Background

Austria generally has good water resources, but they are not evenly distributed. In eastern Lower Austria in particular, some areas face limited water availability and low levels of precipitation, a situation that is expected to worsen due to climate change. While irrigation can temporarily address water shortages, in the long term it puts pressure on local water supplies and increases energy costs and carbon emissions. Agriculture therefore needs new strategies and technologies that support efficient water use and allow water consumption to be accurately monitored. This is where the current project comes in. We aim to optimise plant water use and reduce overall water inputs in vineyards. As viticulture is a high-value sector in Lower Austria, it was chosen as a key area of research. The solutions developed can later be transferred to other agricultural contexts.

Aims

The overarching aim of the project is to develop a low-cost, multimodal 3D canopy capture system. This means that it uses several data channels simultaneously, combining colour (RGB) information with thermal and multi-/hyperspectral imaging. By leveraging computer vision and machine learning techniques, the system assesses key plant traits such as leaf area and transpiration. Plant-specific traits and canopy architecture form the basis for water consumption models that enable more accurate and sustainable water management in vineyards.

The project addresses the following research questions

• How accurate are 3D reconstructions of grapevines generated from outdoor data?
• How can multimodal information (e.g. thermal and hyperspectral images) be effectively integrated into a 3D plant model?
• How challenging is it to train robust models for segmenting plant organs (e.g. leaves and grape clusters), and how can label-efficient methods, domain adaptation, and foundation models support this process?
• Which extracted plant traits are essential for accurately parameterising a grapevine water budget model, and how does its performance compare with classic Leaf Area Index (LAI) methods?
• Can the model capture changes in plant transpiration in response to variations in plant architecture and soil water availability?

Results and Innovation

In this project, we lay the groundwork for a low-cost, handheld 3D plant scanning system. It uses colour images together with thermal and multispectral measurements to assess leaf area, temperature, and key indicators of plant health. These measurements are linked with precise water-use data to reveal how plant structure influences transpiration and irrigation needs. The project also investigates how vineyard practices affect water consumption and yield. The insights gained form the basis for improved water-use models that account for both climate conditions and plant-specific characteristics. While tailored to viticulture, the approach can be transferred to other crops such as orchards or vegetables. Overall, the project paves the way for more efficient irrigation and sustainable resource use, helping agriculture in Lower Austria and beyond prepare for future challenges.

Partner

Funding