Weed Emergence Models

Abstract

Weed emergence models are practical tools that aim to describe the dynamics of emergence in the field. Such models can be conceptualized from two main perspectives: a reductionist/mechanistic approach and an empirical modelling viewpoint. While the former provides a close description of the basic ecophysiological processes underlying weed emergence (i.e. seed dormancy, germination and pre-emergence growth), they usually require a large amount of difficult to estimate species-specific parameters, as well as sometimes unavailable or missing experimental data for model development/calibration/validation. Conversely, the latter aims to describe the emergence process as a whole by seeking a general mathematical description of field emergence data as a function of field environmental variables, mainly temperature and precipitation. As reviewed in the literature, most emergence models have been developed using nonlinear regression (NLR) techniques. NLR sigmoidal type models which are based on cumulative thermal or hydrothermal time have become the most popular approach as they are easy to develop and use. However, some statistical and bioecological limitations arise, for example, the lack of independence between samplings, censored data, need for threshold thermal/hydric parameter estimation and determination of ‘moment zero’ for thermal/hydrothermal-time accumulation, among other factors, which can lead to inaccurate descriptions of the emergence process. New approaches based on soft computing techniques (SCT) have recently been proposed as alternative models to tackle some of the previously mentioned limitations. In this chapter, we focus on empirical weed emergence models with special emphasis in NLR models, highlighting some of the main advantages, as well as the statistical and biological limitations that could affect their predictive accuracy. We briefly discuss new SCT-based approaches, such as artificial neural networks which have recently been used for weed emergence modelling.