Spatiotemporal Models and Inference (Seminar)

Spatiotemporal Models and Inference (Block Seminar Fall 2025)

---

In many scientific fields, data is not just tied to a single location or a single moment in time—it evolves over both space and time. Whether tracking the spread of infectious diseases, predicting climate patterns, modeling traffic flows, or analyzing environmental pollution, understanding how variables change across both dimensions is crucial. This is where spatiotemporal modeling comes in.

Spatiotemporal models aim to capture dependencies and relationships in data that vary over time and across geographical locations. Unlike purely spatial or purely temporal models, these approaches integrate both dimensions, allowing us to uncover trends, make predictions, and quantify uncertainties in dynamic systems. Spatiotemporal models can be of different types:

• Deterministic Models: Based on physical laws or equations (e.g., fluid dynamics for weather modeling).
• Statistical Models: Use probability distributions to describe uncertainties (e.g., Gaussian processes, Bayesian hierarchical models).
• Machine Learning Models: Learn patterns from large datasets using neural networks, random forests, or deep learning approaches.

The latter two will be the main focus of this seminar. Applications of spatiotemporal modelling can be found across the sciences and include

• Epidemiology: Forecasting the spread of diseases (e.g., COVID-19 modeling).
• Climate Science: Analyzing temperature changes over decades.
• Urban Planning: Optimizing transportation networks and predicting traffic congestion.
• Ecology: Tracking animal migration or deforestation patterns.

Throughout this seminar, we will explore the foundations of spatiotemporal modeling, discuss cutting-edge methodologies, and learn about real-world applications. To a varying extent, we will cover:

•  Theoretical underpinnings of spatiotemporal statistics.
•  Bayesian approaches and hierarchical models.
•  Machine learning techniques for handling complex spatiotemporal data.

Possible topics are listed below. Students are also welcome to suggest topics and papers of their own choosing if they fit the scope of the seminar.

We're looking forward to meeting you in the seminar!

Lecturer: Jonas Wahl

Time:  September 22-26. Preparations start earlier.

Kick-off Meeting: TBD

Max. number of participants: 10

Prerequisites: You have successfully participated in a machine/deep learning course, or you have a background in statistical modeling, for instance thanks to a course on statistics or stochastic processes.

Grading (Tentative):

50%: Presentation.

25%: Participation in talks of fellow participants and contribution to the discussion.

25%: Final report on your topic.

 

Preliminary list of topics:

• Statistics for Spatiotemporal Data, Chapter 3: Fundamentals of Temporal Processes (2 talks). 
• Statistics for Spatiotemporal Data, Chapter 4: Fundamentals of Spatial Random Processes (2 talks).
• Introduction to Gaussian processes (see the lecture notes here, and here, as well as this famous book)
• Coding tutorial on spatial Gaussian processes (you can base yourself on this blog post and have a look at this blog post)
• Introduction to point processes (see the lecture notes here and the slides here and here).
• Coding tutorial on point processes (take us through some of the examples in the tick package and the simulator here. You can also have a look at these Jupyter notebooks. Alternatively, if you prefer working with R code, you can take use through the section on spatial point patterns here)
• Scalable spatiotemporal prediction with Bayesian neural fields
• Coding tutorial on Bayesian neural fields
• A Bayesian machine-learning approach for spatiotemporal prediction of COVID-19 cases
• Deep learning and process understanding for data-driven Earth system science

I might add more topics to the list before the kick-off meeting. Students are also welcome to suggest topics of their own. Your presentation should last between 60 and 75 minutes.