Variance Component Re-Estimation and Model Simplification in a Single-Step Random Regression Test-Day Model for Nordic Red Dairy Cattle

Abstract

The Nordic Red dairy cattle random regression test-day model (RRM) has been updated to reflect increased genetic similarity among the original country subpopulations (Finland, Sweden and Denmark) and changes in data structure over the past two decades. Data prior to 2005 were excluded, and previously country-specific models were harmonized into a simplified nine-trait multi-trait single-step RRM using a metafounder approach. The objective of this study was to describe the implementation of the updated model. This included re-estimation of variance components for parameter-reduced covariance functions for random effects identified as optimal in a previous study, accounting for heterogeneous measurement error (co)variances across eight milk recording schemes, and evaluation of genomic estimated breeding value reliability using forward cross-validation. Variance components were estimated from a harmonized subset of data sampled across Finland, Sweden and Denmark, applying the updated model. Compared with previously used variance components derived from Finnish-only data, heritability estimates for 305-day yields increased by 8 to 46% across all traits. Substantial differences in measurement error variances and residual correlations among daily milk, protein, and fat yields were observed across the eight milk recording schemes, reflecting heterogeneous milking systems and historical protocol changes in automated milking systems in Finland. For Finland, four residual covariance matrices were required to model measurements from conventional milking systems and three automated milking system protocols. In contrast, two residual classes (automated and conventional milking systems) were sufficient for Denmark and Sweden. The selected covariance functions enabled a substantial reduction in variance component parameters while maintaining model stability and predictive ability. The model was solved using a fully component-wise ssGTBLUP approach with metafounders. The updated model provides a computationally efficient basis for routine genetic evaluation of Nordic Red dairy cattle.