geoAI

Module on geospatial data prediction and remote sensing of the Honour Degree Programm „AI und Entrepreneurship"

Authors of the module

Lisa Bald, Kevin Frac, Thomas Nauss, Christoph Reudenbach and Dirk Zeuss.

Available basics as a basis for the module developed in the project

Those responsible for the module have been using and developing digitally-supported forms of teaching/learning for more than 10 years and have already developed online formats for the training of geoinfomation content and skills at an early stage, both in theory and in practice, which are action- and competence-centred (Ammoneit et al. 2019, Schulze et al. 2012, Reudenbach et al. 2010). The resulting teaching materials are available under various open licences. A selection can be found, for example, in the area of geography on the Philipps University OER server (https://uni-marburg.de/myKnp). As a basis for the module development, the existing modules of the B.Sc. and M.Sc. Geography, designed and tested by the module leaders, here in particular "Vegetation Remote Sensing" (https://uni-marburg.de/vvomY), "Data Analysis" (https://uni-marburg.de/WGUqu), "Geo Information Science" (https://uni-marburg.de/VWS57), "Remote Sensing" (https://uni-marburg.de/fPKzC), will be both subject-focused and further developed in terms of didactics and media technology. Up to now, the modules have mainly served to provide information parallel to classroom lectures. In particular, the AI-based spatially and temporally resilient prediction of point data for the derivation of quasi-continuous spatiotemporal information represents a major challenge of the GeoKI module, since the students are integrated into the current research work. Both the sampling of the training data and the performance of the learning algorithms place extended demands on the students in the spatio-temporal and subject-specific context. For this purpose, the module is closely linked to the ongoing research of the LOEWE priority programme Natur4.0 (https://www.uni-marburg.de/de/fb19/natur40) in the Marburg Open Forest (MOF) in the sense of inquiry-based learning (Meyer et al. 2019, Gottwald et al. 2019, Meyer et al. 2018).

Learning objectives

Students will be able to

• Classify spatial data (e.g. from point measurements, surveys, remote sensing) based on their properties and understand the resulting characteristics of spatial or spatio-temporal predictions,
• select, adapt and apply machine learning methods to solve spatio-temporal problems in human-environment research,
• design, apply and evaluate training/testing strategies for reliable error and validity estimation,
• pre-process and standardise highly diverse, large (remote sensing) data sets in a question-specific manner,
• select and create appropriate representations for visual analysis and presentation of results
• use digital platforms for collaborative project management, software development and active data management,
• work together in teams in an agile manner,
• document their approach and results in a comprehensible and transparent manner, and analyse and critically evaluate the results.

Syllabus

The course spans 10 sessions, with the exception of session 10 on a weekly basis. Spatial issues and spatial data

1. Many of humanity's major challenges are spatial. Learn about the importance of spatial configuration using the IPBES framework Nature's Contributions to People as an example.
2. spatio-temporal information. Learn about and appreciate the importance of spatial and temporal scales of data for environmental science issues.
3. remote sensing. Refresh your knowledge of spatially continuous and spatially explicit measurement techniques. Spatial data and spatio-temporal predictions.
4. "Everything is related to everything else, but near things are more related than distant things." (Tobler 1970) On the invalidity of the first law of geography in heterogeneous spaces. From data reproduction to spatial prediction. Why latitude and longitude are (almost) never good predictive variables.
5. spatial variable selection and validation strategies. Randomly good or resilient
6. random good. Prediction of spatial features with machine learning and random validation.
7. spatial good. Resilient spatial predictions using appropriate selection and validation strategies. Final project in a team: Spatio-temporal prediction.
8. development, documentation and discussion of a spatio-temporally resilient prediction strategy on the basis of selectable, exemplary research questions.
9. examination performance (2-3 weeks after unit 9): presentation of the results (poster, webinar or similar according to a concrete blended learning concept in the specific study programme).