Prediction of 100-d and 305-d Milk Yields in a Multibreed Dairy Herd in Thailand Using Monthly Test-Day Records 1/

Transcription

1 Predicion of -d and 35-d Milk Yields in a Mulibreed Dair Herd in Thailand Using Monhl Tes-Da Records / Skorn Koonawooririron *, Mauricio A. Elzo /, Sornhep Tumwasorn, and Wiro Sinala 3/ Deparmen of Animal Science, Facul of Agriculure, Kasesar Universi, Bangkok 9, Thailand. Absrac The abili of eigh procedures o predic -d and 35-d milk ields using monhl esda records was esed using 8,45 dail ields from 88 cows in a mulibreed dair herd provided b he Sakon Nakhon Agriculural Research and Training Cener. The eigh procedures were: es inerval mehod, gamma funcion, mixed log linear, and second, hird, fourh, fifh, and sixh degree polnomial models. The breed groups represened in he mulibreed herd were HF, /HF /RS, and 3/4HF /4RS. Predicion of -d and 35-d milk ields b he eigh procedures were compared wih acual -d and 35-d milk ields wihin breed group x lacaion number x calving age and breed group x lacaion number x calving season subclasses. Leas squares means of individual cow differences prediced and acual -d and 35-d milk ields were compued for each subclass. Number of significan leas square means of differences and ranking of models wihin and across subclasses for -d and 35-d were used o evaluae he predicive abili of he eigh procedures. The highes-ranking model for -d was model 4 (hird degree polnomial) and for 35-d was model 3 (second degree polnomial). However, no procedure was uniforml beer across all subclasses. Thus, perhaps several models migh be needed for a geneic evaluaion of he animals in his mulibreed populaion. If compuaional simplici were he primar goal, hen perhaps a single model (model 3) migh suffice. However, he resuls of his sud appl onl o he daa se and he mulibreed populaion used here. To obain resuls of naional relevance, his sud needs o be repeaed wih a larger mulibreed populaion ha more accurael represens he Thai mulibreed populaion. Ke words: dair cale, milk ield, es-da ield, predicion, mulibreed Inroducion Recording of milk ields is essenial for geneic improvemen and herd managemen in dair cale. Under increasing pressure o reduce cos, numerous milk-esing schemes have been developed in man counries. One of he mos widel used is monhl recording. In Thailand, monhl es-da records are used o compue cumulaive producions of milk and fa o -d and 35-d for dair geneic evaluaion purposes. The Dair Promoion Organizaion (DPO), and probable oher organizaions in Thailand, compue monhl milk ields using a single es-da milk ield sample, and hen, hese monhl esimaes are used o compue he accumulaed -d and 35-d milk ields. This procedure is no appropriae for individual animals because i will, in / This research was suppored b he Florida Agriculural Experimen Saion and a gran from he Thailand Research Fund under he Roal Golden Jubilee Projec, and approved for publicaion as Journal Series No.R-865. / Deparmen of Animal Sciences, Universi of Florida, Gainesville, FL 36-9, USA * Correspondence: Presen address 3/ Sakon Nakhon Agriculural Research and Training Cener, P.O. Box 3, Pungkone, Sakon Nakhon 476, Thailand. skornk@homail.com

2 mos cases, eiher overesimae or underesimae accumulaed milk ields. Noice ha his procedure is differen from he es-inerval mehod (Sargen e al., 968; Norman e al., 999), which compues oal milk ields of inervals beween wo consecuive es das using he average milk ield of hese wo es-das. Alhough he es-inerval mehod is he mos widel used procedure o compue cumulaive milk producion rais, predicion of hese milk rais could poeniall be improved b using a linear or a nonlinear funcion. Koonawooririron e al. () showed ha he second degree polnomial was he bes ou of seven models o predic dail and 35-d milk ields and he sixh degree polnomial model was he bes for he predicion of -d milk ield wihin breed group x lacaion number x calving age and breed group x lacaion number x calving season subclasses in a Holsein Friesian-Red Sindhi herd in he Norheas of Thailand, when using all dail lacaion records. Milk recording organizaions in Thailand sample milk producion rais on a monhl basis. These are he records used for geneic animal evaluaion in dair cale. Thus, milk predicion models need o be revalidaed under a es-da sampling sraeg, and hen compared o he es-inerval mehod for heir abili o predic -d and 35-d milk producion ields under Thai condiions. Thus, he objecives of his sud were o assess he predicive abili of he es-inerval mehod and of seven models (gamma, mixed log linear, second o sixh degree polnomial models) o predic -d and 35-d milk ields based on monhl es-da records relaive o he acual -d and 35-d milk ields of individual cows wihin breed group x lacaion number x calving age and breed group x lacaion number x calving season subclasses. Maerials and Mehods Animals, Managemen, and records This sud used he same daa se as Koonawooririron e al. (). Thus, onl a reduced descripion of i will be given here. Dail lacaion ields (8,45, 5 o 35 d) from 75 Holsein Friesian (HF), 8 ½ HF ½ Red Sindhi (RS), and 5 ¾ HF ¼ RS dams were colleced a he Sakon Nakhon Agriculural Research and Training Cener (SARTC) beween 997 and 999. Cows were assigned o hree breed groups according o heir breed composiion (HF, /HF /RS, and 3/4HF /4RS). Animals of all breed composiions were milked wice a da, and raised under he same nuriional (grass and concenrae plus minerals) and managemen condiions. Cows were arificiall inseminaed up o hree imes, and if no pregnan a 6 d afer las inseminaion, he were placed wih a clean up bull. Pregnan cows were dried off wo monhs prior o calving. Lacaion number was classified as firs, second, hird, and fourh and laer lacaions. Calving seasons were defined as winer (November o Februar), summer (March o June), and rain (Jul o Ocober). Two calving ages per lacaion were defined. This resuled in eigh lacaion x calving age subclasses: ) calving age less han 3 monhs x lacaion, ) calving age equal o or greaer han 3 monhs x lacaion, 3) calving age less han 44 monhs x lacaion, 4) calving age equal o or greaer han 44 monhs x lacaion, 5) calving age less han 6 monhs x lacaion 3, 6) calving age equal o or greaer han 6 monhs for he hird lacaion, and 7) calving age greaer han 6 monhs x lacaion 4 and greaer. Models and Daa Analsis To accomplish he objecives of his research lacaion records from he SARTC daa se had o be sampled according o he curren milk sampling procedure used in Thailand. Monhl

3 3 sampling of milk producion rais is he prevalen ssem in Thailand. Because of colosrum, sampling usuall begins 5 d posparum. Thus, dail milk ields from he SARTC daa se were sampled on das 5, 35, 65, 95, 5, 55, 85, 5, 45, 75, and 35 of each lacaion. These measuremens per lacaion were used as monhl es-da records o assess he predicive abili of he es-inerval mehod (TIM) and he seven predicion equaions o predic -d and 35-d milk ields used b Koonawooririron e al. (). Firsl, he monhl es-da records were used o predic individual cow lacaion dail milk ields (5 o 35d) wihin breed group x lacaion x calving age and breed group x lacaion x calving season subclasses using he es-inerval mehod and he seven predicion equaions. Secondl, accumulaed -d and 35-d milk ields for individual lacaions were compued using hese eigh predicion procedures. The prediced values for -d and 35-d milk ields from each procedure were deviaed from heir corresponding acual -d and 35-d milk ields. Leas squares means of -d and 35-d milk ield deviaions (prediced minus acual milk ields) for he es-inerval mehod and he seven predicion procedures were obained separael for each se of breed group x lacaion x calving season and breed group x lacaion x calving age subclasses. The saisical model used was: where d ijk = µ + subclass i + model j ijk d ijk = difference beween prediced and acual milk producion of cow k a - d or a 35-d of lacaion, wihin subclass i and model j, µ = overall mean, subclass i = i h breed group x lacaion x calving season or breed group x lacaion x calving age, model j = j h predicion model and, e ijkl = residual. All effecs in he model were assumed o be fixed, excep for he residual erm ha was assumed o be independen, idenicall disribued wih mean zero and a common variance. T- saisics were hen used o es if he prediced and acual -d and 35-d milk ields differed significanl. For compleeness, he predicion equaions used o compue accumulaed milk ields a d and 35 d are briefl described below. For furher deails on predicion models o 7, see Koonawooririron e al. (). ) Tes-inerval mehod: P + P TMY = (P [] k i i + + D) [( ) Di] (Pk+ Dk+ ) i= where TMY is oal milk ield, P is he milk ield of he firs es-da record, D is inerval beween five das afer calving and he firs record, P i is he milk ield on es-da i (i =,, k), D i is he inerval beween es-record record i and i (i =,, k), P k+ is he milk ield on he las es-da before dring off, and D k+ is he inerval beween he las es-da and he da a cow was dried off (Sargen e al., 968). ) Predicion model : Gamma funcion (Wood, 967): b c = a e []

4 4 where is he milk ield on da in each subclass, a is he iniial ield of lacaion, b represens he increasing slope, and c represens he decreasing slope. Compuaions were done using he naural logarihm of equaion, i.e., ln = ln a ln c + ε where ε is he residual. Thus, he prediced ield on da was, = exp (ln ). 3) Predicion model : mixed log second-degree polnomial (Ali and Schaeffer, 987), = b γ γ w w [3] 3 where = milk ield on da, γ = /35, w = ln(35/), = das since calving or das in milk, b, b, b, b 3, and b 4 are regression coefficiens, and e is he residual. 4) Predicion models 3, 4, 5, 6, and 7: second, hird, fourh, fifh, and sixh polnomial regression models, Model 3: = b [4] Model 4: Model 5: Model 6: Model 7: 3 b = e [5] 3 4 b b b = e [6] = e [7] = e [8] where = milk ield on da, = das since calving, b, b, b, b 3, b 4, b 5 and b 6 are regression coefficiens, and e is he residual. Program PROC REG of he SAS saisical ssem (SAS, 99) was used o compue he regression coefficiens in models hrough 7 and o obain he prediced dail milk ields (5 o 35 d) for all models. Accumulaed ( d and 35 d) individual cow acual and prediced milk ields, and prediced (es-inerval mehod, models o7) minus acual accumulaed milk ield deviaions were compued using he general SAS program (SAS, 99). Leas squares means of accumulaed milk ield deviaions per subclass were obained using he LSMEANS saemen of PROC GLM (SAS, 99). T-ess were used o assess he significance of he accumulaed deviaions per model wihin and across subclasses. Resuls and Discussion Prediced -d milk ields b eigh procedures relaive o acual ields Calving age subclasses. Leas squares means of -d milk ield deviaions (prediced minus acual milk ields) of he es-inerval mehod and seven predicion models b breed group x lacaion x calving age subclasses are presened in Table. Models, 4, 5, 6, and 7 had nonsignifican -d milk ield deviaions (prediced minus acual milk ield (P >.5) for all subclasses. The es-inerval mehod (TIM) had wo, and model had hree significan differences of LS means for -d milk ield deviaions (a leas P <.5). Model 3 had four significan differences (a leas P <.5). Among he models wih nonsignifican deviaions, wo iers can be disinguished: models 4 and 5 (mean absolue deviaion = 6.3 kg), and models, 6, and 7 (mean absolue deviaion of abou 8.6 kg). Thus, models 4 and 5 appear o be good choices for -d milk predicions. However, because i is simpler han model 5, model 4 (hird degree polnomial) should be he model of choice. Noice ha his choice of model for -d milk ields using es-da records here differed from he bes model found (model 7) when using all

5 5 dail records b Koonawooririron e al. (). The number of es-das considered (and probabl he es-da values hemselves) would likel affec -d predicions, and he model ha will have he smalles deviaions from acual -d milk ields. Table. Leas squares means of -d milk ield deviaions (prediced minus acual milk ields) of he es-inerval mehod and seven predicion models b breed group x lacaion x calving age subclasses Breed group / Lacaion number Calving age No. of Acual Yield lacaions (kg) TIM / Model Model Model 3 Model 4 Model 5 Model 6 Model 7 HF < 3 mo 8, HF > 3 mo,.9-4.3* -46.7** ** HF < 44 mo 9, HF > 44 mo 4, * ** HF 3 < 6 mo 6, HF 3 > 6 mo 3, HF > 4 all ages 8, /HF /RS > 3 mo * /HF /RS < 44 mo 3, /HF /RS > 44 mo 3, /HF /RS 3 < 6 mo 3, ** /4HF /4RS < 3 mo, ** /4HF /4RS > 3 mo 3, * All All All 6, * / HF = Holsein Friesian, RS = Red Sindhi / Tes -inerval mehod 3/ Model : b c = a e Model : = b γ γ 3w 4 w Model 3: = b Model 4: 3 = b 3 Model 5: 3 4 = b 3 4 Model 6: = b Model 7: = b Where is milk ield a da in lacaion, = das in lacaion, γ = /35, w = ln(35/) * significan (P <.5) for Ho: LSMEAN=, ** highl significan (P <.) for Ho: LSMEAN=. 6 Calving season subclasses. Table shows LS means of -d milk ield deviaions of he TIM and seven predicion models b breed group x lacaion x calving season subclasses. Resuls for calving season subclasses were similar o hose of calving age subclasses. Models 4 hrough 7 had nonsignifican -d milk ield deviaions (P >.5) for all breed group x lacaion x calving season subclasses. Leas squares means of prediced minus acual -d milk ields from TIM and model were significanl differen (P <.5) for wo ou of wen calving season subclasses. Model 3 had hree significan differences (P <.). The leas accurae model was model ; i had four significan differences (a leas P <.5). As i happened wih calving age subclasses, he model of choice overall was model 4 (hird degree polnomial). All procedures (he es-inerval mehod and he seven predicion models) had LS means of -d milk ield deviaions ha varied in value and level of significance of he wo ses of subclasses above (calving age and calving season). This variabili suggess ha here was no uniforml beer model across all subclasses in his daa se. However, an overall ranking across calving age and calving season subclasses can be consruced using ) he number of significan subclasses found per procedure, and ) he level of significance of he overall LS mean of he prediced minus he acual -d milk ields. The resuling ranking (firs o las) was: ) model 4 (hird degree polnomial), ) model 5 (fourh degree polnomial), 3) model 6 (fifh degree

6 6 polnomial), 4) model 7 (six degree polnomial), 5) model (mixed log second degree polnomial), 6) TIM (he es-inerval mehod), 7) model 3 (second degree polnomial), and 8) model (gamma funcion). Table. Leas squares means of -d milk ield deviaions (prediced minus acual milk ields) of he es-inerval mehod and seven predicion procedures b breed group x lacaion x calving season subclasses Breed group / Lacaion number Calving season / No. of Acual ield lacaions (kg) TIM 3/ Model Model Model 3 Model 4 Model 5 Model 6 Model 7 HF Winer 9, * -6.3** ** HF Summer 3, * HF Rain HF Winer 3, * -8.** ** HF Summer 7, HF Rain 3, HF 3 Summer, HF 3 Rain 7, HF > 4 Winer 9, HF > 4 Summer 4, HF > 4 Rain 5, /HF /RS Winer * /HF /RS Rain /HF /RS Winer, * /HF /RS Summer, /HF /RS Rain 3, * /HF /RS 3 Winer, ** /HF /RS 3 Summer, /4HF /4RS Winer 4, ** /4HF /4RS Summer, All All All 6, * / HF = Holsein Friesian, RS = Red Sindhi / Winer = November Februar, Summer = March June, Rain = Jul Ocober 3/ Tes -inerval mehod 4/ Model : b c = a e Model : = b γ γ 3w 4 w Model 3: = b Model 4: 3 = b 3 Model 5: 3 4 = b 3 4 Model 6: = b Model 7: = b Where is milk ield a da in lacaion, = das in lacaion, γ = /35, w = ln(35/) * significan (P <.5) for Ho: LSMEAN=, ** highl significan (P <.) for Ho: LSMEAN=. This ranking indicaed ha he abili of TIM o predic individual cow -d milk ields was inermediae compared o ha of he seven predicion models. In addiion, he overall LS mean of he prediced minus acual -d milk ield for TIM was he onl significan value (P <.5). These resuls indicae ha he TIM is no as accurae as an of he oher seven procedures o predic individual cow -d milk ield using monhl es-da records. The ranking of -d milk ield predicive abili of models hrough 7 using monhl es-da records (4, 5, 6, 7,, 3, ) was similar o he ranking of he same models using all dail lacaion records (7, 5, 6,, 4, 3, ; Koonawooririron e al., ). Onl wo procedures changed heir ranking: model 4 ranked firs, and model 7 ranked fourh here, and heir corresponding rankings when using all dail records were fourh and firs, respecivel. These changes in ranking were solel due o he use of differen numbers of dail records for he

7 7 predicion of lacaion curves. Probabl a differen se of monhl es-da would also have produced a differen ranking. If a single lacaion model were o be chosen for -d milk ield geneic evaluaions, a hird degree polnomial (model 4) would probabl be an appropriae model for boh calving age and calving season subclasses in his populaion. However, if a more accurae accounabili of -d milk ield per cow wihin calving age and(or) calving season subclasses were desired, perhaps a se of lacaion models raher han a single model would be required for he geneic evaluaion ssem. Prediced 35-d milk ields b eigh procedures relaive o acual ields Calving age subclasses. Leas squares means of individual cow 35-d milk ield deviaions using TIM and he seven predicion models wihin breed group x lacaion x calving age subclasses are presened in Table 3. Table 3. Leas squares means of 35-d milk ield deviaions (prediced minus acual milk ields) for he es-inerval mehod and seven predicion models b breed group x lacaion x calving age subclasses Breed group / Lacaion number Calving age No. of Acual Yield lacaions (kg) TIM / Model Model Model 3 Model 4 Model 5 Model 6 Model 7 HF < 3 mo 8, HF > 3 mo, HF < 44 mo 9, HF > 44 mo 4 3, * -69.* * HF 3 < 6 mo 6 3, HF 3 > 6 mo 3 3, HF > 4 all ages 8, /HF /RS > 3 mo 4,4..* /HF /RS < 44 mo 3, * ** /HF /RS > 44 mo 3, /HF /RS 3 < 6 mo 3, /4HF /4RS < 3 mo 3, * 3/4HF /4RS > 3 mo 3 3, * 35.8** ** All All All 6, * / HF = Holsein Friesian, RS = Red Sindhi / Tes -inerval mehod 3/ Model : b c = a e Model : = b γ γ 3w 4 w Model 3: = b Model 4: 3 = b 3 Model 5: 3 4 = b 3 4 Model 6: = b Model 7: = b Where is milk ield a da in lacaion, = das in lacaion, γ = /35, w = ln(35/) * significan (P <.5) for Ho: LSMEAN=, ** highl significan (P <.) for Ho: LSMEAN=. 6 Four predicion models (models 3, 4, 5, and 6) had nonsignifican LS means of 35-d milk ield deviaions for all calving age subclasses. The TIM and he oher 3 models had a leas one significanl differen subclass (a leas P <.5). Model had one, model had wo, he TIM had hree, and model 7 had four significanl differen subclasses.

8 8 Calving season subclasses. Table 4 shows he LS means of individual cow 35-d milk ield deviaions (prediced minus acual milk ields) for TIM and he seven predicion models wihin breed group x lacaion x calving season subclasses. The paern of calving season subclasses wih saisicall significan LS means of individual cow 35-d deviaions was quie similar o he one found for calving age subclasses. Models 3 hrough 6 showed no significan differences, models and had onl one significan difference (P <.5), he TIM had wo subclasses wih significan differences (P <.5), and model 7 had five significanl differen subclasses (a leas P <.5). Table 4. Leas squares means of 35-d milk ield deviaions (prediced minus acual milk ields) for he es-inerval mehod and seven predicion procedures b breed group x lacaion x calving season subclasses Breed group / Lacaion number Calving season / No. of lacaions Acual ield (kg) TIM 3/ Model Model Model 3 Model 4 Model 5 Model 6 Model 7 HF Winer HF Summer * HF Rain HF Winer HF Summer HF Rain * -63.6* HF 3 Summer HF 3 Rain HF > 4 Winer * HF > 4 Summer HF > 4 Rain /HF /RS Winer * /HF /RS Rain /HF /RS Winer * ** /HF /RS Summer /HF /RS Rain /HF /RS 3 Winer ** /HF /RS 3 Summer /4HF /4RS Winer * 3/4HF /4RS Summer All All All * / HF = Holsein Friesian, RS = Red Sindhi / Winer = November Februar, Summer = March June, Rain = Jul Ocober 3/ Tes -inerval mehod 4/ Model : b c = a e Model : = b γ γ 3w 4 w Model 3: = b Model 4: 3 = b 3 Model 5: 3 4 = b 3 4 Model 6: = b Model 7: = b Where is milk ield a da in lacaion, = das in lacaion, γ = /35, w = ln(35/) * significan (P <.5) for Ho: LSMEAN=, ** highl significan (P <.) for Ho: LSMEAN=. 6 The overall ranking of 35-d milk ield predicive abiliies of hese procedures, based on he number of nonsignifican differences and heir overall LS means of he 35-d milk ield deviaions in calving age and calving season subclasses, was (firs o las): ) model 3 (second degree polnomial), ) model 5 (fourh degree polnomial), 3) model 4 (hird degree polnomial), 4) model 6 (fifh degree polnomial), 5) model (mixed log second degree

9 polnomial, 6) model (gamma funcion), 7) TIM (he es-inerval mehod), and 8) model 7 (sixh degree polnomial). As wih -d milk ields, he abili of TIM o predic 35-d milk ields ranked close o he boom among he eigh predicion procedures. Also, he TIM procedure was he onl one o produce overall significan differences (prediced minus acual) for -d and 35-d (P <.5). These resuls sugges ha using TIM o predic eiher -d or 35-d milk ields from monhl es-da records would probabl produce biased prediced milk ields, which in urn, ma bias geneic predicions for hese rais. The ranking of he seven predicion models for 35-d milk ields here (3, 5, 4, 6,,, 7) was ver similar o ha obained using all dail milk ields (3,, 5, 4, 7, 6, ; Koonawooririron e al., ). Again, onl wo models changed ranking. Models was fifh and model 7 was he boom model here, whereas model shared he op spo wih model 3, and model 7 was fifh for he all dail records case. Thus, models (mixed log second degree polnomial) and model 7 (sixh degree polnomial) appeared o be more sensiive o he number of dail records used o predic dail lacaion ields. Considering boh sampling sraegies (all dail records and monhl es-da records), model 3 (second degree polnomial) could be considered he mos appropriae model o predic individual cow 35-d milk ields for his daa se. The simplici and compuaional ease of model 3 (quadraic equaion), makes i ideal for large-scale compuaions such as hose in naional geneic evaluaions. Thus, model 3 could be used insead of TIM for he compuaion of 35-d milk ields in naional geneic evaluaions in Thailand. Before a final decision is reached in his regard, however, his sud needs o be repeaed wih a larger mulibreed populaion ha is more represenaive of he acual naional Thai cale populaion. Random regression models (Schaeffer and Dekkers, 994; Jamrozik e al., 997; Jamrozik and Schaffer, 997) are currenl a popular alernaive o he classical oal producion for geneic evaluaion of dair cale. Should random regression models be applied o he mulibreed Thai dair cale populaion in he fuure, model 3 could be he equaion used in boh he fixed and in he random porion of a random regression model using monhl es-da ields. However, because here was no uniforml bes model for all calving age and calving season subclasses in his HF-RS mulibreed herd, a more accurae geneic evaluaion ssem ma need o consider several lacaion predicion equaions for he various breed group x lacaion number x calving age and on breed group x lacaion number x calving season subclasses. Using a large number of lacaion predicion models, however, ma undul increase he complexi of a geneic evaluaion ssem, paricularl in a mulibreed populaion. Thus, as an approximaion, onl a small number of lacaion predicion equaions could be defined (e.g., 3 or 4), which ma be appropriae o predic individual cow lacaion dail ields (and accumulaed milk ields) wih sufficien accurac wihin calving age or calving season subclasses. The curren Thai cale populaion has more han differen breeds Bos indicus and Bos aurus represened boh in purebred and in crossbred form. The populaion is largel unsrucured and has a large number of crossbreds composed of up o seven breeds. The onl aemp o develop a large-scale sire evaluaion began in 996 hrough a collaboraion beween Kasesar Universi and he Dair Promoion Organizaion of Thailand (DPO, 996). The procedure used was a unibreed bes linear unbiased predicion, and he model was a single -rai (-d and 35-d milk and fa ield) animal model (Henderson, 973; Quaas and Pollak, 98). Considering he amoun of available dair daa and he complexi of he Thai mulibreed populaion, a poenial research work could consider accumulaed ields and a mulibreed model ha uses model 4 for -d and model 3 for 35-d accumulaed milk and fa ields. This nex piece of research will help revalidae he dail milk ield models used here, and i will deermine heir usefulness in a larger, more complex, mulibreed populaion. 9

10 Conclusions The abili of he es-inerval mehod and seven predicion models o predic -d and 35-d milk ields using monhl es-da records of cows in a mulibreed Holsein Friesian-Red Sindhi herd of SARTC varied b breed group x lacaion number x calving age and breed group x lacaion number x calving season subclasses. None of he eigh predicion procedures was uniforml beer across all subclasses for eiher d or 35 d milk ields. In addiion, he eigh procedures ranked differenl for -d and 35-d accumulaed milk ields. The mos appropriae model for -d milk ield predicions was model 4 (hird degree polnomial), whereas for 35-d, model 3 (second degree polnomial) was he bes performer. I should be sressed however, ha hese resuls appl sricl o his mulibreed populaion and his daa se. To obain broader conclusions applicable naionall, a subsaniall larger mulibreed daa se ha capures he diversi of breeds and crossbred groups of he Thai cale populaion will be needed. However, his sud provides a good indicaion of he pes of lacaion predicion models ha migh be applicable under Thai condiions. Because he performance of he eigh models here differed across calving age and calving season subclasses, several lacaion predicion models migh be needed for a naional Thai geneic evaluaion. This aspec also needs o be revisied wih a large naional Thai mulibreed daa se, wih he purpose of finding a model, or a se of models, ha is compuaionall simple, and gives reasonabl accurae geneic predicions. Acknowle dgemens The auhors are hankful for he financial suppor from he Thailand Research Fund under he Roal Golden Jubilee Projec. The auhors are graeful o he saff of he Sakon Nakhon Agriculural Research and Training Cener for making heir daa se available for his research. The auhors hank T. A. Olson, J. Rosales, and C. E. Whie for reviewing he manuscrip. Lieraure Cied Ali, T. E., and Schaeffer, L. R Accouning for covariances among es da milk ields in dair cows. Can. J. Dair Sci. 67: DPO Dair Promoion Organizaion Sire and Dam Summar. Annual Repor 996. p 53. Henderson, C. R Sire evaluaion and geneic rends. pp -4. In Proc. Anim. Breed. Gene. Smp. in Honor of Dr. Ja L. Lush. ASAS and ADSA, Champaign, IL. Jamrozik, J. and Schaeffer, L. R Esimaes of geneic parameers for a es da model wih random regressions for ield rais of firs lacaion Holseins. J. Dair Sci. 8: Jamrozik, J., Schaeffer, L. R., and Dekkers, J. C. M Geneic evaluaion of dair cale using es da ields and random regression model. J. Dair Sci. 8:7-6. Koonawooririron S., Elzo, M. A., Tumwasorn, S., and Sinala, W.. Modeling lacaion curves and predicing cumulaive milk ields in a mulibreed dair herd in Thailand using all lacaion records. Thai. J. Agri. Sci. (Submied).

11 Norman, H. D., VanRaden, P. M., Wrigh, J. R., and Cla, J. S Comparison of es inerval and bes predicion mehods for esimaion of lacaion ield from monhl, ampm., and rimonhl esing. J. Dair Sci. 8: Quaas, R. L. and Pollak, E. J. 98. Mixed model mehodolog for farm and ranch beef cale esing programs. J. Anim. Sci. 5: Sargen, F. D., Lon, V. H., and Wall Jr., O. G Tes inerval mehod of calculaing Dair Herd Improvemen Associaion records. J. Dair Sci. 5: Schaeffer, L. R. and Dekkers, J. C. M Random regressions in animal models for es-da producion in dair cale. pp In Proc. Fifh World Congr. Gene. Appl. Lives. Prod., vol. 8. SAS. 99. SAS/STAT User s Guide. 4h ed. SAS Insiue Inc., Car, NC. Wood, P. D. P Algebraic model of he lacaion curve in cale. Naure Lond. 6:64-65.