A two years study results in the use of artificial neural networks to forecast Plasmopara viticola infection in viticulture.

This paper describes the further results of the study that has been described in session 5 of the 58th International Symposium on Crop Protection (Ghent 2006). Since then our attention has been focused on verifying the previous communication results working on a two years basis data set belonging to a specific farm. The choice of using data from a single farm derives from the considerations that have been explained in the previous study in which it was clear that an efficient forecasting Artificial Neural Network (ANN) model can be created only in restricted (or at least comparable) pedoclimatic areas. On the basis of the matured experience, at the moment we have realized an ANN which, being trained on 2005 year data, elaborating the following year data is capable of correctly predicting the real Plasmopara viticola (Berk. et Curt.) Berl. et De Toni outbreak, never giving false negative signals (no alarm in presence of infection on the field) and, finally, giving few other alarms which are totally comparable with the ones given by the most common statistical instrument used in this field trials. We confirm the advantages of this approach in terms of: (a) Management and optimization improvement of agricultural activities. (b) Reduction of plant protection products use. (c) Quality improvement of the final product for a real lowering of plant protection products use. (d) Reduction of environmental impact. (e) A more efficient management of the climate changes.

Artificial intelligence approach with the use of artificial neural networks for the creation of a forecasting model of Plasmopara viticola infection.

Most of the forecasting models of Plasmopara viticola infections are based upon empiric correlations between meteorological/environmental data and pathogen outbreak. These models generally overestimate the risk of infections and induce to treat the vineyard even if it should be not necessary. In rare cases they underrate the risk of infection leaving the pathogen to breakout. Starting from these considerations we have decided to approach the problem from another point of view utilizing Artificial Intelligence techniques for data elaboration and analysis. Meanwhile the same data have been studied with a more classic approach with statistical tools to verify the impact of a large data collection on the standard data analysis methods. A network of RTUs (Remote Terminal Units) distributed all over the Italian national territory transmits 12 environmental parameters every 15 minutes via radio or via GPRS to a centralized Data Base. Other pedologic data is collected directly from the field and sent via Internet to the centralized data base utilizing Personal Digital Assistants (PDAs) running a specific software. Data is stored after having been preprocessed, to guarantee the quality of the information. The subsequent analysis has been realized mostly with Artificial Neural Networks (ANNs). Collecting and analizing data in this way will probably bring us to the possibility of preventing Plasmospara viticola infection starting from the environmental conditions in this very complex context. The aim of this work is to forecast the infection avoiding the ineffective use of the plant protection products in agriculture. Applying different analysis models we will try to find the best ANN capable of forecasting with an high level of affordability.