Volume: 5, 2026
5th International PhD Students’ Conference at the University of Life Sciences in Lublin, Poland:
ENVIRONMENT – PLANT – ANIMAL – PRODUCT
Abstract number: E012
DOI: https://doi.org/10.24326/ICDSUPL5.E012
Published online: 22 April 2026
Comparative analysis of a hybrid CA–XGBoost model and the FLUS model for urban growth simulation: Lublin, Poland
Mücahit Kaya* and Barbara Sowińska-Świerkosz
Department of Hydrobiology and Ecosystems Protections, University of Life Sciences in Lublin, 37 Dobrzańskiego St., 20-631 Lublin, Poland
* Corresponding author: mucahit.kaya@up.edu.pl
Urban growth simulation is critical for understanding spatial development dynamics and supporting sustainable planning. However, the non-linear nature of urban expansion poses significant challenges for traditional Cellular Automata (CA)-based models. This study proposes a hybrid CA–XGBoost approach to improve the spatial prediction of urban growth.
The model was applied to Lublin, Poland, using 10 m resolution Sentinel-2 data for 2015, 2020, and 2025 to generate land use/land cover maps and produce projections for 2030. The proposed method integrates machine learning-based suitability modeling with CA-based spatial allocation, enhancing spatial realism by considering neighborhood effects and patch-level constraints. Results show that the hybrid model achieves a Figure of Merit (FoM) of 0.3067 for the 2025 simulation, outperforming the FLUS model (FoM = 0.091) in capturing urban growth patterns. Both models demonstrate high classification accuracy, with the hybrid model reporting Overall Accuracy (OA) = 0.9618 and Kappa = 0.9496, while the FLUS model achieves OA = 0.9656 and Kappa = 0.9548. The 2025 urban area predictions indicate that FLUS estimated 62.11 km2, the hybrid model predicted 62.47 km2, while the observed urban area was 63.02 km2. Projections for 2030 suggest the hybrid model predicts urban areas reaching 64.41 km2, slightly higher than the FLUS estimate of 63.65 km2.
Overall, the findings indicate that the hybrid CA–XGBoost approach provides a more realistic spatial representation of urban growth in Lublin compared to the FLUS model. Nevertheless, these differences should be interpreted considering the structural distinctions between the models, as the FLUS model simulates transitions across all land use classes, with temporary classes such as “construction” potentially reducing predictive accuracy.
Keywords: CA; FLUS; Lublin; urban growth; XGBoost
How to cite
Kaya M., Sowińska-Świerkosz B., 2026. Comparative analysis of a hybrid CA–XGBoost model and the FLUS model for urban growth simulation: Lublin, Poland. In: 5th International PhD Students’ Conference at the University of Life Sciences in Lublin, Poland: Environment – Plant – Animal – Product. https://doi.org/10.24326/ICDSUPL5.E012