Reanalyses of the historical series of UK variety trials to quantify the contributions of genetic and environmental factors to trends and variability in yield over time

Historical datasets have much to offer. We analyse data from winter wheat, spring and winter barley, oil seed rape, sugar beet and forage maize from the UK National List and Recommended List trials over the period 1948-2007. We find that since 1982, for the cereal crops and oil seed rape, at least 88% of the improvement in yield is attributable to genetic improvement, with little evidence that changes in agronomy have improved yields. In contrast, in the same time period, plant breeding and changes in agronomy have contributed almost equally to increased yields of forage maize and sugar beet. For the cereals prior to 1982, contributions from plant breeding were 42, 60 and 86% for winter barley, winter wheat and spring barley, respectively. These results demonstrate the overwhelming importance of plant breeding in increasing crop productivity in the UK. Winter wheat data are analysed in more detail to exemplify the use of historical data series to study and detect disease resistance breakdown, sensitivity of varieties to climatic factors, and also to test methods of genomic selection. We show that breakdown of disease resistance can cause biased estimates of variety and year effects, but that comparison of results between fungicide treated and untreated trials over years may be a means to screen for durable resistance. We find the greatest sensitivities of the winter wheat germplasm to seasonal differences in rainfall and temperature are to summer rainfall and winter temperature. Finally, for genomic selection, correlations between observed and predicted yield ranged from 0.17 to 0.83. The high correlation resulted from markers predicting kinship amongst lines rather than tagging multiple QTL. We believe the full value of these data will come from exploiting links with other experiments and experimental populations. However, not to exploit such valuable historical datasets is wasteful.