Towards Modeling Geographical Processes with Generative Adversarial Networks (GANs)

We have presented ground breaking research at the COSIT’19 congress in Regensburg.

Generative Adversarial Networks (GANs) have recently demonstrated great potential for a range of Machine Learning tasks, including synthetic video generation, but have so far not been applied to the domain of modeling geographical processes. In this study, we align these two problems and – motivated by the potential advantages of GANs compared to traditional geosimulation methods – test the capability of GANs to learn a set of underlying rules which determine a geographical process. For this purpose, we turn to Conway’s well-known Game of Life (GoL) as a source for spatio-temporal training data, and further argue for its usefulness as a potential standard training data set for benchmarking generative geographical process models.

Specifically, we have evaluated the capability of GANs to learn the underlying rules of Conway’s Game of Life (GoL).

Fig. 1: conditional GAN setting

For this, a conditional GAN was provided with a series of 4 frames of a GoL simulation, and tasked with accurately predicting the cell configuration at the subsequent next frame. Indeed, the results demonstrate that a network architecture based on convolutional long-short-term memory (convLSTM) layers was successful when trained for 50 epochs.

Fig. 2: Test results for traditional GoL

A similar model, however, failed in a second experiment when being confronted with an altered GoL version with changed neighborhood definitions. This is in line with the limited ability of convolutional layers to capture longer term relations and already suggest alternative models to investigate in future work.

Fig. 3: Cross entropy loss for traditional (GoL I) and manipulated game (GoL II)

Specifically, we will in future work explore

  • Extending the standardized benchmarking dataset by altered versions of the GoL which reflect attributes of real-world geographical processes
  • Developing and benchmarking deep generative geographical process models based on these data
  • Experimentally evaluate the validity of the results when translated to real-world geographical data

Check out our COSIT paper for further details or learn more about Deep Generative Geographical Process Models.


Leave a reply

Your email address will not be published. Required fields are marked *



Imprint | Privacy Policy

Stay in the know with developments at IARAI

Select list(s)

updates from the Institute.
You can later also tailor your news feed to specific research areas or keywords (Privacy)

Log in with your credentials


Forgot your details?

Create Account