Abstract
This research aims at developing a cartographic understanding and methodology for producing precision maps to support local, tech-minded farmers, scientists, and authorities in making spatial decisions. Despite being a widely used cartographic product, the precision map contemporarily exists outside of a theoretical cartographic framework. Geographical Information Science and Remote Sensing on the other hand have already found entry in agricultural practices (Usery, Pocknee, & Boydell, 1995; Papp et al., 2021), laying the foundation for precision agriculture (Sishodia et al., 2020). Making use of sensing systems to obtain (near real-time) environmental data, precision agriculture relies on the extraction of relevant, high-resolution geoinformation to be communicated visually to the end-user (Sishodia et al., 2020). Cartography as an interdisciplinary domain bears a huge potential in meeting the increasing demand for the decision-supporting visualisation of Big Data (Coetzee et al., 2020; Robinson et al., 2017), and thus may contribute to the development of more sustainable agriculture. The focus of this research is on plant disease detection, as plant pests endanger tree health and harvest, thus highlighting farmers' vulnerability to plant diseases and ultimately undermining Sustainable Development Goals, such as Zero Hunger and food security (Savary et al., 2012). In contrast to traditional plant disease management, plant health monitoring through hyperspectral data would enable stakeholders to detect plant stress outside the range of visible light, and hence, at an early disease stage. Particularly, near-and shortwave infrared bands provide crucial information on plant health, allowing early intervention in plant disease control (Lowe, Harrison, & French, 2017). Apple Proliferation (AP), a phytoplasma disease transmitted through insects, will be used as a case study in the Upper Adige Valley in South Tyrol (Italy) to propose a technology-and data-driven strategy for an early, non-destructive plant disease detection and monitoring based on hyperspectral data. The work in progress consists of the following research objectives:-developing a cartographic understanding and definition for a precision map,-identifying relevant spectral bands (SB's) and vegetation indices (VI's) that allow meaningful discrimination between leaves infected with AP and uninfected leaves (binary discrimination),-implementing a robust and reproducible image classification procedure based on a machine learning approach specifically for the identification of AP on varying levels of detail (at the leaf, tree, and orchard level),-establishing a mapping technique to produce precision maps dedicated to AP on different levels of detail Following a multi-modal and multi-scale data acquisition approach, several remote sensing techniques were applied: adopting a ground sensing approach, individual leaf spectral signatures were obtained through a spectroradiometer equipped with a leaf clipping tool extension; airborne proximal sensing was conducted on the orchard level using UAV-mounted multi-and hyperspectral cameras. Insights gained through the spectral data analysis served as the basis for the subsequent multi-and hyperspectral waveband selection and composition of adequate vegetation indices. Additionally, machine learning techniques were adopted for image classification to delineate infected from uninfected trees. A hexagonal grid structure assimilating the tree rows is used for cartographic generalisation of the image classifications as well as for abstraction of the tree canopy shape by the individual hexagonal cell. Finally, the high-resolution classification on the leaf level and the aggregated spatial information regarding the AP infection on the tree and orchard level (i.e. as gridded abstraction) are presented as interrelated map-layers.