Modelling spatial-temporal covariance structures in monocropping barley trials

In long-term trials, not only are individual plot errors correlated over time but there is also a consistent underlying spatial variability in field conditions. The current study sought the most appropriate covariance structure of errors correlated in three dimensions for evaluating the productivity and time-trends in the barley yield data from the monocropping system established in northern Syria. The best spatial-temporal model found reflected the contribution of autocorrelations in spatial and temporal dimensions with estimates varying with the yield variable and location. Compared with a control structure based on independent errors, this covariance structure improved the significance of the fertilizer effect and the interaction with year. Time-trends were estimated in two ways: by accounting the seasonal variable contribution in annual variability (Method 1), which is suitable for detecting significant trends in short data series; and by using the linear component of the orthogonal polynomial on time (year), which is appropriate for long series (Method 2). Method 1 strengthened time-trend detection compared with the method of Jones and Singh [J. Agri. Sci., Cambridge 135 (2000), pp. 251-259] which assumed independence of temporal errors. Most estimates of yield trends over time from fertilizer application were numerically greater than the corresponding linear trends estimated from orthogonal polynomials in time (Method 2), reflecting the effect of accounting for seasonal variables. Grain yield declined over time at the drier site in the absence of nitrogen or phosphorus application, but positive trends were observed fairly generally for straw yield and for grain yield under higher levels of fertilizer inputs. It is suggested that analyses of long-term trials on other crops and cropping systems in other agro-ecological zones could be improved by taking spatial and temporal variability into account in the data evaluation.