24th Int. Symp. “Animal Science Days”, Ptuj, Slovenia, Sept. 21st−23rd, 2016. Acta argiculturae Slovenica, Supplement 5, 71–75, Ljubljana 2016 COBISS: 1.08 Agris category code: L10 EFFICIENT COW – ESTIMATION OF FEED INTAKE FOR EFFICIENCY TRAITS USING ON-FARM RECORDED DATA SESSION II: CATTLE BREEDING AND PRODUCTION Maria LEDINEK 1, Leonhard GRUBER 2, Franz STEININGER 3, Birgit FUERST-WALTL 4, Karl ZOTTL 5, Martin ROYER 6, Kurt KRIMBERGER 7, Martin MAYERHOFER 8, Christa EGGER- DANNER 9 Efficient Cow – Estimation of feed intake for efficiency traits using on-farm recorded data 1 University of Natural Resources and Life Sciences, Dep. Sustainable Agricultural Systems, Div. of Livestock Sciences, Gregor-Mendel-Str. 33, 1180 Vienna, Austria, e-mail: maria.ledinek@boku.ac.at 2 Agricultural Research and Education Centre Raumberg-Gumpenstein, Raumberg 38, 8952 Irdning, Austria, e-mail: leonhard.gruber@raumberg-gumpenstein.at 3 ZuchtData EDV-Dienstleistungen GmbH, Dresdner Str. 89/19, 1200 Vienna, Austria, e-mail: steininger@zuchtdata.at 4 Same address as 1, e-mail: birgit.fuerst-waltl@boku.ac.at 5 LKV Niederösterreich, Pater Werner Deibl-Str. 4, 3910 Zwettl, Austria, e-mail: karl.zottl@lkv-service.at 6 Same address as 2, e-mail: martin.royer@raumberg-gumpenstein.at 7 Same address as 2, e-mail: kurt.krimberger@raumberg-gumpenstein.at 8 Same address as 3, e-mail: mayerhofer@zuchtdata.at 9 Same address as 3, e-mail: egger-danner@zuchtdata.at ABSTRACT Increasing efficiency is necessary to cope with scarce resources and higher costs especially for energy and con- centrate. The Federation of Austrian Cattle Breeders (ZAR) started the project “Efficient Cow” in 2013 to evaluate ef- ficiency traits in cattle breeding under Austrian conditions. Data of approximately 5400 cows, i.e. 3100 Fleckvieh (dual purpose Simmental), 1300 Brown Swiss, 1000 Holstein kept on 167 farms were recorded over a whole year. Feed intake was predicted by a model considering animal and ration specific parameters. The observation of the individual feeding information considering the variety of feeding systems and ration compositions was the biggest challenge. A novel data encoding system for ration components was established to reflect different on-farm feeding situations correctly and to ensure a successful and structured further processing for intake prediction. A total of 1960 different rations could be reduced to 16 different ration types and therefore calculation methods depending on the way ration components were offered, namely mixed together, separately or without known amount or proportion in diet like pasture. Key words: cattle breeding, efficiency, phenotypes, dairy cows, feed intake prediction, on-farm data collection produce more milk to be as efficient as smaller cows. But feed intake capacity did not develop in parallel with milk production. Therefore higher concentrate diets are nec- essary to meet demand. Steinwidder (2009) calculated a proportion of concentrate of 18 % for a cow with 550 kg but of 27 % for a cow with 850 kg. In case of insufficient nutrient supply, a negative energy balance especially in early lactation leads to a higher risk for diseases and in- fertility (Martens, 2012). Results of a survey among Austrian dairy farmers 2012 (Steininger et al., 2012) revealed their rising interest in health and efficiency traits. Beside this the discussion about greenhouse gas emissions was another reason for 1 INTRODUCTION Increasing efficiency is necessary to cope with scarce resources and higher prices especially for energy and concentrate, but also when prices for products are under pressure. Because total costs in dairy production consist of feed costs for more than 50 % (de Haas et al., 2014), feed efficiency is an important part to increase herd profitability. Feed efficiency is however only one as- pect of efficiency. Efficiency should also include aspects of health, fertility and longevity. Over the past decades milk production and therefore live weight of dairy cows has increased (Krogmeier, 2009). Heavier cows have to Acta agriculturae Slovenica, Supplement 5 – 201672 M. LEDINEK et al. starting the project “Efficient Cow” in Austria in 2012, headed by the Federation of Austrian Cattle Breeders (ZAR). As the possibilities of recording efficiency related traits in research herds are limited in Austria, the project aims at on-farm recording. Aside from that a reasonable number of animals also enables genetic analyses of the new defined efficiency traits. Because of the lack of in- dividual measurements of feed intake, novel strategies for recording and estimating feed intake had to be devel- oped considering the big variety of diet composition and feeding systems in Austria. Furthermore information of characteristics of diets and of feed and nutrient intake is essential for modelling greenhouse gas emissions. This paper focuses on the methodical way from data collection to the feed intake prediction models. 2 MATERIAL AND METHODS 2.1 DATA COLLECTION The Federation of Austrian Cattle Breeders (ZAR) initiated the project “Efficient Cow” at the end of the year 2012 with a one-year data collection in 2014. The objec- tive of “Efficient Cow” was to develop efficiency parame- ters in cattle breeding considering Austrian circumstanc- es. Efficiency combines already used traits like milk, beef, health and functional traits, and other traits which are relevant for feed efficiency. Therefore beside data, which are included in the routine performance recording, ad- ditional parameters like live weight, body measurements and parameters describing diets, feed quality and health were collected at each performance testing during the whole year 2014. Data of nearly 5400 cows (3100 Fleck- vieh – dual purpose Simmental, 1300 Brown Swiss, 1000 Holstein) kept on 167 farms were collected. Farms were selected to cover the diverse production environments in Austria ranging from mountainous regions to intensive farms in climatically favourable regions. Despite this, the herd size with 32.6 cows is approximately twice as high as the Austrian average (Steininger et al., 2015). 2.2 FEEDING SYSTEMS AND PARAMETERS The observation of the individual feeding informa- tion considering different feeding systems and ration compositions was the biggest challenge. The information on ration composition needs to be structured in such way that it can be used for feed intake estimation with the prediction model no. 1 of Gruber et al. (2004). This equa- tion considers, inter alia, the influence of forage quality (NELForage, MJ NEL kg DM −1) and of the total amount of concentrate (kg cow−1) in the diet. The amount of concentrate can be measured rela- tively accurately, if it is offered separately per automation, but less precisely if it was offered manually. Despite these inaccuracies, the separately fed amounts of concentrate were assumed to be fed without residues. The challenge was to find a method to calculate the concentrate intake, if a total mixed ration (TMR) or partial mixed ration (PMR) is fed. The amount of concentrate depends on the intake of mixed ration, but at the same time the intake of mixed ration depends on the total amount of concen- trate. So methods to calculate concentrate supplementa- tion had to be developed depending on the type of ration (TMR, PMR, separately supplemented concentrate SEP) and special characteristics. Another challenge was to integrate farmers’ state- ments that a little amount of forage had been fed sepa- rately to the main (mixed) ration. For example, 1 kg hay was scattered over the TMR (80  % forage, 20  % con- centrate) to motivate cows to eat more. This separately offered component must not be integrated into the re- maining mixed ration if calculating feed intake. Strictly speaking, this ration is not a TMR anymore, but the way to calculate feed intake is equal to a TMR combined with a special formula to integrate the separately fed forage. Therefore in this study such a ration is still understood as TMR but with a special calculation module. So the amount of the separated hay is assumed to be known like the separately fed concentrate, but the intake of the TMR is depending on individual parameters like milk yield and live weight and is therefore estimated with the feed intake model considering the hay. These separately fed amounts of a ration component are defined as “fixed” components. Fixed components are assumed to be eaten without feed residues, so that the accurate amount is not an unknown variable. Contrary to these fixed parts of ra- tion, feed intake of the 80 % forage and 20 % concentrate of the TMR is not known, but it can be estimated because of the known composition of the mixed ration. The ratio of forage intake of the fixed components to the total for- age intake makes it possible to weight and mathemati- cally express NELForage now considering both, mixed and fixed forage components in the total ration. The third challenge was to handle components, where no amount or proportion was recorded. For exam- ple the diet consists of pasture with supplementation of preserved food. The offered and known amounts of pre- served food are too much, as that they can be assumed to be eaten fix. The ratio of pasture to preserved food is unknown. This constellation of ration components led to the introduction of “ad  lib”-components. Here the ratio of offered mixed forage components to the poten- Acta agriculturae Slovenica, Supplement 5 – 2016 73 EFFICIENT COW – ESTIMATION OF FEED INTAKE FOR EFFICIENCY TRAITS USING ON-FARM RECORDED DATA tial mixed forage intake was used to assume a ratio for NELForage calculation. Therefore the data had to be ex- pressed in kg cow−1. So the ration components were partitioned into mixed, fixed and ad lib components, which describe the component type. Each main ration type (TMR, PMR and SEP) can thus be modified with a fixed and/or ad lib for- age component. The simplest diet consists only of mixed forage components. Overall, 16 different combinations of the component types mixed forage, mixed concentrate, fixed forage and fixed concentrate and ad lib forage were defined. So a standard TMR only consists of mixed for- age and mixed concentrate, the PMR has additionally fixed concentrate, and a SEP only mixed forage and fixed concentrate. The encoding of the ration components ac- cording their component type reflects the different feed- ing systems and diets of the dairy cows in a transparent way. To ensure high data quality, completed forms of the farmers had to be checked across different form types and dates within each farm before finally entering data into the database. Implausibilities were clarified directly with the farmers or the person responsible for the on- farm data collection. The following data had to be recorded: – start date of ration and used concentrate mix- tures – three feeding groups: lactation, additional high lactation if necessary and dry cows – ration type: TMR, PMR and SEP – component type: mixed, fixed and ad lib compo- nents – category of forage considering botanical ori- gin (grassland, legumes, forage maize, straw), conservation (hay, silage, fresh) and number of mowing – concentrate composition (proportion of barley, wheat, …) – commercial compound feed and nutrient content – feed samples for analysis of forage in the labora- tory for feed analyses of the chamber of agricul- ture in Austria – individual amount of concentrates fed separately from forage (kg/cow and day) 2.3 ESTIMATION OF FEED INTAKE The individual daily feed intake estimation was con- ducted in cooperation with the Austrian Agricultural Re- search and Education Centre Raumberg-Gumpenstein. As individual feed intake was impossible to measure on-farm, the total feed intake (DMI) prediction model no. 1 for separated concentrate supplementation of Gru- ber et al. (2004) was used for calculation: DMI = 3.878 + Country * Breed + Parity + Day in Milk + bBW * BW + bMilk * Milk + bConcentrate * Concentrate + 0.858 * NELForage This empirical model considers the fixed effects of breed and country, management level, parity, stage of lactation depending on day in milk and the regression coefficient for the energy content of forage (NELForage). Depending on the day in milk the regression coefficients for body weight (bBW), milk performance (bMilk) and for amount of concentrate (bConcentrate) have to be calculated. This shows the influence of the stage of lactation on milk performance, live weight and on forage substitution (Gruber et al., 2004). The original model no. 1 only covers diets, where concentrate is supplemented separately from forage. For calculating with a TMR and PMR, the input parameters concentrate amount had to be expressed mathemati- cally depending on feed intake, concentrate proportion in mixed ration (mixed concentrate) and separately fed fixed concentrate. If the ration additionally had a fixed forage component, NELForage had to be expressed accord- ing the characteristics of the ration type. The adaption of the chosen equation was preferred to take advantage of the high coefficient of determina- tion (R2 = 86.7 %) and the low residual standard devia- tion (RSD = 1.32 kg DM) compared to prediction model no. 5 for TMR (R2 = 83.5 %, RSD = 1.46 kg DM) (Gru- ber et al., 2004). Jensen et al. (2015) evaluated the up-to- date feed intake models of NRC (2001), of Volden et al. (2011), TDMI-Index (Huhtanen et al., 2011), Wagenin- gen-DCM (Zom et al., 2012a, 2012b) and TMR-Model no. 5 (Gruber et al., 2004) for dry matter intake by dairy cows fed TMR and found the Gruber model to be the most accurate one. 3 RESULTS AND DISCUSSION Approximately 1960 different diets were recorded, 1932 were potentially relevant for intake estimation, but under consideration of data quality, only 1890 could fi- nally be used for further processing. On the whole 1260 forage analyses were available for calculating the nutri- ents of 570 forage components without analyses. This method ensures site- and management adapted assump- tions of nutrients instead of using tabulated data. Ap- proximately 2280 different feeds including 1830 forage components and 438 concentrates as well as compound feeds were needed for describing the diets. Finally the Acta agriculturae Slovenica, Supplement 5 – 201674 M. LEDINEK et al. 1960 diets could be reduced to 16 different types of ra- tions due to the possible combinations of the component types. For each ration type another mathematical adap- tion of feed intake model had to be developed. These numbers show the diversity and complexity of feeding systems and ration compositions of the present investi- gation. For this reason a prediction model had to be cho- sen, which reflects ration composition and forage quality parameters besides animal individual factors like parity, stage of lactation, live weight and milk yield. Furthermore estimation should be individual and as accurate as possible to enable calculation of efficiency traits. The feed intake equation by NRC (2001) considers similar animal related criteria, but not feed-specific pa- rameters like forage quality or concentrate level. The feed intake model by Volden et al. (2011) belongs to a semi- mechanistic feeding model and represents a fill-factor system. It combines the feed intake capacity, which is de- termined by live weight, stage of lactation, parity, breed and milk yield with the filling effect of the feed. Simi- larly the Wageningen-Dairy Cow Model (DCM) (Zom et al., 2012a, 2012b) works, but without considering milk yield and live weight for feed intake capacity. The TDMI- Index (total dry matter intake) system (Huhtanen et al., 2011) combines the silage-DMI (SDMI)-Index and the concentrate-DMI (CDMI)-Index. While the SDMI-In- dex pictures the forage quality including parameters like digestibility and fermentation quality, the CDMI-Index considers amount and composition of concentrate. Model no. 5 for TMR (Gruber et al., 2004) in- cludes concentrate proportion of mixed ration instead of amount like in equation no. 1 for separate concentrate supplementation. Although Jensen et al. (2015) found the model no. 5 for TMR (Gruber et al., 2004) to be the most accurate one compared with the before mentioned up-to-date models, it was not chosen for estimating TMR in this project. Instead model no. 1 for separated con- centrate supplementation was applied, and modified for PMR and TMR. Because a specific prediction model for PMR of Gruber et al. (2004) does not exist, model no. 1 had to be adapted to it anyway. A PMR is a more general type of a TMR, because of the additional separately fed concentrate. Furthermore using the same equation only with adaptions to the ration type guarantees a uniform estimation of feed intake. Another advantage of the models by Gruber et al. (2004) is the special consideration of the influence of stage of lactation on the regression coefficients for live weight, milk yield and concentrate level. Therefore they vary with day in milk. Thus, the changes of physiologi- cal stage from early to late lactation are taken in account, i.e., the change from a catabolic to an anabolic metabo- lism (Korver, 1982). Forage substitution by concentrate is higher at the end of lactation, and the influence of live weight decreases due to gained body fat (Gruber et al., 2004). 4 CONCLUSIONS The recording of novel phenotypes from about 5300 cows on 167 farms, especially of feeding information per individual, was a big challenge. The feeding data base had to be designed using the experiences with the survey forms of the first half year of data collection. Without this experience revealing the diversity of feeding systems of the 167 farms, ration compositions and the way to de- scribe this could not have been considered for data enter- ing and feed intake estimation. Rations had to be partitioned into mixed, fixed and ad  lib components, which reflect the way the feed was offered like fixed concentrate with automation or manu- ally, mixed concentrate together with mixed forage com- ponents in a TMR or PMR. Ad lib components had to be inserted into the data encoding system, because mostly the amount or proportion of pasture was not known. This system of encoding ration components was the only possibility, to make the variety of diets and feeding systems handy for feed intake estimation. The estimation model had to be adapted to the 16 cases of ration types, which results from the 16 possible combinations of dif- ferent categories of component Without this novel system of handling the on-farm information, the estimation of feed intake and calcula- tion of mostly individual nutrient contents in finally in- dividual total rations would not have been possible with data observed on-farm. 5 ACKNOWLEDGEMENTS We gratefully acknowledge funding by the Fed- eral Ministry of Agriculture, Forestry, Environment and Water Management (BMLFUW) in Austria, the Federal States of Austria and the Federation of Austrian Cattle Breeders (Project 100681, Efficient Cow). Further we wish to thank all project partners for their support. 6 REFERENCES De Haas, Y., Pryce, J. E., Berry, D. P., Veerkamp, R. F. (2014). Ge- netic and genomic solutions to improve feed efficiency and reduce environmental impact of dairy cattle. In Proceed- ings of 10th Word Congress on Genetics Applied to Livestock Production, Vancouver 2014. Retrieved from https://event. crowdcompass.com/wcgalp2014/activity/krfCbPBxzB Acta agriculturae Slovenica, Supplement 5 – 2016 75 EFFICIENT COW – ESTIMATION OF FEED INTAKE FOR EFFICIENCY TRAITS USING ON-FARM RECORDED DATA Gruber, L., Schwarz, F. J., Erdin, D., Fischer, B., Spiekers, H., Steingass, H., Meyer, U., et al. (2004). Vorhersage der Futteraufnahme von Milchkühen – Datenbasis von 10 Forschungs- und Universitätsinstituten Deutschlands, Ös- terreichs und der Schweiz. In Qualitätssicherung in land- wirtschaftlichen Produktionssystemen. 116. VDLUFA-Kon- gress, Rostock 2004 (pp. 484–504). Speyer: VDLUA-Verlag. Huhtanen, P., Rinne, M., Mäntysaari, P., Nousiainen, J. (2011). Integration of the effects of animal and dietary factors on total dry matter intake of dairy cows fed silage-based diets. Animal, 5, 691–702. Jensen, L. M., Nielsen, N. I., Nadeau, E., Markussen, B., Nør- gaard, P. (2015). Evaluation of five models predicting feed intake by dairy cows fed total mixed rations. Livestock Sci- ence, 176, 91–103. Korver, S. (1982). Feed intake and production in dairy breeds de- pendent on the ration. Doctoral thesis. Wageningen: Land- bouwhogeschool te Wageningen, 139 p. Krogmeier, D. (2009). Zusammenhänge zwischen Nutzungs- dauer und Körpergröße unter besonderer Berücksichti- gung des Stallsystemes bei Braunvieh und Fleckvieh. Züch- tungskunde, 81, 328–340. Martens, H. (2012). Die Milchkuh – Wenn die Leistung zur Last wird! In Milchproduktion – Status quo und Anpassung an zukünftige Herausforderungen. 39. Viehwirtschaftliche Fachtagung, Irdning 2012, Bericht LFZ Raumberg-Gumpen- stein (pp. 35–42). Irdning: AREC Raumberg-Gumpenstein. NRC (National Research Council). (2001). Nutrient require- ments of dairy cattle. 7th edition. Washington D. C.: National Academy Press. Steininger, F., Fuerst-Waltl B., Pfeiffer, C., Fuerst, C., Schwarzen- bacher, H., Egger-Danner, C. (2012). Participatory develop- ment of breeding goals in Austrian dairy cattle. Acta Agri- culturae Slovenica, Supplement 3, 143–147. Steininger, F., Fuerst, C., Fuerst-Waltl, B., Gruber, L., Mayer- hofer, M., Ledinek, M., Weissensteiner, R., et al. (2015). Efficient Cow – Strategies for on-farm collecting of pheno- types for efficiency traits. ICAR Technical Workshop, Kra- kow 2015. ICAR Technical Series 19, 167–173. Steinwidder, A. (2009). Modellrechnungen zum Einfluss der Lebendmasse von Milchkühen auf Futtereffizienz und Kraftfutterbedarf. In Werte – Wege – Wirkungen: Biolan- dbau im Spannungsfeld zwischen Ernährungssicherung, Markt und Klimawandel. Beiträge zur 10. Wissenschaftsta- gung Ökologischer Landbau, Zürich, 11.−13. Februar 2009. Vol. 2 – Tierhaltung, Agrarpolitik und Betriebswirtschaft, Märkte und Lebensmittel (pp. 30−33). Berlin: Verlag Dr. Köster. Volden, H., Nielsen, N. I., Åkerlind, M., Larsen, M., Havrevoll, Ø., Rygh, A. J. (2011). Prediction of voluntary feed intake. In H. Volden (Ed.), The Nordic Feed Evaluation System. EAAP Publication No. 130 (pp. 113−126). Wageningen: Wageningen Academic Publishers. Zom, R. L. G., Andre, G., Van Vuuren, A. M. (2012a). Develop- ment of a model for the prediction of feed intake by dairy cows: 1. Prediction of feed intake. Livestock Science, 143, 43–57. Zom, R. L. G., Andre, G., Van Vuuren, A. M. (2012b). Develop- ment of a model for the prediction of feed intake by dairy cows: 2. Evaluation of prediction accuracy. Livestock Sci- ence, 143, 58–69.