Plant growth analysis: An evaluation of experimental design and computational methods

Abstract

Various aspects of the experimental design and computational methods used in plant growth analysis were investigated. This was done either analytically, or by repeatedly simulating harvests from theoretical populations upon which were imposed the underlying growth curves as well as the variability in plant material. In the first part the consequences of neglecting an In-transformation of the primary weight data were considered. T-tests are affected in such a way that significant differences between treatments show up less readily than in transformed data. A more fundamental point is that most hypotheses on plant weight concern proportional effects rather than absolute ones. In these cases, an In-transformation prior to a statistical test is required anyway. Secondly, the accuracy of average RGR estimates was evaluated. Variability in RGR estimation increases linearly with variability in the plant material. It is also strongly dependent on the time interval between harvests and the number of replicates per harvest. Even with conservative values for plant weight variability, the chances of arriving at aberrant RGR estimates are rather high. Therefore, it is suggested that the variability in the population is decreased deliberately, unless variability within the population is itself of biological interest. Thirdly, three computational methods to fit dry weight progressions and describe time trends in RGR and related growth parameters were evaluated. Although complicated to calculate, the Richards function was superior to polynomials fitted through either the weight data ('polynomial' approach), or the classically derived RGR values ('combined' approach).