Use of various functions to analyse the fertiliser responses of maize ( Zea mays L.) hybrids in long-term experiments

Abstract

The effect of various fertiliser treatments on the yield of maize hybrids was studied on the basis of 26 years of data obtained in a long-term bifactorial split-plot experiment set up in 1967. The seven treatments (NPK ratio 2:1:1) applied were as follows (rates per hectare): 1. Control (no fertiliser), 2. 100 kg NPK, 3. 200 kg NPK, 4. 300 kg NPK, 5. 400 kg NPK, 6. 600 kg NPK, 7. 800 kg NPK. The maize was grown with the conventional cultivation techniques in continuous cropping. The results of analyses carried out with three different methods (analysis of variance, cumulative yield analysis and regression analysis) all indicated that under the given conditions the yield of maize hybrids was highest at an NPK fertiliser rate of 200-400 kg ha -1 . The effect of fertilisation on the maize yield was significant in 21 of the 26 years. Combined analysis of variance for the years showed that the year effect (quantity of rainfall) had the greatest effect on the maize yield, but although the year effect had a fundamental effect on the yield level it did not influence the fertiliser response pattern. The fertiliser responses of the maize hybrids were described by fitting four types of functions (quadratic, square root, inverse exponential, linear-plateau) to the yield data. It was found that when selecting the best function a consideration of the regression deviations (measured yield - calculated yield) was just as important as the coefficient of determination (R 2 ). In 12 of the 26 years the fitting of the quadratic function was not significant and overestimated the fertilisation optimum. The fertiliser response curve generally has a broad maximum which is far better described by the square root function than by the quadratic. If the fertiliser response pattern includes a depressive phase, a square root function should definitely be used in place of the quadratic function. If the maximum of the response surface forms a plateau (as opposed to a maximum point) a linear-plateau function or an inverse exponential function can be recommended. In the present work the linear-plateau function gave the best results.