VitiProtect–Development and testing of a downy mildew AI forecasting model for Swiss viticulture

Abstract

Downy mildew (Plasmopara viticola) is a fungal pathogen that causes a destructive disease in grapevines (Vitis vinifera). The dissemination of the disease exhibits considerable regional variability and is influenced by the prevailing microclimate. To control downy mildew, the use of substantial quantities of plant protection products (PPPs) is needed. The development of reliable forecasting models that can provide early warning of impending infection events is of paramount importance for the targeted application of PPPs. However, the prediction of downy mildew is a highly complex undertaking. Current models, which have been tested and proven effective, are mechanistic in nature and have been applied so far for regional contexts. As is becoming increasingly apparent, these models are not able to quickly take into account the changes caused by climate change and microclimatic conditions. New methods from the field of machine learning can increasingly be used and offer more flexible solutions for the future. Such tools could process environmental changes in real time, incorporate them into the risk forecast, and thus enable more individualized, farm-based risk predictions at the vineyard level.

The VitiProtect project is investigating the potential of machine learning for the prediction of diseases in viticulture, with downy mildew serving as a case study. Over the past two years, data on the phenology of the vines and downy mildew infestation were collected on a weekly basis between May and August on 107 unsprayed plots. In addition, meteorological parameters such as rainfall, humidity, air temperature, leaf wetness, and relative soil moisture were measured hourly at all plots. Following data cleansing, a prototype risk prognosis model for downy mildew is developed with the help of artificial intelligence. Currently, this prototype is tested in the field and will be further improved based on new infection data. Field experiments are carried out to examine whether the new forecasting model can be used to reduce the use of PPPs and thus contribute to achieving the sustainability goals.