Big dairy data to disentangle the effect of geo-environmental, physiological and morphological factors on milk production of mountain-pastured Braunvieh cows

The transhumance system, which consists in moving animals to high mountain pastures during summer, plays a considerable role in preserving both local biodiversity and traditions, as well as protecting against natural hazard. In cows, particularly, milk production is observed to decline as a response to food shortage and climatic stress, leading to atypical lactation curves that are barely described by current lactation models. Here, we relied on five million monthly milk records from over 200,000 Braunvieh and Original Braunvieh cows to devise a new model accounting for transhumance, and test the influence of environmental, physiological, and morphological factors on cattle productivity. Counter to expectations, environmental conditions in the mountain showed a globally limited impact on milk production during transhumance, with cows in favourable conditions producing only 10% less compared to cows living in adverse conditions, and with precipitation in spring and altitude revealing to be the most production-affecting variables. Conversely, physiological factors as lactation number and pregnancy stage presented an important impact over the whole lactation cycle with 20% difference in milk production, and may therefore alter the way animals respond to transhumance. Finally, the considered morphological factors (cow height and foot angle) presented a smaller impact during the whole lactation cycle (10% difference in milk production). The present findings can help farmers to establish sustainable strategies for alleviating the negative effects of transhumance on productivity and preserving this important livestock practice.

1. Summary 1 2 The transhumance system, which consists in moving animals to high mountain pastures during summer, 3 plays a considerable role in preserving both local biodiversity and traditions, as well as protecting against 4 natural hazard. In cows, particularly, milk production is observed to decline as a response to food shortage 5 and climatic stress, leading to atypical lactation curves that are barely described by current lactation 6 models. Here, we relied on five million monthly milk records from over 200,000 Braunvieh and Original 7 Braunvieh cows to devise a new model accounting for transhumance, and test the influence of 8 environmental, physiological, and morphological factors on cattle productivity. Counter to expectations, 9 environmental conditions in the mountain showed a globally limited impact on milk production during 10 transhumance, with cows in favourable conditions producing only 10% less compared to cows living in 11 adverse conditions, and with precipitation in spring and altitude revealing to be the most production-12 affecting variables. Conversely, physiological factors as lactation number and pregnancy stage presented 13 an important impact over the whole lactation cycle with 20% difference in milk production, and may 14 therefore alter the way animals respond to transhumance. Finally, the considered morphological factors 15 (cow height and foot angle) presented a smaller impact during the whole lactation cycle (10% difference in 16 milk production). The present findings can help farmers to establish sustainable strategies for alleviating 17 the negative effects of transhumance on productivity and preserving this important livestock practice. Transhumance, which consists in moving livestock to high mountain pastures in the summer months, 22 provides both ecological and socio-cultural services to the human populations living in the mountainous 23 regions of many European countries [1][2][3]. Indeed, transhumance-annexed grazing sustains and preserves 24 endemic plant communities [4], feed local cattle to produce traditional alpine cheese, and attract many 25 tourism-related activities [5]. Further, it counteracts land abandonment in mountain areas and therefore 26 contributes preserving landscape against scrubs growth and vegetation encroachment [6], as well as natural 27 hazards such as avalanches [7] and wild fires [5]. The term "alping" (a translation of the German word 28 "Alpung" or its French equivalent "montée à alpage") will be used here to describe the approximately 100 29 days that dairy cattle spend on alpine pastures during the summer months. Similarly, animals brought to 30 mountain pastures will be referred to as "alped" cows, and the alpine summer pastures will be called 31 "alps". 32 33 Despite such ecological and social benefits, the surface dedicated to alping decreases each year (~2400 ha 34 per year [8]), and a questionnaire-based study revealed in 2010 that one third of the participating breeders 35 intend to probably abandon the transhumance practice in the following decades. In summer 2018, 107'000 36 dairy cows were alped in Switzerland during approximately 100 days [9]. A steep drop in milk production 37 is observed during this period, which hampered the evaluation of lactation curves through standard models 38 that assume a linear decrease in production [10] after the maximum milk yield is reached (i.e. ~100 days 39 after calving) [11]. Among the explanations proposed to interpret such a detrimental effect on productivity 40 are the food deficit intake due to the meagre grassland as found in high alpine pastures, as well as the need 41 to tackle environmental stress due to new and sometimes harsh habitat conditions [ Protein (kg and %), somatic cell count (1000 cells/ml), but our study specifically focused on milk 84 production in terms of quantity (milk yield). Out of the total number of records, 1,481,387 were taken in 85 the alps, whose altitude were systematically stored in the database, while their precise location were 86 documented in 95% of the cases (Fig. 1). The first record in the alp is usually taken within the first four 87 days after arrival, and is followed by three more records in the alp to encompass the entire alping period 88 (typically 100 days). Moreover, to morphologically describe animals, linear type description and 89 classification of cows are scored during the first lactation of all cows of the database. In our study we 90 considered the body height at withers and the scores (1-9) for foot angle. In addition, insemination data for 91 each lactation (date, sire's name) are also available. 92 93 A stringent data quality control procedure was applied prior to analysis to remove: 1) incomplete years 94 (which resulted in removing beginning of 2000 as well as end of 2015 due to missing lactation records); 2) 95 cows with average interval between first and last insemination longer than 100 days (as computed over the 96 first three lactations); 3) cows that had their first calf while being younger than two years, or older than 97 four years; 4) cows belonging to breeds different from the Braunvieh or Original Braunvieh; 5) cows with 98 parents other than Braunvieh or Original Braunvieh; 6) lactations shorter than 270 days; 7) lactations with 99 calving interval shorter than 290 days; 8) lactations with alps below 1100 meters above sea level (masl) or 100 above 2600 masl; 9) lactations with calving happening between March and August; 10) lactations from 101 cows that had already calved more than nine times; 11) lactations with the first record taken after the 42 nd 102 day after calving; 12) lactations with records taken before calving; 13) records taken before the 5 th day and 103 after the 500 th day after calving; 14) the second alping season (i.e. final part of lactation curves) from 104 animals that are alped twice in the same lactation.
where Y t is the observed variable (milk yield), t is the DIM, and a, b, c and k are the parameters to 180 estimate. However, k is usually set to 0.1 to make this equation linear [40]. 181 182 Here, we introduce additional terms to Eq. 3 in order to explicitly account for the transhumance effect. 183 Particularly, alping has been observed to severely affect milk production, with alped animals showing a 184 steeper linear decrease than before alping (Fig. 2). Further, alped cows usually experience a small yet rapid 185 boost shortly after their return to the lowland farm, followed by a softer decline in milk production. 186 Tacking these observations into account, we then propose to adapt Eq. 3 as follows: 187 Where t 1 is the DIM at which the cow is alped, and t 2 is the DIM at which the cow is brought back to the 192 lowland farm. Importantly, the expression d*max(0,t-t 1 ) is the expected linear decrease during alping, so 193 that the d-parameter reflects the effect of alping. The f*max(0,ceiling(t-t 2 )/305 captures the expected boost 194 in production after alping and g*max(0,t-t 2 ) represents the linear decrease in milk yield after alping; in the 195 latter arguments, the max() term ensures the model to be only affected during and after alping respectively, 196 while the ceiling expression (i.e. round to the upper integer) constructs a binary operator (0/1) to recreate 197 the instantaneous boost after the return to the lowland farm. In our case, t 1 and t 2 were determined 198 independently for each calving month. The proposed equation only works for a standard lactation period of 199 305 days. 200 201 The d-parameter enables the estimation of the loss in milk yield associated with alping over a given period 202 of time. Indeed, the amount of milk lost during alping for a period of x days can be approximated with 203 204 However, it is essential that the model fits well the beginning of the curve for this equation to work, which 207 can be achieved by artificially increasing the weight of point measurements before the transhumance. 208 Thus, weights before alping were multiplied by 100 when investigating the d-parameter depending on the 209 calving month ( Fig. 2 and 3). Furthermore, as older cows tend to calf later in the season, thereby creating a 210 correlation between lactation number and calving month, the impact of alping according to the calving 211 month is entangled with lactation number. Therefore, when examining milk production and the impact of 212 alping for each calving month, only cows in their first lactation are considered (Fig. 3). 213 214 Ordinary linear regression models were then computed in R using the lm() function of the stats package 215 [41] to estimate parameters in Eq. 4. 216 217 Measuring the effect of influencing factors 218 For sake of interpretation, all influencing factors (i.e., explanatory variables) were grouped into 219 environmental, physiological and morphological categories (Tab. 1). The effect of influencing factors was 220 tested by comparing milk records produced in conditions as dissimilar as possible. Importantly, since the 221 low number of measurements per animal imposed the use of averages, effect determination was not 222 possible through classical regression models. Consequently, groups were created according to the first and 223 third tertile of the distributions, in order to include animals from the most contrasted situations 224 (environmental, physiological and morphological) while retaining enough observations to guarantee a 225 sufficient statistical power. Since productivity is known to be optimized with mild weather conditions [34], 226 exceptions were made for THI and CSI where the second and the third tertiles were used as the two 227 contrast groups instead of the first and third tertile. 228 229 Group membership was assessed through the creation of a dummy variable assuming the value of 1 if 230 belonging to the group considered, 0 otherwise. Then, the impact of influencing factors was computed by 231 adding an interaction term to Eq. 4 that allows chosen parameters to vary as a function of the group. The 232 here defined environmental variables affect milk production during the alping stay only. Accordingly, 233 lactation curves were modelled only until the end of the alping season (meaning the f and g parameters not 234 to be estimated), with the sole d-parameter varying as a function of the group. In contrast, physiological 235 and morphological factors influence the whole lactation cycle, so that all terms of Eq. 4 (coefficients a, b, 236 c, d, f and g) are allowed to vary as a function of the group. 237 238 Within-group production was estimated both at the lowland farm and in the alps for physiological and 239 morphological factors or during alping season only for environmental factors, by integrating the area under 240 the lactation curve. The between-group difference was then assessed by computing the percentage of the 241 difference in milk production with respect to the reference group, this group being arbitrarily chosen as the 242 one with the highest milk production during alping. The difference in the d-parameter (Δd) between the 243 two groups is then also displayed to show how differently the concerned groups were impacted by alping. 244 As Lactation curve modelling 262 Overall, the proposed equation fit both the drop in milk production due to alping and the tail of the 263 lactation curve, as illustrated here for the calving months of September and February (Fig. 2). In particular, 264 the terms added to the Wilmink equation (Eq. 4) significantly increase the full model performance (p-265 value<10 -16 ). In the case of autumn calving (Fig. 1a), the proposed equation fits the entire lactation cycle. 266 For winter calving (Fig. 2b), the beginning and the end of the transhumance season appear to be the most 267 challenging periods to be fitted because of a non-linear slope. The use of Eq. 5 can be illustrated with the 268 autumn calving, with a d-parameter of -0.08, which is translated by a loss of 144 kg over 60 days. 269 270 Total milk production and milk production during alping is reported for the calving months of September 271 and February (Fig. 3). For the sake of comparison among months, only cows in their first lactation are 272 considered in this graph, as lactation number and calving month are correlated. Cows calving in autumn 273 produce on average 6033 kg during their first lactation, among which 1320 kg are produced in the alp. In 274 contrast, total milk production turns out to be lower for cows calving in winter (5155 kg during their first 275 lactation), while milk production during alping is increased (1755 kg). The d-parameters for the two 276 calving seasons being markedly different (-0.08 and -0.02 for autumn and winter calving respectively) 277 indicates that productivity is more impacted by alping when calving occurs in autumn than when occurring 278 in winter.

280
Effect of influencing factors 281 The significance of the interaction between the group variable and the d-parameter is reported (Sup. Mat. 282 S1). Hereunder, only factors with at least one calving month having a significant Δ d (i.e. a significantly 283 different impact of alping between the two contrast groups) are presented. 284 285 Among environmental conditions, THI, spring precipitation, biogeography and altitude turned out to show 286 a significant effect on milk production during alping (Fig. 4a-l). Particularly, precipitation in spring and the 287 biogeographical region showed the most important difference on milk production during alping, followed 288 by altitude and altitude difference. Further, calving period appears to interact with environmental 289 conditions, with bigger differences between groups being present in autumn. 290 291 The effect of environmental factors are small compared to those of physiological factors, where the biggest 292 effect is found for pregnancy stage for winter calving with a difference in milk production during alping of 293 20%. Although third and higher lactation cows produce more milk during the whole lactation cycle 294 including alping (Fig. 4m), they also appear to be more impacted by alping than the first lactation cows as 295 highlighted by negative Δ d-values (Fig. 4o). The influence of pregnancy stage appears to affect milk 296 production during alping, especially for cows calving in autumn (Fig 4p and 4r). Further, higher cows 297 and/or with steeper foot angle produce more milk both before and during alping than lower ones with 298 gentle foot angle (Fig. 4s, t, v, w). However, alping appears to negatively impact such cows, especially 299 higher ones (Fig. 4u and 4x). The importance of calving season 305 The proposed model succeeded in quantifying the impact of alping on milk production by assuming a 306 Wilmink pattern for cows experiencing the same conditions ( Fig.2; [40]). As expected, total milk 307 production resulted globally higher for cows with alping occurring at the end of the lactation, since the 308 drop in production happens later in the cycle. Anyway, winter calving might still be financially attractive 309 for farmers since milk produced in the alps will have a higher economic value on the market and 310 productivity will be higher during alping (Fig. 3). 311 312 Calving season also influences the way an animal is prompt to respond to environmental stress, with a 313 greater impact of transhumance (i.e. greater d-parameter in absolute value) for cows calving in autumn and 314 therefore alping at the end of their lactation cycle. Increased feed intake is known to have distinct effects 315 on milk production depending on the lactation stage [44], and from what we observe it appears that milk 316 production at the end of the lactation cycle is more sensitive to environmental changes. Similarly, when 317 studying the effect of the considered factors, we showed that the between-group difference in milk 318 production during alping is almost always greater for autumn calving. 319 320 Effect Lactation number has long been known to strongly influence milk production [49], and this also holds for 337 milk production during alping (Fig 4m-o). Even more important, pregnancy stage was found to have a 338 significant impact on milk production during alping, especially when calving occurs in autumn (Fig. 4p- performances even at alping, higher cow appear to be more impacted when moved to high mountain 350 pastures (Fig. 4s-u). As for foot angler, steep angle is associated with a smaller risk of developing hoof 351 diseases [52]. Cows with steeper foot angle were observed to produce more milk both in lowland farm and 352 during alping, but this factor appears to be have limited on the d-parameter (Fig. 4v-x). 353 354 Limitations 355 Traditionally, lactation modelling is performed on an individual basis, and usually relies on daily or weekly 356 milk records [53]. Here, we based our work on a database composed of monthly milk records, which 357 required the transformation of the data into daily averages over thousands of cows to avoid over-358 parameterisation in the model. This averaging might have diluted the strength of the effect we investigated. 359 360 Moreover, the proposed approach still misses validation, which could be achieved by relying on individual 361 observations recorded daily or weekly and belonging to different breeds from the one used here. 362 363 Next, the amount of observations among calving months was not constant in the dataset, which possibly 364 made the estimates from the winter months less robust. Further, a hidden age effect -as older cows tend to 365 calf later in the season-could have biased the observed differences in milk productions among groups. 366 367 Last but not least, the model does not explicitly take into account cow feeding during alping, which is 368 likely to affect milk production [24]. Indeed, the use of concentrate feeding varies among alps and among 369 cows of the same alp. Particularly, differences in milk yield with different calving season could be globally 370 influenced by varying concentrates feeding, with cows at an early stage in the lactation cycle -and thus 371 producing a substantial amount of milk -potentially receiving more concentrates.

401
The data was provided from the Braunvieh-CH association, under the explicit conditions that it will not be shared nor used for 402 other studies. However, a partial dataset is available with the average milk production during alping from 20000 cows, together 403 with lactation information (calving date, lactation number) and environmental data at the location of alping. Cows were chosen 404 randomly, with equal number of animals per year, lactation number and calving month. Furthermore, researchers interested in 405 performing studies on these data may contact directly the association (see contact information homepage.braunvieh.ch).

406
The code used for this article is available at https://github.com/SolangeD/lactModel 407 408    Table 1: List of factors included in the present study with supposed influence on lactation during alping. Factor-specific cut-off values are reported in the last column. These values are used to assess factorspecific effects on lactation (see Methods for an exhaustive explanation).    The effect of influencing factors is tested by investigating the difference in productivity between two groups of animals coming from contrasted conditions (first and third tertiles, except for THI where second and third tertile are chosen). Each factor is here reported in a separate column. At the top of each column, the factor name as well as the contrasted groups are reported; the group with highest milk yield during alping is chosen as the reference group, highlighted in red. In each barplot, the first bar shows the result for autumn calving, and the second for winter calving. The between-group difference in milk production during alping is displayed in the top panel, the between-group difference in milk production during the whole lactation in the intermediate panel, the change in the d-parameter at the bottom. The Δ d values are plotted in black, while grey indicates non-significance. Environmental factors affects production during alping only, making a comparison of the whole milk production redundant (which is why no graph is present in the intermediate panel of the concerned variables). To facilitate the understanding of this graph, the example of lactation (Lact #) is detailed here, where we refer to cows in their third or higher lactation as third lactation cows: third lactation cows produce 5% more milk during alping than first lactation cows when calving in autumn and even 20% more when calving in winter (m). When considering the whole lactation, third lactation cows produce 15% more milk than first lactation cows when calving in autumn and 19% when calving in winter (n). Third lactation cows calving in autumn are slightly less negatively impacted by alping (positive Δ d) than first lactation cows; an inverse behaviour is observed for winter calving, although both relationships are not significant (o). THI: Temperature Humidity Index, as averaged over 3 (THI-3d) or 30 (THI-30d) days. Prec sp: precipitation in spring. B-region: biogeographical region. Alt (diff): (difference in) altitude, lact #: lactation number, Preg: pregnancy stage, height: height at withers and ft ang: foot angle.