RESEARCH TEAMS:
• Optimisation of settlement and road network generalisation at small scales using artificial intelligence and graph theory
Project Leader: Eng. Izabela Karsznia, DSc
Team: Eng. Karolina Wereszczyńska, PhD
Albert Adolf, MA; Iga Ajdacka, MA
International team:
Prof. Dr Arzu Çöltekin (FHNW, Switzerland)
Prof. Dr Stefan Leyk (University of Colorado Boulder, USA)
Prof. Dr Robert Weibel (University of Zurich, Switzerland)
NCN OPUS Project: Optimisation of settlement and road network generalisation at small scales using artificial intelligence and graph theory,
No. UMO2020/37/B/HS4/02605 (2021–2025)
• Optimisation of map legend design as a component of geovisualisation tools in the context of efficiency and information acquisition strategies
Project Leader: Prof. Izabela Gołębiowska, DSc.
Team: Robert Piątek, MA
Prof. Tomasz Opach, (NTNU, Norway; WGSR UW)
Prof. Arzu Çöltekin (FHNW, Switzerland)
Jolanta Korycka-Skorupa, PhD (WGSR UW)
Prof. Jan Ketil Rød (NTNU, Norway)
NCN SONATA Project: Optimisation of map legend design as a component of geovisualisation tools in the context of efficiency and information acquisition strategies, No. 2018/31/D/HS6/02770 (2019–2023)
Can a computer think like a cartographer, and can a user trust a map that speaks in many voices? At the Faculty of Geography and Regional Studies, two independent projects were conducted exploring the boundaries and capabilities of modern cartography. One teaches algorithms to select map content in the same way that experienced cartographers do. The other examines how the presentation of information—especially in interactive, multi-panel visualisation tools—influences our ability to comprehend it. Despite the methodological differences, both projects share a common goal: to create maps that not only present data, but that also enable its understanding.
Cartographic generalisation is a key stage in the production of small-scale maps, involving the deliberate simplification of content to maintain legibility and usability. Until now, no tool has existed that could fully automate this process by replacing expert knowledge with an algorithm. The project undertaken at our faculty fills this gap by developing a comprehensive method for automatic selection of geographic features—such as settlements, road and river networks—for small-scale maps. The research employed advanced machine and deep learning tools, graph theory, and spatial analysis. Models were trained on datasets from Poland, Switzerland and the United States, using expert maps as references. This enabled the algorithms to “learn” which features should be retained on the map and which could be omitted. In the future, these models may be used by national cartographic institutions, significantly accelerating and improving map production processes.
In parallel, research was conducted on the perception of complex, interactive geovisualisation tools—so-called geo-dashboards—that combine various forms of data presentation: maps, charts and tables. Although these tools enhance analytical capabilities, their complexity can lead to cognitive overload for users. The aim of the second project was to investigate the role of legend design in the perception of such tools. An eye-tracking study was carried out, comparing two types of legend layouts: a combined legend (in a single interface location) and a distributed legend (assigned to individual panels). The results showed that users preferred the localised solution—consistent with the Gestalt principle of proximity—even though it was visually more dispersed. The findings were used to formulate specific design recommendations for geovisualisation tools.
Upper graphics:
↑ Eye-tracking data showing that legend layout influenced more frequent and shorter gaze transitions to elements of the distributed legend compared with the combined legend.
↑ Selection results for watercourses within the Nysa Kłodzka catchment. A – source data from the General Geographic Objects Database; B – reference atlas map used for machine learning; C – selection based on regulation using priority stream classification; D – selection using decision trees with genetic algorithms (DT-GA).
Red percentage values indicate the extent to which the generalisation result matches the reference map. Source: Ajdacka, Karsznia (2023).
Lower graphics:
↑ FROM THE LEFT. 1
Example of a multi-element geovisualisation tool ViewExposed, composed of a thematic map, parallel coordinates chart, and tables. The tool presents data on the exposure and resilience of Norwegian municipalities to natural hazards such as floods, landslides and storms. Source: Gołębiowska, I., 2021.
Maps of Chojnice and Bytów counties – selection results for 1:500,000 scale.
a) Source data from the General Geographic Objects Database
b) Reference atlas map used for machine learning
c) Selection result from Random Forest model
d) Selection result from Deep Learning model
e) Selection result from DT-GA model (decision trees + genetic algorithms)
f) Selection result from decision tree model Red percentage values show agreement with the reference map. Source: I. Karsznia (2023).
The legend layout not only influenced the gaze frequency and user evaluations, but it also affected the sequence of information reading: the thicker the ribbon was, the more often the participants shifted their gaze between the elements of the interface.
Using the distributed legend (left diagram), significantly more users moved their gaze from the bar chart to the data selection function compared with the combined legend (right): compare the width of the red ribbon on the left and right. Source: Gołębiowska et al., 2023 Map of Kępno county – selection result for 1:500,000 scale.
a) Source data from the General Geographic Objects Database
b) Reference atlas map used for machine learning
c) Basic selection result based on official generalisation rules
d) Automatic selection result using DT-GA model (decision trees and genetic algorithms)
e) Selection result using Random Forest (RF) model Red percentage values indicate the degree of agreement with the reference map. Source: I. Karsznia et al. (2022)
RESEARCH LOCATION: North America / Europe (SWITZERLAND / POLAND / NORWAY / UNITED STATES)