Scope and topics
Earth observation (EO) is key in providing relevant information to observe the surface of the earth at different spatial and temporal details from complementary sensing modalities (optical, microwave, and thermal). Coupled with recent advances in computer vision algorithms mainly artificial intelligence (AI) for computer vision applications in image processing have paved the realization of novel information extraction pipelines and fusion of multi-sensor imagery. By leveraging the availability of computational resources, it is also common to see regional, continental and global scale information layers extracted from earth observation imagery. Even though there are promising developments in AI models for image processing, there are technical and operational limitations demanding further research including but not limited to intensive demand for properly annotated training and validation datasets, generalization across space and time (given there is obvious spatial heterogeneity on objects and phenomena across geography and different time and season), computational resource and ease of use. Therefore, in this session, we welcome submissions that try to address EO-AI research themes including deep learning as well as symbolic expert-based or hybrid AI in analyzing EO data and derivatives related to:
- Application of EO-AI models (machine learning, deep learning, knowledge based) in various applicationcases including flood mapping, damage assessment, building and or settlement change detection and monitoring, building, settlement and population mapping, burned area and fire risk mapping, mapping food (in)security, spatially explicit migration mapping and modeling
- EO-AI algorithmic developments in label-efficient training strategies like self-supervised learning, unsupervised learning including generative models, few-shot learning, learning from noisy and or incomplete labels
- Adapting language and foundation models for EO-based operational tasks in disaster management and humanitarian response like visual question answering, object detection and localization tasks
- Spatio-temporal transfer learning strategies like domain adaptation, meta-learning and any other novel transfer learning strategies enhancing transferability of EO-AI models across geography and time
- Multi-modal hazard prediction and mapping like drought, earthquake, disease outbreak and risk hotspot mapping, flood inundation and return periods that are relevant for the early warning sys-tem
- Datasets tailored for training and validation of (geo)AI models and systematic evaluation of open
source EO-based information layers focused on bias, reliability, quality, scope(covering geographic,
thematic and temporal aspects) - Integrated hazard and vulnerability mapping using model based and multi-indicator approaches
- Industry use cases beast practices showcasing application of Geo-AI in humanitarian emergency re-
sponse(rapid mapping, integrated vulnerability and risk mapping, field operations and challenges)
Organisers
Prof. Dr. Stefan Lang: Paris Lodron University of Salzburg, Department of Geoinformatics, stefan.lang@plus.ac.at
Dr. Getachew Workineh Gella: Paris Lodron University of Salzburg, Department of Geoinformatics, getachewworkineh.gella@plus.ac.at