#### 3.2. Carbon Footprint Analysis

We started by determining the GWP of concentrated feeds. Out of 83 feeds used in the 25 farms, we were only able to obtain the specific GWP for 3.

Table 2 shows the unitary composition, assumed region of production and associated GWP of each ingredient present in the average concentrated feed (in only three concentrated feeds was it possible determine the amount of each ingredient in the feed)—which was used for the feeds with unknown ingredient composition. The GWP of this average feed is 0.64 kg CO

_{2}e/kg.

As seen previously, the characteristics of the farms varied considerably within the sample. Due to these differences in farm inputs and outputs, GWP results also vary significantly between farms.

The full breakdown of GWP per source of emissions for each of the 25 farms is shown in

S3 of the Supporting Materials files.

Table 3 presents the mean, minimum and maximum GWP of the 25 farms for six FUs, namely farm, farm area (ha), cow (number of animals), kg of milk, kg of ECM and kg of FCPM. Enteric fermentation, on average, is the main source of impact, with 46% of the impact on average (33–52%—

Figure 3). However, it has the smallest standard deviation as percentage of the mean value (59%) for most FUs. This is because some of the sources are correlated, for example, unit emissions from enteric fermentation per cow are constant for all farms and therefore depend only of the number of cows. In this case, the variation observed in results is justified by the number of cows by farm (the grazing period does not change significantly among farms). The contribution of concentrated feeds has the highest variation between minimum and maximum values (16–37%, average: 27%—

Table 3 and

Figure 3). Despite the fact that average unit emissions indicated in

Table 2 were used for most farms, the variability of feed consumption between farms causes the greatest standard deviation of the mean value (139%). The standard deviations of the other sources range between 62% (manure management: 10–19%, average: 15%) and 6% (cleaning agents) and are connected mostly to variation in usage at each farm. In spite of higher standard deviation, cleaning agents and energy consumption contribute only 4% to the total impact (on average). Energy, agricultural operations and cleaning agents never represent more than 8% of total impact (mean: 5%; min: 1%).

Results for farm-level and farm area FUs have higher between-farm standard deviations (

Table 3). These results show that some FUs dilute the variability verified at the level of the source data, that is, when emission from concentrated feed consumption is higher the fertilizer application emission tends to be lower and vice-versa and therefore the final GWP is less variable than the inputs and outputs collected in the survey.

Table 4 presents the multivariate regression results, that is, coefficients value and units and respective

p-value, which indicate coefficient significance. From the four independent variables, only concentrated consumption is not statistically significant as explanatory variables of GWP (at 5%). The Repeating the multivariate regression only with milking cows, dry cows and heifers (significant variables), the coefficients did not change significantly. The r-squared in both multivariate regressions is 0.78. Regarding the interval of variance of significant variables, obtained by multiplying the specific parameter by the range of values observed by the variable, milk yield is the variable with highest contribution to the total impact (between 0.17 and 0.55 kg CO

_{2}e/kg milk). The second highest contribution is from concentrated feed consumption, which ranges between 0.09 and 0.23 kg CO

_{2}e/kg milk.

Figure 4 shows the univariate dependency of the CF on milk yield per cow. We calculate the CF as a function of the inverse of milk yield but present it with milk yield in the x-axis. Therefore, the linear regression was depicted as a curve (represented in yellow). We did not consider farms 3 and 23 in the univariate regression. These farms are outliers as their CFs are 0.2 kg CO

_{2}e/kg milk higher above average. The r-squared is 0.59 (considering the two outliers it would drop to 0.44). The regression constant (0.45 kg CO

_{2}e/kg milk) can be interpreted as the life cycle emissions that cannot be eliminated by increasing milk yield (mainly from fertilizer application and concentrate use). The regression slope (i.e., the average emissions per additional milk unit) can be interpreted as the penalty for lack of full efficiency of the milking cows (i.e., an efficiency gap). This value (3352 kg CO

_{2}e/cow) is similar to the basal emissions of the milking cow (i.e., enteric methane and emissions from manure management), which are about 3872 kg CO

_{2}e/cow (considering the average grazing period—ranging between 18 and 21 h per day—for the emissions from manure management).

The most noticeable outliers where the CF is higher than predicted by its current milk yield are farms 3 and 23. Farm 3 has the highest consumption of concentrate feed of all farms (6.5 t/head yearly, or 0.61 kg/kg milk). Farm 23 is also among the highest consumers of concentrate as well as silage. The two farms that have by far the highest milk yield (Farm 10,13) have some of the lowest CFs. Farm 22 is the main outlier with lower emissions than expected given the univariate model for milk yield. Farm 22 has one of the lowest levels of feed consumption (0.36 kg/kg milk), as well as low levels of organic fertilization. The milk yield is highest in the farm 13 and the GWP has the greatest distance (downwards) to the regression line; this farm has the second highest concentrated feed consumption per milking cow (6042 kg/cow—similar consumption to Farm 10) but Farm 13’s milk yield is more than 3000 kg/cow higher (Farm 13: 15,800 kg milk/cow.year; Farm 10: 13,200 kg milk/cow.year) and lower number of dry cows and heifers (Farm 13: 39%; Farm 10: 49%).

#### 3.3. Comparison with the Literature

We found a relatively small number of agri-food LCA studies on cow milk production that met our criteria (25 papers). However, some of those studies included multiple production systems (e.g., [

51]), multiple farms/production countries (e.g., [

11]) or even comparisons of different production systems in different countries (e.g., [

15]). The sample used for comparison thus included 84 individual CFs, collected from the 25 studies. The full list of references, as well as the main production characteristics of the studies (concentrated feed consumption and milk productivity per cow, land area, number of cows and stocking rate) and related impact are included in the

S3 file of the Supplementary Materials files.

Only one article was published for milk production in Portugal [

32], as also observed by Morais et al. [

52] in a revision of LCA studies in Portugal. Castanheira et al. [

32] indicate an impact of 1.02 kg CO

_{2}e/kg milk, which is 23% higher than the “Vacas Felizes” production system. However, the production system is considerably different. It assumes similar concentrated feed consumption (about 2000 kg/cow), about three quarters of milk yield (~6 t/cow versus ~8 t/cow in our study), half the number of cows (53 cows versus 93 cows) but the main difference is farm area and associated stocking rate. The farm area considered in Castanheira et al. [

32] is 20% less than in “Vacas Felizes” farms, while the stocking rate is three times higher. For Europe, we found 50 CFs in 13 studies. The average impact is 1.16 CO

_{2}e/kg milk (median 1.09 CO

_{2}e/kg milk), which is 40% higher than the “Vacas Felizes” production system. Average concentrated feed consumption and milk yield are lower, about 1600 kg/cow and 7400 kg/cow, respectively. Average land area and number of cows per farm are similar, 80 ha and 109 cows, respectively. However, stocking rate is about 50% higher. Taking into account all 25 studies, the average impact is 1.22 CO

_{2}e/kg milk (median 1.05 CO

_{2}e/kg milk), which is 50% higher, indicating that non-European production systems tend to have higher impacts. Average concentrated feed consumption and milk yield are similar to the European average and lower than the production system studied (concentrated feed consumption is 1650 kg/cow and milk yield is 7300 kg/cow). Average land area and number of cows are slightly higher than Europe’s average but clearly higher than the production system studied (land area: 86 ha and 123 cows). Regarding stocking rate, the global average is slightly lower than the European average but higher than the stocking rate of the production system studied.

Twenty-one studies (84%) included co-product allocation of the impacts. The average milk allocation factor is about 88% (minimum 63% and maximum 93%). Meat is the main co-product but also crop production (in two studies) and manure export (only one study). Economic allocation was the most used allocation method (about 90% of the studies which considered allocation) and it leads also to a higher allocation of impacts to milk (about 90%). Highly specialized farms in milk production (i.e., co-products are not determinant for the economic viability of farms) as are “Vacas Felizes” farms, had milk allocation factors of approximately 90%.

The heifer replacement rate is absent in a significant number of studies (7 CFs, in 3 studies [

53,

54,

55]). In studies that included this rate in calculations, it was reported as a percentage of the herd and number of animals. In the studies which considered replacement, it ranges between about 5% (in highly specialized farms, for example, Guerci et al. [

15]) and 40% of the herd. The replacement rate of the herd in “Vacas Felizes” is about 40%. There is only one exception, farm 3, where the replacement rate is 150% but which also produces cattle for meat. Only 3 studies (13 CFs) report the number of lactations per cow. The average number of lactations per animal in those studies is about 3. The study with the highest number of lactations (more than 3) per animal reports a low CF (average 0.99 kg CO

_{2}e/kg) [

14]. The same study tested increasing the number of lactations per animal and discovered it led to a reduction in the CF of about 5% [

14]. The other two studies report a lower number of lactations per animal (less than 3) [

17,

56] and higher CFs (average: 1.55 kg CO

_{2}e/kg). “Vacas Felizes” cows can have three or four lactations throughout their productive cycle.

Only four cases from four papers had lower CFs than the milk from Azores [

12,

15,

56,

57]. The first case is also pasture-based milk production in New Zealand [

12], with a CF of 0.80 kg CO

_{2}e/kg, similar to “Vacas Felizes” milk. This is a more intensive system, with higher farm area 165 ha (average in this study is 74 ha) and also higher stocking rate, 2.7 cows/ha. In the New Zealand system, the animals eat more concentrated feed (about 6 t/animal/year). The milk yield is significantly lower, about 5000 kg milk/animal. Nevertheless, fertilizer application is almost inexistent. The second case is a very low-intensity production system in Germany [

15]. In this farm, concentrated feed quantity per cow during a year is almost negligible. In this system, the animals are almost exclusively fed with grass and the fertilizer application is also almost negligible. This low level of inputs also leads to low milk output (5 t/animal/year). Land area in this farm is 19 ha and stocking rate is 1.1 LU/ha. The third case is an intensive milk production system in Brazil [

57], with lower area and number of milking cows, 17 ha and 55 cows but significant higher stocking rate (3.2 cows/ha) and concentrated feed consumption (6 t/animal/year). The milk yield is similar (8000 kg milk/cow.year). The system is based on animal housing, with no inorganic fertilizer application, which leads to low N

_{2}O emissions from fertilization. The forth case is a highly intensive 71 ha farm in Canada [

56]. The system is based on animal housing, with no inorganic fertilizer application, which leads to low N

_{2}O emissions from fertilization. Most of animal feed is produced in the farms and thus concentrated feed consumption is very low considering milk production per cow. Livestock density in this farm is 0.9 LU/ha.

The “Vacas felizes” production system R ratio is 0.28 kg/kg milk. The world average R ratio is 0.25 kg/kg milk, which means that, despite spending the vast majority of each day in the pasture, cows in Azores consume an above-average quantity of feed relative to milk production.

Table 5 present an equal distribution of the studies reviewed per class of R ratio (only in 69 studies was it possible to calculate the R ratio). Production systems with extreme R ratios tend to have the highest GWP, while intermediary classes have the lowest GWP. “Vacas Felizes” milk is featured in the ]0.21 –0.37] class, which is the class with the lowest mean CF (1.11 CO

_{2}e/kg milk) and lower variation between minimum and maximum values.