Engormix/Pig Industry/Technical articles
Animal Nutrition Conference of Canada 2021
The following technical article is related to the event::
Animal Nutrition Conference of Canada 2021

Basic Elements, Difficulties and Pitfalls on the Development and Application of Precision Nutrition Techniques for Smart Pig Farming

Published on: 4/25/2022
Author/s : Candido Pomar and Aline Remus / Sherbrooke Research and Development Centre, Agriculture and Agri-Food Canada, Sherbrooke, QC.
Summary

Smart farming (SF) is an emerging concept that refers to managing farms using modern information and communication technologies where interconnected smart sensors and devices, big data analysis, and conventional and artificial intelligence models will provide a new edge for innovation and productivity. SF is part of the precision livestock farming (PLF) concept that by automatically and continuously monitoring and controlling farm processes will help farmers to improve management tasks such as monitoring of animal performance and health, and optimization of feeding strategies. An important component of SF and PLF is precision livestock nutrition (PLN). PLN consists in providing in real time to individuals or group of animals the amount of nutrients that maximizes nutrient utilization and farm performance. The use of PLN in growing-finishing pig operations can decrease protein intake by 25%, nitrogen excretion into the environment by 40%, while increasing profitability by nearly 10%. The success of the development of SF and PLN depends on the automatic and continuous collection of data, data processing and interpretation, and the control of farm processes. The advancement of PLN requires the development of new nutritional concepts and mathematical models able to estimate individual animal nutrient requirements in real time. Further advances for these technologies will require the coordination of different experts (e.g., nutritionists, researchers, engineers, technology suppliers, economists, farmers, and consumers) and stakeholders. For the adoption of SF and PLN the development of integrated user-friendly systems and the end-user training is imperative. The development of PLN will not just be a question of technology, but a successful marriage between knowledge and technology in which improved and intelligent mathematical models will be essential components.

Introduction
Swine production systems have dramatically changed in the last three decades, and the livestock industry is facing many new challenges today. Among others, this industry has to continually adapt to the international trade rules uncertainties, the gradual reduction of governments support, to a society that increases its animal welfare and environmental standards, to the increase of feed and production costs, and to a decrease in the availability of common protein-providing feed ingredients. The pork industry must also adapt to new realities such as the reduction of the workforce, a workforce that is looking for working conditions more adapted to their abilities and family-work balance, and other emerging realities like the working conditions imposed by Covid19. The adoption of digital technologies can help farmers to face all these challenges.
Smart livestock farming (SLF) and precision livestock farming (PLF) are emerging concepts that refers to managing farms using modern information and communication technologies where interconnected smart sensors and devices, big data analysis, and mathematical and artificial intelligence models will provide a new edge for innovation and productivity. Although the definition of SLF and PLF may have some subtle differences for some authors, they are both used as synonymous in this document. One of the objectives of PLF systems is the on-line continuous and automatic monitoring of animals to support farmers in the management of animal production such as feeding strategies, the control of growth rate, and health management (Berckmans 2004; Maselyne et al. 2013). The main purpose of PLF is, however, to enhance farm profitability, efficiency, and sustainability (Banhazi et al. 2012a). Precision livestock nutrition (PLN) and precision feeding are considered in this document as part of the PLF approach and involves the use of feeding techniques that allow the proper amount of feed with the suitable composition to be supplied in a timely manner to a group of animals (Parsons et al. 2007; Cangar et al. 2008; Niemi et al. 2010; Pomar et al. 2014) or to individual animals within a group (Pomar et al. 2009; Andretta et al. 2014; Andretta et al. 2016; Zuidhof 2020). The on-farm application of PLF and PLN requires the design and development of measuring devices (e.g., to determine the animal’s feed intake and weight), computational methods (e.g., estimating nutrient requirements based on the actual animal’s growth and feed intake), and feeding systems able to provide the required amount and composition of feeds that will generate the desired production response.
Precision livestock nutrition can have a major impact in livestock profitability and environmental impacts. Feed is the most important cost component in commercial growing-finishing pig production systems and represents between 60 and 70% of the overall production costs (Patience et al. 2015). Similar figures hold for broilers and other livestock. Given that nutrients that are not retained by the animal or in animal products are lost, and that the efficiency by which domestic animals transform dietary nutrients into animal products is generally low, improving nutrient efficiency can largely contribute to reduce production costs and to improve the sustainability of livestock production systems. For instance, protein (i.e., nitrogen), which is among the most limiting and expensive nutrient in livestock feeds, is retained by growing pigs with efficiencies normally ranging from 15% (Flachowsky and Kamphues 2012) to 33% (Dourmad et al. 1999; Poulsen et al. 1999). Similar figures are found for converting dietary protein into meat protein in beef cattle and broilers where the efficiency ranges from 10% to 20%, and from 30% to 40%, respectively (Flachowsky and Kamphues 2012). The protein in the feed that is not incorporated into animal products is excreted in the form of N and can result in environmental problems such as nitrate pollution of surface water with problems such as algal bloom. Feed itself accounts for 70% of the environmental impact of pig and poultry production (Andretta et al. 2021). Given the challenges that are associated with the expected increase in the human population, the limited arable land, and the environmental problems that are associated with livestock production (Niemann et al. 2011) improving nutrient efficiency should be one of the first priorities for the development of sustainable livestock production systems.
Pigs, broilers, and other livestock animals are typically raised and fed in groups, usually with the same feed that is supplied to all animals in the group during long periods. The difficulty of feeding animals in groups is that nutrient requirements vary largely among animals (Pomar et al. 2003; Brossard et al. 2009) and that these requirements evolve over time following individual patterns (Hauschild et al. 2012; Andretta et al. 2014). When the objective of a commercial production system is to maximize growth, nutrients have to be provided at a level that will allow the most nutrient demanding animals in the group to express their growth potential (Hauschild et al. 2010). In this situation, almost all animals receive more nutrients than they need most of the time. Providing animals with high levels of nutrients to maximize herd performance is common practice in commercial livestock operations even though maximum growth does not ensure maximum economic efficiency (Hauschild et al. 2010; Niemi et al. 2010). Besides the estimated 33% nitrogen loss associated with digestion, maintenance, and production inefficiencies, an additional 30% is lost from the protein given in excess to optimize group production responses. Nutritionists have to include safety margins when formulating diets to account for the variability among animals but also among feed ingredients and other uncontrolled factors (e.g., environment, health) (Patience 1996). The need of these safety margins can be seen as an admission of our inability to precisely estimate the nutrient availability in feed ingredients, and the dietary nutrients concentration required to optimize herd responses. Precision nutrition will play an important role in future animal production systems because innovative monitoring approaches simplify the determination of nutrient requirements which, when estimated in real-time, allow for the possibility of feeding animals, individually or in groups, according to specific production objectives. These objectives can include optimal growth rate, minimal supply of excess nutrients and reduced environmental impacts. Safety margins are not required for PLN. Compared to a 3-phase feeding program for growing pigs, precision nutrition can reduce protein intake by 25% and nitrogen excretion by 40% while feed cost can be reduced more than 10% (Andretta et al. 2014; Andretta et al. 2016). Because animals and feed distribution are monitored and controlled automatically, PLN will reduce the time that nutritionists and farm staff will spend on animal observation, decision-making, and applying production strategies, enabling them to work on other aspects of farm management. The objective of this document is to describe the basic concepts of precision nutrition, its essential elements, and difficulties and pitfalls encountered in the development and application of PLN.
The Basic Concepts of Precision Livestock Nutrition
Precision livestock nutrition concerns the use of feeding techniques that provide animals with diets tailored according to the production objectives (i.e., maximum or controlled production rates), including environmental and animal welfare issues. Precision livestock nutrition is presented in this document as the practice of feeding individual animals, or groups of animals, while accounting for the changes in nutrient requirements that occur over time, and for the variation in nutrient requirements that exists among animals (Pomar et al. 2019). The accurate determination of available nutrients in feed ingredients, the precise diet formulation, and the determination of the nutrient requirements of individual animals or groups of animals should be included in the development and use of PLN systems (Sifri 1997; Van Kempen and Simmins 1997; Pomar et al. 2009; Pomar et al. 2017). The operation of precision nutrition in commercial farms requires the integration of three types of activities: 1) automatic collection of data, 2) data processing, and 3) actions concerning the control of the system (Berckmans 2004; Banhazi et al. 2012a; Pomar et al. 2019). Application of precision nutrition at the individual level is only possible where measurements, data processing, and control actions can be applied to the individual animal (Wathes et al. 2008).
Automatic data collection. Measurements on the animal, the feeds, and the environment are essential for PLN and these parameters have to be measured directly and frequently (if possible, continuously). In fact, we cannot manage and control a production system without appropriate measurements. Essential measurements for PLN in growing pig operations include feed intake and body weight of individuals or group of pigs. The availability and the rapid development of new devices and emerging sensor technologies for PLF and PLN offer a great potential for farm monitoring and management. Available technologies and sensors listed in the literature include ear plastic button tags containing passive transponders (RFID) or visual image analysis for individual animal identification, low-cost cameras which, in combination with image analysis, can be used to quantify animal behaviour and estimate body weight (Wathes et al. 2008; Banhazi et al. 2012a; Halachmi et al. 2019). Real-time sound analysis and audio-visual observations have been proposed to monitor health status and welfare in pigs (Maselyne et al. 2013; Vranken and Berckmans 2017; Norton et al. 2019) and behaviour in laying hens (Berckmans 2004).
Data processing. Collected data has to be processed to establish optimal farm control strategies. There are several potential control strategies available for the application of PLN in livestock operations. In animals offered feed ad libitum, nutrient intake can only be controlled by varying the composition of the served feed. In this situation, both the between-animal and the overtime nutrient requirements variation can be controlled. In contrast, in animals that are offered feed restrictively the amount and the composition of the feed can be controlled (Pomar et al. 2019).
Mathematical modelling is a methodology used to understand and to quantify complex biological phenomena involved in animal production and it is today the basis for data processing for PLN. Mathematical models developed for PLN, however, have to be designed to operate in real-time using real-time system measurements. Therefore, they are structurally different from traditional nutrition models, which are developed to work in a retrospective manner and to simulate known production situations. The first mathematical model developed to estimate in real-time individual pigs nutrient requirements was proposed by Hauschild et al. (2012). The required daily concentration of lysine is estimated in this model using individual feed intake and body weight information. Using these data, an empirical model component estimates the expected body weight, feed intake, and weight gain for the starting day, whereas a mechanistic model component uses these three estimated variables to calculate with a factorial method the optimal concentration of lysine that should be offered that day to each pig of the herd to meet its requirements. Other amino acids and nutrient requirements are assumed proportional to lysine requirements using an ideal protein concept (Remus et al. 2019b; Remus et al. 2020d).
There are several potential control strategies available for PLN in farm operations. Feed restriction has been often proposed to control growth and body composition in growing pigs (Niemi et al. 2010) and poultry (Cangar et al. 2008; Zuidhof et al. 2017) or to monitor real-time individual pigs’ feed efficiency in finishing pigs (Peña Fernández et al. 2019). Nutrient requirements are in this situation assumed related to the expected body weight gain (Pomar et al. 2017). Nonetheless, PLN is in full development today and new control approaches are being proposed. We should be expecting important developments with new data-driven modelling methods in which machine learning and deep learning algorithms combined with conventional mathematical models will be used to efficiently examine farm data patterns and produce accurate predictions (Ellis et al. 2020).
Control of the system. The information collected and processed in farms with PLN systems is used to control the production system. In the context of PLN, automatic precision feeders are used to provide individual pigs with the right amount and composition of the feed at any given time. At least 2 feeds (named A and B) are needed for PLN. These 2 feeds should be formulated on the basis of net energy, standardized ileal digestible amino acids and other essential nutrients. Feed A (high nutrient density feed) is formulated for the most demanding pigs at the beginning of the growing period, whereas feed B (low nutrient density feed) is formulated for the less demanding pigs at the end of the finishing period. Blending feeds A and B at different proportions allows the feeders to provide individual pigs with the feed that has the right concentration of nutrients (Pomar and Remus 2019). The feeders consist of a single space trough in which precision Archimedes's screw conveyors deliver and blend simultaneously volumetric amounts of two feeds contained in independent feed containers. The feeder identifies each pig when their head is introduced into the feeder and the feeds are blended and delivered upon the animal request according to the estimated optimal lysine concentration. A serving is composed of the amount of feed delivered upon each effective serving request. A time lag is imposed to ensure that pigs eat each serving before requesting a new one. Serving size is progressively increased and ranges between 15 and 25 g (Pomar et al. 2011; Andretta et al. 2014). A meal includes all the servings delivered during each feeder visit. Pigs tend to leave the feeder trough empty or leave very small amounts of feed after each visit, thus ensuring that each pig receives the assigned amount of blended feed.
The Impact of Using Precision Nutrition in Growing-Finishing Pig Operations
In growing animals, maintenance and production requirements change over time and so do nutrient requirements (NRC 2012). Farm animals are normally raised in groups and they all eat the same feed, although within the group, they have different nutrient requirements due to their differences in body weight, feed intake and growth rate (Pomar et al. 2003; Wellock et al. 2004; Brossard et al. 2009). The dynamic (i.e., time-dependent) and the between-animal variation are the two main sources of variation in nutrient requirements that can be controlled in PLN systems. Thus, real-time modelling-control approach was used by Pomar et al. (2014) to control the time-dependent variation of group-housed pigs offered feed ad libitum. Moving from the conventional three-phase feeding system to the daily-phase feeding systems in growing-finishing pig operations, these authors demonstrated that protein intake can be reduced by 7% while nitrogen excretion by 12%. Controlling the time-dependent and the between animal variation can further help the reduction of nutrient intake and excretion. The modelling approach proposed by Hauschild et al. (2012) was used to estimate real-time nutrient requirements in individual pigs were calibrated in two animal trials (Zhang et al. 2012; Cloutier et al. 2015) and the overall approach of estimating real-time amino acid requirements was challenged in 2 validation trials (Andretta et al. 2014; Andretta et al. 2016). The latter authors showed that daily adjustment of the diet resulted in a 27% reduction in total lysine supply, without detrimental effects on growth. This additional 20% reduction in lysine intake in relation to group-fed pigs could be obtained by feeding the animals individually and thus controlling simultaneously the time-dependent and the between-animal variation. Although feed cost reduction depends to a great extent on feed prices, it is expected that feed cost can be reduced by 1-3% when only controlling the time-dependent variation while an 8-10% reduction can be obtained when controlling both sources of variation.
Restricting feed or nutrient intake has been proposed in several PLN systems with the objective to minimize feed cost, ammonia emissions, or to maximize the return per pig space. Demmers et al. (2012) used an automated feeding system to provide the desired amount of feed of fixed composition to each pen. Daily body weight was estimated using a commercial visual image analysis system. The controller was based on a recursive neural network of growth and ammonia emission models, which were calibrated from previous experiments. The system was used to control the amount of feed delivered to pens and the ambient temperature to optimize growth and reduce ammonia emissions. A PLN system was also used by Niemi et al. (2010) to study multiphase and two phases feeding systems and growth patterns in terms of economic return per pig space. In this multi-phase feeding system, the amount of feed, the protein concentration in the diet, and the time to reach slaughter weight were optimized on a daily basis. The controller included a stochastic dynamic model that estimated nutrient requirements as a function of body weight and evaluated the different scenarios to maximize the return on capital investment. The authors concluded that producers would benefit from adjusting diet composition on a daily basis but that the optimal production strategy and the return on investment are affected by the variation among pigs and the variation in feed and carcass prices.
A real-time system for the integrated control of population pig growth and pollutant emissions was also proposed (Whittemore et al. 2001; Parsons et al. 2007) using an automatic daily feed intake recording device and a visual image analysis system to estimate daily body weight (Schofield et al. 1999; White et al. 2004). Pigs were fed ad libitum in this precision feeding system with diets varying in crude protein concentration. A high and a low-protein diets were manually blended to obtain the desired level of protein in served feed. The authors concluded that weight gain in pigs can be controlled through the proposed ad libitum precision feeding system and that some control of body fatness may also be possible.
Difficulties and Pitfalls on the Development and Application of Precision Nutrition Techniques
Precision livestock nutrition is an important component of PLF and the successful on-farm application of PLN will face similar challenges than the development of many other PLF system components. Precision livestock farming is as an embryonic technology with great promise but few PLF have been implemented successfully so far. There are important challenges than developers will face to successfully develop PLN systems in commercial farms.
The utilization of mechanistic mathematical models in PLN systems has been criticized because these models are overly complex and the required information to simulate practical conditions is not always available (Aerts et al. 2003; Wathes et al. 2008; Peña Fernández et al. 2018; Norton et al. 2019). The simplicity of empirical models is counteracted by the difficulty to represent interactions between nutrients and animals. Despite the fundamental structural differences between empirical and mechanistic models in the way they predict the response of animals to nutrient supply, both types of models have to be calibrated a priori using data collected from reference populations (Pomar et al. 2015) in which the phenotypic performance potential of the animal is quantified. Indeed, mechanistic growth models for pigs use intrinsic characteristics of a reference population, either to describe the potential (phenotypic) protein deposition and feed intake patterns (Dourmad et al. 2008; van Milgen et al. 2008; NRC 2012) or potential body protein and lipid deposition (Emmans 1981; Black et al. 1986; Ferguson et al. 1994; Wellock et al. 2004) while empirical models have the animal responses embedded into the model. To be used for PLN, empirical and mechanistic models are, therefore, challenged by the difficulty of identifying the right reference population for its calibration, the fact that actual populations and individual animals may follow feed intake and growth patterns different than the ones observed in the reference population (Pomar et al. 2015).
The computational power and reliability of modern information technologies empower the utilization of advanced recursive technologies in the development of PLF and PLN applications (Wathes et al. 2008; Ellis et al. 2020). These modelling techniques (e.g., artificial neural networks) estimate unknown model parameters of an abstract mathematical model, based on on-line input and output measurements. Model parameters are estimated on-line during the process, resulting in a model that continuously adapts its response to on-line process inputs and outputs. There are few examples in which these models have been used in PLF or PLN applications (Thomson and Smith 2000). The limitation of using the recursive approach in PLN is related to the fact that model parameters and model structure do not provide biological insight in the causal mechanisms implicated in animal responses, that animal response and input parameters may have unsymmetrical variation, and that the animal responses to input variation does not evolve in the same timeframe (Pomar et al. 2019). For example, when animal processes are modelled for which there is a significant time lag between the effects of varying input parameters (e.g., dietary lysine intake) and the response (e.g., body weight gain and composition), the autocalibration capability of these recursive models is limited and they will generate irregular control signals (Cangar et al. 2008). Rapid animal responses such as a behavioural response to inputs such as temperature and light intensity may be easily controlled by recursive models in PLF applications (Aerts et al. 2000).
There may be a long time path between experimental development and commercial application. For example, the milking robot was developed in the 80s and has been commercialized since the early 90s but, despite 25 years of availability, it is today revolutionizing the dairy industry. In an article with the provocative title “Is precision livestock farming an engineer’s daydream or nightmare, an animal’s friend or foe, and a farmer’s panacea or pitfall?”, Wathes et al. (2008) discussed the development and adoption of PLF systems. Others (Groot Koerkamp et al. 2007; Banhazi et al. 2012a; Banhazi et al. 2012b) have expanded on these ideas and the main issues in the development and successful adoption of PLF (and thus PLN) can be summarized as suggested by Pomar et al. (2019) as follows:
  • There is a strong need for coordination and to involve different experts and stakeholders (researchers, engineers, technology suppliers, economists, farmers, consumers, and citizens) in the development and practical implementation of PLF and PLN systems.
  • With the rapid development of sensors, more focus should be given to data interpretation and automatic control mechanisms than to sensor utilization.
  • The benefits of PLF systems should be verified on the farm.
  • Appropriate deployment of PLF systems and training, service and support for farmers should be assured. The latter may imply the development of a new service industry. As indicated by Banhazi et al. (2012b), farmers are biologists by nature and only technologists occasionally. Although they do invest in technology, it is typical machinery that they look forward to buying as opposed to software, sensors or services.
An increasing concern is the adaptability and training required by farmers using PLF systems. Some authors (Van Hertem et al. 2017) believe that use of appropriate data visualization tools can facilitate the farmer acceptance and adoption of PLF applications. These authors tested and evaluated PLF systems on ten fattening pig farms and five broiler farms. Data of production, climate and behaviour was continuously measured, analyzed daily and made available on a web-based tool. Nearly 50% of the farmers took the training, but only 28% of the trained farmers actively used the tool. According to the authors, the success of the training seemed to be dependent on the complexity of the system installed on the farm (e.g., environmental sensors) and the training/education of the end user. They conclude that training is fundamental for the adoption of such systems.
Future Perspectives
We should be expecting that PLN will have major impact in livestock profitability and sustainability. Indeed, the nutrients consumed which are not retained by the animal or in animal products are excreted, and therefore, improving the efficiency of nutrient utilization can greatly contribute to reducing production costs and improving the sustainability of animal production systems. Several strategies to develop PLN in livestock have been developed, but the most promising ones are those that can control de variation in nutrient requirements that occurs over time and between the animals. To further develop PLN systems, it is necessary to improve our actual understanding of several animal metabolic processes. Precision livestock nutrition is still based on mathematical models and nutritional concepts developed to optimize population responses. When feeding individual animals with daily tailored diets, these traditional nutritional concepts are not accurate and even sometimes incorrect (Remus et al. 2020a; Remus et al. 2020c; Remus et al. 2020b). It is necessary to distinguish the nutritional requirements of a population from those of an individual (Pomar et al. 2003; Brossard et al. 2009). Individual animals are able to modulate growth and the composition of growth according to the level of available amino acids (Remus et al. 2019a). Also, pigs can respond differently to the same amount of ingested amino acid, due to differences in the efficiency of utilization. These aspects are not considered in current nutritional models, which assume that the efficiency by which animals use the available amino acids is constant. Similarly, the amino acid composition of whole body protein is assumed to be constant as well, whereas it has been shown that it can vary. Understanding the metabolic processes responsible for the observed variation between individual animals in their ability to use dietary nutrients is challenging nutritionists and modellers but is required to further improve the efficiency of livestock production. Advances in PLN rely on the development of sound nutritional concepts and comprehensive biological models developed to more precisely estimate individual real-time nutrient requirements. The new understanding of individual metabolism and nutrition will allow animal science to move forward, opening up new opportunities for individual nutrition. This will ultimately enhance farm profitability, efficiency, and sustainability of the overall production system (Banhazi et al., 2012b). Precision livestock farming, or smart farming, as well as its components as PLN, should not be seen as just being a question of technology, but a successful marriage between knowledge and technology. For PLN, the synergy between mathematical mechanistic and data-driven models (Ellis et al. 2020) will play an essential role in this marriage that will drive the next revolution in livestock farming.
       
Presented at the 2021 Animal Nutrition Conference of Canada. For information on the next edition, click here.

Bibliographic references

 
Author/s :
 
Views53Comments 0StatisticsShare