Animal activity to improve the welfare and energy and productive efficiencies in intensive production buildings of piglets

TESE DE DOUTORAMENTO ANIMAL ACTIVITY TO IMPROVE THE WELFARE AND ENERGY AND PRODUCTIVE EFFICIENCIES IN INTENSIVE PRODUCTION BUILDINGS OF PIGLETS Roberto Besteiro Doval ESCOLA DE DOUTORAMENTO INTERNACIONAL PROGRAMA DE DOUTORAMENTO EN ENXEÑARÍA PARA O DESENVOLVEMENTO RURAL E CIVIL LUGO 2018 AUTORIZACIÓN DOS DIRECTORES DA TESE Animal activity to improve the welfare and energy and productive efficiencies in intensive production buildings of piglets D. M. Ramiro Rodríguez Rodríguez, en calidade de titor e director e Dna. M. Dolores Fernández Rodríguez en calidade de directora INFORMAN: Que a presente tese, correspóndese co traballo realizado por D. Roberto Besteiro Doval, baixo a miña dirección, e a utorizo a súa presentación, considerando que reúne os r equisitos esixidos no R egulamento de Estudos de Doutoramento da USC, e que como director desta non incorre nas causas de abstención establecidas na Lei 40/2015. De acordo co artigo 41 do Regulamento de Estudos de Doutoramento, declara tamén que a presente tese de doutoramento é idónea para ser defendida en base á modalidade de COMPENDIO DE PUBLICACIÓNS , nos que a participación do/a doutorando/a foi decisiva para a súa elaboración. A utilización destes artigos nesta memoria, está en coñecemento dos coautores, tanto doutores como non doutores. Ademais, estes últimos teñen coñecemento de que ningún dos traballos aquí reunidos poderá ser presentado en ningunha outra tese de doutoramento. En Lugo, ..... de ......... de 2018 Asdo.................................. Asdo.................................. DECLARACIÓN DO AUTOR DA TESE Animal activity to improve the welfare and energy and productive efficiencies in intensive production buildings of piglets D. Roberto Besteiro Doval Presento a miña tese seguindo o procedemento adecuado ao Regulamento, e declaro que: 1) A tese abarca os resultados da elaboración do meu traballo. 2) No seu caso, na tese faise referencia ás colaboracións que tivo este traballo. 3) A tese é a versión definitiva presentada para a súa defensa e coincide coa versión enviada en formato electrónico. 4) Confirmo que a tese no incorre en ningún tipo de plaxio de outros autores nin de traballos presentados por min para a obtención doutros títulos. En Lugo, ..... de ......... de 2018 Asdo: Roberto Besteiro Doval Agradecementos institucionais Os traballos realizados para completar esta tese de doutoramento foron financiados polo Goberno da Xunta de Galicia a través do “Programa de consolidación e estruturación de unidades de investigación competitivas” (GPC2014/072). O autor agradece tamén á Consellería de Cultura, Educación e Ordenación Universitaria da Xunta de Galicia a concesión da axuda de apoio á etapa predoutoral no marco do “Plan galego de investigación, innovación e crecemento 2011-2015 (Plan I2 C)”. Doutra banda, agradecer á Universidade de Santiago de Compostela, e en especial ao Grupo de Investigación GI-1720 “Modelización, Enerxía e Mecanización en Biosistema” (BioMODEM), a cesión dos seus espazos e equipamentos para poder levar a cabo a presente tese. Agradecementos persoais O remate dunha tese non é soamente a presentación dun documento científico, non se trata tan so da consecución dun novo título académico. Rematar unha tese vai moito máis aló, é por fin a un período cheo de momentos inesquecibles, de recordos e experiencias que de seguro enriquecen a quen decide apostar por este camiño. Pero por sorte, un final soamente significa un novo comezo. Sen lugar a dúbidas, nesta nova etapa sería todo un orgullo poder contar con toda a xente que en maior ou menor medida me aportou o seu gran de area para alcanzar os meus obxectivos. Refírome a xente indispensable como os meus directores, compañeiros e amigos, a Dra. M. Dolores Fernández e o Dr. M. Ramiro Rodríguez. Ro-Ro e Dolores, parella inquebrantable que tanto loitou e seguen a loitar polos que dunha forma ou outra caemos nas vosas mans. Grazas por tanto esforzo e dedicación, polos consellos e conversas que axudaron sobre todo a facer máis levadeira e amena esta viaxe. Pero o traballo nese despacho non sería o mesmo sen a amizade de Tamara Arango. Desa xente que sabes que che porán a túa disposición toda a súa experiencia, que sabes que te comprenderá e te apoiará. Dese tipo de xente que todos debemos coidar de ter ao noso carón. Tampouco me quero esquecer de todos os que fostes aparecendo neste proceso. Falo de xente como Marcos e Cándido, sempre dispostos a colaborar e facilitar o noso traballo na explotación. Xente como Carmen Muíños ou Ismael Moeda, compañeiros e axuda indispensable na etapa experimental. Grazas especiais aos membros da Aarhus University, ó Dr. Guoqiang Zhang, Dr. Hao Li, Qianying Yi e Xiaoshuai Wang, pola acollida no voso grupo e os consellos compartidos. Por suposto, grazas aos compañeiros de Nørresø Kollegiet, sobre todo a Romain, Silvestre e Pablo, pero tamén a Pedro Paulo, Daniela e Vianney por facilitarme a convivencia e facerme máis entretida a miña estadía. Do mesmo xeito, ás persoas da Escola Superior Agraria de Ponte de Lima que me recibistes como un integrante máis da vosa familia, Dr. José P. Araújo e Dr. Joaquim Lima, moitas grazas pola vosa xenerosa hospitalidade. E por suposto enormemente agradecido por ambos os métodos, con valores do Coeficiente de Correlación de Concordancia de 0,86 e correlación de Spearman de 0,90. Con todo, a precisión non foi constante ao longo de ciclo, senón que esta foi maior ao principio do mesmo, cando os animais eran máis pequenos e lixeiros. Posteriormente procedeuse a realizar unha análise espectral da serie completa de actividade mediante a Transformada Rápida de Fourier e a Transformada Wavelet Continua. Deste proceso confirmouse a influencia dos distintos factores ambientais e de manexo sobre a actividade animal, como os períodos de luz/escuridade ou o horario de repartición de alimento. Ademais, establecéronse unha serie de ecuacións sinusoidales que definen a actividade dos leitóns. En xeral, durante o ciclo os animais mostraron dous picos de actividade ás 10:00 e ás 18:00 h, cun período de repouso nocturno entre as 22:00 e as 07:00 h. Con todo durante a primeira metade do ciclo detectouse un patrón de comportamento máis simple, cun único pico de actividade, debido principalmente á menor competencia polo espazo no curral. Finalmente, estes patróns de actividade foron empregados nun modelo híbrido que combinou a Transformada Wavelet Discreta con Redes Neuronais Artificiais, denominado modelo WaveletNeural Network. Os resultados deste modelo, empregado para a predición da concentración de CO2 no aloxamento, demostran que a inclusión da actividade xunto coa temperatura da sala e a temperatura exterior predín o seguinte valor de concentración cunha raíz do erro cuadrático medio de 154 ppm. Ademais detectouse un retardo de case unha hora entre o momento de actividade e a súa resposta na concentración do gas. Os mellores resultados predictivos alcanzáronse a través dun modelo autorregresivo, que emprega os valores pasados da serie de CO2, indicativo da elevada inercia desta variable. A modo de conclusión confirmouse que o sistema de medición da actividade a través de sensores de infravermello pasivo é un método fiable e económico que permite rexistrar os patróns de comportamento dos animais. Estes patróns, ou os propios valores de actividade, poden ser empregados en modelos predictivos do clima en aras a lograr un control máis eficiente dos sistemas de climatización. Resumen En la actualidad el sector ganadero se enfrenta a dos grandes retos: alcanzar una producción más respetuosa con el medio ambiente y menos contaminante y asegurar unas condiciones de bienestar animal en el alojamiento, sin olvidarse de su finalidad productiva y económica. Una de las maneras más eficaces de abordar esta doble pretensión en las explotaciones intensivas es a través de los sistemas de climatización y control ambiental. En este sentido, y gracias al desarrollo del Big Data y de conceptos como el Smart Farming, surgen nuevos mecanismos que pueden ser empleados como una herramienta indispensable para lograr los mencionados objetivos. Entre estas nuevas herramientas se incluyen nuevos algoritmos de control de los sistemas de climatización que permiten la inclusión de distintas variables. Y precisamente, de entre la enorme variedad de variables que se podrían emplear dentro de una explotación ganadera, la actividad o el comportamiento de los animales es una de las más prometedoras. Esto es así porque permite integrar en una única variable aspectos relativos al bienestar animal y a la calidad del aire del alojamiento. En esta tesis, presentada bajo la modalidad de compendio de publicaciones, se persigue conseguir una mejora en el bienestar animal y la eficiencia energética y productiva de las explotaciones de ganado porcino en fase de transición. Para ello se evalúan las posibilidades de integración de la variable actividad animal en los incipientes sistemas de control ambiental predictivos. En el desarrollo de este trabajo, y para alcanzar los objetivos deseados, se registraron distintas variables ambientales y de carácter animal en una sala de una explotación de lechones destetados de 6 a 20 kg durante dos ciclos completos de cría. Se almacenaron en un colector de datos valores diez-minutales de hasta 30 variables, entre las que destacan la temperatura, humedad, concentración de CO2, velocidad del aire o la actividad animal. Es precisamente la actividad animal, variable de estudio en esta tesis, para la cual se estableció un método de medición robusto, fiable y de bajo coste a través de sensores de infrarrojo pasivo. Se comprobó la validez de estas mediciones mediante su comparación con una observación directa de los animales. Se pudo confirmar la concordancia entre los valores registrados por ambos métodos, con valores del Coeficiente de Correlación de Concordancia de 0,86 y correlación de Spearman de 0,90. Sin embargo, la precisión no fue constante a lo largo de ciclo, sino que ésta fue mayor al principio del mismo, cuando los animales eran más pequeños y ligeros. Posteriormente se realizó un análisis espectral de la serie de actividad mediante la Transformada Rápida de Fourier y la Transformada Wavelet Continua. De este proceso se confirmó la influencia de los distintos factores ambientales y de manejo sobre la actividad animal, como los períodos de luz/oscuridad o el horario de reparto de alimento. Además, se establecieron una serie de ecuaciones sinusoidales que definen la actividad de los lechones. En general, durante el ciclo los animales mostraron dos picos de actividad a las 10:00 y a las 18:00 h, con un período de reposo entre las 22:00 y las 07:00 h. Sin embargo, durante la primera mitad del ciclo se detectó un patrón de comportamiento con un único pico de actividad, debido principalmente a la menor competencia por el espacio en el corral. Finalmente, estos patrones de actividad fueron empleados en un modelo híbrido que combinó la Transformada Wavelet Discreta con Redes Neuronales Artificiales. Los resultados de este modelo, empleado para la predicción de la concentración de CO2 en el alojamiento, muestran que la inclusión de la actividad junto con la temperatura de la sala y la exterior predicen el siguiente valor de concentración con una raíz del error cuadrático medio de 154 ppm. Además se detectó un retardo de casi una hora entre el momento de actividad y su respuesta en la concentración del gas. A pesar de todo, los mejores resultados predictivos se alcanzaron con un modelo autorregresivo, que empleaba valores pasados de la serie de CO2, indicativo de la elevada inercia de esta variable. A modo de conclusión se confirmó que el sistema de medición de la actividad a través de sensores de infrarrojo pasivo es un método fiable y económico que permite registrar los patrones de comportamiento de los animales. Estos patrones, o la propia actividad, pueden ser usados en modelos predictivos del clima en aras a lograr un control más eficiente de los sistemas de climatización. Index CHAPTER I: BACKGROUND ...........................................................................................1 1. INTRODUCTION ............................................................................................................ 3 1.1 Socioeconomic context ..................................................................................... 3 1.2 Pig sector .......................................................................................................... 5 2. JUSTIFICATION OF THE THESIS ........................................................................................ 10 2.1 Environmental impact of swine production .................................................... 10 2.2 Animal welfare in the pig sector ..................................................................... 12 2.3 Environmental control in pig farms ................................................................ 15 2.4 Animal activity as an indicator ....................................................................... 17 3. THEMATIC COHERENCE OF THE THESIS ............................................................................. 18 4. OBJECTIVES ............................................................................................................... 20 CHAPTER II: MATERIALS AND METHODS ................................................................... 23 1. LIVESTOCK BUILDING ................................................................................................... 25 2. MEASURED VARIABLES ................................................................................................. 26 3. TECHNIQUES USED ...................................................................................................... 29 CHAPTER III: AGREEMENT BETWEEN PASSIVE INFRARED DETECTOR MEASUREMENTS AND HUMAN OBSERVATIONS OF ANIMAL ACTIVITY ................................................. 31 1. INTRODUCTION .......................................................................................................... 33 2. MATERIAL AND METHODS............................................................................................. 34 2.1 Animal and husing .......................................................................................... 34 2.2 Passive infrared detector measurements ........................................................ 34 2.2.1 Data acquisition ................................................................................ 34 2.2.2 Sensor location .................................................................................. 15 2.3 Human observation ........................................................................................ 35 2.4 Statistical analysis .......................................................................................... 35 3. RESULTS .................................................................................................................... 36 4. DISCUSSION ............................................................................................................... 36 5. CONCLUSIONS ............................................................................................................ 37 CHAPTER IV: ESTIMATION OF PATTERNS IN WEANED PIGLETS' ACTIVITY USING SPECTRAL ANALYSIS ....................................................................................... 39 1. INTRODUCTION .......................................................................................................... 41 2. MATERIAL AND METHODS ............................................................................................ 42 2.1 Animal and husing .......................................................................................... 42 2.2 Animal activity measurements ........................................................................ 43 2.3 Data analysis .................................................................................................. 43 3. RESULTS ................................................................................................................... 43 4. DISCUSSION .............................................................................................................. 45 4.1 Evolution of animal activity during the cycle ................................................. 45 4.2 Periodicity of animal activity ......................................................................... 45 4.3 Wavelet analysis ............................................................................................. 46 5. CONCLUSIONS ........................................................................................................... 47 CHAPTER V: PREDICTION OF CARBON DIOXIDE CONCENTRATION IN WEANED PIGLET BUILDINGS BY WAVELET NEURAL NETWORK MODELS .............................................. 49 1. INTRODUCTION .......................................................................................................... 51 2. MATERIAL AND METHODS ............................................................................................ 52 2.1 Experimental test ............................................................................................ 52 2.1.1 Animals and housing ......................................................................... 52 2.1.2 Input variables for the model ............................................................ 52 2.2 Wavelet neural network .................................................................................. 52 2.2.1 Data Pre-processing .......................................................................... 52 2.2.2 Artificial neural network ................................................................... 53 2.2.3 Wavelet transform .............................................................................. 53 2.2.4 Design of the model ........................................................................... 53 2.3 Model performance evaluation ...................................................................... 53 3. RESULTS ................................................................................................................... 54 3.1 Description of the cycles ................................................................................ 54 3.2 Selection of variables ..................................................................................... 54 3.3 WNN model building ...................................................................................... 55 4. DISCUSSION .............................................................................................................. 55 5. CONCLUSIONS ........................................................................................................... 56 CHAPTER VI: GENERAL DISCUSSION .......................................................................... 59 CHAPTER VII: CONCLUSIONS ..................................................................................... 65 1. CONCLUSIONS ........................................................................................................... 67 2. FUTURE WORKS ......................................................................................................... 68 CHAPTER VIII: REFERENCES ....................................................................................... 69 Chapter I: Background ROBERTO BESTEIRO DOVAL 8 Figure 1: Distribution of other pigs by type of pig farm. Small fatteners: no sows and fewer than 10 other pigs; Large fatteners: no sows and at least 400 other pigs; Large breeders: at least 400 pigs and 100 sows. Source: Marquer, Rabade and Forti (2014). This trend towards specialization continues to this day, since between 2007 and March 2018 the number of total farms decreased by 13%. Specifically, the number of small farms, according to RD 324/2000 (less than 4.8 UGM), were reduced by 45%. Meanwhile, the largest holdings belonging to group 3 (between 360 and 864 UGM) grew by 43% in the same period, although they still represent only 2.3% of the total. This process of concentration occurs in farms of an intensive nature. However, during the last years, there is an increase in the number of extensive farms located especially in Andalusia and Extremadura. The specialization of the sector was the great bet of the Spanish meat industry, partly pushed by the reduced profit margins of the products. This situation forced the producers to reduce production costs, mainly by increasing the size of the farms. As can be seen in Figure 2, the production costs per kg of carcase were progressively reduced throughout the historical series, while the purchase prices CHAPTER I: BACKGROUND 9 continued to fluctuate in a cyclical manner with a global trend to the downside. For example, the crisis in the sector in 2015, which lasted until mid-2016, accelerated the closing of small farms and the least efficient ones. The year 2017 was a relatively good year of prices, although in the first quarter of 2018 the profit margins returned to negative values. As a consequence, in order to increase the viability of the farms, it is required an improvement in their efficiency, by reducing costs mainly related to food, prophylaxis and energy. Figure 2: Profit margin of Spanish pig farms. Source: Subdirección General de Productos Ganaderos (2018b). ROBERTO BESTEIRO DOVAL 10 2. JUSTIFICATION OF THE THESIS 2.1 Environmental impact of swine production The foreseeable increase in the demand for food products will bring with it an increase in pressure on natural resources, from deforestation of lands for cultivation to excessive use of water or environmental pollution. Currently, the livestock sector has already a high impact, since it uses 35% of the farmland and 20% of the water available in the soil (Macleod et al., 2013). In 2015, agriculture generated 426473 kilotonnes of CO2 equivalent of greenhouse gases other than CO2 itself, representing 10% of total EU emissions. Although at the European level, emissions have been reduced by 20% since 1990, Spain was along with Cyprus, the only member states that increased their emissions, mainly due to the growth of the bovine, pig and poultry sectors (EUROSTAT, 2016). On the other hand, in the EU in 2015, agriculture was responsible for 94% of the ammonia emissions (3751 kilotons). Being this gas a precursor of the formation of aerosols in the atmosphere. Livestock is the source of 39% of global emissions (Galloway et al., 2004). Once again, there has been a reduction in emissions of 24% since 1990 at European level, except Spain, which increased its amount of ammonia emissions by 12%. The global estimates of emissions of greenhouse gases in the pig sector are 9% of the emissions of the livestock sector, which places the sector as the second largest livestock originator of emissions, well behind the bovine. (EUROSTAT, 2016). The analysis of the life cycle in the pig sector chain indicates that animals are the main contributors to gas emissions (62.1%), compared to transport, distribution and commercialization (14.4%) and to consumers (23.5%). Of the activities related to animal production, the production of the concentrates is the main responsible for the emissions, followed by the handling of the slurry (Noya et al., 2017). It is precisely the anaerobic degradation of organic matter by bacteria, both in the digestive tract and in the slurry, the main source of methane (EUROSTAT, 2016). For nitrous oxide, it is originated only from slurry, and its formation occurs during an incomplete nitrification/denitrification processes performed by microorganisms that normally convert NH3 into non-polluting molecular nitrogen (N2) (Philippe and Nicks, 2015). CHAPTER I: BACKGROUND 11 Similarly, pig farms are responsible in Europe for 25% of NH3 livestock emissions. This gas comes mainly from the rapid hydrolysis of urea in the urine and faeces of animals. The hydrolysis of urea is catalysed by the urease enzyme, originating carbonic acid and ammonia (Sigurdarson, Svane and Karring, 2018). Ammonia emissions are concentrated mainly in livestock housing (50%) and at the time of distribution of slurry (30%). The fattening farms are responsible for the greatest amount of emissions (70%), while the sows and the piglets produce 20% and 10% of the emissions respectively (Philippe, Cabaraux and Nicks, 2011). The formation and emission of these gases in farms is influenced by numerous factors, such as the type of facilities, management, indoor climate or animal activity. Blanes-Vidal et al. (2008), found a high correlation between ammonia emissions and daily animal activity of pigs, in part due to the relationship of this with the air flow ventilation and partly to the urinating behaviour of animals. Jeppsson (2002) also found a correlation between temperature and animal activity with NH3 and CO2 emissions, in such a way that the increase in ammonia emissions was exponentially related with the increase in temperature. Other factors, such as high air velocity over the surface of the pit, increase the transfer rates of gases to the environment (Blanes-Vidal et al., 2012). Therefore, in a farm, several actions can be carried out to reduce the emission of harmful gases. One of them is to optimize the use of the food, adjusting the composition of the diets to the specific requirements of the animals. The improvement of slurry management systems by separating the solid and liquid phase or the use of acidifiers can contribute also to this reduction. The reduction in energy consumption from non-renewable sources contributes positively too. Another action that should be implemented to reduce polluting emissions by pig farms is the establishment of an adequate air conditioning system. The equipment involved must be coordinated properly avoiding energy losses. These systems must also adapt their operation to the conditions of the pen, controlling environmental parameters dynamically according to the real needs of the animals. Even though these actions may have a limited direct impact on the production of polluting gases, they do have on the emission patterns. So much so that, for example, a greater air flow ventilation brings with it an increase in emissions ROBERTO BESTEIRO DOVAL 12 (Hamon, Andrès and Dumont, 2012). In addition, proper climate regulation has benefits on the health, performance, welfare and behaviour of the animal, causing an indirect effect on the level of emissions. For example, a bad thermal regulation will originate a response in the animals altering their preferences for the excreting and lying areas (Haeussermann, Hartung and Jungbluth, 2005). This relationship can reach such a point that, at high temperatures, the dirt of the pens increases linearly with the temperature, while increasing the level of air pollution (Aarnink et al., 2006). There are recent emission mitigation techniques that consist in the extraction of the air from the head of the pits. This way, up to 43% of the ammonia emitted can be retained and prevented from spreading through the housing (Saha et al., 2010). With this system, it is possible to extract air with concentrations up to 20 and 30 times higher than those achieved with ventilation in the room, especially in ammonia and amines, even using low ventilation flows (Van Huffel et al., 2016). In this way the air quality in the housings is improved, and allows a more efficient treatment of the extracted air by means of filtering techniques with the aim of reducing its polluting power (Hansen et al., 2012). 2.2 Animal welfare in the pig sector Joined with the reduction of the environmental impact of the farms, the protection of animals and their welfare will be one of the main challenges for the sector. In addition to the growing and restrictive legislation on animal welfare, market pressure has guied, and will continue to do so, the transformations that the sector must adopt. This is reflected in the 2015 Special Eurobarometer "Attitudes of Europeans towards Animal Welfare", which shows that 94% of Europeans agree with the idea that it is important to protect the welfare of farm animals (57% they consider it very important). Even 82% of European citizens consider that it would be necessary to increase this protection. Finally, 59% say they are willing to pay an extra price for products derived from production systems that respect welfare. Aware of the progress of these mindset, the European Commission has been collaborating with Member States for over 40 years with the intention of improving the living conditions of farm animals. In 1974, the first legislation on animal welfare in slaughterhouses took place. In 1998, the Council Directive CHAPTER I: BACKGROUND 13 98/58/EC was passed, which legislated the basic matters for the protection of intensive livestock. The concept of animal welfare on which this legislation is based and which is widely accepted as a definition is the so-called "Five Freedoms": 1. Freedom from hunger and Thirst; 2. Freedom from discomfort; 3. Freedom from, pain, injury or disease; 4. Freedom to express normal behaviour; 5. Freedom from fear and distress. The Treaty on the Functioning of the EU, in its Article 13, also includes the need for the Union and the Member States to take into account the requirements regarding the welfare of animals during the planning and application of policies. The welfare of animals in swine production is currently ensured by the Council Directive 2008/120/EC, addressing issues such as the characteristics of facilities, spaces, feeding, management or training of operators. In Spain, the current legislation on animal welfare in the pig sector is regulated by the RD 1135/2002, amended by RD 1392/2012 in order to adjust it to the new regulations. All these regulations place special emphasis on the control of the indoor environment and air quality, as well as the availability of space for animals. The provision of materials that allow them expressing their behavioral needs and the limitation of actions on animals that may affect their welfare, especially mutilations (castration or tail docking), are also regulated. These conditions and requirements are especially compromised in the current intensive systems. Of the so-called "Five Liberties", several studies have focused on the influence of their limitation on the welfare of pigs and therefore on the productive performance of the same. The deprivation of food or water has a direct and clear effect on the welfare and performance of animals. Although it is not the interest of any farmer to subject their animals to hunger or sedentary periods, these situations can appear in cases of strong social competition, weak healthy condition, extreme environmental conditions or even associated with abrupt changes such as the weaning period (Quiniou et al., 2001; Dybkjaer, Jacobsen and Togersen, 2006). The absence of diseases, pains or injuries also has a direct ROBERTO BESTEIRO DOVAL 14 impact on productive performance and animal behavior, from the reduction of mortality ratios to better food indexes (Escobar et al., 2007, Pastorelli et al., 2012). To continue with, having the ability to express pig natural behaviors, such as digging into the ground, reduces injuries and aggressive behavior among animals (Nannoni et al., 2016, Bracke, 2017), thus achieving better welfare and social cohesion. In this case, good management of the enrichment objects is of great importance in pigs (Scott et al., 2006, Da Silva, Manteca and Dias, 2016). On the other hand, the breeding of animals under conditions of continuous stress or fear, such as those created by hostile management, can have repercussions on reproductive and productive parameters (Sommavilla, Hötzel and Dalla Costa, 2011). Finally, creating the right conditions of comfort, providing the animals with an adequate environment with areas of refuge and rest, seems like a key element to ensure welfare situations in animals. Here, the design of the facilities and the management of the environmental parameters will define the achievement or not of an adequate well-being. It is logical to think, for example, that very high temperatures or humidities will reduce the comfort of pigs, and therefore these will increase the response mechanisms against stress situations, mechanisms that normally go through alterations in the normal behaviors (Collin et al., 2001; Debreceni et al., 2014). Since animal welfare is a multidimensional concept, which includes both physical and mental aspects, the determination of the level of well-being is a task that can involve a high degree of subjectivity. This is because the importance given to different aspects of well-being may be different for each person. This is why there are numerous works that attempt to make an overall assessment of well-being based on objective measurements (Welfare Quality®, 2009, Jongman, Hemsworth and Skuse, 2014, Brandt et al., 2017), based mainly on indicators of animal origin and not exclusively in the facilities or in the environment as it had traditionally been done. Therefore, it is the animal itself who should tell us about the welfare state through measurements of its behavior. However, these indices are calculated based on stady observations, it would be interesting to make valuations for longer periods of time. Thus, together with the climate data of the buildings, information such as the consumption of water and feed or the levels of animal activity, would play a fundamental role in that determination. CHAPTER I: BACKGROUND 15 2.3 Environmental control in pig farms As is evident in the previous sections, the environmental conditions of the livestock buildings and their management have direct repercussions, both on the levels of polluting emissions and on the welfare of the animals. Consequently, it influences the productive performance of farms. This statement is especially important in intensive farms with forced ventilation systems, which will be the type of farm object of study in this doctoral dissertation. Therefore, ensuring adequate environmental conditions according to the needs of the pigs at all times is a priority task. The control techniques for these variables depend on the air conditioning systems available on the farm: natural or forced ventilation, hot water or electric heating systems, cooling systems and even air filtering equipments. Logically, it is easier to achieve better control in those buildings where there are forced ventilation equipment accompanied by heating and/or cooling systems. Even so, the key part of these systems is the controller that must coordinate and manage the different facilities. This control can be done in different ways, from the classic techniques of "local control" such as the allor-nothing systems or the popular PID control algorithms (Proportional, Integral and Derivative), to more sophisticated "supervisory control" techniques such as Fuzzy Logic Control or Predictive Control (Model Predictive Control, MPC) (Afram and Janabi-Sharifi, 2014). Due to the relative simplicity of the air conditioning systems used in pig farms, PID control is the most widespread method among current controllers, since it allows basic control and automation functions that largely satisfy the objectives (Dong and Lam, 2014). On the other hand, the industry was traditionally reluctant to adapt complex control methods, it is also for this reason that most air conditioning systems still employ very simple controllers (on/off, PID) instead of more modern controllers like the MPC (Killian and Kozek, 2016). It is precisely the MPC, together with artificial intelligence, one of the most promising techniques due to its ability to integrate the rejection of disturbances, the management of constraints, a dynamic control and strategies of energy conservation in the approach of the controller (Afram and Janabi-Sharifi, 2014). Research has shown that the use of an MPC controller in a plant can mean from 7% to 50% savings in energy consumption with respect to the most basic ROBERTO BESTEIRO DOVAL 16 controllers. The MPC control algorithm is based on the use of a system model to predict its future state, and then generate control vectors that minimize a certain cost function over the prediction horizon, taking into account the presence of restrictions and disturbances (Afram et al., 2017). In the case of livestock facilities, these restrictions could be related to the reduction of the level of emissions, an increase in thermal comfort and a reduction in energy consumption. Figure 3: Diagram of Model Predictive Control. Source: Afram et al. (2017). In all predictive control algorithms, identification and modeling are the most expensive and determinant steps of the process. Achieving reliable predictions are crucial for good MPC performance. These models must meet basic conditions that are: simplicity, good estimation of system dynamics and good predictive properties (Prívara et al., 2013). The need for specific models for each system is probably the reason why the MPC is not yet an algorithm widely adopted by the industry. The so-called black-box models, or "data-driven models", can greatly simplify this process. These models are developed by in situ measurements of the necessary inputs and outputs of the system and the subsequent adjustment of a CHAPTER I: BACKGROUND 17 mathematical function to estimate the functioning of the system (Killian and Kozek, 2016). Therefore, the research efforts for the effective development of predictive control algorithms in the primary sector should focus on the design of simple, versatile and precise models, as is already happening in other fields (Liang and Du, 2005; Huang, Chen and Hu, 2015; He et al., 2016). In this sense, in the livestock sector, it would be especially useful to include exogenous variables of animal origin, such as activity levels, water and feed consumption, weight or age of the animals. In this way, the control algorithms could be able to understand the state of the animal at any time. Thus, they would adapt their responses to the needs of the livestock, without doing that only based on a set of ventilation instructions at the beginning of the production process. 2.4 Animal activity as an indicator Animal activity, understood as the movement of animals, varies throughout the day (de Sousa and Pedersen, 2004, Blanes and Pedersen, 2005, Schauberger et al., 2013). This, together with the animal weight and temperature, constitutes one of the main responsible for the modification of the pen environmental conditions due to its influence on the production of CO2 or NH3, dust, heat, humidity, etc. Due to the aforementioned relationship with the health status and welfare of the animals (Broom and Corke, 2002; Huynh et al., 2005), the environment of the buildings (Brown-Brandl et al., 2004; Pedersen, Jørgensen and Theil , 2015), and with the emission of polluting gases (Blanes-Vidal et al., 2008, Estellés et al., 2010, Ngwabie, Nimmermark and Gustafsson, 2011), activity or animal behavior is one of the most promising variables for inclusion in the climate control systems of livestock housing (Youssef, Exadaktylos and Berckmans, 2015). Despite these efforts, few studies have set their objective to define the physical activity of the animal, even knowing that its behavior follows biological cycles of around 24 hours (Villagrá et al., 2007), including periods of rest and activity. In the last one, the movement of animals is not constant, instead of that, it presents peaks of activity, usually influenced by environmental factors, management, health, age or social behavior (Quiniou et al., 2001, Pedersen, Jørgensen and Theil, 2015). CHAPTER II: MATERIAL AND METHODS 25 1. LIVESTOCK BUILDING All the measurements and analyzed parameters in this doctoral thesis were taken during two breeding cycles in a swine farm of weaned piglets from 6 to 20 kg located in the northwest of the Iberian Peninsula. Specifically, the farm is located in Toques, A Coruña. It has a maximum capacity of 5971 piglets of 20 kg, distributed in two contiguous buildings in "L" form and with five and three rooms of weaning respectively. The edifices have a total dimensions of 69.8 x 14.5 m2 and 31.9 x 14.9 m2 respectively, following a similar interior arrangement both of them, with a lateral corridor of 1.2 m wide with access to the successive weaning rooms. Figure 4: Aerial view of the piglet farm where the measurements were taken. The configuration of the rooms is similar in all of them, and as described in depth in Chapters III, IV and V, each one has two corridors that serve four corrals on each side, with PVC dividing panels and a fully polypropylene slatted floor. Each room contains 16 pens of 3.2 x 3.2 m2, with maximum capacity for 50 piglets each. In addition, each room is equipped with an air extraction system and a heating system. The renewal of air is done from the side windows placed on both sides of the rom. The air exit the room through a chimney arranged in the central ROBERTO BESTEIRO DOVAL 26 area of the chamber. The air inlet and outlet sections are regulated automatically. A heating plate of hot water (3.20 x 0.50 m) under a PVC roof placed at a height of 0.70 m composes the pen heating system. Piglets were fed ad-libitum with compound feed. Water was supplied at a nipple drinker placed close to the feeder. In addition, the farm has other additional facilities such as: 1. double perimeter closure, 2. changing rooms, showers and toilets, 3. drive-through disinfection bath 4. four vertical feed storage silos of polyester, 5. groundwater collection point with chlorination system, 6. warehouse of products and office, 7. outer slurry storage pit. 2. MEASURED VARIABLES In this farm, measurements were made of different parameters, both environmental and animal, that were used to a different extent in this PhD thesis. Table 1 shows the variables measured together with their position, measurement equipment and periodicity of the measurements. In total information was recorded of two complete weaning cycles of 40 (June 9 - July 19) and 38 (February 27 - April 6) days of duration. The external conditions were also measured. A total of 33 environmental and animal variables were measured. From the set, 30 were measured continuously, making measurements every second and storing the ten-minute average. The feed intake was obtained daily, and the two remaining variables, body mass of the animals and height of the level of slurry in the interior pit, were measured weekly. CHAPTER II: MATERIAL AND METHODS 27 Figure 1: Weaning room for piglets where measurements were made with the location of the sensors used. a) air inlet 1 (AI 1); b) PIR and Video Camera; c) animal zone 1 (AZ 1); d) animal zone 2 (AZ 2); e) air inlet 2 (AI 2); f) air outlet by chimney located at 2.8 m height (AO). Between the 30 variables recorded, 23 of them were taken inside the building. Specifically, the measurements were made in a farm room in three different positions: the animal zone, differentiating two subzones within it, one below the climatic cover (AZ2), and another outside the cover (AZ1); in the entrances of fresh air from the lateral corridor of access to the rooms (AI1) and from the outside (AI2); and in the air outlet of the room (AO). To protect the sensors installed in AZ1, they were placed inside a cage and at a height of 20 cm from the ground. All ten-minute measurements were stored in two data collectors, HOBO® and Campbell Scientific Ltd. CR-10X. The remaining 7 variables correspond to data from the outside of the farm (EX), recorded by a weather station located on the plot itself. ROBERTO BESTEIRO DOVAL 28 Table 1. Measured variables with their position, equipment and measurement range. Variable Position Model Range CO2 concentration AI 1 AZ 1 AO Delta Ohm HD37BTV.1 with infrared tecnology (NDIR) with double wave length 0 – 5000 ppm Temperature Relative Humidity AI 1 AI 2 AZ 1 AZ 2 AO ONSET® S-THB-M002 -40 – 75ºC 0 – 100% NH3 concentration AZ 2 Murco © MGS 150. Electrochemical sensor. 0 – 100 ppm Air speed AZ 1 Delta Ohm HD103T.0 omnidirectional hotwire probe. 0,08 – 5 m/s Animal activity 1,3 (2 pulses in 20 s) Animal activity 2,4 (4 pulses in 20 s) AZ 1 AZ 2 AZ1 AZ2 OPTEX© RX-40QZ. Passive infrared detector. Cover area 12 x 12 m2, with 78 subzones Water intake Feeder water intake - Elster M170 con pulse emitter Retrofit M170. 0 – 3 m3/h Temperature Relative humidity Wind speed Wind direction Gust speed Pressure Solar radiation EX HOBO® Weather Station equiped with S-THB-M002, SWSB-M003, S-WDA-M003, MCAA, RS3-B. -40 – 75ºC 0 – 100% 0 – 76 m/s 0 – 355º Noise AZ 1 Cirrus Research plc CR:822B 21 dB – 140 dB Lighting AZ 1 Nesa srl Lux-A 0,02 – 2 klux Body mass - Digital scales MI2000 0 – 500 kg Feed intake - Manual - Slurry level - Manual - CHAPTER II: MATERIAL AND METHODS 29 It is outside the object of this thesis the analysis of all the variables collected during the trial. However, a great majority of them were taken into account in the initial processes of characterization of the cycles, in the search for interactions between variables and in the analysis prior to the publications finally achieved. As a result of this work and other complementary works, other publications arose parallel to the development of this document. These are three articles in popular science magazines of the sector, one in Portugal, and six publications in national and international congresses. 3. TECHNIQUES USED Throughout the Chapters of this document, different techniques of analysis of the collected data series will be used, whose basis is detailed in the corresponding sections. However, since it is outside the scope of this doctoral thesis, there is no in-depth development of the different methods. The following is a brief enumeration of the techniques used, their purpose within each work and the bibliography consulted. Chapter III deals with the validation of the PIR measurement system in front of a human observation of the animals. For validation, only statistical techniques were used in order to analyze the concordance between the data obtained by the two measurements. Human observation was taken as a reference method. The statistical techniques used, Spearman's rho correlation coefficient (ρ), the Concordance Correlation Coefficient (CCC) and Bland Altman's graphic method are perfectly defined in the Material and Methods section of the mentioned chapter (Bland and Altman, 1986, Lin, 2016). In Chapter IV, in which the animal activity patterns were established, two different techniques of spectral analysis were used, the Wavelet Continuous Transform (CWT) and the Fast Fourier Transform (FFT). Through the CWT the series of animal activity in the continuous time-frequency spectrum was analyzed. This way, their behavior was studied throughout the entire cycle. Numerous technical documents were consulted to gain an in-depth knowledge of the fundamentals of this analysis, among which we can highlight: (Marchant, ROBERTO BESTEIRO DOVAL 30 2003, Neild, McFadden and Williams, 2003, Singh et al., 2010, Rosch and Schmidbauer, 2014). The FFT for its part was used to obtain the sinusoidal equations that define animal activity patterns, consulting more theoretical information about it (Bloomfield, 2000, Wilks, 2006, Shumway and Stoffer, 2011). Finally, in Chapter V, a hybrid model was designed to predict the concentration of CO2 in the livestock buildings. For this, a previous filtering technique was combined with a predictive model of the branch of artificial intelligence. Specifically, the Discrete Wavelet Transform (DWT) and the Artificial Neural Networks (ANN) were employed. The previous filtering of the series used in the hybrid model was done by the DWT. It differs from the CWT in the way of discretizing the scale parameter, in a manner that the CWT allows a more detailed analysis than the DWT, although more expensive (Nason, 2006; Liu, 2010; Joo and Kim, 2015). Finally, the ANN takes advantage of the filtering done by the DWT on the data series to predict the CO2 concentration. Thus, achieving the socalled Wavelet-Neural Network (WNN) hybrid model. (Rojas, 1996, Nourani et al., 2014, Alizadeh and Kavianpour, 2015). To carry out the aforementioned data processing and analysis, a computer with an Intel® Core ™ i3-2100 processor of 3.10 Ghz, with 4 GB RAM memory and Windows 10 operating system was used. In addition, a free version of the RStudio software, an integrated development environment for the programming language R, was used as a basic support for this thesis. It was also used the license of the University of Santiago de Compostela for the use of commercial software Microsoft® Office (MO Word, Excel and Access) and the license managed by the Spanish Foundation for Science and Technology (FECYT) for access to the Web of Science and SCOPUS platforms. Finally, the free Mendeley bibliographic management tool was also used. Chapter III: Agreement between passive infrared detector measurements and human observations of animal activity Chapter IV: Estimation of patterns in weaned piglets' activity using spectral analysis ROBERTO BESTEIRO DOVAL 62 2014). Also, new algorithms have been developed to estimate the body mass or water consumption (Kashiha et al., 2013, 2014). The system generates a series of ten-minute data that defines the total time of activity recorded in that period. The analysis of this data series again justifies the validity of the method through the activity patterns obtained, which are very similar to others described in the literature for the same species (CIGR, 2002, Villagrá et al., 2007; Schauberger et al., 2013). This analysis reflects the influence of weaning on animals, which show high levels of activity during the first five days post-weaning, caused by the stress generated by the separation from the mother, the change of the diet type or the establishment of new social hierarchies. After this initial stage, activity levels stabilize until the last third of the cycle. In this last period, the activity is progressively reduced again, due to the increase in the kg m-2 ratio, and the consequent reduction of the available space per animal. Knowledge of the levels of animal activity throughout a whole cycle is of interest from the point of view of animal welfare, or also for the early detection of diseases in situations of abnormal behavior (Kristensen and Cornou, 2011), as Madsen and Kristensen (2005) has already done with water intake. But the greatest interest lies in the analysis of daily behavior. The decomposition in sinusoidal waves of the daily behavior of the animals. In that sense, using the FFT, reflects the influence of the periodicities of 24, 12 and 8 hours in the composition of a pattern with two peaks of daily activity, at 10:00am and 6:00 pm, and a clear resting period from 10pm to 7am. These periodicities correspond to the day/night cycles (Debreceni et al., 2014) in the case of the 24-hour harmonic; with the influence of the midday high temperatures in the case of the harmonic of 12 h (Costa et al., 2014); and with the food distribution schedules for the third harmonic. This reflects the importance of three basic factors that define the activity, the circadian rhythms of the animal, the climate and the animals' management by the farmer. But this pattern with two peaks was not constant during the whole cycle. Instead, CWT has shown that during the first half of the weaning stage, the animals show only one peak of daily activity, and two peaks in the second half cycle. This fact has a multifactorial explanation, such as the greater availability of space during the first weeks together with a greater CHAPTER VI: GENERAL DISCUSSION 63 attitude towards the game (Spoolder et al., 2012), or a greater sensitivity to high temperatures when animals are younger (Bracke, 2011 ). Obtaining sinusoidal activity equations allow the definition of daily animal behavior patterns, susceptible to be used in environmental control algorithms or real-time management systems. For example, for the prediction of the CO2 concentration inside livestock buildings, the activity plays a fundamental role since it is the main source of its generation (Zong et al., 2014). In the WNN hybrid model established in this thesis, animal activity was selected as a predictive variable by the algorithms of variable selection. Specifically, the activity values were selected with a delay of 40 and 50 minutes for the prediction of the next CO2 value. This fact shows an inertia of approximately one hour in the concentration of this gas. That is, current animal activity will have an impact on the gas concentration one hour later. In addition to the animal activity, the model also takes into account the current and the outdoor temperature with one, two and five delays. Aslo, the current and 20 minutes earlier temperature in the animal zone were chosen. The importance of the outdoor temperature in this CO2 prediction model is related to its close connection with the ventilation flows, responsible for renewing the air (Jeppsson, 2002). In the case of the temperature in animal zone, it can also be explained by the ventilation flows, in addition to the influence on the slurry emission ratio (Ni et al., 1999). Both variables have a more immediate effect on air quality, symptomatic of an efficient ventilation. The predictive results of this model, which employs DWT as a filter, are very promising, with a Pearson correlation of 0.89 and root mean square error (RMSE) of 154 ppm. However, due to the high autoregressive condition of the CO2 series, the best predictive results are achieved with a WNN hybrid model in which the current and immediately previous value of the series itself is used, as well as the current outside temperature. With this model, the predictive results improve up to a correlation of 0.99 and RMSE of 26 ppm. This model surpasses the performance of the predictive model of Sun et al. (2008) for air quality in pig farms. The good performance of this hybrid model lies in the pretreatment of the data series with the DWT and the robustness of the ANNs. The decomposition of series into different subseries allows the model to cope with greater guarantees of success to non-stationary or trend series (Nourani et al., 2014). Chapter VII: Conclusions CHAPTER VII: CONCLUSIONS 67 1. CONCLUSIONS The work developed in this doctoral thesis allows expanding the knowledge of the animal activity in the weaned piglet’s farms. It had been taken as starting point the developed and the validation of the performance of the measurement system, PIR sensors. Then it developed activity patterns based on a long period of measurement, analysing the factors and causes of animal behaviour. Finally, the possibilities of including animal activity in the incipient control systems were evaluated, that it could be able to include the animal status in the decision making process. The specific conclusions that it can be drawn from this thesis are: 1. The use of a system for measuring animal activity based on the use of PIR sensors allows precise data of animal behavior, in a simple, economical and robust way. Even so, as happens with other methods, the accuracy is slightly affected in conditions of high activity and high density of animals. 2. The data obtained through the PIR detectors allow to establish generic sinusoidal patterns of behavior, with two peaks of activity and a night rest period. However, parameters such as lack of space, thermal requirements or the animals' age made the behavior of the animals change throughout the cycle studied. It evolved from the initial purely sinusoidal daily patterns, with a single peak of activity, to the patterns, at the end of the cycle, with two daily maximums. 3. The inclusion of the animal activity variable in predictive models of the indoor environment of the livestock housing helps to improve its performance, as demonstrated in the hybrid model Wavelet - Neural Network hybrid developed of CO2 prediction. It was also found that the effect of the activity did not immediately affect the CO2 concentration in the livestock housing, contrary the influence was perceived almost an hour later. Even so, due to the high inertia of the gas in the housing, the autoregressive hybrid model was the one that obtained the best predictive results. ROBERTO BESTEIRO DOVAL 68 2. FUTURE WORKS Given the good response of the PIR measurement system, and the mentioned importance of animal activity in intensive farms, it is proposed to continue the work done in this thesis through the development of predictive climate models, trying to get more predictions reliable in the long term and adaptable to changing conditions. Once this has been achieved, the inclusion of these models in climate control algorithms such as Model Predictive Control (MPC). This fact will allow to optimize the use of energy, reduce emissions to the environment and maximize environmental comfort, also, it will help the development of controllers more efficient and effective in livestock buildings. In addition, more emphasis should be placed on the need of including in these controllers the status and actual requirements of livestock. This can be achive using animal variables such as animal activity, health status, age, water or feed intake or animal weight. Regarding this topic, an opportunity to design and implement systems of early disease detection through the analysis of the animal activity disturbances is also opened. Chapter VIII: References CHAPTER VIII: REFERENCES 71 Aarnink, A. J. A. et al. (2006) ‘Temperature and body weight affect fouling of pig pens’, Journal of Animal Science, 84(8), pp. 2224–2231. doi: 10.2527/jas.2005521. Afram, A. et al. (2017) ‘Artificial neural network (ANN) based model predictive control (MPC) and optimization of HVAC systems: A state of the art review and case study of a residential HVAC system’, Energy and Buildings. Elsevier B.V., 141, pp. 96–113. doi: 10.1016/j.enbuild.2017.02.012. Afram, A. and Janabi-Sharifi, F. (2014) ‘Theory and applications of HVAC control systems - A review of model predictive control (MPC)’, Building and Environment, 72, pp. 343–355. doi: 10.1016/j.buildenv.2013.11.016. Alizadeh, M. J. and Kavianpour, M. R. (2015) ‘Development of wavelet-ANN models to predict water quality parameters in Hilo Bay, Pacific Ocean’, Marine Pollution Bulletin. Elsevier Ltd, 98(1–2), pp. 171–178. doi: 10.1016/j.marpolbul.2015.06.052. Bland, J. M. and Altman, D. G. (1986) ‘Statistical methods for assessing agreement between two methods of clinical measurement’, Lancet, 327, pp. 307– 310. doi: 10.1016/S0140-6736(86)90837-8. Blanes-Vidal, V. et al. (2008) ‘Emissions of ammonia, methane and nitrous oxide from pig houses and slurry: Effects of rooting material, animal activity and ventilation flow’, Agriculture, Ecosystems and Environment, 124(3–4), pp. 237– 244. doi: 10.1016/j.agee.2007.10.002. Blanes-Vidal, V. et al. (2012) ‘Emissions of NH3, CO2and H2S during swine wastewater management: Characterization of transient emissions after air-liquid interface disturbances’, Atmospheric Environment, 54, pp. 408–418. doi: 10.1016/j.atmosenv.2012.02.046. Blanes, V. and Pedersen, S. (2005) ‘Ventilation flow in pig houses measured and calculated by carbon dioxide, moisture and heat balance equations’, Biosystems Engineering, 92(4), pp. 483–493. doi: 10.1016/j.biosystemseng.2005.09.002. ROBERTO BESTEIRO DOVAL 78 Prívara, S. et al. (2013) ‘Building modeling as a crucial part for building predictive control’, Energy and Buildings, 56, pp. 8–22. doi: 10.1016/j.enbuild.2012.10.024. Puppe, B., Schön, P. C. and Wendland, K. (1999) ‘Monitoring of piglets’ open field activity and choice behaviour during the replay of maternal vocalization: A comparison between Observer and PID technique’, Laboratory Animals, 33(3), pp. 215–220. doi: 10.1258/002367799780578110. Welfare Quality ®. (2009) Welfare Quality ® Assessment protocol for pigs, Welfare Quality® Assessment Protocol for Pigs. Quiniou, N. et al. (2001) ‘Modelling heat production and energy balance in group-housed growing pigs exposed to low or high ambient temperatures’, British Journal of Nutrition, 85(01), p. 97. doi: 10.1079/BJN2000217. Rojas, R. (1996) Neural Networks. A systematic Introduction, Springer. doi: 10.1016/0893-6080(94)90051-5. Rosch, A. and Schmidbauer, H. (2014) WaveletComp : A guided tour through the R-package. Saha, C. K. et al. (2010) ‘Effects of a partial pit ventilation system on indoor air quality and ammonia emission from a fattening pig room’, Biosystems Engineering, 105(3), pp. 279–287. doi: 10.1016/j.biosystemseng.2009.11.006. Schauberger, G. et al. (2013) ‘Empirical model of odor emission from deep-pit swine finishing barns to derive a standardized odor emission factor’, Atmospheric Environment, 66, pp. 84–90. doi: 10.1016/j.atmosenv.2012.05.046. Scott, K. et al. (2006) ‘Influence of different types of environmental enrichment on the behaviour of finishing pigs in two different housing systems’, Applied Animal Behaviour Science, 99(3–4), pp. 222–229. doi: 10.1016/j.applanim.2005.10.013. Shumway, R. H. and Stoffer, D. S. (2011) Time series analysis and its applications. Third Edit. Springre. doi: 10.1007/978-1-4419-7865-3. CHAPTER VIII: REFERENCES 79 Sigurdarson, J. J., Svane, S. and Karring, H. (2018) ‘The molecular processes of urea hydrolysis in relation to ammonia emissions from agriculture’, Reviews in Environmental Science and Biotechnology, 17(2),pp.241–258. doi:10.1007/s11157-018-9466-1. Da Silva, C. A., Manteca, X. and Dias, C. P. (2016) ‘Needs and challenges of using enrichment materials in the pig industry’, Semina:Ciencias Agrarias, 37(1), pp. 525–536. doi: 10.5433/1679-0359.2016v37n1p525. Singh, C. B. et al. (2010) ‘Wavelet analysis of signals in agriculture and food quality inspection’, Food and Bioprocess Technology, 3(1), pp. 2–12. doi: 10.1007/s11947-008-0093-7. Sommavilla, R., Hötzel, M. J. and Dalla Costa, O. A. (2011) ‘Piglets’ weaning behavioural response is influenced by quality of human-animal interactions during suckling’, Animal, 5(9), pp. 1426–1431. doi: 10.1017/S1751731111000358. de Sousa, P. and Pedersen, S. (2004) ‘Ammonia emission from fattening pig houses in relation to animal activity and carbon dioxide production’, Agricultural Engineering International: the CIGR Journal of Scientific Research and Development, VI(Manuscript BC04 003), pp. 1–13. Spoolder, H. A. M. et al. (2012) ‘Effect of increasing temperature on space requirements of group housed finishing pigs’, Applied Animal Behaviour Science, 138(3–4), pp. 229–239. doi: 10.1016/j.applanim.2012.02.010. Subdirección General de Productos Ganaderos, D. G. de P. y mercados A. (2018a) El Sector de la Carne de Cerdo en Cifras. Principales Indicadores Económicos 2017. Subdirección General de Productos Ganaderos, D. G. de P. y mercados A. (2018b) Informe Trimestral Indicadores de Porcino. Available at: https://ec.europa.eu/agriculture/marketobservatory/meat_en%0Ahttps://ec.europa.eu/agriculture/sites/agriculture/files/ market-observatory/meat/pigmeat/doc/dashboard-pig_en.pdf. ROBERTO BESTEIRO DOVAL 80 Sun, G. et al. (2008) ‘Forecasting daily source air quality using multivariate statistical analysis and radial basis function networks’, Journal of the Air and Waste Management Association, 58(12), pp. 1571–1578. doi: 10.3155/10473289.58.12.1571. United Nations (2017) ‘World population rospects: The 2017 revision’, Volume I:, pp. 1–377. Available at: https://esa.un.org/unpd/wpp/Publications/Files/WPP2017_VolumeI_Comprehensive-Tables.pdf. Villagrá, A. et al. (2007) ‘Modelling of daily rhythms of behavioural patterns in growing pigs on two commercial farms’, Biological Rhythm Research, 38(5), pp. 347–354. doi: 10.1080/09291010600950131. Wilks, D. S. (Department of E. and A. S. C. U. (2006) Statistical methods in the atmospheric sciences, International Geophysics Series. doi: 10.1002/met.16. Youssef, A., Exadaktylos, V. and Berckmans, D. A. (2015) ‘Towards real-time control of chicken activity in a ventilated chamber’, Biosystems Engineering, 135, pp. 31–43. doi: 10.1016/j.biosystemseng.2015.04.003. Zong, C. et al. (2014) ‘Carbon dioxide production from a fattening pig building with partial pit ventilation system’, Biosystems Engineering, 126, pp. 56–68. doi: 10.1016/j.biosystemseng.2014.07.011.