best prediction of actual lactation yields
Post on 31-Dec-2015
31 Views
Preview:
DESCRIPTION
TRANSCRIPT
2004
2004
2007
John B. Cole
Animal Improvement Programs Laboratory
Agricultural Research Service, USDA, Beltsville, MD
john.cole@ars.usda.gov
Best prediction of actual lactation yields
ATA 2007: Best prediction of yield (3) Cole
Best PredictionVanRaden JDS 80:3015-3022 (1997), 6th WCGALP XXIII:347-350 (1998)
Selection IndexSelection Index Predict missing yields from measured Predict missing yields from measured
yields.yields. Condense test days into lactation yield Condense test days into lactation yield
and and persistency.persistency.
Only phenotypic covariances are needed.Only phenotypic covariances are needed. Mean and variance of herd assumed Mean and variance of herd assumed
known.known.
Reverse predictionReverse prediction Daily yield predicted from lactation yield Daily yield predicted from lactation yield
and persistency.and persistency.
Single or multiple trait predictionSingle or multiple trait prediction
ATA 2007: Best prediction of yield (4) Cole
History
Calculation of lactation records for milk (M), fat (F), protein (P), and somatic cell score (SCS) using best prediction (BP) began in November 1999.
Replaced the test interval method and projection factors at AIPL.
Used for cows calving in January 1997 and later.
ATA 2007: Best prediction of yield (5) Cole
Advantages
Small for most 305-d lactations but larger for lactations with infrequent testing or missing component samples.
More precise estimation of records for SCS because test days are adjusted for stage of lactation.
Yield records have slightly lower SD because BP regresses estimates toward the herd average.
ATA 2007: Best prediction of yield (6) Cole
Users
AIPL: Calculation of lactation yields and data collection ratings (DCR).
DCR indicates the accuracy of lactation records obtained from BP.
Breed Associations: Publish DCR on pedigrees.
DRPCs: Interested in replacing test interval estimates with BP.
Can also calculate persistency.May have management applications.
ATA 2007: Best prediction of yield (7) Cole
Restrictions of Original Software
Limited to 305-d lactations used since 1935.
Used simple linear interpolation for calculation of standard curves.
Could not obtain BP for individual days of lactation.
Difficult to change parameters.
ATA 2007: Best prediction of yield (8) Cole
Enhancements in New Software
Can accommodate lactations of any length (tested to 999 d).
Lactation-to-date and projected yields calculated.
BP of daily yields, test day yields, and standard curves now output.
The function which models correlations among test day yields was updated.
ATA 2007: Best prediction of yield (9) Cole
How does BP work?
Inputs: TD and herd averages in
Statistical wizardryStandard curve calculatedCow’s lactation curve based on TD deviations from standard curve
Outputs: Actual lactation yields, persistency, and daily BP of yield
ATA 2007: Best prediction of yield (10) Cole
Specific curves
Breeds: AY, BS, GU, HO, JE, MS
Traits: M, F, P, SCS
Parity: 1st and later
ATA 2007: Best prediction of yield (11) Cole
Modeling Long Lactations
Dematawewa et al. (2007) recommend simple models, such as Wood's (1967) curve, for long lactations.
Curves were developed for M, F, and P yield, but not SCS.
Little previous work on fitting lactation curves to SCS (Rodriguez-Zas et al., 2000).
BP also requires curves for the standard deviation (SD) of yields.
ATA 2007: Best prediction of yield (12) Cole
Data and Edits
Holstein TD data were extracted from the national dairy database.
The data of Dematawewa et al. (2007) were used.
1st through 5th paritiesLactations were at least 250 d for the 305 d group and 800 d for the 999 d group.
Records were made in a single herd.
At least five tests reported.Only twice-daily milking reported.
ATA 2007: Best prediction of yield (16) Cole
Modeling SCS and SD
Test day yields were assigned to 30-d intervals and means and SD were calculated for each interval.
Curves were fit to the resulting means (SCS) and SD (all traits).
SD of yield modeled with Woods curves.
SCS means and SD modeled using curve C4 from Morant and Gnanasakthy (1989).
ATA 2007: Best prediction of yield (22) Cole
Correlations among test day yields
Norman et al. JDS 82:2205-2211 (1999)
Rather than calculate each correlation separately we use a formula to approximate them.
Our model accounts for biological changes and daily measurement error.
The programs were updated to use a simpler formula with similar accuracy.
ATA 2007: Best prediction of yield (29) Cole
ST versus MT estimation
MT uses information on 3 or 4 traits simultaneously.
Provides more accurate estimates of components yield.
Advantage greatest with infrequent testing or missing component samples.
Canavesi et al. (2007)
ATA 2007: Best prediction of yield (33) Cole
Uses of Daily Estimates
Daily yields can be adjusted for known sources of variation.
Example: Daily loss from clinical mastitis (Rajala-Schultz et al., 1999).
This could lead to animal-specific rather than group-specific adjustments.
Research into optimal management strategies.
Management support in on-farm computer software.
ATA 2007: Best prediction of yield (38) Cole
Validation: new versus old programs (ST)
Trait Parity CorrelationMilk 1 0.999
2+ 0.999Fat 1 0.999
2+ 0.998Protein 1 0.999
2+ 0.998SCS 1 0.995
2+ 0.960
ATA 2007: Best prediction of yield (39) Cole
Validation: new versus old programs (MT)
Trait Parity CorrelationMilk 1 0.999
2+ 0.999Fat 1 0.999
2+ 0.998Protein 1 0.999
2+ 0.999SCS 1 0.995
2+ 0.960
ATA 2007: Best prediction of yield (40) Cole
Validation: first 3 versus all 10 TD
Trait Parity CorrelationMilk 1 0.931
2+ 0.934Fat 1 0.925
2+ 0.920Protein 1 0.937
2+ 0.938SCS 1 0.790
2+ 0.791
ATA 2007: Best prediction of yield (41) Cole
Validation: sums of 7-d averages and daily BP
Daily milk yields from the on-farm system (7-d averages) were summed and compared to BP.
Correlations:First parity: 0.927Later parities: 0.956
Quist et al. (2007) reported that actual yields are overestimated with the Canadian equivalent of BP.
ATA 2007: Best prediction of yield (42) Cole
Implementation
When testing is complete:Yields in the AIPL database will be updated.
Data will be submitted to Interbull for a test run.
The BP programs have been sent to 4 DRPCs for testing.
ATA 2007: Best prediction of yield (43) Cole
Conclusions
Correlations among successive test days may require periodic re-estimation as lactation curves change.
Many cows can produce profitably for >305 days in milk, and the revised BP program provides a flexible tool to model those records.
Daily BP of yields may be useful for on-farm management.
top related