Banner
Home      Log In      Contacts      FAQs      INSTICC Portal
 
Documents

Keynote Lectures

Climate Change: Impacts, Trends and Extremes
Ricardo M. Trigo, Faculty of Sciences, University of Lisbon, Portugal

Novel Approaches to Model Assessment and Interpretation in Geospatial Machine Learning
Alexander Brenning, Geography, Friedrich Schiller University, Germany

 

Climate Change: Impacts, Trends and Extremes

Ricardo M. Trigo
Faculty of Sciences, University of Lisbon
Portugal
 

Brief Bio
Ricardo Trigo is an Associated Professor at the Geophysics, Geographical Engineering and Energy Department and Director (between 2018 and 2021) of the Associated Laboratory Instituto Dom Luis (IDL), both at the Faculty of Sciences, Univ of Lisbon. He has published extensively (>280 papers, h=71 Scopus, h=83 Google Scholar) in complementary areas of research with particular emphasis Natural Hazards such as Floods, Droughts, Landslides, Heatwaves and Wildfires. In particular how these phenomena are related to climate change or to large-scale patterns such as NAO, Blocking or Atmospheric Rivers. In recent years he has been working increasingly on longer time scales, studying the impact of solar and volcanic variability with both proxy and early meteorological data. Ricardo Trigo has participated in more than 20 national projects, mostly funded by FCT but also by large companies related to energy (REN, EDP, E-Redes) and paper production (Portucel and the Navigator company). He has participated in 12 EU funded projects, including the ongoing ROADMAP and the recent IMDROFLOOD and INDECIS. In 2008 he was awarded with the International Journal of Climatology Prize endorsed by the Royal Met. Society and in 2017 he received the prize UL/CGD for the scientific production in Earth Sciences. In 2011 he edited the book "Hydrological, Socioeconomic and Ecological impacts of the North Atlantic Oscillation in the Mediterranean Region" published by Springer. He is currently supervising two PhD students and has supervised successfully 10 PhD students.


Abstract
Available soon.



 

 

Novel Approaches to Model Assessment and Interpretation in Geospatial Machine Learning

Alexander Brenning
Geography, Friedrich Schiller University
Germany
 

Brief Bio
Alexander Brenning is an applied mathematician (Technical University of Freiberg, Germany) and geographer (Ph.D., Humboldt-Universität zu Berlin) with research interests in geospatial machine learning and geostatistics for modeling of Earth surface processes, including landslide hazards, mountain permafrost, and environmental pollution. He joined Friedrich Schiller University Jena, Germany in 2015 as a Full Professor of Geographic Information Science after previously holding a faculty position at the University of Waterloo, Canada, since 2007. He has visited the University of Heidelberg as a Humboldt Research Fellow, and the Pontifical Catholic University of Chile as a Distinguished Visiting Professor, and is a member of the ELLIS Unit Jena, a research cluster in the field of machine-learning for Earth system science.


Abstract
The increasing interest in the interpretability and explainability of artificial intelligence (AI) decisions requires innovative model diagnostic tools that account for the unique challenges of geospatial and environmental data, notably spatial dependence and high dimensionality. Leveraging the geostatistical paradigm rooted in distance-based metrics, spatial prediction error profiles (SPEPs) and spatial variable importance profiles (SVIPs) as well as derived summary statistics based on spatial cross-validation are novel, model-agnostic tools for assessing and interpreting models across various prediction horizons. Additionally, to tackle the complexities of deciphering joint effects in environments abundant with strongly correlated or high-dimensional features, interpretation tools that distill aggregated relationships from complex models are required. The efficacy of these techniques is demonstrated in two case studies: the regionalization of environmental pollution based on point measurements of concentrations, and a classification task from multitemporal remote sensing of land use. In these case studies, SPEPs and SVIPs effectively highlight differences and unexpected similarities of geostatistical methods, linear models, random forest, and blended algorithms. With 64 correlated features in the remote-sensing case study, the transformation-based interpretation approach successfully summarizes high-dimensional relationships in a small number of diagrams for effective science communication. These innovative diagnostic tools enrich the toolkit of geospatial data science, offering potential enhancements to the interpretation, selection, and design of geospatial machine-learning models.



footer