Matera, Roberta (2024) Evaluation of the application of Precision Livestock Farming (PLF) technologies to improve production performance in large dairy ruminants. [Tesi di dottorato]

Documento PDF PhD_Thesis_Roberta_Matera_XXXVII_cycle.pdf Visibile a [TBR] Amministratori dell'archivio Download (5MB) | Richiedi una copia

Abstract

In recent years, population growth has led to increased livestock farming intensity, impacting consumer perceptions regarding food safety, sustainability, and animal welfare. In response, Precision Livestock Farming (PLF) has emerged. The latter is a system based on advanced technologies such as sensors, real-time data analysis, predictive algorithms, and robotic milking systems, enabling continuous and automated monitoring of animal health, welfare and production. The goal is to improve efficiency and reduce environmental impact. Among the major diseases affecting dairy ruminants, mastitis is one of the most significant. It compromises animal welfare and causes substantial economic losses, making early diagnosis and effective management essential. In dairy cows, PLF technologies have already shown great potential in improving milking practices and diagnosing mastitis, with significant advantages in reducing milk losses and targeted antibiotic use. However, the adoption of these technologies presents challenges related to costs, data complexity, and the need for trained personnel. Moreover, the application of PLF in buffalo farming is still in its early stages, despite the economic importance of this species in Italy. The different morphology and behavior of buffaloes compared to cattle indeed require adaptations to existing technologies. This thesis aims to explore the implementation of PLF technologies to improve milking practices and mastitis diagnosis in dairy cattle and buffaloes. The objective was to evaluate the effectiveness of sensors and management systems to enhance farm sustainability and animal welfare, with particular attention to adapt these technologies in buffalo farming. In the first experiment, the effectiveness of measuring milk electrical conductivity (EC) was evaluated as a routine method for diagnosing mastitis in Italian Mediterranean buffalo (IMB) farms. Using a dataset of 4,530 recordings from 741 buffaloes, two mixed models were constructed to analyze the relationships between EC and milk qualitative-quantitative characteristics (MY, PP, FP) and somatic cell counts (SCC). Five EC parameters were considered, measured on the functional check day (EC_day0), three days before (EC_day3), and five days before (EC_day5). Additionally, three additional parameters were calculated from the EC recorded one, three, and five days before each check: the mean EC, the standard deviation of EC, and the slope. The results showed that all effects were significant except the effect of parity order on FP. The relationship between EC and SCS was positive and influenced by parity order, with the best results obtained using the average EC in the five days preceding the check. In the second experiment, machine learning (ML) analyses were applied to predict subclinical mastitis in 1,038 IMB from six farms, with 3,891 records. Predictive models were developed using four different ML algorithms (Generalized Linear Models-GLM, Support Vector Machines-SVM, Random Forest-RF, and Neural Network-NN) and two approaches for data set division to create a training and validation set (80% of the data) and a testing set (20%). Data division was performed by records (same animals could be included in both training and testing sets) or by animal ID (training and testing sets included different animals). The SVM model was found to be the best method for predicting high or low somatic cell values in the next check, identifying SCS, DSCC, EC, and MY as the most important variables for predicting subclinical mastitis, along with climate parameters like temperature and humidity. The highest accuracy was obtained with the NN model (76.2% accuracy and a kappa value of 0.518). The slight improvement in predictive performance observed with the data division by animal ID suggested this approach as the most appropriate in the presence of repeated measures. The results confirmed that ML methods are promising tools for improving the prevention and monitoring of subclinical mastitis, leveraging the large amount of data currently available. The third experiment aimed to identify non-genetic factors influencing DSCC, EC, and SCS, considered indirect indicators for diagnosing mastitis. The dataset used for the analysis included 14,571 test-day (TD) records from 1,501 animals from six farms and climate information from the sampling sites. The statistical model included various fixed effects: herd (6 classes), days in milk (10 classes of 30 days each, with the last class open up to 360 days), parity order (6 classes, from 1 to 6+), year-season of calving (11 classes, from summer 2019 to winter 2021/2022), year-season of sampling (9 classes, from spring 2020 to spring 2022), production level (4 classes based on quartiles of herd average milk production), and THI (4 classes based on quartiles calculated using the average temperature and relative humidity of the five days before sampling). The results showed higher EC values at the beginning and end of lactation, while SCS values were slightly lower around the lactation peak when DSCC values were lowest. Increased levels of EC, SCS, and DSCC were observed with increasing parity order. Year-season of calving and year-season of sampling had only a slight effect on the variation of the studied traits. Milk from high-production buffaloes was characterized by lower average EC and SCS values, albeit with slightly higher DSCC percentages. Buffaloes grouped in higher THI classes (classes 3 and 4) showed, on average, higher EC, SCS, and DSCC compared to lower classes, particularly class 2. The results of this study represent a preliminary yet necessary step towards the possible future inclusion of EC, SCS, and/or DSCC in breeding programs aimed at improving mastitis resistance in dairy buffaloes. The fourth experiment studied the influence of different milking parameters on the performance of IMB. Data on milking parlor settings (Milking Dry Test [MDT]) and on the test day (Test Day-TD) were collected. Each TD data included MY, FP, PP, LP, and SCS, while each MDT recorded working vacuum level (VL), pulsation ratio (PR), automatic cluster removal system (AC), and effective vacuum reserve (EVR). A total of 558 MDT and 217,967 TD were collected from 43,593 buffaloes in 198 buffalo farms. The data were analyzed using a mixed linear model and a logistic regression model to evaluate the relationship between VL and EVR. The analysis of MY and milk quality revealed that incorrect milking settings (inadequate EVR) were responsible for higher SCS and lower MY and LP, along with higher FP and PP compared to farms with adequate EVR. Except for PP, the height of the milking system significantly affected all milk parameters. Conversely, VL influenced all milking characteristics. A higher PR (70:30) was responsible for significantly lower SCS and higher FP compared to a PR of 60:40. Similarly, the presence of AC significantly affected FP and MY, with a slight reduction in SCS. Finally, a pipeline diameter and length (DL) below 2.5 m were associated with a decrease in SCS. These results suggest that dairy buffaloes require specific milking settings rather than those adapted from cattle. In the fifth experiment, improvements were assessed in an IMB farm in southern Italy that transitioned from an older automated milking system (AMS) model (Classic) to a next-generation model (VMS 300). The experiment lasted six months during which both robot models were present on the farm. A total of 315 lactating buffaloes were assigned to two homogeneous groups based on days in milk (DIM) and parity (ODP) and were milked with either the Classic or VMS 300 system. Alongside the monthly functional checks carried out by the Breeders Association, individual milk quality and quantity parameters were recorded. Daily records of functional parameters of both robots were collected, including milk flow, EC, and peak flow. Statistical analyses were performed using SPSS: the groups were compared in terms of milk quality and quantity and milking system functional parameters using analysis of variance (ANOVA, generalized linear mixed model). Based on the comparative analysis of performance between the VMS 300 and Classic automated milking systems, it was concluded that buffaloes milked by the VMS 300 system achieved significantly higher MY as well as higher actual fat and protein content compared to those milked by the Classic system. Furthermore, the VMS 300 system also showed lower milk EC and higher total peak flow. These results suggested that the VMS 300 system could be a more efficient and effective option for buffalo milking compared to the Classic system. The sixth experiment aimed to expand existing knowledge on the use of AMS in the buffalo species and evaluate the efficiency of new AMS models, focusing on milk production and quality. Two data sources were analyzed: data collected during each milking session by the AMS software over 22 months, and monthly milk productions and qualitative traits. Statistical analysis was performed with R software. A mixed repeated measures model was adopted, using DIM and parity as subject factors and visit/sampling day as repeated measures. A linear regression model was also used to study the relationship between the number of daily milkings per buffalo (as an independent variable), lactation persistence, DIM, parity, somatic cell score, and milk production (as dependent variables). The effects of parity, DIM, and their interaction were significant for all milk quality traits, except for their interaction on SCC and SCS. An average of 2.35 milkings per cow was recorded, with an average duration of 7.36 minutes. Average milk production was 9.15 kg/day, with fat and protein content of 7.97% and 4.81%, respectively, and the lactation peak was reached at 38.17 ± 1.31 DIM. Differences between multiparous and primiparous animals were evident for both milk production and energy-corrected milk. Furthermore, parity and lactation stage significantly influenced both milk flow rate (1.49 kg/min on average) and peak milk flow (2.68 kg/min on average): the latter showed a comparable trend, with higher values during the early lactation days followed by a decrease throughout lactation, along with lower milk production. The results recorded in this trial are comparable to those obtained in other studies conducted on dairy cows and dairy buffaloes with older AMS systems. It is concluded that the new AMS model represents a profitable alternative for the buffalo species, reducing labor and improving routine. The seventh experiment aimed to verify the effect of milking permission (MPE) and concentrate supplementation (CS) on milking frequency (milkings/cow/day) and MY (kg/cow/day) in a pasture-based AMS farm. Sixty-eight cows were randomly assigned to one of four homogeneous groups by ODP, DIM, and MY. The treatments were free MPE (milking allowed after 6-8 hours) or restricted MPE (milking allowed after 9.6-14 hours) and low (LC) or high (HC) CS, 0.5 kg or 3.5 kg/cow/day, respectively. The combination of the two MPE levels and the two CS levels produced the four treatment combinations (HC free [FHC], HC restricted [RHC], LC free [FLC], and LC restricted [RLC]). This study was designed as a 2 × 2 factorial design with a treatment crossover (one week of pre-treatment and four weeks of treatment; after each five-week period, groups were switched to another treatment combination until they experienced all combinations). A mixed model with repeated measures was used to evaluate the effect of MPE, CS, and their interaction on MY/cow/day, milking frequency, box time, milking time, and average milk flow rate. MY/cow/day and milkings/cow/day were significantly higher with free MPE compared to restricted MPE (1.5 kg and 0.65 milkings, respectively). MY/cow/day and milkings/cow/day were significantly higher with HC compared to LC CS (3.1 kg and 0.25 milkings, respectively). Moreover, MY/cow/day was influenced by the interaction of MPE and CS and was higher with the FHC treatment combination (20.1 kg), followed by the RHC treatment combination (18.2 kg). The number of milkings/cow/day was also influenced by the interaction of MPE and CS. The highest estimated number of milkings per cow per day was recorded for the FHC (2.12) and FLC (1.77) treatment combinations, followed by the RHC (1.38) and RLC (1.23) combinations. Similarly, the milking interval was longer by 2.5 hours for the RLC treatment combination compared to RHC. The shortest milking interval was observed for the FHC (11 hours) and FLC (12.8 hours) treatment combinations. In conclusion, the study demonstrated that allowing robot access between 6 and 8 hours after the previous milking was sufficient (even with minimal CS levels) to achieve acceptable milk production and milking performance in a pasture-based AMS. In conclusion, the research conducted demonstrated that the adoption of PLF technologies represents an effective and sustainable strategy to improve health management and productivity in both buffalo farming and pasture-based systems. These results provide a solid foundation for future research aimed at further refining these technologies, highlighting the potential of PLF technologies as innovative and non-invasive tools for the livestock farming of the future.

Downloads

Actions (login required)

Modifica documento