Background

Austrian winegrowers cultivating the Blauer Zweigelt grape face an unpredictable and serious threat from the ripening disorder berry shrivel (BS; "Traubenwelke" in German). In some regions, this disorder caused yield losses of up to 40% by 2023. Despite four decades of research and recent advances in analytical methodologies, the causes of BS remain unclear. Moreover, no standardized method exists for assessing BS in vineyards, nor is there a dedicated database to track its occurrence.

Project Content and Aims

In the BAISIQ project, we aim to develop a standardized method for reliably detecting berry shrivel (BS) in vineyards. To achieve this, we combine modern image-processing techniques with explainable artificial intelligence (XAI). Additionally, we investigate whether hyperspectral data can help identify BS in its early stages.

The most important steps and aims of the project comprise:

• Testing whether image processing technologies are sensitive enough to differentiate between healthy and BS-affected grape clusters. We will also use multi-sensor phenotyping equipment in controlled environments to identify potential hidden signatures and use them to detect BS before it fully develops.
• Developing a Geometric Deep Learning (GDL) method for the data generated by the image processing and sensor technologies to identify key features that distinguish healthy from BS-affected grape clusters.
• Using remote sensing equipment to apply the developed technology in vineyards, enabling the quantification of BS in the field and the creation of a robust database.
• Validating the methods and technologies applied, ensuring they are ready for future use in vineyards.

Method

We employ multi-sensor phenotyping systems to detect hidden signatures of BS, using Geometric Deep Learning (GDL) to analyze sensor and image data. For monitoring BS in vineyards, remote sensing technology—such as UAVs equipped with LiDAR and multispectral sensors—is used. The project also leverages multi-modal sensor systems to integrate data from various sources. This helps to optimize AI and machine learning models. Furthermore, the models' predictive capabilities are enhanced by an image processing pipeline that includes advanced registration techniques and dimensionality reduction for hyperspectral data.

Data is collected in vineyards located in Carnuntum and the Thermenregion. All collected data will be made accessible for future research.

Impact

BAISIQ brings together a multi-disciplinary team to address a complex, yet unresolved, problem of high economic relevance in viticulture. The team aims to develop a methodology to assess expected yield loss due to berry shrivel (BS), a ripening disorder primarily affecting the Blauer Zweigelt grape. Currently, farmers are unable to take action until harvest. By detecting BS earlier in the ripening process, they could better adapt their harvesting schedules and cellar capacities. In other words, winegrower benefit as the expected harvest could be predicted with higher certainty. The project also lays the groundwork for establishing a BS database, which is essential for identifying environmental risk factors and for developing a potential BS insurance model.