Geospatial Artificial Intelligence (GeoAI) represents a transformative force in spatial analysis, combining artificial intelligence with geospatial data to address critical urban and environmental challenges. This study seeks to answer the research question: how can GeoAI contribute to reducing territorial inequalities and promoting more sustainable cities? The research examines GeoAI applications in urban planning, infrastructure monitoring, and environmental risk management, highlighting its potential to revolutionise how spatial data is analysed and used for decision-making.

The theoretical approach is grounded in an interdisciplinary framework, synthesising insights from human geography, computational sciences, and sustainability studies. It draws on concepts of spatial justice and resilience to explore the socio-economic and ecological dimensions of GeoAI’s integration into urban systems. By bridging these fields, the study aims to uncover the transformative capacity of GeoAI in addressing disparities between urban centres and peripheral regions.

Methodologically, the study employs a mixed-method approach. Quantitative analysis leverages advanced machine learning models, including deep learning algorithms for satellite image processing, spatial data visualisation, and predictive modelling. These techniques are applied to big geodata to uncover patterns and dynamics in urban development and environmental change. Complementing this, qualitative methods involve case studies and a critical literature review to explore ethical implications, operational challenges, and governance frameworks for GeoAI adoption. Data sources include open-access geospatial datasets and high-resolution satellite imagery from programs such as Copernicus, ensuring robust and diverse data inputs.

The findings are expected to contribute to the growing body of research on GeoAI’s role in shaping urban futures. The study highlights opportunities for GeoAI to optimise urban planning, enhance disaster preparedness, and address territorial inequalities. However, it also underscores the need to address challenges, such as data biases, accessibility issues, and regulatory gaps. By critically assessing these dimensions, this research seeks to provide actionable insights for policymakers and practitioners aiming to harness GeoAI for sustainable and inclusive urban development.

Land-use/land-cover (LULC) change prediction depends on input variables, but the importance of each variable may vary based on geographical location, time, and the targeted changes in the study area. Most environmental variables are derived from remote sensing data. With advancements in artificial intelligence (AI), particularly machine learning (ML), various change detection models have been developed that effectively assess the importance of input variables based on these criteria. Such weightings are critical for layer embedding in Markov models to predict future changes. In this research, we predict land-use changes over a 30-year period using the Land Change Modeler (LCM) with different transitional models, including Multi-Layer Perceptron (MLP), Decision Forest (DF), and Support Vector Machine (SVM), for the Krkonoše Protected Area in the Czech Republic. We utilized satellite images from the TM and ETM+ sensors of Landsat 5 and 7 and the OLI sensor of Landsat 8 for the years 1990, 2000, 2010, and 2020, respectively. Additionally, the ASTER Digital Elevation Model (DEM) and spatial data related to human and natural factors were incorporated as comprehensive input variables for our transitional ML models. Satellite images were classified in Google Earth Engine into eleven classes. The LCM Markov model predicted land-use changes, comparing our applied ML models. According to the results, from 1990 to 2020, the classes of deciduous, coniferous, and mixed forests increased respectively by +0.74%, +7.11%, and +6.15%. Conversely, transition forests and clearings, areas with sparse vegetation, decreased by -9.81% and -1.20%, respectively. Land-use change predictions for 2020, generated by the applied models, were validated against the actual 2020 land-use map using the overall accuracy method. The models achieved accuracy levels of MLP: 83.38%, SVM: 86.72%, and DF: 89.88%. This study demonstrates the effectiveness of combining machine learning techniques with the LCM Markov model for LULC prediction. The findings emphasize the dynamic nature of forest ecosystems and the need for accurate models to ensure sustainable resource management.

The MOSAIK project introduces an innovative framework to capture and analyze socioeconomic transformations in Vienna at the micro level. By harnessing diverse image data—from Google Street View and Kappazunder to aerial photography—we employ AI-based techniques to automatically extract urban features such as facade structures, roof conditions, and ground floor uses. Neural networks detect building facades and roofs, while supervised machine learning models classify ground-level activities. This enables the comprehensive mapping of diverse built-functional areas across the entire urban space: the usage patterns of ground floor zones, the structural condition of facades, as well as roof conditions and attic expansions. Geostatistical analyses, including hotspot and cluster detection, further delineate areas of urban appreciation and depreciation. By integrating these spatial metrics with socioeconomic indicators like education, household income, and migration background it obtains a comprehensive overview of the socioeconomic transformation in Vienna. Moreover, MOSAIK aims to clarify whether social structures and dynamics can be automatically extracted from image data. This presentation will discuss our methodology, first findings, and the broader implications of street view image analysis for human geography.

This project explores the use of machine learning to investigate state-led cultural assimilation policies in the People's Republic of China, with a particular focus on urban public spaces affected by the nationwide “Sinicization” campaign. These policies, aimed at aligning ethnic, religious, and linguistic minority cultures with a dominant Han Chinese identity, manifest visibly through architectural modifications: altered shop signs, erased ornaments, and defaced building facades. Documenting such spatial changes is increasingly challenging due to restricted access to affected regions and a lack of comprehensive data.

To address this gap, the project leverages Baidu’s “Total View” platform—China’s equivalent of Google Street View—which continuously captures panoramic images across urban and rural environments. By harnessing computer vision and pattern recognition techniques, the project develops an AI-based framework to algorithmically detect architectural and spatial alterations over time. These changes, though subtle—like glue stains where signage was removed or inconsistencies in wall textures—reveal the systematic implementation of assimilation policies in everyday urban environments.

The methodology involves comparing multiple iterations of street-level imagery from identical locations to identify visual discrepancies. This AI-driven temporal analysis enables the detection and quantification of thousands of instances of visual erasure or modification across a vast geographic scale. The approach transforms Baidu’s regularly updated image streams into a dynamic, searchable archive for visual evidence of cultural suppression.

This work not only contributes a novel methodological approach to geographic inquiry—blending AI with critical urban studies and political geography—but also demonstrates the potential of machine learning to uncover patterns of socio-political transformation that might otherwise remain obscured. The project highlights both the power and ethical responsibility of using AI in politically sensitive contexts, especially where traditional fieldwork is restricted. By making visible the hidden geography of state control, this research underscores the role of AI in extending the reach and depth of human geographic analysis into contested and inaccessible spaces.