| |
(2012). Santiago Raúl Caminotti Raspo – 27 de diciembre de 1934 – 8 de agosto de 2011. Revista Argentina de Produccíón Animal, 32(1).
Texto Completo: Como compañeros y amigos sentimos la necesidad de transmitir un emocionado recuerdo del Prof. Santiago Raúl Caminotti, quien falleció el pasado 8 de agosto de 2011. Toda su trayectoria profesional fue una rica historia de vida profundamente comprometida con el desarrollo del sector porcino regional y nacional. El impulso decidido y la fuerza que Santiago brindara a productores y profesionales ligados a esta actividad a través de una tarea cotidiana sin fisuras, compartida con su grupo de trabajo, fueron una impronta activa e inseparable desde la misma creación del INTA Marcos Juárez.
Santiago tuvo un pasaje inicial por la Agencia de Extensión Rural Marcos Juárez como asesor técnico y luego de haber sido quien plantara en la Estación Experimental las primeras semillas de un entonces desconocido cultivo de soja en la temprana década del ’60, se capacitó decididamente en producción porcina para poder asistir la demanda de información de los productores. La tecnificación de esta actividad y la iniciativa de Santiago promovieron que se fuera ampliando el grupo de trabajo y esta dinámica culminó con la creación en 1978 de la Unidad Demostrativa Agrícola Porcina la cual, durante mucho tiempo, fue un verdadero punto de encuentro de productores y técnicos deseosos de encontrar un espacio para informarse, intercambiar ideas y discutir sobre los avatares de una actividad con más oscilaciones que mesetas de bonanza. La gran concurrencia a jornadas informativas, fue una de las razones que lo llevaron a ser un impulsor de Fericerdo, muestra integral del sector porcino, cuya consolidación tuvo y mantiene para Santiago el sentido de un reconocimiento para una carrera digna, entusiasta y generosa.
Su compromiso con la actualización y transferencia de la tecnología y el conocimiento lo llevaron a ser miembro de numerosas asociaciones profesionales como AAVEPP, AADER y AAPA entre otras. En nuestra Asociación son muchos quienes recuerdan a este técnico afable, de perfil bajo pero que nunca pasaba desapercibido en sus distintos roles y participaciones en cuanto espacio dedicado a la especie porcina se proyectara desde AAPA en Congresos, Simposios y eventos regionales. Nos sostiene el recuerdo de su entrega por una actividad históricamente rozada por el desaliento y su compromiso por la consolidación de las instituciones a partir del reconocimiento y acción sobre las necesidades del sector productivo.
Área de Producción Animal INTA Marcos Juárez
|
|
Machado, C. F. y B., H. (2012). Uso de modelos de simulación para asistir decisiones en sistemas de producción de carne. Revista Argentina de Produccíón Animal, 32(1), 85–105.
Texto Completo: Resumen
En la experiencia internacional en el desarrollo de sistemas de soporte a la toma de la decisión (DSS en inglés), es recurrente la referencia que la naturaleza fuertemente “cualitativa e intuitiva” de las decisiones conspira con la adopción de dichos sistemas. Lo anterior motiva la fuerte recomendación de la bibliografía del tema de emplear “estrategias participativas” para el uso eficiente de simuladores complejos, y la aplicación de metodologías orientadas a usuario (UCD en inglés) para el desarrollo de DSS más específicos y acotados y por ende mejorar su usabilidad y adopción. En la sección I se hace una breve reseña de algunas connotaciones sobre el proceso de toma de decisión, y posteriormente en la sección II se menciona la evolución de los DSS y causas de su baja adopción. En la sección III, se consignan algunas consideraciones para mejorar la adopción de DSS y en la sección IV se describen dos experiencias propias de los dos tipos de modelos mencionados. La primera, sobre el desarrollo y aplicación de un simulador agropecuario biofísico (SIMUGAN), donde se muestran particularmente, las diferentes estrategias para favorecer los nexos a la investigación de campo y a asesores agropecuarios para el chequeo permanente de su pertinencia y confiabilidad. La segunda experiencia se orienta a la descripción de una herramienta específica de planificación forrajera, productiva y económica (Planificador Ganadero de cría 1.1.1 ®). Se describe el diseño y desarrollo, la estructura, lógica interna y la experiencia de su uso en un curso virtual de agronegocios del Instituto de Promoción de la Carne Vacuna Argentina (IPCVA). En la sección V se concluye sobre la importancia de la complementariedad de los enfoques participativos, por un lado con el objetivo de mejorar la comprensión y análisis y diseño de los sistemas en nexo a la investigación de campo incorporando diferentes visiones, y por el otro favorecer mayor disponibilidad nacional de herramientas “amigables” para diferentes tipos de decisiones en la ganadería vacuna.
Palabras clave: software, ambiente de aprendizaje, participativo, utilidad, usabilidad.
Summary
In the international experience in the development of decision support systems (DSS), it is frequently mentioned that the highly “qualitative and intuitive” nature of decisions is one of the causes of low adoption of such systems. Consequently, there is a strong recommendation to use participatory approachs in the development and use of complex biophysical simulators and the application of User Center Design methodology (UCD) for the development of more specific tools. In the section I, a brief review of decision making process is developed, and in section II a short evolution of DSS history is developed and some causes of their low adoption are mentioned. Literature recommendations to improve DSS adoption are consigned in section III, and in section IV two local experiences about two types of DSS are described. In the first case, it is mentioned the development and use of a whole-farm simulator (Simugan) and how it is linked to field research and to the opinion of consultants to ensure its pertinence and accuracy. The second example is about a tool for feed and economic planning of cow-calf systems, and it describes its design, structure and how it was used in a virtual training course for farmer and consultants sponsored by the Argentinean Institute for Beef Promotion (IPCVA). Finally, in the section V it is concluded about importance of the use of participatory approach to improve system analysis and UCD to make available more friendly tools for different decisions in cattle systems.
Key words: Software, learning environment, participatory, usefulness, usability.
Introducción
En la actualidad, es normal observar la creciente disponibilidad pública y privada de información facilitada por sistemas automáticos de recolección, con acceso en tiempo real posibilitado por diferentes formas de comunicación y dispositivos (por ejemplo teléfonos, PC, palms, notebooks, netbooks, y tablets). El sector agropecuario no es ajeno a eso, pero en este nuevo contexto puede presentarse desinformación por el “exceso” de datos disponibles, donde la limitación empieza a ser la capacidad de análisis de los mismos. En este sentido, las Tecnologías de la Información y la Comunicación (TICs), dentro de las cuales se encuentran los modelos de simulación, ofrecen en su conjunto oportunidades de aporte positivos de análisis para el sector (Albornoz, 2006; MINCYT, 2009).
Los modelos constituyen herramientas que permiten la integración de distinta información y diversos procesos permitiendo el estudio de sus interacciones y la evolución del impacto de modificaciones en el sistema global (McKinion, 1980). Si bien surgen de la aplicación industrial, su utilización en procesos biológicos (Haefner, 1997) y agropecuarios en particular es de crecimiento exponencial (Ahuja et al., 2002; Bryant y Snow, 2008; Gouttenoire, 2011; Johnson, 2003), incluso atendiendo necesidades ambientales como es el caso de la cuantificación de emisión de gases de efecto invernadero por sistemas de producción pecuaria (Crosson, 2011). Por lo expuesto, a nivel internacional los modelos de simulación son herramientas complementarias indispensables en proyectos de investigación de sistemas agropecuarios (Ahuja et al. 2002; Newman et al. 2003; Romera et al. 2008; Thornton y Herrero, 2001). A nivel nacional, y específicamente para el caso de producción de carne, hay también diferentes experiencias positivas (Berger et al., 2002a; Feldkamp, 2004; Machado et al., 2010b; Romera et al., 2004; Ruiz et al., 2000).
Los procesos implicados en la toma de decisión agropecuaria se componen de fases analíticas e intuitivas que involucran el análisis de distintos tipos de información, por un lado datos cuantitativos y por otro datos más cualitativos, orientados a la identificación de tendencias (Öhlmér, 2008). Aunque la aplicación de la simulación como elemento de soporte de investigación es ampliamente aceptada, su contribución a la toma de decisión es un tema de mayor controversia (McCown, 2002a; Parker, 2005).
Esta presentación se orienta a describir el aporte posible de los modelos de simulación a la mejora de la toma de decisiones en sistemas de producción de carne. A partir de algunas experiencias locales iniciales, pero haciendo particular referencia a las recomendaciones internacionales en la materia, este documento está organizado en cinco secciones: I. Breve reseña sobre el proceso de toma de decisión agropecuaria; II. Evolución de los sistemas de apoyo a la decisión agropecuaria y causas de su baja adopción; III. Estrategias propuestas para la mejora de adopción de los sistemas de apoyo a la decisión (DSS), IV. Ejemplos locales de simulación en sistemas ganaderos y V, consideraciones finales y conclusiones.
I. Breve reseña sobre el proceso de toma de decisión agropecuaria
Este punto se orienta a contextualizar la toma de decisión agropecuaria para comprender sus implicancias generales y como facilitarla con modelos de simulación bajo la concepción de DSS. No es motivo de esta sección hacer una revisión exhaustiva de este tópico ya que, en ese sentido existen distintas fuentes posibles de consulta (Anderson et al., 1977; Dane y Pratt, 2007; McCown, 2012; Nuthall, 2011).
Desde un punto de vista didáctico, al proceso de toma de decisión se lo presenta frecuentemente como una secuencia de pasos que implican definir un problema, analizarlo, evaluar y elegir alternativas de mejora, aplicar una (“que puede ser incluso no hacer nada”), elegir criterio de valorización de opciones, y finalmente evaluar el impacto (Baker et al., 2002). Sin embargo, las decisiones son el producto final de un largo proceso de análisis y aprendizaje, realizado en grados variables de manera conciente y deliberada, pero también de manera inconciente, (Attonaty et al., 1999; McCown, 2002a; Sterk, 2011) intuitiva, heurística y cualitativa (Dane y Pratt, 2007). Este proceso se conforma de reglas que resumen el conocimiento y entendimiento que necesita el decisor para el control de todas las interacciones que afectan, por ejemplo a los cultivos y la producción animal a nivel del sistema de producción (McCown, 2002b), ya que los productores usan métodos empíricos para evaluar la información relativa a nuevas alternativas productivas o tecnologías (Llewellyn, 2007; McCown, 2012).
En la Figura 1 se observa que además de los insumos y productos cuantitativos considerados en la toma de decisión (líneas contínuas), hay igualmente insumos y también productos de más difícil cuantificación (líneas punteadas), pero que a su vez tienen gran relevancia en el proceso decisional. Este punto resulta evidente cuando analizamos la consideración de un asesor experimentado con su cliente, al momento de definir una recomendación técnica. Ej., “Mi sugerencia a Pedro, en consideración de su fuerte vocación ganadera (perfil del decisor), es hacer la agricultura por arrendamiento (renunciando a parte del margen si fuera por administración) para concentrar los esfuerzos en articular mejor los planteos de recría y engorde a corral estratégico …”
En la Figura 2 se observa el contraste o brecha entre la toma de decisión empresaria basada mayormente en la experiencia y prácticas propias, versus la mejor recomendación técnica basada en el análisis y el conocimiento de sistemas de apoyo a la decisión (McCown, 2002b). Estas dos dimensiones se acercan en la medida en que las recomendaciones técnicas incluyen también el conocimiento de las necesidades, intereses, valores y objetivos del decisor (se puede hacer referencia al mismo ejemplo de relación asesor y su cliente expuesto previamente), y cuando el decisor va incorporando a las prácticas propias más conocimiento técnico y análisis. En este sentido, se suele mencionar que aquellos decisores que han sido entrenados en procesos analíticos como los que utiliza la educación formal o la que se generan en el trabajo grupal de productores, son más proclives a este cambio (Öhlmér, 2008).
II. Evolución de los sistemas de apoyo a la decisión agropecuarios y causas de su baja adopción
Los modelos de simulación son una herramienta matemática descriptiva y cuantitativa que permite describir de forma simplificada un sistema actual (hipótesis cuantitativa), y además proponer y cuantificar mejoras (Ahuja et al., 2002; Mitra, 1988). Por lo tanto, regularmente deben estar sujetos a evaluaciones para saber si son adecuados para el uso que queremos darle (Oreskes, 1998; Tedeschi, 2006).
Es posible identificar etapas marcadas en la historia de la aplicación de los DSS al sector agropecuario. Se resumen aquí los aspectos más generales de esas experiencias, de manera de identificar cuáles son las condiciones y características que deberían evitarse y aquellas que deberían potenciarse para mejorar la adopción de estas herramientas.
En el enfoque inicial de investigación operativa, derivada de su aplicación industrial, se pretendía sustituir el pensamiento intuitivo basado en prácticas propias del decisor por recomendaciones óptimas sustentadas en el análisis (Hutton, 1965) tal como lo mostrado en la Figura 2, resultando el mismo un enfoque demasiado sesgado para permitir un uso más generalizado (Dreyfus, 1986).
Posteriormente se toma conciencia más claramente del “mecanismo” decisional (McCown, 2000b). Un productor maneja su campo de forma integral; sin embargo, cuando toma una decisión particular en un contexto complejo e incierto como es lo habitual, lo hace en base a reglas simplificadas que resumen su criterio y conocimiento del sistema sobre ese tema (Plant y Stone, 1991). De manera posterior al enfoque de investigación operativa surgieron de la investigación a nivel internacional numerosos e importantes desarrollos de simulación bajo la forma de DSS orientados a explorar opciones del tipo “que pasa si”, pero también con baja adopción general. Esta situación ha merecido la atención de diversos autores (Arnott y Pervan, 2005; McCown, 2002a; McCown et al., 2009; Newman et al., 2000; Parker, 2005) y resumiendo, de estos trabajos se puede consignar los siguientes puntos con respecto a los DSS:
a) En general, los decisores no están involucrados en su diseño y desarrollo, por lo que la forma de plantear alternativas y el formato de los resultados normalmente no ajusta al estilo y preferencias de los usuarios/beneficiarios potenciales.
b) La predominancia de los investigadores en su diseño y desarrollo, los vuelve muy demandantes en información que no siempre está disponible para los decisores, y si estuviera accesible requeriría mucha dedicación de carga.
c) Mayormente, mantienen el sesgo sustitutivo sobre el juicio del decisor o del asesor basado en la mejor recomendación técnica (Figura 2), siendo particularmente crítica la falta de relevancia o pertinencia del DSS ante los ojos del decisor considerando el punto a). Sobre este punto, es importante mencionar que es difícil que un decisor renuncie a la aplicación de su propio juicio cuando tienen que tomar una decisión que es de su responsabilidad, a favor de la recomendación de un modelo del que desconoce sus supuestos y limitaciones (de Geus, 1994).
d) Los objetivos del uso de los modelos no suelen estar claramente explicitados, sobre todo dentro de procesos de investigación-transferencia (con quién, hasta dónde, para qué, cuándo y cómo); del mismo modo es deficitaria la cuantificación del impacto de su uso.
e) Debida a la baja adopción (mercado), es difícil sustentar un equipo de soporte, mantenimiento y mejora (que a su vez permita ampliar usuarios), si no es mediante un aporte subsidiado de la investigación.
III. Estrategias propuestas para la mejora de adopción de los DSS
Entre los modelos de simulación se diferencian por un lado aquellos simuladores biofísicos, los cuales suelen ser relativamente complejos y motivados desde la investigación y destinados al estudio mayormente estratégico de tecnologías u opciones productivas, y por otro aquellas herramientas más simples y de uso más directo como apoyo a la toma de decisiones específicas (Le Gal et al., 2010; McCown, 2002b). En consideración de las diferencias de origen y de objetivos de los simuladores biofísicos y del desarrollo de herramientas más específicas y directas (identificados en este documento como DSS específicos para facilitar la diferenciación), a continuación las estrategias para ambos tipos de DSS se describen por separado.
III.a. Simuladores biofísicos:
En el caso de simuladores más complejos, la transformación del conocimiento científico en información útil para la toma de decisiones (manejo del sistema productivo) requiere poder establecer un nexo entre los procesos biológicos simulados, los procesos y tecnologías desarrollados en el sistema, provisto con indicadores adecuados para la toma de decisión. Esto normalmente pone en evidencia la necesidad de lograr información relevante y de fácil acceso al usuario (Cash et al., 2003; Gluck, 1996; Gouttenoire, 2011; Martin et al., 2011; McNie, 2007). Los modelos permiten hacer análisis de prospectiva en base a la utilización de escenarios posibles (Godet, 2000), que se definen como una visión internamente congruente de lo que podría ser el futuro (Porter, 1980) y por lo tanto nos facilita cuantificar el impacto sobre el sistema de diferentes incertidumbres en la que se encuentra nuestra actividad/negocio. Dicho de otro modo, esto nos permite cuantificar como un determinado esquema productivo (incluyendo el modo que se lo maneja) se comporta ante uno o más escenarios. Esto resulta relevante no solo para cuantificar el impacto de escenarios más probables, sino también para cuantificar escenarios excepcionales.
Este tipo de herramientas más potentes pero más complejas, requieren cierta habilidad para su utilización que hace que en algunos ejemplos de análisis participativo de posibles escenarios agropecuarios (Cabrera et al., 2007; Machado et al., 2010a; Martin et al., 2011; Vayssières et al., 2011), la tradicional figura de usuario en software, sea reemplazada por la de beneficiario, ya que el usuario es un técnico o grupo técnico facilitador de la aplicación del simulador. En estos trabajos, la temática y el enfoque experimental fue seleccionada por los productores y asesores agropecuarios (beneficiarios), pero las simulaciones fueron desarrolladas por un equipo de soporte (usuario facilitador). Este tipo de enfoque se denomina modelación participativa; donde se rescata la experiencia de quien vive y trabaja día a día en un sistema productivo dado, recuperando información y características de los procesos, así como los fenómenos que pueden no haber sido captados por los investigadores, quienes no están inmersos en el trabajo diario, pero que manejan otro tipo de información complementaria a la generada en los sistemas in situ (Girard y Hubert, 1999). En resumen, los procesos participativos sirven para integrar todo tipo de conocimientos (empírico, técnico y científico) de una variedad de disciplinas y fuentes de información (Cabrera et al., 2007; Voinov y Bousquet, 2010), lo que favorece un dialogo “cuantitativo” entre personas con distintos puntos de vista, criterios o percepciones, tales como investigadores, asesores y productores, conformando una “herramienta de comunicación” entre diferentes concepciones o ideas (Jakku y Thorburn, 2010; Jakku et al., 2004; Sterk, 2011).
En términos generales, todas las técnicas participativas están relacionadas con el método educativo conocido como aprendizaje colaborativo o aprendizaje social, donde el proceso de aprendizaje tiene lugar a través de la comunicación entre los participantes (Craps, 2003; Maurel et al., 2007; Voinov y Bousquet, 2010). Bajo esta lógica, las simulaciones permiten indagar cómo estructuras y mecanismos clave del sistema pueden estar interaccionando para producir los eventos observados; sirven como laboratorio de aprendizaje (Figura 3) para explorar y comprender la situación bajo análisis, superando algunas de las limitaciones de los experimentos reales, como las demoras en la aparición de los efectos de los cambios en las variables, y las variaciones biológicas no controladas en los ensayos, la incertidumbre ambiental y el grado de precisión de las mediciones (Woodward et al., 2008). Sin embargo, el proceso participativo per se no es efectivo si no se toman los recaudos necesarios que pueden condicionar los resultados de esta metodología, como es el caso de la constitución y el tamaño del grupo, el propósito para el cual han sido convocados, la metodología de coordinación de las actividades grupales (Renger et al., 2008) y los principios o recomendaciones que guían el proceso participativo (Voinov y Bousquet, 2010).
El nuevo rol de los simuladores como laboratorio de aprendizaje (Figura 3) en vez de generador de recomendaciones técnicas específicas para la toma de decisión (Figura 2), hace énfasis en el cambio cognitivo cuando alguien actúa en un entorno determinado, pero que además lo adopta como conocimiento nuevo y lo utiliza en sus prácticas futuras (Leeuwis, 2004). Esta forma indirecta de mejorar las futuras decisiones a través del aprendizaje en un entorno virtual, encuadra dentro de los DSS denominados de tipo pasivo (Haettenschwiler, 1999). En general se mencionan dos tipos generales de aprendizaje el que permite mejorar la forma como hacemos las cosas (relacionado a la eficiencia), y uno más medular, que implica la modificación de las certezas básicas (Argyris y Schön, 1996). Es importante destacar que la predisposición a aprender de un modelo de simulación, implica que el mismo es desde el punto de vista de los beneficiarios, creíble, relevante y legítimo para el problema que se pretende abordar (Cash et al., 2003 ). Sobre este mismo punto, se ha observado más predisposición de los decisores en situaciones donde, por condiciones extraordinarias (ej sequías prolongadas, nuevas políticas, nuevos cultivos, etc.), se pierde el sustento normal de las referencias históricas de los productores y sobre las que basan sus reglas de decisión (Louis y Sutton, 1991). Para que el simulador biofísico resulte creíble, relevante y legítimo, requiere un proceso iterativo gradual de mucha apertura entre decisores y analistas investigadores, y que estos últimos estén dispuestos a brindar el apoyo al proceso, aprendiendo junto con los decisores e identificando como mejorar su propia actividad. Esto es la esencia de la Investigación-acción (McCown, 2002b).
III.b. Diseño de DSS específicos:
A diferencia de los simuladores previamente mencionados que surgen de la investigación, en este caso nos referimos a sistemas que se originan por una necesidad directa del mercado. A partir de los simuladores biofísicos como los descriptos en III.a, han surgido excepcionalmente necesidades traducibles en una oportunidad de mercado (Hochman et al., 2009), pero no es lo habitual, ni el objetivo de los mismos. Particularmente en este tipo de herramientas más específicas, nos referimos a aquellas destinadas a resolver necesidades puntuales, frecuentemente asociados a decisiones operativas y tácticas (aunque no exclusivamente) de un usuario, como podría ser por ejemplo el ajuste de una dieta por parte de un asesor nutricional o el análisis una serie de precios ganaderos para apoyar una decisión particular. Aunque es evidente el rol central que tiene la usabilidad (específica para cada tipo de usuario) en cualquier software y más aún en un DSS, la falta de integración formal de la opinión continúa del “usuario” en el diseño/desarrollo del producto, está dentro de las causas frecuentes de fracasos en la industria del software aplicado a diferentes sectores productivos e industriales (Charette, 2005). Por lo tanto, en este tipo de DSS el usuario debe ser el centro del proceso de desarrollo (Parker, 2001; Shneiderman, 1998) y entonces se recomienda la utilización de metodologías específicamente diseñadas para subsanar este déficit, como los son centradas en el usuario, UCD en inglés (ISO-13407 1999). A partir de la necesidad detectada de un DSS, esta metodología contempla el análisis de los requerimientos del usuario (identificando usos y perfiles de usuarios) y el prototipado y la evaluación de la usabilidad (heurística y mediante usuarios) que reatrolimenta el diseño y funcionalidad del prototipo (Sears, 2007). Para entender la adopción del mercado de estas herramientas se han desarrollado enfoques sofisticados que incluyen elementos de facilitación e influencia del entorno, entre otros (Venkatesh, 2003), la misma se sustenta principalmente tanto en la percepción por parte del usuario de utilidad como de la facilidad de uso de esta tecnología (Davis, 1989; Keil, 1995).
IV. Ejemplos de uso local de simulación en sistemas ganaderos
A continuación se describen dos experiencias propias de cada uno de los tipos mencionados previamente. La primera es el desarrollo y aplicación de un simulador agropecuario (SIMUGAN) en el marco de la investigación interinstitucional de sistemas agropecuarios del sur de la Provincia de Buenos Aires. La segunda experiencia se trata una herramienta de planificación forrajera, productiva y económica (Planificador Ganadero de cría 1.1.1), diseñada y desarrollada con usuarios en los últimos cuatro años.
IV. a. Desarrollo y aplicación de un simulador biofísico de empresas agropecuarias como ambiente de aprendizaje e investigación:
Evolución y funcionalidad actual:
La necesidad de disponer de alguna herramienta que permitiera modelar y cuantificar situaciones de empresas agropecuarias surge tempranamente en el grupo, iniciándose los primeros prototipos con planillas electrónicas (Machado y Ponssa, 1995; Ponssa y Machado, 1995). Posteriormente se fueron refinando los diseños y la posibilidad de evaluaciones más complejas con uso de modelos implementados en sistemas de simulación como Extend® y Stella® (Berger et al., 2002a; Berger et al., 2002b), pero en el ánimo de mejorar la usabilidad, escalabilidad y el rendimiento del simulador, se avanzó el trabajo conjunto con ingenieros de sistemas (Arroqui et al., 2009) que permitió finalmente disponer de una versión web denominada SIMUGAN (Machado et al., 2010b). El desarrollo de la misma se integró paralelamente con investigación de campo para calibrar componentes específicos que resultaban clave a la capacidad predictiva del mismo (Machado et al., 2006; Machado et al., 2007). Antes de describir cómo se está utilizando el simulador, se efectúa una descripción general de su funcionalidad (Figura 4). El usuario accede vía web y utiliza un planteo productivo-económico (PPE) disponible ó genera uno nuevo desde cero o a partir de otro existente. El PPE se construye a partir de la información necesaria de la situación inicial, que se carga en diferentes pantallas a las que se accede desde botones por temas (Figura 5), incluyendo el diseño de diferentes reglas de manejo destinadas a darle flexibilidad a las simulaciones, que son órdenes condicionales que se desencadenan dependiendo de las condiciones presentes (si…tal cosa….en tal momento, hacer tal otra). Por ejemplo, si las vacas presentan una condición corporal debajo de 2 (Escala 1-5) a los X días del mes X, aplicar destete precoz a las mismas…Dentro del planteo se define también la duración de la simulación, que puede ser días, meses ó varios años. Cuando se manda a simular, el modelo simula diariamente, y cuando termina la simulación se genera un archivo de resultados productivos y económicos en planilla electrónica que el usuario recibe por correo electrónico (Figura 4).
Uso actual y futuro del simulador:
El desarrollo local más reciente de SIMUGAN permitió capitalizar las recomendaciones que surgen de la bibliografía internacional (Ahuja et al., 2002; McCall et al., 1994; McCown, 2002b; McCown et al., 2009; Morley, 1972; Woodward et al., 2008). De manera sintética, se han fomentado estrategias participativas como las mencionadas en III.a para identificar los planteos a estudiar con el simulador, con la participación de extensionistas, investigadores y estudiantes en diferentes talleres (Figura 6). Un desafío importante de estas experiencias para el equipo más ligado al simulador es explicar su funcionamiento, potencialidades y limitaciones y sus capacidades a técnicos no familiarizados con modelación. Esto ha dado un balance positivo, permitiendo disponer de pautas para la mejora de la pertinencia de la herramienta e identificar casos de interés (Machado et al., 2010b). A modo de ejemplo, en un taller se identificó la necesidad de estudiar en sistemas de cría el impacto combinado de la superficie destinada a sorgo diferido, el mes de inicio de utilización y la carga animal sobre la producción de carne anual por ha, como se muestra en la Figura 7 (Stefanazzi et al., 2011). De los resultados de la simulación, se observa que los mismos sugieren un uso temprano del diferido (Abril), con una superficie de sorgo entre el 6 y 9%, asimilable a prácticas encontradas en sistemas comerciales de la Cuenca del Salado. El margen bruto ganadero presentó la misma tendencia que la producción de carne (Stefanazzi,
I. datos sin publicar).
La estrategia de uso de SIMUGAN orientada a la potenciación del mismo como laboratorio de aprendizaje colectivo (Jakku y Thorburn, 2010; Woodward et al., 2008), ha permitido una mejor interacción con la investigación de campo, formación interdisciplinaria de recursos humanos (Figura 8) y de lo que se denomina investigación-acción (McCown et al., 2009). Como es recomendable que este tipo de estrategia de trabajo tenga un buen sustento institucional (van Delden, 2011), por lo mismo se ha avanzado en la formalización y puesta en marcha de una red interinstitucional para la modelación de sistemas agropecuarios de la región, MODASUR (integrada por INTA CERBAS, Facultad de Ciencias Veterinarias UNCPBA, la Comisión de Investigaciones Científicas (CIC, Provincia de Buenos Aires), el Instituto de Promoción de la Carne Vacuna Argentina (IPCVA), la Facultad de Ciencias Agrarias – UNMdP, y la Facultad de Ciencias Exactas UNCPBA). Adicionalmente SIMUGAN está siendo utilizado en dos proyectos ganaderos nacionales (INTA PNCAR 011172, INTA AUDEAS CIAC 940106) y uno regional (720142 INTA CERBAS). El doble rol de investigación y extensión del INTA, ha facilitado la articulación orgánica de proyectos y acciones de experimentación y de discusión de casos como se muestra en la figura 8 que evalúan las simulaciones bajo un criterio estadístico, pero también en términos de pertinencia del modelo utilizado para los casos y de credibilidad de las proyecciones de acuerdo a la visión de asesores según su experiencia en el territorio (Machado et al., 2010b). El camino de construcción de la confianza en la herramienta por los diferentes actores e instituciones es reciente, pero los resultados preliminares son promisorios, de manera que con una estrategia local gradual y equilibrada permitirá capitalizar mejor las experiencias internacionales de la temática.
IV.b. Uso de DSS específicos:
Sobre este tipo de desarrollos se pueden encontrar diversos casos en Argentina orientados a producción de carne, por ej. REQNOV Plus®, Huellas®, entre otros. Para este punto se podría considerar alguna de las herramientas mencionadas, sin embargo, por mayor disponibilidad de información propia se describe un ejemplo de nuestro grupo, pero haciendo énfasis en los elementos recomendados en la bibliografía para este tipo de desarrollo.
Planificador ganadero 1.1.1®
Este software surge de necesidades detectadas con profesionales veterinarios en el marco del programa de Educación continua de la FCV-Tandil. Para facilitar el proceso de planificación de sistemas alternativos de cría mediante indicadores físicos y económicos, se desarrolló participativamente un prototipo en MS Excel® que fue perfeccionado y utilizado durante 4 años. Posteriormente, se desarrolló en un trabajo de 1900 horas hombre y con el apoyo del FONSOFT-MINCyT, un prototipo de escritorio denominado Planificador ganadero 1.1.1® Beta (Ponssa et al., 2009), mediante la utilización de metodologías centradas en el usuario (ISO-13407, 1999) y de desarrollo ágil (Arroqui et al., 2009).
De acuerdo a las preferencias de los usuarios, el proceso de carga de un escenario como el ejemplificado es paso a paso, alternando etapas de carga de datos y de salidas parciales, permitiéndole al usuario avanzar y retroceder entre interfaces (pasos) en todo momento (Figura 9). Para ilustrar su funcionalidad, se generó un planteo productivo de 700 ha dedicadas a cría vacuna a partir de tasas de eficiencia reproductiva obtenidas de un monitoreo de empresas ganaderas de la Cuenca del Salado (Maresca et al., 2007). Se estimó el ingreso neto (IN) del mismo y se plantearon dos niveles de mejora logrables de tasas reproductivas, evaluando el IN adicional en ambos casos (Cuadro 1), que resultó en una mejora del IN de +16% y de + 27% sobre el escenario actual. A diferencia de SIMUGAN, el montaje de un ejercicio como el descripto puede realizarse muy intuitivamente en menos de 5 minutos (en este ejemplo solo hace falta llegar al punto 5 de la Figura 9), lo que permite que un usuario inexperto pueda rápidamente familiarizarse con análisis productivos y económicos como el descripto o similares.
En una primer prueba de usuarios, esta herramienta se usó intensivamente para reforzar, en forma de trabajos prácticos, a los conceptos teóricos disponibles en un curso virtual sobre Gestión de la información ganadera y agronegocios del IPCVA y la FCV-UNCPBA, desarrollado a distancia mediante un aula virtual durante 2011 (Figura 10). El mismo tuvo una duración de 3 meses y 60 hs totales, y fue finalizado por 37 profesionales y productores, con realidades muy diversas tanto en la formación técnica e informática previa, como por la realidad productiva zonal. Como parte final del curso cada participante debió desarrollar un trabajo propio de su zona. Los trabajos presentados incluyeron por ejemplo cría en zona de riego del Río Colorado, mejora forrajera en zonas de monte, e incorporación de destete precoz. A pesar de las condiciones heterogéneas de las zonas y de los perfiles de participantes, en general hubo muy buena aceptación de la herramienta, dado que en una encuesta anónima, el 93% de participantes (n total=37) indicaron que el Planificador Ganadero resultó muy amigable y flexible para el análisis de casos (FCV-IPCVA 2011).
V. Consideraciones finales y conclusiones
En un contexto de gran dinamismo de los sistemas agropecuarios, los simuladores biofísicos utilizados para la exploración cuantitativa de impactos de diferentes escenarios resultan en una alternativa relativamente rápida como parte de una primera barrera de análisis de información. Esto puede resultar de utilidad para el sector científico tecnológico de manera de explorar preliminarmente impactos potenciales de diferentes tecnologías, que deben ser confirmadas en trabajo de campo para proponer mejoras a los sistemas reales. En general, la contribución directa de estas herramientas a la toma de decisión ha sido muy baja. Más recientemente estos desarrollos parecen haber encontrado su nicho con un enfoque diferente, dando la posibilidad de un intercambio cuantitativo entre los interesados bajo la forma de laboratorio de aprendizaje y no como generador de recetas tecnológicas específicas. El éxito de las estrategias participativas requiere mucha apertura de los analistas exponiendo claramente los supuestos y limitaciones del simulador, de modo de ir generando confianza gradual en la herramienta por parte de los diferentes actores, fundamentalmente en los investigadores y extensionistas con gran formación en campo y en sistemas reales, tradicionalmente no integrados a los equipos de modelación.
La adopción de los DSS específicos como el Planificador ganadero 1.1.1® Beta, está ligada a su facilidad de uso y a la medida en que este tipo de herramientas potencian las capacidades operativas y/o profesionales de los usuarios (Matthews et al., 2008). El planificador presentado está en una etapa inicial sin lanzamiento como producto, pero los resultados de su uso en un curso virtual son muy alentadores. En relación a este tipo de herramientas más específicas, existe un déficit nacional de oferta, uso y soporte de TICs para la ganadería (MINCYT, 2009). Este déficit resulta más evidente si se lo compara con lo que ocurre en el sector agrícola, en un marco de creciente incorporación de TICs en todas las actividades humanas (Internet, celulares, tablets) que posibilitan el acceso y procesamiento de información de potencial uso para la toma de decisión.
En resumen, los modelos de simulación biofísicos tienen gran posibilidad de aportar información para la toma de decisión en sistemas de producción de carne de forma indirecta, a través de facilitar el intercambio de experiencias y visiones de forma cuantitativa a través de estrategias participativas, potenciando el aprendizaje colectivo y la integración de la modelación con la investigación de campo. Lo anterior no es posible sin un buen soporte institucional que garantice el mantenimiento y mejora continua de todo el desarrollo. Finalmente, el diseño y desarrollo de DSS específicos que ofrezcan una solución tecnológica eficiente en costo y preferencias del usuario sumado a un buen mantenimiento y soporte, podrían tener una buena oportunidad de contribuir a una ganadería nacional de precisión, potenciando las capacidades de los distintos actores.
Agradecimientos
Al Ing. Agr. Alberto Garcia Spil, al Med. Vet. M.Sc Julio C. Burges y a los Drs. Guillermo Milano y Cristian Feldkamp, por las valiosas sugerencias a una versión inicial de este documento. A la ANPCyT – MINCYT, que a través de diferentes líneas del Foncyt, Fonsoft y Fonarsec ha posibilitado la continuación varias de las actividades mencionadas en el documento.
Palabras Clave: software, ambiente de aprendizaje, participativo, utilidad, usabilidad
|
|
Barioni, L. G., Zanett Albertini, T., Tonato, F., Raposo de Medeiros, S. and de Oliveira Silva, R. (2012). Running head: Computer models for beef systems. Using computer models to assist planning beef production: experiences in Brazil. Revista Argentina de Produccíón Animal, 32(1), 77–86.
Texto Completo: Summary
Brazil is a major beef producer and exporter with most of its production obtained from tropical grazing systems based on Urochloa (Brachiaria) pastures and Nellore (Bos indicus) animals. Despite of major differences to the production systems of developed countries, adaptation of foreign models has been the most successful strategy to develop our own decision support systems (DSSs). In Brazil, DSSs have been more straightforwardly developed and adopted in feedlot operations, probably due to the easier tuning of the available models and the better information and control related to feedlots. An example of successful Brazilian computer model for feedlots is RLM, which includes a model of animal growth based on NRC adjusted for Nellore and crossbred animals. Further, RLM includes a modified feed intake equations, a Brazilian feed library and diet optimization methods for least cost of dry matter or minimum production cost. The development of computer models for our grazing conditions has been comparatively slower. Greater changes in model structure and parameter values were required as most process-based grazing system models have been developed based on temperate or rangelands pasture species. Brazilian research on pasture production, feed intake and diet selection seems more distant to modeling, possibly due to the absence of a reference model to help driving the experimental procedures. From the manager perspective, planning occurs on a longer time horizon in grazing systems and much more risk is associated to the effects of variable climate conditions on the dynamics of pasture production and quality. Besides, higher costs and lower accuracy in monitoring the pasture may also discourage adoption. However, great interest by the Brazilian extensionists and farmers has been perceived for Invernada, a dynamic DSS model recently released Brazilian for grazing systems. Experiences with Invernada training courses indicated that adoption may be slowed down due to the higher complexity of this tool compared with feedlot DSSs and due to the lack of acquaintance of the users with other similar tools. Drivers and future options for the development of DSSs for beef production are discussed.
Key words: computer models, livestock planning, beef production, feedlot, grazing systems.
Introduction
Managing beef production systems aims at making the best use of resources to achieve an objective (Parker et al., 1993) or the maximum satisfaction of the owner (Olson, 2004). According Keating and McCown (2001) Management System is a key component of farming systems and is defined as a compartment which monitors and fit controlled inputs and outputs of Production System. Computer-based decision support systems (DSSs), are tools that may provide valuable information from data available in order to improve and facilitate management (e.g. Donnelly et al., 2002; Cros et al., 2004; Sorensen et al., 2010).
The wide availability of personal computers today have turned applicability of DSS feasible for on-farm decision making, although adoption rates are still limited in relation to the original expectations (McCown, 2002; Sorensen et al., 2010). There are, nevertheless, perspectives of new challenges and opportunities for DSSs as result of the fast and continue development of monitoring tools (particularly remote sensing and other on-farm electronic sensors), databases and computers. It is even expected that information and decision support systems may become essential as the scope and complexity of farm management growths with new environmental issues and technological options (Sorensen et al., 2010).
One of the approaches most widely applied to develop DSSs for planning purposes is mathematical modeling, particularly applications of mathematical programming and process-based dynamic models.
Until the 90’s, about all computer-based tools for on-farm beef decision support were developed by teams in North America, Europe and Oceania. The adoption of those foreign tools was negligible in Brazil, particularly regarding dynamic simulation models of production processes. The NRC system (NRC, 1996) and diet optimization software were, however, successfully adopted by nutrition companies and consultants in Brazil.
It was only in the late 90’s that software for ration balancing and animal performance predictions on feedlots were developed and released in Brazil [e.g. RLM (Lanna et al., 2011); Super crac (TDSoftware)]. A grazing system model for the Brazilian conditions became available just in 2011, with the release of Embrapa Invernada.
Nowadays, RLM is the DSS most used by feedlot beef cattle industry in Brazil (Millen et al., 2009; C. Costa Júnior, University of São Paulo, 2011, personal communication). Furthermore, over 2.200 users made download of Embrapa Invernada since their publication in February 2011. Experiences with RLM and Invernada development and adoption will be further discussed in the following sections.
Decision support for feedlots
Models to calculate nutritional requirements and optimize ration formulation are among the best examples of successful decision support tools in beef production. Input data required (e.g. animal breed, live weight, gender, feed composition, etc) was available at low cost and time demand besides having an acceptable of precision. Also those systems were very consistent with the existing knowledge of nutrition taught in universities. Despite the language barrier, some models, particularly the NRC for beef cattle (NRC, 1996), were found to be useful and widely adopted by nutritionists.
Those models require feed composition data, which would imply in higher costs and time demand. However, DSSs as the NRC for beef cattle software soon integrated feed libraries with expected nutritional values for a collection of feeds. Although the original NRC feed libraries included most of the feeds used in the Brazilian feedlots it was not complete. Brazilian bromatological laboratories have then provided feed standards for conditions (Tedeschi, et al., 2002; Valadares Filho et al., 2002). Valadares Filho also developed ta- bles of feed composition available online with regular updates (available at http://cqbal.agro- pecuaria.ws/webcqbal/index.php).
Experiences with the NRC beef nutritional requirements model in Brazil made identification of opportunities for software development and change requests quite clear. Among them were incorporating tropicalized feed library and model parameters and a friendlier user-interface in Portuguese.
Supercrac and RLM (Rações de Lucro Máximo – Maximum profit rations, with versions in Portuguese, English and Spanish – Lanna et al., 2011) are two DSSs which can be reported successful cases in the Brazilian market. Supercrac focus on least dry matter cost ration formulation which implements the NRC model for estimation of nutrient requirements (i.e. optimization constraints). RLM includes, besides least dry matter, least production cost optimization through parametric linear programming (Glen et al., 1980). This is very important when defining most profitable diets in a wide range of grain/roughage prices ratios (Lanna et al., 1999).
RLM also made adjustments to the NRC model, particularly for Nellore animals, including a re-parameterized dry matter feed intake equation (DMI, Almeida, 2005). The resultant DMI equation [DMI (kg/d) = (SBW0.75*(0.2039*NEm – 0.03844*NEm2 – 0.07376))/NEm] was modeled using meta-analytical data from animals evaluated in experimental institutions. Sequentially, the DMI equation was evaluated using a different data set from Nellore young bulls. The Almeida equation accounted for 77.2% of variation in actual DMI and had less overprediction bias compared with NRC 1984 and 1996 equations (1.3% vs. 6.1 and 3.2%). Further, actual intakes and the predicted estimates did not differ from each other (based on t test, p>0.10).
According to NRC (1996) fat provides small quantity of energy to the rumen microorganisms, thus TDN correction it is necessary when fat level is highest than 3.5% (DM basis). RLM includes TDN fat fit (TDNfit) to predict more accurately the bacterial crude protein synthesis (BCP), as follow: TDNfit (%) = TDN – (Fat – 3.5)*2.25, and BCP (g/d) = (13%*NDTfit*DMI*10)/100. In the previous equations, TDN is multiplied by the factor 2.25 to diets with fat concentration highest than 3.5%, and 13% is assumed to BCP synthesis averaged percentage, according to NRC, respectively.
Most recently, Hoffman (2007) evaluates RLM updating the model considering: Almeida DMI equation; Brazilian feed library, maintenance energy requirements fit (differences of Bos indicus and crossbreeds, previous nutrition over compensatory gain, ionophores and antibiotics). RLM updated was evaluated using independent experimental results published in Brazilian Journals (n = 21) and commercial feedlots (n = 892 pens) of Brazilian Central Region to two features: DMI and shrunk daily BW gain (SBWG). In general, RLM accounted for 67 or 68 and 30 or 53% of variation in actual DMI and SBWG in experimental or commercial situation, respectively. Moreover, RLM predicted data in experimental and commercial feedlot situation had equal or less bias compared with NRC (1996) for both features evaluated. According Hoffman (2007) in commercial feedlots, the SBW bias probably was influenced by orts, compensatory gain and frame size. In order to exemplify the orts influence over bias, when the SBW was predicted from data set with “real” DMI (i.e. removing orts) the residual was improved (SBW observed = 1.35, SBW predicted on DMI observed = 1.49, and real DMI = 1.37). Nowadays, RLM penetration in the feedlot industry is associated to several features, but parameterization in order to improve the model accuracy and precision based on experimental studies is extremely relevant.
Decision support for grazing systems
Modal Brazilian beef production systems present major differences to those of developed countries where most DSSs for grazing systems were developed. Tropical pastures are the base of feeding, comprising over 95% of the total dry matter intake of the herd (Bürgi and Pagoto, 2002). According to the Brazilian National Institute for Geography and Statistic (IBGE), based on the 2006 Agricultural Census data, Brazil has over 172 million ha of grasslands, more than 140 million ha allocated to beef animals (Ferraz and Felício, 2009). Extensive low-input systems predominate (Ferraz and Felício, 2009), with average stocking rates is around 1.08 heads/ha (IBGE, 2011). Grasses of the Brachiaria (Urochloa) genus are predominant in the pastures. Panicum, Cynodon and Penisetum grasses are adopted in some regions, particularly in the most intensive systems. Zebu breeds, predominantly Nellore, have major contribution to the Brazilian beef cattle genetics (Pereira, 2004).
Some of the particularities of the Brazilian livestock production have been reported as reasons for its competitive production costs (da Silva and Sbrissia, 2000; Ferraz and Felício, 2010). However, these differences also hamper the as-is adoption of DSSs from developed countries. This resulted in the absence of important cases of adoption of foreign DSSs for beef grazing systems up to date.
The Invernada DSS (Barioni et al., 2011a) was developed targeting this gap. It aims at helping farm consultants planning beef production in the stocking and finishing phases and addresses decisions at the tactical level. However, during model development, several issues related to adapt models to the Brazilian conditions were faced.
The main knowledge barrier to the development was modeling pasture production and selective grazing for the major pasture species [Brachiaria (Urochloa) spp., Panicum spp., Cynodon spp. and Penisetum spp.]. Although Soto (1981) had adjusted thermal sum models for Panicum maximum and Paspalum atratum, no other important development was made in the 80’s and early 90’s. It was only the 2000’s that a reliable set of information became available for the main pasture species cultivated in Brazil. Among the main contributions are Villa Nova et al. (1999); Moreno et al. (2000); Medeiros et al. (2001); Tonato (2003); Moreno (2004); Rodrigues (2004); Detomini (2004); Lara (2007, 2011) and Cruz (2011).
Tonato et al. (2010) compiled primary data of several of the experiments above and developed a database in order to parameterize models to estimate potential productivity of the main cultivars of the pasture species currently used in Brazil.
Tonato’s potential production model was coupled with a soil water balance model and incorporated to the Invernada DSS in 2011. The current perception is that the model produces reliable predictions of seasonal patterns of pasture accumulation rates for the central part of Brazil. However, as it neglects some important information such as soil fertility and grazing management, magnitude of production has to be adjusted for individual production systems (Barioni et al., 2011b). Also, better evaluation in different regions of Brazil is needed as parameter estimation was based on experiments with relatively limited geographic distribution (Tonato et al., 2010).
Modifications of the grazing models described in the literature were also necessary for the Brazilian production systems. The grazing model included in Invernada was developed based on concepts of Woodward et al. (1997; 2001) and Freer et al. (1997) and calibrated using pre and post-grazing evaluations of forage mass and composition Brachiaria brizantha cv. Marandu (Gimenes, 2010). The Davis Growth Model (Oltjen et al., 1986) was adopted for predicting animal growth and composition of the animals in Invernada, with parameters estimated also for Nellore animals (Sainz et al., 2004). The NRC model for energy and nutrient requirements with some of the new developments included in RLM was also incorporated to that model.
Invernada also includes ration optimization routines (both least cost of dry matter and production). An overview of Invernada’s model structure is presented in Figure 1.
Software development was also found to be a challenge. Efficient and modular software architecture was designed for the development (Torres et al., 2007, Figure 2).
Unfortunately users were not included in early stages of the development of Invernada, but four workshops were organized where prototypes and preliminary versions were presented to a partner company, Bellman Nutrição Animal, and to a private farm consultancy office, Boviplan. Typically, those workshops included a short training period, and informal validation of user interface, performance and outputs. Software testing was carried out in a period of two weeks after the workshops. In each of the iteration, change requests were analyzed and a new version produced.
Those workshops revealed that our focus as researchers was much narrower than that of the end-user. First it was felt that problem identification was not very clear in the first moment as researchers focused only in accurate model predictions and the end-user on how the model predictions would be useful to solve their problems. In this respect, the interactions with consultants have enabled us to introduce some useful features to more specific cases such as better evaluation of responses to supplementation (e.g. accounting for the effect of low protein diets; evaluation of generic technologies to change animal intake and performance). For the reason that user evaluation came late in software development, excessive work of redesigning and reprogramming was necessary. Therefore other methods of analysis and development, particularly agile methods (Arroqui et al., 2009) and user-centric software design (Sorensen et al., 2010) should be used in future software development in our group.
In the first six month (march to September/2011) from release over 2,200 downloads have been recorded and geographic dispersion of the visits to the download site show the software (Figure 3) indicate potential user all around the country. The demand for DSS tools for grazing systems, such as Invernada, seems to have a substantial potential.
Decision support for whole farm systems
Optimization models based on linear programming (LP) or mixed integer linear programming (MIP) are excellent tools for strategic planning, particularly regarding the allocation of farm resources.
In Brazil, several mathematical programming models have been developed and applied to the optimization of beef systems (e.g. Balverde, 1997; Veloso et al., 2009). Those models can provide optimized strategies taking into account stocking rates, quantity of supplements, financial loan and risk, among others. Unfortunately, none of them have led to the development of a tool available for the end-user yet.
Whole farm systems simulator based on biological processes are still not available in Brazil. Comparing to the existing mathematical programming models, a whole farm simulator may have the advantage of allowing more detailed description of the biological processes related to production and environmental impact. In comparison to Invernada, a whole farm systems simulator would have the advantage to enable simulations of crop-livestock and other mixed systems besides a better description of the whole farm business.
Perspectives for future development
Agricultural systems modeling is a growing field in Brazil and our research team foresees considerable opportunity to improve the beef cattle production efficiency, increase the profit and reduce environmental impact through the adoption of decision support systems. In our evaluation, some of the most active streams of research DSSs for beef production are:
1. Evaluation of environmental impact, particularly greenhouse gases (GHG) net balance in beef production systems;
2. Developing whole beef cattle farm systems DSS;
3. Using new technologies of data acquisition (GPS, sensors and mobile computers) and communication to monitor individual animal performance to provide inputs for simulation models.
In stream 1, a large research project, with more than 300 members, led by Embrapa called PECUS (Sustainable Cattle Farm. Successful Cattle Operations, Embrapa SEG 011.006.001) has the objective of quantifying GHG net balance aiming at supporting the selection of mitigation alternatives and implementation of public policies. In that project, dynamic models of processes related to GHG fluxes will be incorporated to the Invernada simulator.
The PECUS project will also include stream 2 research through the development of Mixed Integer Programming Model of a beef cattle production system. The goal of this model is to establish optimal strategic plans for beef production systems by applying a bio-economic model incorporating environmental constraints and evaluating GHG mitigation policies, such as carbon credits and governmental incentives.
This model aims to maximize the producer's net revenue at the end of planning period taking into account: land areas allocated to each activity in the production system, pasture production and grazing animals intake, pasture production decline, supplementation and pasture recovery policies besides the credit and dynamics of financial resources. The GHG emissions will be attached in the model as a constraint, but also will be studied the financial return by CO2 equivalent abatement by means of governmental incentives and carbon credit.
In stream 3, another multi-institutional research project, including the University of São Paulo and Embrapa was conceived with the objective of optimizing the economic slaughter endpoint of Nellore steers using online monitoring of individual performance and a dynamic model (Embrapa call (SEG MP2 01/2011) and FAPESP. To achieve this, there will be an effort to improve a mechanistic dynamic animal growth model and composition, parameterized from experiments based on evaluation of genotypes adapted in the Brazilian conditions, particularly Zebu germplasm. A second aspect is the integration of monitoring technologies to evaluate the growth of individual animals in real time with the dynamic simulation model in order to allow, through dynamic filtering (Oltjen and Owens, 1987) fitting model parameters to individual animals. Finally optimization methods will be applied to support decision making by cattle producers, slaughterhouses and consultants.
Concluding remarks
The development and application of DSS for beef production in Brazil is much more recent than that in several developed countries. However considerable level of adoption is noticeable with the release of the first DSSs in Brazil. This makes us optimistic about the future of DSSs for beef production. Demands and opportunities for improvement of current systems are expected to be driven by user feedback, evolution of scientific knowledge, data availability (particularly with online monitoring methods and remote sensing) and improvement of software development methods.
Environmental issues will also demand new models and DSSs for beef as those may have to be included in planning and evaluating production systems in the future. In that case, besides farmers and extensionists, the government may become an important client of such solutions.
Also, opportunities generated by new technologies, particularly regarding remote sensing and real-time monitoring are expected to drive development of DSSs for beef systems in the future.
Palabras Clave: computer models, livestock planning, beef production, feedlot, grazing systems
|
|
Beauchemin, K. and M. G., E. (2012). Life Cycle Assessment – A Holistic Approach to Assessing Greenhouse Gas Emissions from Beef and Dairy Production. Revista Argentina de Produccíón Animal, 32(1), 69–76.
Texto Completo: Introduction
There is growing pressure on agriculture to reduce its emissions of greenhouse gases (GHG), including the release of methane (CH4) from ruminant livestock. It has been estimated that ruminant livestock account for 8-10% of global anthropogenic (i.e., human derived) GHGs, using the International Panel on Climate Change (IPCC) methodology, or as much as 18% of global GHGs when additional emissions from land use change are also included in the analysis (O’Mara, 2011). Because of the significant contribution of livestock to GHG emissions, animal scientists have become increasingly interested in the potential for mitigation.
In particular, there is increasing interest in finding ways to reduce enteric methane (CH4) production from cattle and sheep. Methane gas is produced by ruminants during feed digestion, with higher CH4 produced as feed intake increases and as forage proportion in the diet increases. However, CH4 is not only a potent GHG but also represents a loss of energy for the animal. Thus, a number of mitigation strategies have been proposed, as reviewed elsewhere (Beauchemin et al., 2008, 2009; Martin et al., 2010; Grainger and Beauchemin, 2011). Some of the proposed options include: diet supplementation with lipids, feeding more concentrates, use of feed additives such as monensin, enzymes, yeasts, direct fed microbials, plant essential oils and tannins, improved feed conversion efficiency, selection of low CH4 emitting animals, and manipulation of the rumen microflora.
However, a change in diet composition or farm management designed to reduce enteric CH4 production does not necessarily guarantee a reduction in the net GHG emissions from the farm due to a potential increase in other GHG emissions. To adequately assess the true impact of CH4 mitigation strategies, animal scientists need to use a whole system modeling approach, such as life cycle assessment (LCA). LCA enfolds the inputs and outputs in a farming system, accounting for all changes in GHG emissions arising from a prospective mitigation practice. The purpose of this paper is to review the basic concepts of LCA for evaluation of GHG emissions arising from beef and dairy production, highlighting the advantages and limitations associated with this approach, and offering some perspective on future direction.
The Importance of Holistic Thinking
Using an LCA approach can help animal scientists avoid making recommendations that might reduce enteric CH4 at the expense of increasing the other GHGs (nitrous oxide [N2O] and carbon dioxide [CO2]). Although CH4 mitigation strategies target a reduction in total CH4 production (g/animal/day), it is possible that the change in farm management or diet could lead to an increase in other GHGs or a decline in animal productivity. Expressing total GHG emissions in units of intensity, such as kg CO2 equivalents (e) per kg of beef or milk produced, helps ensure that reduced CH4 is transferred through the entire system. For example, with improved pasture quality, a logical response by the farmer is to increase stocking rate. However, this is likely to further increase CH4 yield/unit area and total GHG emissions from the farm (g/d), even though it may reduce GHG emissions/kg milk or meat produced. Similarly, high yielding dairy cows typically consume more feed than low yielding cows, and as a result produce more CH4 (g/cow/day),yet high yielding cows produce less total GHG (and CH4)/kg of milk produced. This apparent contradiction illustrates the need to assess the true impact of CH4 emission reduction strategies using an LCA approach.
Life Cycle Assessment Simplified
An LCA determines the potential environmental impact of producing meat or milk by examining inputs relative to outputs. Specifically, the objective is to estimate the total GHG emissions (CH4 + N2O + CO2) associated with producing meat and milk. International standards such as ISO Life Cycle Assessment 14040 and 14044 (ISO 2006a,b) are used to ensure the LCA is conducted according to approved methodology.
The total emissions are expressed in units of GHG intensity, which provides an estimate of the emissions per unit of product. The GHGs are expressed in terms of CO2 equivalents (e) to account for the global warming potential of the various gases (CH4 = 25, N2O= 298, CO2= 1; IPCC, 2007). The output or functional unit describes the product produced, in terms of kilograms of beef or fat and protein corrected milk or other bases of comparison (e.g., 100 g of protein, mj energy, hectares of land, etc).
To conduct an LCA, one must first determine the system boundary. This specifies the components of the production chain to be included in the analysis. For beef and dairy production the components might include inputs into the farm (such as purchased feed, electricity, fuel, herbicides, pesticides, machinery, etc.), the farm itself, transportation of the product from the farm, processing of the product, transportation of the processed product, the retail sector, and finally, the consumer. However, in most LCA of agricultural products, the retail and consumer components are not considered. An example of a system boundary used in an LCA of milk production is shown in Figure 1. Processing of meat and dairy products usually contributes less than 20% of the total GHGs. For example, in an LCA of lamb produced in New Zealand and consumed in the United Kingdom, GHGs from the farm represented 80% of the total GHG, as shown in Figure 2 (Ledgard et al., 2010). For this reason, there is value in conducting partial farm-based LCA that focus only on the GHG associated with producing meat and milk as it leaves the farm (i.e., from cradle to farm-gate), as the greatest potential for GHG reduction is at the farm level.
Estimating Farm GHG Emissions
A number of models have been proposed to estimate GHG emissions from farms. One such model is Holos (www.agr.gc.ca/holos-ghg), developed by scientists from Agriculture and Agri-Food Canada. Holos is an empirical model, with a yearly time-step, based primarily on IPCC (2006) methodology, modified for Canadian conditions and farm scale. The model considers all significant emissions and removals on the farm, as well as emissions from manufacture of inputs (fertilizer, herbicides, pesticides) and off-farm emissions of N2O derived from N applied on the farm. Holos estimates a whole-farm GHG emission, calculating emissions for soil-derived N2O, enteric CH4, manure-derived CH4 and N2O, CO2 from on-farm energy use and the manufacturing of inputs, and CO2 emission/removal from management induced changes in soil carbon stocks. This systems approach allows net whole-farm emissions to be calculated from management changes on any part of the farm. We have used Holos to conduct an LCA of beef and dairy production in Canada.
LCA Example: Beef Production
Beef cattle production in Canada typically involves two systems, managed in separate operations: cow–calf ranching and beef feedlot finishing. Thus, to account for emissions from the entire beef production cycle, we needed to simulate a farm that included both a cow–calf operation and a feedlot, along with cropland needed to supply all feed and bedding for the cattle. The LCA was conducted over multiple years to fully account for all components of the beef production system from birth to slaughter. Our aim was to evaluate the beef production system, rather than to document emissions from a specific commercial operation in a particular year.
The simulated farm consisted of a beef production operation, a cropping operation, and native prairie pasture for grazing. The beef herd was comprised of 120 cows, 4 bulls, and their calves, with the progeny fattened in a feedlot. In this system, the cows nurse their calves on pasture for 7 months. The weaned calves are then fed a high forage backgrounding diet in the feedlot for 110 d (240 to 350 kg), followed by a finishing period of high grain feeding for 170 days (350 to 605 kg). The LCA was conducted over an 8-year period to represent the period from birth to slaughter of the breeding stock. Further details of the farm management are given in Beauchemin et al. (2010). By examining an entire cycle, our analysis estimated average emissions for the full system, rather than from segments thereof. Total GHG emissions from the beef herd were calculated using Holos by summing the emissions for the cows, bulls, and feedlot cattle for one complete 8-year cycle. Total liveweight was for all animals, except for replacements, and carcass yield was assumed to be 60% of total liveweight.
Our LCA yielded an estimated GHG intensity of 22 kg CO2 eq/kg carcass, which is at the lower range (17–43 kg CO2 eq/kg carcass) of estimates for beef production from LCAs of other systems (as summarized by Capper et al., 2011, and Crosson et al., 2011). However, it must be cautioned that the range in GHG intensities reflects not only differences among farming systems, but also different assumptions, approaches, and algorithms in calculating emissions, so direct comparison among studies is not recommended.
The major GHG for beef production was enteric CH4, comprising 63% of the total (Figure 3). Within the beef production cycle, the cow-calf system accounted for about 80% of total GHG emissions and the feedlot system for only 20%, dividing roughly equally between the backgrounding and finishing phases. In reality, this breakdown might vary somewhat, depending on the specific management practices used. In western Canada, for example, weaned calves are sometimes fed forage-based diets over winter and placed back onto pasture the following summer before entering the feedlot, rather than being placed directly into feedlots as in our simulation. Such variations, however, are unlikely to displace the cow–calf system as the primary source of GHG emissions resulting from beef producing farms.
About 84% of enteric CH4 was from the cow–calf system, mostly from mature cows. In contrast to some perceptions, the feedlot system accounts for a relatively small fraction of enteric CH4 from beef production. The lower CH4 emission from the feedlot is due mainly to its relatively brief duration and, to a lesser extent, to the use of grain-based finishing rations.
LCA Example: Milk Production
Similar to the beef production LCA, this assessment required the development of a simulated farm. In Canada, the province of Quebec has the highest number of dairy cows constituting 37.5% of the national herd (Canadian Dairy Information Centre, 2011), thus the simulated farm was based in this region. The approach taken in this analysis mirrored that of the beef production LCA by encompassing emissions generated by all on- and off-farm processes that contribute to dairy production.
The LCA was conducted over a 6 year period to represent the typical lifespan of dairy cows under Quebec conditions. The assessment was initiated with the birth of 65 female Holstein calves. Sixty heifers survived to first calving at 27 months of age and following calving, cows were retained for, on average, 2.75 lactations. A calving interval of 14 months, with 12 months lactation was assumed. Milk yield was set at 9,742 kg/animal/lactation for primiparous animals and 10,450 kg/animal/lactation for multiparous animals. Male calves were finished as grain-fed veal at 6.5 months of age at approximately 270 kg, with female calves in excess of replacement requirements also finished in this manner. Livestock were housed for the duration of their lifetime, with all feeds produced on farm. HOLOS was used to calculate the GHG emissions for the system for a single 6-year cycle, which were then expressed relative to one kilogram of fat and protein corrected milk (FPCM).
This assessment yielded an estimated GHG intensity of 0.91 kg CO2e/kg FPCM, which is marginally lower than estimates reported for a number of other dairy LCAs (as summarized by Rotz et al., 2010). However, as discussed previously, comparison between studies is inadvisable and may be potentially misleading given the differences between the systems under investigation.
In this assessment, the major GHG emitted was CH4 (56% of the total GHG emissions), with 86% of this figure arising from enteric fermentation (Figure 4.). Although less than CH4, emissions of N2O were also notable, accounting for 40% of the total, thus indicating the importance of nutrient cycling and management in dairy systems. Assessment of emissions by cattle group showed that 64% of the total GHGs arose from lactating animals. The contribution of the veal calves was negligible (3%), with animals less than 12 months of age contributing only 10% of the total system emissions.
Advantages and Future Perspectives of Life Cycle Analysis for Ruminants
The major advantage of conducting a farm-level LCA is the ability to evaluate the impact of changes in farm management in terms of the GHG intensity of meat and milk production. In our case, once the baseline LCA was established for beef production, mitigation practices aimed at reducing enteric CH4 were applied and their impacts on the intensity of GHG emissions assessed (Beau- chemin et al., 2011). Mitigation practices included dietary modifications aimed at reducing CH4 emissions (i.e., changed forage inclusion levels, dietary supplementation with lipids, use of corn distillers dried grains, improved forage quality) and improved animal husbandry (i.e., increased longevity of breeding stock, improved reproductive performance of the herd). Strategies applied to the cow-calf herd individually reduced total farm GHG intensity by up to 8% with up to a 17% total reduction possible by combining strategies. In comparison, strategies applied to the feedlot had only a small impact on GHG emissions; reducing total GHG intensity by less than 2% when applied individually or by 3–4% when applied in combination. Although the North American beef production system is already highly efficient, a number of mitigation strategies could be implemented to further lower GHG emissions associated with producing beef, with a total reduction of about 20% attainable if multiple strategies are applied to both the cow herd and the feedlot. However, the biggest reductions in GHG emissions are achieved when mitigation practices target reducing enteric CH4 from the breeding herd. When the grassland in the baseline scenario was newly seeded onto previously cropped land, its soil carbon gain more than offset all GHG emissions, changing the beef production system from a net emitter to a net sink of carbon. Although such estimates of soil carbon gain have uncertainty, this scenario demonstrates that the net GHG balance of a beef production system is powerfully influenced by carbon dynamics in the associated land base, emphasizing the importance of including these dynamics in assessments of mitigation potential.
Such a whole-farm vantage can only be achieved using a GHG model, such as Holos. Like other models, Holos still has many limitations and its outputs carry significant uncertainty. However, such a model can help identify research areas and questions that merit particular emphasis and it can thereby guide future research and policy recommendations. Further work is still needed to bolster the model’s reliability and expand its applicability.
A major disadvantage of conducting an LCA focused only on GHG is that the analysis does not consider other potential benefits of maintaining ruminants on grasslands. The GHG intensity of producing beef is considerably higher than other livestock products (Table 1), and consequently many consumers world-wide are choosing to consume less beef for reasons of perceived environmental impact. However, GHG intensity is only one environmental consideration, and LCAs of livestock products need to look beyond GHGs. For example, the cow–calf system has potential ancillary benefits because it relies extensively on pastures and forage crops which, in many cases, can preserve or augment soil carbon, thereby mitigating CO2 build-up in the atmosphere. Furthermore, because the nutrient cycle in grazing systems is relatively closed, excreted nutrients are returned directly to the land from which they came so emissions from manure may also be suppressed. In the future, we need to look beyond GHG emissions, to also include other environmental questions, such as water quality, ammonia emissions from intensive livestock operations, nutrient cycling, biodiversity and so forth.
Conclusions
While continued research to reduce enteric CH4 emissions from beef and dairy production is merited (if only to improve efficiency), the net mitigative effect of the change in management can only be assessed using GHG model and an LCA approach. However, focusing only on GHG emissions does not fairly assess the many benefits of ruminants, particularly pastured cattle. Perennial forages and pastures provide many ancillary environmental benefits. Such lands not only preserve or build soil carbon reserves, thereby withholding CO2 from the air, but also have many other ecosystem services including the conservation of biodiversity, water quality, wildlife habitat, and aesthetic value. The environmental merits of prospective production systems can only be fully assessed by considering all of these benefits. The continued development of robust, ecosystem-level models is essential for such broadly-based LCA analyses.
|
|
Oltjen, J. W. (2012). Bioeconomical model for best slaughter endpoint for maximum profit. Revista Argentina de Produccíón Animal, 32(1), 63–68.
Texto Completo: Introduction
As feedlots are used to finish more cattle, improved strategies evolve. One of the most important management questions once cattle are in the feedlot is “How long to feed them to maximize profit?” However, complex interaction between type of cattle, market demand and price, ownership of the feedlot and/or the cattle, and application of marketing and management tools make profit prediction difficult. Addressing these interactions requires proper application of economic principles, identifying the relevant biology of the animals being fed, and tools such as bioeconomic models to estimate animal performance. When used properly to answer the appropriate question, these tools provide insight into improved management of feedlot cattle.
Economic Considerations
One of the basic principles of microeconomics is profit maximization. However, for the feedlot case one needs to determine who is trying to maximize profit on a pen of cattle—the owner of the cattle or the feedlot. If the feedlot also owns the cattle, there are different constraints and objectives than if the cattle are owned, or partly owned by another person. Also, whether a pen of cattle can be replaced with another, as opposed to one pen per year or time period, makes a difference.
In the case of an owner independent of the feedlot, and assuming this person’s capital is not limited by a particular pen of cattle, the profit maximizing strategy is to feed the pen of cattle until the costs for that day exceed the pen’s gain in value, that is, the marginal net revenue becomes negative. Hence, the cattle are fed until the feed and other costs for the last day exceed that day’s cattle gain multiplied by the cattle’s value per unit weight. Special attention is warranted in this scenario with discounts in cattle price for increasing carcass weight or decreasing yields, as this decrease in the pen’s value may be sudden. For this reason the more variable the cattle in a particular pen, the shorter is the optimum feeding period for profit maximization (Smith et al., 1988).
In the case of a feedlot owning the cattle, and recognizing that the feedlot’s profit maximizing objective is to make money on the pens over time, not just for any particular pen of cattle, then the economic principle is to feed cattle until their marginal net revenue (daily increase in value minus daily cost) no longer exceeds the average daily net revenue for an average animal in a pen in the feedlot. Average daily net revenue is the profit for an animal divided by the number of days that animal was in the feedlot. As long as the average net revenue is positive (the feedlot is making a profit on the cattle, as well as on the feedlot enterprise), cattle owned by the feedlot will be fed fewer days than those owned by others. Again, as in the case above, more variable cattle will be fed fewer days than more uniform ones, but this is less important in the feedlot owning the cattle scenario since the shorter days on feed reduce the chance of discounts.
There are two exceptions or alternatives in the above scenarios. If the cattle owner cannot feed additional cattle until a pen of cattle is sold, then the objective is profit maximization over time, not for a pen of particular pen of cattle. Therefore, their cattle should be fed as in the case of the feedlot owning the cattle above. If the feedlot owns the cattle, and for some reason cannot use the pen again after the cattle are sold/removed (often in case where only one set of cattle are fed in each physical pen annually), then their profit maximization objective is as first scenario above where the cattle are owned independently of the feedlot—indeed the cattle profit is then independent of the feedlot profit.
The above depends on the marginal net revenues as cattle progress in a feeding period. Note that in the early days after a pen of cattle is put on feed, their total value is less than the money invested and the early feed and processing costs (which would result in negative returns if sold at that point). However, the marginal net revenue is usually positive—the value of their daily gain exceeds the daily feed cost. Hence profits are increasing, or losses decreasing. If this is not the case, and marginal returns are negative, prolonging the feeding period increases the loss, and the cattle should be sold. This is not unusual for chronically sick individuals. Even for well animals, a pen of cattle can loose money if fed to the proper endpoint—it is just that they will loose less money if fed to that endpoint.
Marketing is a major consideration—that is the relative difference between the price paid for cattle and that for which they are sold. In the case of an independent owner, it is often the difference between profit and loss, independent of the discussion above on proper time of marketing the animals. For the feedlot owning the cattle (or the capital limited outside owner), it is more interesting. Since the optimal feeding strategy is to feed until marginal net revenue decreases to average daily revenue for typical pens in the feedlot, the average daily revenue is important—and depends more on the difference between the prices paid and received for cattle. If the feedlot does an exceptionally good job of buying cattle low and selling them high, then the average daily revenue is high, and cattle are fed shorter times. In fact, if it is quite high days on feed approaches zero, and the feedlot simply becomes a holding pen for cattle being transferred in ownership—a cattle broker’s location. Understanding how a feedlot’s average daily net revenue may change over time is thus important to optimizing profit for cattle owned by that feedlot. Thus if average daily net revenue is projected to decrease in the coming months after a set of cattle may be sold (incoming cattle prices too high, market cattle prices declining, feed prices increasing), the argument is to feed the cattle longer.
Biological Constraints
The above economic discussion seems to ignore the resulting animal’s product and its growth. However, this is not the case because the animal’s value is quite dynamic, depending on carcass weight, quality, possible defects and other market factors; its profit also depends on efficiency of gain. These biological parameters are complex and have been the focus of much beef cattle research for over fifty years. Rather than summarize all the literature, an overview of the major factors affecting how animal value changes as feedlot animals approach slaughter endpoints follows.
Carcass weight is a major driver of revenue, and animal value increases in direct proportion unless other factors interact to decrease value per unit carcass weight. Thus, in most analysis, feeding animals to heavier weights usually increases profit. Constraints due to excessive carcass size in slaughter plants, or undesirably large muscle cuts limit carcass size by decreasing value of the carcass. Increasing cost of gain as animals age also constrain carcass size, usually as a result of an increasing proportion of the feed being used for animal maintenance (related to body weight) instead of gain. Hyer et al. (1986) also showed that as steers reached or exceeded normal market weight, feed intake decreased, further exacerbating the above effect of less feed available for gain.
Carcass weight and the yield of retail cuts in the carcass change with increasing body weight, and result in value differences as well. Pricing cattle on a live weight basis requires consideration of the relative increase in carcass weight as a proportion of live weight. But pricing cattle on either a live weight or carcass weight basis must also consider the decrease in retail cut yield as carcass fatness increases. As animals finish in a feedlot fatness increases, so beef yield as a proportion of carcass decreases, particularly so for genetically fatter animals. In the US this is called Yield Grade, and steep discounts for animal with higher Yield Grades effectively limit time on feed.
Although Yield Grades, or carcass yield, become less desirable with time on feed, carcass quality, Quality Grade in the US, generally improve. Genetics and feeding strategy affect carcass quality with certain breeds (and sires) exhibiting greater marbling and other improved meat qualities. Steroid status of the animal often affects marbling, and interacts with age the animal enters the feedlot. Aggressive anabolic implant use earlier in life seems to decrease marbling; the younger the animal is when entering the feedlot enhances marbling. This probably depends on the endpoint at which marbling is measured—calves are often fed longer before slaughter at a lighter weight than yearling or older cattle. Backfat of calves reaches a given level at a lighter body weight than for older cattle, so they are often slaughtered younger and lighter, to avoid carcass yield discounts and possibly resulting in decreased total value due to lighter carcasses. Yearlings may be more profitable if the cost of gain is high, and the trend to feed these older cattle increases with feed and grain prices.
An interesting interaction between frame size (mature weight of the animal) and optimal feeding period exists, with larger frame cattle benefitting from earlier feedlot entry, or smaller frame cattle benefitting by being grown on forage diets or pastures before feedlot entry. On forage diets, backfat does not increase with body weight as it does on feedlot rations (Sainz et al., 1995), thus the smaller frame animal can be fed to larger, more profitable weights after a period of restricted growth on a lower energy diet. The NRC (2000) accounts for this using an equivalent weight concept, the weights at which different animals reaches 28% body fat. Thus, one makes an adjustment on body weight to account for different frame size and management effects.
Useful Models
There are multiple interacting factors that affect the dynamic costs and returns for the finishing beef animal. A method to integrate the biology of the animal, its management, and market prices is needed to project animal and carcass characteristics through time, and associated costs and potential revenue. While the NRC (2000) can be used to make point estimates of animal performance and a good spreadsheet implementation adequate for a budget projection, it is severely limited in evaluating and projecting animal compositional and value changes through time. More dynamic tools are needed, and in the past decade, the Cornell group has developed the Cornell Value Discovery System to assist in decisions for individual growing cattle management (Guiroy et al., 2001; Fox et al., 2004; Tedeschi et al., 2004). The Cornell Value Discovery System software provides the following: 1) predicted daily gain, incremental cost of gain, and days to finish to optimize profits and marketing decisions while marketing within the window of acceptable carcass weights and composition; 2) predicted carcass composition during growth to avoid discounts for under- or overweight carcasses and excess backfat; and 3) allocates feed fed to pens to individual animals for the purpose of sorting of individuals into pens by days to reach target body composition and maximum individual profitability. This allows mixed ownership of individuals in pens, determination of individual animal cost of gain for the purposes of billing feed and predicting incremental cost of gain, and providing information that can be used to select for feed efficiency and profitability.
Oltjen et al. (1986a) developed a similar system, the Davis Growth Model (DGM), and implemented it in ration formulation and profit projection software, TAURUS (Ahmadi et al., 1994; Dunbar et al., 1994) which shows the daily costs and returns throughout a feeding period. The DGM is based on general cell number and size mechanisms of growth to predict net protein synthesis. It is integrated into the same net energy system used in the Cornell System to estimate gain of fat and lean tissue. The model was evaluated first with respect to its ability to predict growth and composition of steers as affected by nutrition, initial condition, frame size, and use of growth-promotants. Using 2 independent data sets, the model predicted empty BW and fat content with standard deviations of predicted minus observed of 14 and 10 kg, respectively (Oltjen et al., 1986b). No systematic biases were evident with respect to composition, frame size, or energy intake. However, fat gain was underpredicted (p<0.01) at high feed energy concentrations.
Although the DGM accounted for variations attributable to initial body composition and mature size, the model did not always yield acceptable estimates of fat gain. This was not unexpected, because fat accretion was computed after energy requirements for maintenance and protein gain were satisfied. Thus, any errors in estimates of maintenance or protein gain resulted in biased fat gain predictions. Garcia et al. (2007) compared the DGM with a dynamic French model (IGM) also developed to predict protein and fat deposition in growing cattle (Hoch and Agabriel, 2004). Both models gave accurate and precise predictions of body protein. They also performed well for prediction of body fat in continuously growing animals. However, DGM tended to underestimate body fat deposition during feed restriction periods. This suggests that DGM overestimated heat production during periods of low MEI. The IGM was not sensitive enough to MEI, because it overestimates body fat at low MEI and it underestimates body fat at high MEI. Also, IGM does not take into account ME concentration of the diet and thus did not simulate different growth trajectories for same MEI but different ME concentrations. These results suggest that model’s structure and equations for protein accretion in DGM and IGM are valid. These limitations require a focus on prediction of heat production during feed restriction periods for DGM, confirming the need for a variable maintenance component, and on mathematical formulation of feed energy utilization for fat synthesis for IGM to improve model sensitivity to MEI.
To improve the accuracy of the predictions of these systems, a more mechanistic approach is required to account for variable maintenance energy requirements and thus reduce the errors of the NRC (2000) and the Cornell System. Sainz and Bentley (1997) showed that the observed changes in maintenance energy expenditures were closely related to changes in visceral protein mass. A collaborative effort between scientists in New Zealand, Australia, and the United States developed a dynamic model of the visceral protein (v), muscle protein (m), and fat (f) pools (Soboleva et al., 1999). In the model, muscle and viscera each have an upper bound (m* and v*, respectively). For muscle, m* is genetically fixed, although the possibility of reaching this level depends on both the current intake (MEI) and nutritional history of the animal. However, v* is also affected by energy intake and depends on previous nutrition. Net energy for gain drives the growth of muscle and viscera. Heat production for maintenance depends on MEI and changes asymptotically to new levels when MEI changes resulting in a lag in change of maintenance requirements after intake changes. Additional information regarding kinetics of the growth model is given by Oltjen et al. (2000). The heat production per unit of protein mass of viscera is about 10 times that of muscle. Also, viscera responds faster than muscle to changing energy intake by the animal, but this change has some time lag. Therefore, maintenance requirement becomes a dynamic variable depending on nutritional history as well as current energy intake. Thus, the static form of maintenance function used in traditional feeding systems is probably inappropriate, especially for dynamic situations. One of the advantages of the way the model is formulated is that the performance of different functions describing animal heat production can be investigated. That is, the fit of the model to data, using either traditional NE concepts and maintenance energy, or more general functions for HP, can be compared with choose the best functional description.
We have recently refined this prediction system for ruminant animal growth and composition. With a new equation for viscera, the multiple regression prediction of heat production using m, v, and their accretion (Oltjen and Sainz, 2001) is also improved, as is the prediction of body fatness. New additions refine predictions at levels of energy intake at or below maintenance. Although the model provides the structure for predicting composition of growing cattle, not all its parameters have been estimated and evaluated. Barioni et al. (2006) added the variable maintenance representation from the sheep model to the DGM for beef cattle. Fitting beef cattle growth data, variable instead of fixed maintenance requirements for each experimental group significantly improved the precision of the model for fat and RE, confirming the conclusions of Sainz et al. (1995) that previous nutrition had substantial effects on maintenance energy expenditures and indicates that variable maintenance can significantly improve model predictions.
McPhee et al. (2007a,b) has extended the DGM to 4 fat depots: intermuscular, intramuscular, subcutaneous, and visceral, again based on DNA and cell size concepts. Fat depot parameters were estimated, and no differences between implant status and frame size were detected. The model currently underpredicts fat in all 4 fat depots for finishing steers fed high concentrate diets, which suggests that a secondary phase of hyperplasia may be occurring, which is not represented in the DGM. Future efforts will incorporate these more refined estimates of carcass quality into value systems and profit projection.
Most recently, Barioni et al. (2009) designed a hybrid algorithm to efficiently find optimal solutions to the time on feed and feedlot rationing problem. The algorithm has, in an internal loop, a linear diet optimizer and, in an external loop, a non-linear evolutionary algorithm (Eiben and Smith, 2003) to maximize profit, constrained by capital and feed availability. The optimum slaughter time is calculated based on simulations with the Davis Growth Model. The DGM simulates the average growth and body composition of each group of cattle, but intra-group variation is unaccounted for at this stage. For each iteration of the non-linear algorithm, a new least cost diet is formulated with the constraints for diet formulation defined by the non-linear algorithm and optimum slaughter date is then defined by the DGM outputs for the diet and seasonal prices variation informed. Analyses of performance have shown that feeding period and optimum liveweight are strongly affected by the feeding cost in Brazil. For high grain prices, optimum strategies include buying heavier animals and having shorter feeding periods. Diets with minimum cost of gain were not always best because of beef prices seasonality. Results indicate that as important as having low cost of production is to provide liveweight gains that allow slaughter in periods of highter prices. The combination of a linear (simplex) and a non-linear (evolution strategy) and dynamic simulation of animal growth produced robust solutions for the problem of optimizing feedlot operations allowing the identification of more promissing strategies.
|
|
|