Trends over time in annual crop yields potentially provide measures of the likely long-term sustainability of cropping systems. However, where large annual variability in the growth environment is responsible for most of the large year-to-year yield differences, appropriate analytical techniques must be developed to distinguish real long-term trends from the ‘background noise’. This paper presents models for the estimation of time trends in the yield data from crop rotation systems and discusses the results of applying these models to yield values from two types of long-term trial involving barley, each conducted at two sites in northern Syria.

The models used were linear with respect to time (years) and allowed for seasonal effects by means of a quadratic relationship on total rainfall and a linear relationship on planting date. A more complex model might account for more of the variance, but restrictions were imposed by the limited number of degrees of freedom (number of years of data less one) and the choice of meaningful single-valued parameters of growth-season conditions. For many experimental treatments the model accounted for less of the total variance at the wetter site. This may be due to seasonal bufferring by soil moisture stored at depth from one year to the next, and future iterations of the analysis will try to allow for this. The appropriateness of the linear time function is also questioned, and alternative functions will be tested along with alternative structures for plot errors over time.