3.2. Factor Analysis
Due to the high number of variables involved in this work, it was deemed necessary to reduce the dimensionality of the data, in order to make sure that only the relevant information contained in the dataset is used. The main results of that factor analysis are shown in
Table 3. The data adequacy to this multivariate technique appeared to be satisfactory since the value of the KMO measure was higher than 0.6 and the null hypothesis of Barlett’s sphericity was rejected with a
p-value of less than 0.05, coinciding with the criteria of Cuadras [
42]. The analysis identified six relevant factors that accounted for 85.34% of the variance. Each of the variables contributed to a greater or lesser extent to each of the factors; however, the greater contribution of a variable to a specific factor is what will determine the result of the subsequent analyses, i.e., the cluster and the canonical population analysis. The parameters fat content, protein content, dry extract and α-tocopherol were the most important contributors to factor 1; SFA, MUFA, PUFA, CLA and ω6 to factor 2; ω3 and β+γ-tocopherol to factor 3; CO
2 and CO
2eq to factor 4; energy consumed and N
2O to factor 5; and only CH
4 to factor 6.
Factor 1 is thereby related to milk’s basic composition and α-tocopherol, factor 2 is associated with milk’s lipid profile, showing the saturated (SFA) and unsaturated fatty acids (MUFA, PUFA and CLA) have an inverse correlation, and factor 3 is linked to green pasture as this increases the ω3 content, although it showed a negative correlation with the β+γ-tocopherol content. These three factors are consequently associated with milk composition. On the other hand, factor 4 is linked to the farm’s carbon footprint and factors 5 and 6 to the energy needed for operation and other GHG emissions.
3.4. Canonical Population Analysis
All indicators (milk quality and environmental) were reduced to two canonical functions that together cover 94.3% of the total variance.
Figure 2 shows the scatter plot whose axes are represented by these two canonical functions and where the 17 farms are positioned in a bidimensional space. This diagram shows the clear formation of the four different clusters of farms, except for farm 11, which was separated from its cluster. Fat content, α-tocopherol, energy consumed, CO
2 and CO
2eq were selected by the analysis as the five parameters with the greatest discriminant power. The first canonical discriminant function revealed a direct relation with energy consumed, CO
2, N
2O and CO
2eq and an inverse relation with protein content, PUFA, CLA, ω3 and α-tocopherol. On the other hand, the second canonical discriminant function showed a direct relation with fat content, dry extract and SFA and an inverse relation with ω6 and PUFA.
Therefore, according to the interpretation of the scatter plot, these preliminary results would suggest that farms in clusters 1 and 2 would have higher energy consumption and higher GHG emissions and their milks would have low PUFA, CLA, ω3 and α-tocopherol content. In contrast, farms in clusters 3 and 4 would have lower energy consumption and environmental impact, and their milks would have higher contents of the above-mentioned parameters. On the other hand, milks from farms in clusters 1 and 4 will have a higher fat content, higher dry extract and higher SFA content, but lower ω6 and PUFA content. Farms in clusters 2 and 3 will have the opposite characteristics.
3.6. Cluster’s Milk Quality
Milk quality characteristics of the four clusters of farms, previously determined by multivariate methods, are shown in
Table 5. Statistically significant differences were found for protein content, total fat content and dry extract, obtaining in all of them the highest value in cluster 4 and the lowest in cluster 2. The protein content was directly correlated with the grazing level (pasture in total forage) and total forage and inversely correlated to the milk production (
Table 4), showing a Pearson correlation coefficient of 0.50, 0.25 and −0.33 (
p < 0.01), respectively. Previous works point out that dairy cows grazing
ad libitum had higher concentrations of milk protein and casein than animals grazing a restricted pasture allowance [
43] or with a total mixed ratio [
44]. However, these results are more in agreement with previous works that reported lower milk yield and higher protein content in high grazing level ewes [
45] and cows [
9], while there were no differences in protein content with grazing level when milk yield was unaffected [
7] or a higher protein content was observed for feedlots that had lower milk production compared to pasture grazing ewes [
46]. These observations can be explained due to a dilution effect [
47]. This is due to the fact that at a particular level of energy intake, there is a minimum protein intake, and reduction below this protein level will cause a reduction in milk yield [
3].
On the other hand, and regarding the correlation observed with total forage, previous works have shown that there is strong evidence showing an increase in milk protein in dairy cows when the forage:concentrate ratio increases [
48], but it depends on the energy density of the diets [
2] and their composition.
Regarding fat content (
Table 5), usually this component decreases when the amount of ingested concentrate increases [
49], as observed for clusters 2 and 3. There was not a significant correlation between milk yield and fat content, because at certain levels, even when the milk yield increases, the fat content does not decrease significantly [
50]. Moreover, when the objective is to acquire sheep milk with a higher fat content, the strategy adopted is that of supplying a fat-supplemented diet [
51]. That might possibly be reason for the high fat content of cluster 2, in spite of the low forage proportion. Regarding the effect of grazing level, there was a significant correlation (0.161,
p < 0.05) between fat content and pasture in total forage as previously reported by Sales-Duval et al. [
52] and differing from the results found by Delgado-Pertiñez et al. [
7] which did not determine a significant effect of grazing level on fat content. A negative energy balance produced by undernutrition in grazing animals will result in an increase in milk fat content due to an increase in free fatty acids in blood, which is a consequence of body fat mobilization [
53]. Other authors reported that grains can provide a high proportion of starch for digestion in the small intestine, leading to an increase in milk yield and a decrease in milk fat concentration [
54].
Dry extract showed same results as fat content, because protein content is more stable than fat and lactose usually did not show significant variation due to diet [
12], and the differences observed are likely linked to the variations in milk production [
2]. Besides the effect of management system, the effect of breed should be taken into consideration. Previous works have shown that Churra milk had significantly higher protein, fat and dry extract content than that of the Assaf breed [
55,
56], which is directly related to clusters 4 and 2, respectively.
No significant differences among the four clusters were found for SFA, MUFA and PUFA (
Table 5). Indeed, no significant correlations were observed for these parameters with milk production or grazing level, except for PUFA, which showed a significant correlation (0.210
p < 0.01) with the percentage of pasture in total forage. Regarding CLA levels (
Table 5), there were no significant differences among the clusters (
p = 0.070), but Pearson coefficients showed a significant correlation between CLA content and the percentage of pasture in total forage (0.511
p < 0.01) and with grazing time (0.323
p < 0.01). Similar results have been previously reported for grazing ruminants. Then, organic ewe’s milk had a higher content of PUFA and CLA [
12,
57]. Indeed, some works revealed that although CLA content tended to increase with the grazing level [
8,
58], this is not always observed for PUFA [
58]. On the other hand, the lack of statistically significant differences among the four clusters could be related to the high variability of CLA contents within each cluster throughout the year, mainly due to the differences in pasture composition [
11]. Previous results showed that grazing animals had the highest levels of CLA in ewe milk during spring [
59]. A significant correlation between CLA content with both ω3 and ω6 were observed (0.622 and 0.232, respectively,
p < 0.01) because most CLA isomers originate from microbial hydrogenation in the rumen and subsequent enzymatic desaturation of hydrogenated intermediates in the mammary gland, mainly from α-linolenic and linoleic acid [
60], which are the major fatty acids of the total ω3 and ω6.
Finally, ω6 did not show significant differences between the four clusters, but ω3 content was significantly higher in cluster 4 than in cluster 2 (
Table 5). The ω3 showed a direct correlation with % of pasture in total forage (0.498
p < 0.01) and time in pasture (0.561
p < 0.01), as previously observed for CLA, while the ω6 content showed an inverse correlation with these two variables (−0.239 and −0.410,
p < 0.01). This means that the grazing level produced a linear increase in ω3 and a linear decrease in ω6 in milk, as was also previously reported by Couvreur [
58], because the hydrogenation level was similar between diets. The increase in ω3 is due to the higher concentration of these fatty acids, mainly α-linolenic acid, in fresh pasture [
61]. However, not only the amount of pasture, but also the forage species and its phenological phase had a strong influence on fatty acid composition [
62,
63]. On the other hand, this study showed that ω6 content depended on concentrates of the diet that should be the source of these PUFA. Fatty acids containing more than 18 atoms of carbon are inhibitors of the novo fatty acid synthesis. Therefore, the linear increase in these fatty acids induced a decrease in the short- and medium-chain FA contents. As previously observed for fat, protein, dry extract and ω3, the content of α-tocopherol was significantly higher in cluster 4, while cluster 2 showed the lowest value (
Table 5). The Pearson correlation coefficients showed a direct correlation with % of pasture in total forage and grazing time (0.491 and 0.514, respectively,
p < 0.01), as observed for the above parameters. The correlation between grazing level and α-tocopherol was previously reported [
11,
64]. This is linked to the fact that fat-soluble vitamins secreted into ruminant milk depend directly on their level in the ration [
65]. Therefore, the high α-tocopherol content of green pasture results in a higher transference from blood into milk [
11]. Regarding β+ γ-tocopherol, no significant differences were observed among clusters, as previously reported by Gutiérrez-Peña et al. [
11], and no significant correlation with grazing level was observed. A significant correlation with α-tocopherol (Pearson’s coefficient value of 0.207,
p < 0.01) was observed because this is the main active form of vitamin E and is selectively incorporated among eight isomers that are naturally found in plants [
66].
3.7. Cluster’s Carbon Footprint Indicators
Table 6 shows the carbon footprint indicators for the different farm clusters. The energy consumed was significantly higher in cluster 1 in contrast to cluster 4, which showed the lowest value; the latter coincided with the cluster of more extensive farms where concentrate intake is lower and higher in forage and pasture (
Table 4). This is consistent with other studies [
67] where the animals that were fed a high level of concentrate had a higher energy consumption. According to Eldesouky et al. [
68], higher intensive systems have a negative impact on the environment due to greater energy requirements of livestock, as well as higher pollutant emissions, mainly from the transport of raw materials. The CH
4 levels did not show significant differences among the four described clusters. This is consistent with the results of other works [
15,
16], in which the two main emissions contributing to the carbon footprint on farms come from enteric emission and feed. The methane concerned is released as a product of enteric fermentation in the rumen, which depends on the ruminant’s own nature and not on the use of other fossil sources, which are indeed affected by the management system [
68]. CO
2 emissions were significantly lower in cluster 4, while the highest value was found in cluster 2, a result that is inversely related to grazing time. Concerning N
2O emissions, these were significantly higher in cluster 1, showing the lowest value in cluster 4. CO
2 and N
2O, together with CH
4, are the main GHG emissions [
69]. CO
2 and N
2O are mainly produced during fossil fuel combustion, coinciding with the highest values in the clusters where the level of external inputs is the greatest [
16], i.e., clusters 1 and 2. CO
2eq was significantly higher in cluster 2, in contrast to cluster 4, which had the lowest value.
The lowest carbon footprint indicators values were found on farms with a higher grazing level and higher natural resource consumption instead of external inputs, associating this type of farms with a significant reduction of GHG emissions [
17]. This is consistent with other studies where dairy production from pasture-fed animals was associated with a reduction in GHG emissions [
14]. Conversely, other authors claim that intensive farms produce lower GHG emissions than extensive farms [
70]. Regarding the carbon footprint, semi-intensive dairy farms with stabled animals had a significantly lower carbon footprint than semi-extensive dairy farms with free-grazing animals. However, considering soil carbon storage (SCS) in the carbon footprint calculations, SCS is higher on farms with higher grazing levels, consequently reducing their GHG emissions [
27]. This aspect is further supported by the work of Gutiérrez-Peña et al. [
71], where SCS was lower on intensive farms than on extensive farms.
When carbon footprint total emissions are expressed as kg of FPCM, milk production is a very influential factor, as more productive systems reduce carbon footprint. This is exactly what happened in this study, where farms belonging to clusters 3 and 4 showed significantly higher total emissions values than farms from clusters 1 and 2. This is consistent with the results found by Robertson et al. [
72] in dairy goat farms in New Zealand, where grazing goat farms had a significantly higher carbon footprint per kg of FPCM compared to intensive farms.