Plant disease models: from field observations to biological mechanisms

A plant disease model is a simplification of a real pathosystem (i.e., the relationships between a pathogen, a host plant, and the environment) that determine whether and how an epidemic develops over time and / or space. Different approaches have been used for the development of plant disease models, with relevant improvements in recent years. Empirical models have been elaborated using data collected under variable field conditions since the second half of the last century. The so called 3 10 rules for predicting first seasonal infection of grape downy mildew is a precursor of this empirical approach for understanding relationships between pathogens, plants and the environment. By using this approach, the model is developed by searching mathematical or statistical relationships between field collected data and these relationships do not necessarily have cause-effect meaning. Lack of knowledge, accuracy and, especially, robustness are the main weaknesses of these models, which impose accurate validation and, usually, proper calibration when these models are used in different environments or under changing climate. Recent methods of data analysis, like for instance neural networks, improve the capability of searching the mathematical structure of the model but they do not overcome the above mentioned weaknesses. Mechanistic models are a new class of models based on knowledge of biological and epidemiological behaviour of the system under study. These models (also referred to as explanatory, theoretical, or fundamental) explain the pathosystem on the basis of what is known about how the system works in relation to the influencing variables. Mechanistic models are dynamic, because they analyse the changes over time of the components of an epidemic due to the external, influencing variables. Dynamic modelling is based on the assumption that the state of the pathosystem in every moment can be quantitatively characterised and that changes in the system can be described by mathematical equations. These models overcome the weakness of the empirical models. Compared to the 3 10 rule, a mechanistic model for grape downy mildew increased the overall accuracy of the predictions from ~60% to ~90%. Complexity of mechanistic models has been regarded as a problem for the implementation of models in practical disease control, compared to the simplicity of the empirical models. This is a false problem, because confusing complexity of the mathematical framework of the model with complexity of model output is misleading. Indeed, it is possible to use complex models, able to depict the complexity of the biological systems, to produce simply, easy-to-use output for growers. Implementation of the above mentioned mechanistic model for grape downy mildew in a Decision Support System used by viticulturists clearly demonstrates the inconsistency of the “complexity paradigm”.