Artificial Intelligence (AI) has been aiding the geospatial industry in various stages, such as generating data, analysing data, and presenting it through maps. We have previously published a blog post on GeoAI which can be found by visiting this link. In the current blog post, we will discuss the use of AI in the process of map-making, i.e. cartography. The integration of AI technology in cartography is referred to as CartoAI.
AI has accelerated complex cartographic design tasks, making them easier to accomplish. Here are a few examples worth noting:
- Assigning symbology actively: Cartographers put in a lot of effort to create accurate maps with symbols that are indexed through a legend. Choosing the right symbols can be complicated, as it needs to vary depending on the scale of representation. This adds an extra level of complexity to the process. However, recent developments in AI have made it possible for machine learning algorithms to learn from existing maps and create new styles.
- Assigning labels at appropriate locations and with appropriate sizes: Labels have historically been difficult to work with as they require a lot of manual adjustments before the final map can be published. However, there is ongoing research into the use of AI to place appropriate labels on maps. This involves training AI models with maps designed by cartographers and then utilizing CycleGAN and Pix2Pix architectures to generate labels.
- Relief shading: It is a way to represent the terrain features on a map by using light and shadow. The direction of illumination is crucial in creating this effect. In hand-drawn relief shading, the direction of illumination can be adjusted locally to fit the terrain feature’s size and shape. In digital relief shading, mathematical models are used to simulate the effect of terrain elevation, but the direction of illumination is typically kept constant to avoid complexity. However, with the U-Net neural network, it has become possible to create a more human-like effect for shaded relief.
Human-in-the-loop AI for cartography: Those involved in the AI industry are familiar with the concept of “human-in-the-loop”. It involves a joint effort between humans and AI when neither can independently produce the desired outcome. This approach is particularly useful when using AI in cartography, as it allows us to achieve what was previously impossible.
Text-to-map generation using AI: Text-to-image generation models are being used to develop on-demand mapping technology. To make it work, it is necessary to understand the meaning of the request. The DALL-E model has been employed to create maps, but at present, it is only suitable for artistic map generation. The maps it generates are not very reliable, but it has given rise to more creative cartography.
With advanced levels of detail being represented efficiently, the future of maps looks promising.
References:
Kang, Y., Gao, S., & Roth, R. E. (2024). Artificial intelligence studies in cartography: A review and synthesis of methods, applications, and ethics. Cartography and Geographic Information Science, 0(0), 1–32. https://doi.org/10.1080/15230406.2023.2295943
Jenny, B., Heitzler, M., Singh, D., Farmakis-Serebryakova, M., Liu, J. C., & Hurni, L. (2021). Cartographic Relief Shading with Neural Networks. IEEE Transactions on Visualization and Computer Graphics, 27(2), 1225–1235. https://doi.org/10.1109/TVCG.2020.3030456
Kang, Y., Gao, S., & Roth, R. E. (2019). Transferring Multiscale Map Styles Using Generative Adversarial Networks. International Journal of Cartography, 5(2–3), 115–141. https://doi.org/10.1080/23729333.2019.1615729