No abstract

Due to their peculiar characteristics, seed germination and emergence assays may pose problems for data analysis, due to non-normal error distribution and serial correlation between the numbers of seeds counted on different dates from the same experimental unit (Petri dish, pot, plot). Furthermore, it is necessary to consider viable seeds that have not germinated ⁄ emerged at the end of an experiment (censored observations), as well as late germination ⁄ emergence flushes, that relate to genotypic differences within natural occurring seed populations. Traditional methods of data analysis may not be optimal for dealing with these problems. Therefore, survival analysis may represent an appropriate alternative. In this analysis, the time course of germina-tion ⁄ emergence is described by using a non-parametric step function (Ôgermination functionÕ) and the effect of factors and covariates on Ôgermination functionsÕ is assessed by Accelerated Failure Time regression and expressed in terms of Ôtime ratiosÕ. These parameters measure how a change in the explanatory variables changes (prolongs ⁄ shortens) the time to germination of a seed lot. This paper presents four examples of the application of survival analysis on seed germination ⁄ emergence studies. Results are discussed and compared with those obtained with more traditional techniques.

No abstract

For implementation of simple yield loss models into threshold-based weed management systems, © ]996 European Weed Research Society a thorough validation is needed over a great diversity of sites. Yield losses by eompetition wsth Sinapis alba L. {white mustard) as a model weed, were studied in 12 experiments in sugar beet (Beta vutgarLs L.) and in 11 experiments in spring wheat (Triiicum aestivunt L.). Mosl data sets were bet ler described by a model based on the relative leaf area of the weed than by a hyperbolic model based on weed deasity. This leaf area mode] accounted lor (part of) the effect of different emerging times of the S. alba. whereas the density modd did not. A parameter that allows the maximum yield loss to be smaller than K)()% was mostly not needed to describe the effects of weed competition. The parameter that denotes the competitiveness of the weed species wiih respect to the crop decreased the later the relative leaf area of the rnustard was determined. This decrease could be estimated from the differences in relative growth rate of the leaf area of crop and..?. atba. However, the accuracy of this estimation was poor. The parameter value of the leaf area model varied considerably between sites and years. The results strongly suggest that the predictive ability ofthe !eaf area mode! needs to be improved before ii can he applied in weed management systems. Such improvement would require additional information about effects ot abiotic factors on piant development and morphology and ihe definition of a lime window for predictions with an acceptable level of error.

No abstract