Uncertainty evaluation in GeoAI models with sparse training data and Causal Graphs

 Suggested by: Johannes Scholz 

Keywords: GeoAI, Uncertainty Quantification, Sparse Data, Causal Graphs, Bayesian Inference, Machine Learning, Spatio-Temporal Analysis, Environmental Modeling 

 

Objective: Evaluate the uncertainty of GeoAI models when trained on sparse datasets and analyze how uncertainty behaves with varying levels of training data quantity and quality. Additionally, assess the influence of causal graphs in improving model robustness and interpretability. 

 

Short Description: This thesis investigates uncertainty quantification in GeoAI models that rely on sparse training data, such as wildfire occurrence datasets. The research will explore how different amounts and qualities of data impact uncertainty levels and prediction confidence. Furthermore, it will examine the role of causal graphs in enhancing model understanding and reducing uncertainty. By applying statistical and machine learning techniques, the study aims to improve the reliability of GeoAI models in real-world applications. 

 

Start: 

As soon as possible 

 

Prerequisites/qualification: 

• Knowledge in GIS and spatial data processing 

• Interest/Knowledge in GeoAI 

• Interest/Knowledge in Causality