorghum (Sorghum vulgare L.) in dairy buffaloes. These researchers, comparing corn and sorghum silage-based diets, found that the percentage of milk fat, as well as of protein, was not significantly improved by sorghum.
The comparison between ensiled species showed as corn (i) is a good source of crude fiber, ensures (ii) a better dry matter intake, and (iii) an adequate rumen transit.

LCA Impact Categories in Relation to Culled Cows Allocation (CCA)
Results about these LCA categories are indicated in Table 6, while in Table 7, these results are shown giving evidence to their distribution, according to No Culled Cows Allocation and Culled Cows Allocation. GWP = global warming potential; AP = acidification potential; EP = eutrophication potential; ALO = agricultural land occupation; WD = water depletion of four two production systems before and after the economic allocation of culled cows (mean ± SD). NCCA = No Culled Cows Allocation; CCA = Culled Cows Allocation; S = system; A = allocation; A × S = allocation × system.

No Culled Cows Allocation
Global Warming Potential (GWP): GWP is the most investigated descriptor in LCA studies and often unique [14,19] in dairy systems. Pirlo and Lolli [20] showed a regression between sustainability and milk yield expressed either as kg of FPCM per cow per year or as tons of FPCM per hectare. They compared conventional and organic systems and recorded that the emissions of GHGs per kg of FPCM were significantly (p < 0.05) reduced by the increase in average milk production per cow in both systems, suggesting that increasing of yield per head improves sustainability. The authors, in agreement with Guerci et al. [20], also highlighted that methane enteric emissions are strongly mitigated by the size of functional unit, whereas acidification and eutrophication are subjected to the contribution of many sources. Gislon et al. [57] found higher enteric methane emissions in cows fed with hay than those with corn silage.
Pirlo et al. [6,14] argued that 87% of total on-farm CH4, was due to enteric emissions. Moreover, investigating carbon footprint in dairy buffaloes, they indicated similarly to cattle, the increase of milk yield per buffalo is an effective strategy for reducing enteric CH4 emission per kilograms of NBM. NBM yields of NCS and CS farms showed no significant (p > 0.05) differences, with very similar means per lactation (2458 and 2334 kg respectively), and this is consistent with the overall similarity of both the farms clusters in multiproduction efficiency.
Acidification Potential (AP): This LCA descriptor, similarly to eutrophication, is affected by the conduction system and by environmental conditions, such as temperature, relative humidity, or rainfall [9]. Our study involved farms in very similar environments and the manure management was superimposable in both systems. Non-corn silage (NCS) system showed significantly higher (p < 0.01) values of AP than corn silage (CS). SO2 equivalents of NCS (54 g) were comparable with those found by Pirlo et al. [6].
Our study adopted the same milk normalization formula as Pirlo et al. [6], suggested by Di Palo [23], whereas Berlese et al. [2] chose FU 1 kg FPCM according to the IDF formula [58]. These authors found AP of 1 kg of FPCM was 37.3, 20.4, and 35.7 g SO2 eq on average, obtained with no-allocation, physical and economic allocation, respectively. The

No Culled Cows Allocation
Global Warming Potential (GWP): GWP is the most investigated descriptor in LCA studies and often unique [14,19] in dairy systems. Pirlo and Lolli [20] showed a regression between sustainability and milk yield expressed either as kg of FPCM per cow per year or as tons of FPCM per hectare. They compared conventional and organic systems and recorded that the emissions of GHGs per kg of FPCM were significantly (p < 0.05) reduced by the increase in average milk production per cow in both systems, suggesting that increasing of yield per head improves sustainability. The authors, in agreement with Guerci et al. [20], also highlighted that methane enteric emissions are strongly mitigated by the size of functional unit, whereas acidification and eutrophication are subjected to the contribution of many sources. Gislon et al. [57] found higher enteric methane emissions in cows fed with hay than those with corn silage.
Pirlo et al. [6,14] argued that 87% of total on-farm CH 4 , was due to enteric emissions. Moreover, investigating carbon footprint in dairy buffaloes, they indicated similarly to cattle, the increase of milk yield per buffalo is an effective strategy for reducing enteric CH 4 emission per kilograms of NBM. NBM yields of NCS and CS farms showed no significant (p > 0.05) differences, with very similar means per lactation (2458 and 2334 kg respectively), and this is consistent with the overall similarity of both the farms clusters in multiproduction efficiency.
Acidification Potential (AP): This LCA descriptor, similarly to eutrophication, is affected by the conduction system and by environmental conditions, such as temperature, relative humidity, or rainfall [9]. Our study involved farms in very similar environments and the manure management was superimposable in both systems. Non-corn silage (NCS) system showed significantly higher (p < 0.01) values of AP than corn silage (CS). SO 2 equivalents of NCS (54 g) were comparable with those found by Pirlo et al. [6].
Our study adopted the same milk normalization formula as Pirlo et al. [6], suggested by Di Palo [23], whereas Berlese et al. [2] chose FU 1 kg FPCM according to the IDF formula [58]. These authors found AP of 1 kg of FPCM was 37.3, 20.4, and 35.7 g SO 2 eq on average, obtained with no-allocation, physical and economic allocation, respectively. The value with no-allocation, using NBM formula [23], is 62.2 g SO 2 eq, in line with our values and those obtained by Pirlo et al. [6]. Pirlo et al. [6] however, showed heterogeneous values, with a range of equivalents between 34 and 98 g SO 2 . The authors stated that AP was mainly affected by manure management (40-60%), the most representative ammonia source. Moreover, in the farms described in the present study, we recorded a positive correlation between AP and purchased feeds per kilogram of NBM. In addition, they suggested the improvement of the feed self-sufficiency had a positive effect on AP.
Berlese et al. [2] found that manure management gave a very significant contribution to AP, i.e., 80%. These differences among different papers could be due to environmental patterns (temperature, rainfall). Sabia et al. [5] adopted as FU 1 kg of milk, corrected in agreement with Di Palo [23]. They compared two dairy buffalo farms, with two different heifer rearing systems, free-ranging (FR) and confinement (C). The study showed as, in both systems, a significant amount of corn silage was provided to each livestock category (3-8 months calves, gestation, lactating and dry cows), except FR heifers. Interestingly, a very similar amount of equivalents was found in the lactation phase for both modes, 7.01 and 7.40 g in FR and C respectively, about half of the total. Although a farm was partially pasture-based, the very low AP values, 14.17 and 15.58 g SO 2 in FR and C respectively, could be related to the better efficiency, due to corn silage. In our CS system, mean values of AP, while higher, where comparable with the equivalents showed by Sabia et al. [5], consistent also with SO 2 eq that Pirlo et al. [6] showed in the more sustainable farms (34 and 41g).
The impacts are often influenced both by the weight of the functional unit [8,18] and by the overall feed efficiency. Gislon et al. [59] investigated about 46 dairy cattle farms and found (in accordance with other authors, e.g., Gerber et al.) [10] lower environmental impact in the High Quality Forage System, HQFS for almost all the impact categories. As NCS and CS did not show differences in milk production traits, we consider this consistent with the different DM yield per hectare between corn silage and hay.
According to Gislon et al. [59], this gap could explain the high amounts of concentrates and fodders purchased in the farms without corn silage. Although all the diets composition (except NCS2) is based on transoceanic sources, i.e., soybean, and it is based on feeds bought in Northern Italy (maize flour, maize germ meal), the farms without corn silage (NCS) showed a lower feed self-sufficiency than corn silage-based (CS) ones. In particular, NCS1 arable area provides almost 50% of hay only, all other sources are purchased. In the NCS2 one, the origin of all horticulture by-products and ryegrass silage is off-farm, in addition the purchases involve a large amount (60-65%) of barley.
The fava bean on-farm yield is lower than 10% of total requirements. The fodder-based area (hay and ryegrass silage) ensures a better self-sufficiency to other NCS farms, even though the significant amount of purchased feed, in particular, concentrates (NCS3, 4,5) and straw (NCS3).
Eutrophication Potential (EP): In the non-corn silage (NCS) feeding system, we found slightly higher (p = 0.051) PO 4 3− equivalents in comparison with the corn silage based (CS). Berlese et al. [2] investigated six corn silage-based farms, converting the FPCM (IDF 2015) [58] adopted by the authors into the mode used in this study [23]; we are able to say that they found mean values of 12.4-13.6 g PO 4 3− equivalents, comparable with the ones (13.15 g) showed by our CS system. Lactating cows diets indicated by Berlese et al. [2] showed a similar CP share (12%), mainly guaranteed by concentrates, with ours, 14.10 and 13.05% for CS and NCS mode respectively, whereas Pirlo et al. [14] found higher values, i.e., 15.7%.
In a similar research, Pirlo et al. [6] correlated to manure management/application a contribution of approximately 80% to eutrophication, although with a wide variability range among farms. Berlese et al. [2] showed manure management gave the highest contribution to EP (45%), followed by off-farm productive factors (purchased feed), with 39%. In agreement with Nguyen et al. [26], nitrate leaching from agricultural soils is the most important contributor among the group of substances responsible for eutrophication in beef.
Accordingly, Bragaglio et al. [26] recorded that in beef-intensive system the major contributor to EP was the emission from feeding inputs during the fattening phase (48%), heavily based on concentrate-based diets. Guerci et al. [18] investigated 12 dairy cattle farms in Denmark, Germany, and Italy, showing as the high feed efficiency had a strategic role. Three of five with the lowest AP impact farms used corn silage-based diets.
Probably, the management of synthetic fertilizers could explain differences. 233 and 57.5 kg N per hectare were the mean values reported by Pirlo et al. [6] and Berlese et al. [2] respectively, investigating six farms in each study. Our data showed an intermediate administration of N in the CS system, i.e.,~100 kg/Ha; whereas the NCS one had few amounts (32 kg/Ha as mean).
These data are consistent with the requirements of the specific crops. Low synthetic fertilizers inputs were provided to NCS farms; thus, the system showed a low fodder efficiency and consequently the need for purchased feeds, mainly concentrates. Similar to AP, results of the most sustainable farms involved by Pirlo et al. [6] highlighted values (22,25,and 26 g PO 4 3− ) comparable with EP of our NCS system, (20 g). According to data obtained by interviews, the use of phosphate fertilizers was low and it did not affect outcomes.
Agricultural Land Occupation: No differences were found between systems. Dairy buffaloes ALO has been studied in few research, showing poor data on this LCA descriptor. Romano et al. [6] compared farms with arable only focused on feed crops (No Wheat Crop system, NWC) and farms with significant areas dedicated to produce wheat grain (Triticum durum Desf.), With Wheat Crop, WWC system. Higher values of ALO were found for WWC and this conduction mode showed larger mean extent (187 Ha) than NWC system (55 Ha). While allocated, wheat was not enough to decrease m 2 year −1 , lower (p < 0.05) in the NWC conduction.
Land use and analogous categories are mainly affected by farm area (hectares), as suggested by studies focused on beef [24,60,61] and dairy cattle [18]. Farms self-sufficiency and feed productivity also improve these impact categories, as indicated by Gislon et al. [59], in the framework of intensive sustainability, increasing the crop efficiency yield of the Utilized Agricultural Area (UAA).
Sabia et al. [5] found in the pasture-based (FR) and in the confined (C) farm very similar values, 10.37 and 9.64 m 2 year −1 , respectively. Fodder efficiency of corn silage, abundantly supplied to FR animals, could explain so close results. No significant differences of ALO between systems investigated in our study could be explained with higher dry matter (DM) yield per hectare, obtained in the system affected by the larger average UAA: 60 and 142 for non-corn silage and corn silage respectively.
Conflicting results are available for dairy cattle water impacts, probably due to the controversy on the similarity between not completely comparable categories.
Scientific bibliography is larger available on water footprint (WF) than on water depletion (WD) data. Literature supplies some WD values of dairy cattle, FPCM as FU, with 0.52, 0.38-0.55 m 3 respectively found by Noya et al., and Romano et al. [9,17] in a similar Mediterranean environment. Although Sabia et al. [5] stated that no data are available on WD in dairy buffaloes, they found mean values of 4.2 m 3 .
According to the assumption [6,18] described above, these results, that should have been lower, gave those values because they are probably affected by the low samples size. Romano et al. [8] found WD of 1.69 and 1.79 m 3 for With Wheat Crop (WWC) and No Wheat Crop (NWC) system respectively, without significant differences (p > 0.05).
These values are comparable with the results shown in the present study; four times larger than WD of Noya et al., and Romano et al. [9,17].
Although high water requirements of maize silage, CS system did not show to be less sustainable than NCS one. In dairy cattle husbandry both WF and WD were investigated [17].
These researchers found as feed/fodder-related activities contributed almost entirely to water impacts, showing as feed/fodder production stands as the most critical process of dairy system. In the framework of WF green water is the main responsible for the outstanding influence of feed/fodders production, accounting for 88% of the contributions, blue and grey water had a lower influence, with contributions of 7% and 5%, respectively. Moreover, they [17] recorded a much higher share about WF than WD, i.e., 8.65 and 0.52 m 3 respectively. These authors also suggested that WD could be directly related to the blue water, and the assumption was in accordance with their results. In agreement with this standard, water requirements of corn, while larger than other crops would not get worse WD.
Gerbens-Leenes et al. [62] stated that WF is dominated by animal feed and WF of concentrates is roughly five times larger than the WF of fodders in beef production system. This trend could be more evident when considering the blue water category, closer to WD.
As regard WD, no significant effect of system mode was observed by Bragaglio et al. [24]; in particular the authors compared two intensive beef systems (Fattening System and Cow Calf Intensive) with two extensive beef systems (Podolian System and Specialized Extensive). Although all the farms of Cow Calf Intensive were corn silage based, they did not show statistically higher WD values than farms without corn silage, Podolian System, and Specialized Extensive.
In addition, Sabia et al. [5] suggested as buffaloes are kept at pasture (FR heifers), able to drink directly from a stream flowing, decreased WD. While absence of pasture in our scenario, well water (acqua di bonifica) adopted to irrigation, should be considered green water such as stream flowing water. Interviews showed as all farmers adopted drinking water only (human use) for milking routine and hygiene, whereas in some cases, with higher share in the NCS system, this source was also adopted to clean the milking parlor or provided to animals as drinking water.

Culled Cows Allocation
Allocation is commonly adopted to improve sustainability of a functional unit. In the present study the criterion, useful to attribute pollutants to by-product meat, involved culled cows only. As reported in Table 6 all impact categories did not show significant decreases after economic allocation and after interaction allocation/system. Moreover, with reference to different incomes, the Table A1 has been included in the Appendix A.
In agreement with a previous study [9] we chose this strategy to identify high culling rate, comparing results between systems. In this study average percentages of culled cows are low and similar in both the systems. Economic value of each culled cow, the same for all farms, was chosen as standard, given the low income provided by these animals and uncertainty of data. Large difference between milk and meat buffalo incomes, also in comparison with the values in cattle, highlighted by few culled animals, could explain after-allocation results, not significant in this research.
Buffalo and cattle cows had also a different productive scenario. Grandl et al. [63] comparing two feeding modes in Brown Swiss cattle found very similar productive performances of cows. These animals showed 1083 and 925 days of productive life, 3.3 and 3.2 lactation numbers, 3.6 and 3.1 calves born numbers in diets with and without concentrates respectively. Other authors [64] focusing on the length of productive life in German Holstein cows (95,215 heads) found a comparable average length of productive life, i.e., 977 days.
Buffalo cows show different performances: Galeazzi et al. [65] highlighted that in Italy these animals are generally long-lived, with a mean productive life of six to eight calvings (9 to 11 years of age). For buffaloes reared in Bulgaria, Peeva and Ilieva [66] also reported a mean longevity of 7 years. More recently Minervino et al. [67] indicated buffaloes have higher longevity than cattle (they can surpass 30 years of age, while maintaining their reproductive capacity until they are 18-25 years old).
Romano et al. [9], investigating dairy cattle, found as several impact categories were affected by allocation, the trend was conditioned by the following synergy: (i) Higher culling rate, as percentage range, in comparison with buffaloes; (ii) head income as far as 385 EUR, in the farm with higher culling rate; (iii) lower income per kg of bovine milk than buffaloes, although higher yields per head.

Conclusions
This study carried out a life cycle assessment of dairy buffalo farms, comparing standardized (corn silage, CS) with not standardized (non-corn silage, NCS) systems.
The corn silage-based system is recognized to be standardized owing to a significant bibliography. Although these references are mainly focused on dairy cattle, our research has (partially) shown that this kind of dietary profile may characterize dairy buffaloes also.
The assumption is valid despite the fact that the buffalo is recognized as more rustic than bovine.
In fact, in the study some data have shown a positive correlation between corn silagebased diets and results: (i) milk traits, i.e., fat; (ii) LCA categories: acidification and eutrophication.
The first system (CS) recorded higher sustainability than the second one (NCS) for AP and EP categories, whereas no differences in culling cows rate were highlighted by allocation criterion. Dietary corn silage was identified as standard system because it is recognized as strategical to improve sustainability of bovine dairy farm systems. This crop, although commonly adopted in Italian buffalo farming system, is not available in the whole PDO mozzarella cheese area. A not corn-silage based feeding system is not negatively affected in milk yield, but poor fodder efficiency worsens impact categories less influenced by functional unit.
This trend, in dairy cattle farming system, is confirmed by several research; nutritional quality of the fodder base influences the environmental impact also in the buffalo, although less than in the cattle. Corn cultivation is strictly related to water (green water) availability for irrigation purposes, not homogeneously widespread in PDO area. The NCS system is describable as not standardized because farmers without corn silage must adopt fodders or roughages according to different geographical, pedological, climatical conditions. Corn silage ensures greater self-sufficiency, whereas NCS farmers often buy fodders such as hay or by-products as straw; moreover, NCS farms need larger amounts of purchased concentrates. This study has shown that also in dairy buffalo Italian system, good feed efficiency improves sustainability. Allocation, adopted as marker, did not have significant results and this leads us to conclude that dairy cattle and dairy buffalo systems, are characterized by different equilibria between incomes, responding differently to economic allocation. According to our best knowledge, only a study focusing on dairy Italian buffalo has adopted mass allocation, in addition to economic one. Although the authors chose raw milk and mozzarella cheese as the functional units, the target was mozzarella. In future work, the unprocessed product (raw milk) should be studied, and the viability of mass allocation for partitioning the input and output flows between functional unit and co-products should be verified.

Conflicts of Interest:
The authors declare no conflict of interest.
Appendix A Table A1. Yields and economic incomes (%) derived from milk, wheat grain, and culled cows of non-corn silage (NCS) and corn silage (CS).