Geography, geospatial analysis and artificial intelligence: between traditional methods and new spatial paradigms

In 1986, yes 1986, it is not a typographical error, Helen Couclelis, one of the pioneers of Geographic Information Science (GIScience) and an expert in urban and regional modelling and planning, spatial cognition, and the geography of the information society, indicated that artificial intelligence (AI) and geography were interconnected within a broad context that addressed fundamental issues of theory, epistemology, and scientific method. At the time, the still little-known logical-mathematical context of AI was explored, demonstrating substantial implications for modelling in general, prediction, the interpretation of theoretical structures, and, moreover, for our understanding of science as a process. These more abstract aspects of the 'computational revolution', of which AI is a product, could, according to the author, have far-reaching consequences for geographical research (Couclelis, 1986).Since then, AI has been profoundly transforming geography by providing advanced methods for analysing, modelling, and interpreting geospatial data. As a discipline concerned with spatial relations, places, environments and human-environment interactions, geography increasingly benefits from AI methods capable of processing large and complex spatio-temporal datasets (e.g., Big Data), generating new insights into territorial dynamics and complex geographical phenomena (Encalada-Abarca et al., 2023).AI has made significant contributions to geospatial data processing, extending existing capabilities for handling and analysing increasingly large and heterogeneous datasets (Mohamad Alshiha, 2024). Techniques based on machine learning, including deep learning, have become increasingly relevant for processing and interpreting geographic information. These methodologies enable the identification of patterns, the modelling of trends, and the development of new analytical approaches to urban, environmental, and social challenges. AI-based methodologies are now widely explored in geospatial data analysis and interpretation, strengthening the connections between AI principles and geographical research questions (Mohamad Alshiha, 2024). This relationship is also evident when AI is considered in the context of geomatics.Although the use and scope of the term vary across disciplinary and national traditions, geomatics broadly encompasses fields concerned with the acquisition, management, analysis, and representation of geographically referenced information, including GIS, remote sensing, photogrammetry, geodesy, surveying, and digital cartography. These technologies support the collection, analysis, and visualisation of geospatial data, thereby contributing to the study of spatial dynamics (Viana et al., 2021;Viana & Rocha, 2020). Please note that we use the term "geospatial" instead of "spatial" or "geographical" to emphasise that the data and processes considered are explicitly referenced to locations on Earth, whereas "spatial" may refer more broadly to relationships that are not necessarily geographical (Longley et al., 2015). This terminological precision may seem exaggerated, but it reflects the complexity and scope of the phenomena studied by geomatics together with AI.Alongside geomatics, geocomputation shifts the emphasis towards complex, computerbased spatial modelling, simulation, and data analysis. While geomatics has traditionally placed strong emphasis on data acquisition, surveying, representation, and geographic information management, geocomputation focuses on the development and application of computational methods to geographical problems, increasingly incorporating machine learning and large-scale data analysis. Advances in AI technologies have further expanded the ways in which large-scale and spatio-temporal geographical information can be analysed and applied (Y. Li et al., 2024).Geocomputation, as a field dedicated to applying computational methods to geography, plays an essential role in this context. By providing algorithmic and methodological support, geocomputation enables the effective exploration of the vast geographic datasets currently available, expanding analytical and discovery capabilities (Fontanella & Xiao, 2018). The integration of AI and spatial computing offers unprecedented opportunities to understand and address phenomena and communities more effectively and sustainably, providing a promising platform to tackle complex challenges at the intersection of humans and the environment (Chauhan & Shekhar, 2021).From this perspective, this article analyses the emergence of Geospatial Artificial Intelligence (GeoAI) as an interdisciplinary domain spanning AI, GIScience, geomatics, geocomputation, and digital geographies. Beyond its conceptual evolution and main areas of application, the epistemological tensions associated with GeoAI are analysed, alongside its methodological and ethical challenges, and complemented by the contribution of explainable artificial intelligence (xAI) to a more transparent and responsible use.Before discussing GeoAI, let us define how we understand traditional spatial analysis in this context. Conventional GIScience and geocomputational methods have typically relied on spatial statistics, econometrics, geostatistics, interpolation, map algebra, multicriteria analysis, network analysis, supervised image classification, cellular automata, Markov models, and other rule-based or process-based models. In these methodologies, the researcher generally defines, at least in part, the variables, relationships, rules, spatial structures, or functional forms that represent a given geographical process (i.e., defines the context). GeoAI does not turn these established approaches obsolete, it instead, expands the methodological process by enabling the integration of complex, nonlinear, and high-dimensional relationships directly from large and heterogeneous geospatial datasets (i.e., data lakes).GeoAI has emerged as an innovative interdisciplinary field integrating geography and AI. This integration enhances the analysis of spatial processes, intelligent reasoning, and the identification of complex patterns and relationships associated with geographical phenomena (Gao, 2020). Its capacity to perform large-scale analyses, automate specific analytical tasks, and detect complex patterns can offer important advantages in situations where conventional spatial-analysis workflows become difficult to scale. Some GeoAI approaches can also achieve robustness to particular forms of noisy or imperfect data, although their performance remains highly dependent on data quality, model architecture, and the geographical context in which they are applied (W. Li & Hsu, 2022). In Physical Geography, it is frequently used to forecast floods, landslides and wildfires (Rezvani et al., 2024;Z. Wang & Brenning, 2023;Zou et al., 2023). Among its most common applications in Human Geography are predicting urbanisation patterns and assessing sustainable mobility scenarios, thereby promoting efficient urban planning (Y. Wang & Zhu, 2024). In a more specific field of expertise, such as Health Geography, which brings together concepts of both Human and Physical Geography, GeoAI has improved the analysis of spatial patterns related to disease spread and inequalities in access to healthcare infrastructure, contributing to more equitable public management strategies (Ahmad et al., 2024). The emergence of GeoAI brings back an old epistemological tension. On the one hand, we witness the expansion of neogeography and the proliferation of user-generated geographic content, particularly Volunteered Geographic Information (VGI). On the other hand, there is renewed interest in computational and complexity-based approaches, including models inspired by concepts developed in the physical sciences and subsequently applied to social systems (Miller & Tolle, 2016). This duality reflects a historical tension between macrogeography and microgeography. The former focused on formulating general principles and laws (nomothetic knowledge), while the latter emphasised detailed and context-specific description (idiographic knowledge). This division, which characterised important debates in twentieth-century geography, is now re-emerging under new technological and methodological conditions (Miller & Goodchild, 2015). This tension is particularly relevant in research concerned with the role of spatial context in human systems. Some traditions within human geography have been critical of the direct transfer of models and analogies derived from the physical sciences to social phenomena. Nevertheless, complexity research has shown that interactions among components of a system can generate emergent characteristics that are difficult to predict from the behaviour of individual components alone. This phenomenon is particularly relevant in human systems, which exhibit patterns of self-organisation, complexity, and feedback processes that may generate emergent and counter-intuitive outcomes (Pollock, 2016).In this context, the role of space is fundamental. The local pattern of interactions between the components of a system directly influences the resulting dynamics, shaping the system's overall behaviour. This phenomenon, often referred to as aggregate complexity, emphasises the importance of spatial context in the analysis of human systems and their emergent dynamics (Miller, 2017). In GIScience, both perspectives are present: macrogeography in algorithms and methods, while microgeography is in the data. This relationship enables clear connections among physical space, GIS and spatial analysis (Barnes & Wilson, 2014).Although GeoAI still faces some resistance (mainly) in human geography, it already reveals promising applications across various subfields of its more quantitative branch.These include the cultural, economic, political, historical, urban, social, environmental, and transport dimensions (P. Liu et al., 2025). These applications highlight the potential of GeoAI to address critical issues such as population dynamics and urban sustainability (S. Wang et al., 2024). In this context, digital geographies (re)emerge as a field exploring the social, cultural, and economic implications of digital technologies in geography.With increasing digitalisation, the impacts of digital tools and AI on how we understand and interact with space become evident. Digital geographies offer a critical perspective on how digital technologies shape human interactions and space, and how they reflect the complex dynamics between humans and technological environments. Issues such as data governance, digital inequalities, and the impacts of digital platforms on urban space became central to this field of study (P. Liu et al., 2025). Furthermore, digital geographies allow us to analyse how digitalisation redefines concepts of space and place, transforming our perceptions and experiences (Richardson, 2024). For example, the ubiquity of mobile technologies and social networks significantly alters the way we interact with perceived space (Betco et al., 2025;Betco, Viana, et al., 2026). This approach is particularly relevant in contexts of rapid urbanisation and digitalisation, where phenomena such as digital gentrification, platform-mediated tourism, and data-driven urban planning emerge as central topics (Aegerter & Söderström, 2025).In this broader landscape, GeoAI constitutes one important methodological development through which digital data can be transformed into spatial information and, when adequately contextualised, geographical knowledge. Whether applied to urban dynamics, environmental risks, or accessibility, the interaction between AI, geography, geomatics, geocomputation, and digital geographies creates new opportunities for interdisciplinary research. These domains overlap but remain conceptually distinct, combining computational innovation with different traditions of geographical inquiry (Usery et al., 2022).Despite all its significant advances, GeoAI still faces important challenges. One of these is the lack of integration with the methodological contexts of the geographical sciences, which, in turn, can compromise the efficiency and relevance of GeoAI solutions (Scheider & Richter, 2023).GeoAI should not be regarded as a substitute for geographical knowledge through exclusively data-driven approaches. Instead, its primary value resides in the integration of data-driven learning methods with established geographical theory and spatial reasoning. Although AI algorithms are capable of detecting complex and nonlinear relationships or patterns that may not be readily identified by conventional methodologies, the interpretation of these results continues to depend on a thorough understanding of spatial processes, geographical context, scale, spatial dependence, and the interactions between human and environmental systems. Consequently, the methodological innovation of GeoAI arises from the synthesis of computational learning techniques with geographical reasoning, which enables the interpretation of data-driven patterns within frameworks that are both theoretically grounded and spatially explicit. Therefore, GeoAI serves to complement and enhance traditional geographical approaches by introducing new analytical capabilities, while maintaining geographical knowledge as an essential element of the research process.This requires an approach that combines technological advances with a deeper understanding of geographical contexts. A further challenge concerns the capacity of GeoAI models to represent complex human behaviours and nonlinear relationships between behaviours and their determinants (S. Wang et al., 2024). Access to computational resources, high-quality geospatial data, and technical expertise also remains uneven across countries, institutions, and communities, creating technological and scientific inequalities. In addition, GeoAI systems trained on biased or unrepresentative data may reproduce or amplify existing social and spatial inequalities.These models are essential for advancing knowledge and improving practice in the human-environment interface, enabling more robust and applicable solutions in realworld contexts. However, some problems still persist. Developing countries (not only but mainly) face significant limitations, including limited access to advanced technologies and high-quality geospatial data, which exacerbate technological and scientific inequalities. Furthermore, GeoAI systems, when trained with biased data, can perpetuate prejudices and inequalities, reinforcing existing disparities. One of the most frequently discussed issues in the literature is the "black box" problem, a term widely used in AI to describe models whose internal decision processes are difficult for users to interpret. This concern is particularly associated with complex models such as deep neural networks. Such opacity can constitute an obstacle to the critical evaluation and adoption of GeoAI, particularly in high-stakes application domains where the consequences of model-based decisions may be significant (Mersha et al., 2024).It is undeniable that clear explanations of models can facilitate the acceptance of GeoAI technologies by policymakers and communities. Understanding the internal operations of models is fundamental, as it helps clarify doubts, identify potential errors, and ensure appropriate levels of accountability (Letrache & Ramdani, 2023). In high-risk applications, such as medical diagnostics, this clarity is indispensable (Aljalaud & Hosny, 2024).These concerns have contributed to the growing development of explainable artificial intelligence (xAI). xAI encompasses methods designed to make the behaviour and outputs of AI models more interpretable and understandable to users. In the context of GeoAI, such approaches can help clarify which variables, spatial patterns, or interactions contribute to model predictions, thereby supporting critical evaluation and informed use of model outputs (Gaurav & Tiwari, 2023). Some explanation methods are described as model-agnostic because they can be applied to different predictive models without requiring access to, or dependence on, a particular internal model architecture (Avinash et al., 2024;Viana et al., 2021). xAI can help users investigate how GeoAI models process and combine complex information, increasing transparency in model-supported decision-making. By revealing potentially problematic dependencies, uneven model behaviour, or biases in predictions, xAI may support assessments of fairness in applications such as urban planning and resource allocation. However, explainability alone does not guarantee fairness or spatial justice, which also depend on data representativeness, institutional context, decision rules, and the distributional consequences of model use.To achieve the objectives of xAI, various approaches have been developed for different models, users, and contexts. Post-hoc explanation methods such as LIME (Local Interpretable Model-agnostic Explanations) and SHAP (SHapley Additive exPlanations), including model-agnostic variants, can provide information about the contribution of individual features to model predictions (Kim & Joe, 2024;Salih et al., 2024).Yet, other approaches are in use, involving simpler substitute models that mimic the behaviour of complex models to facilitate their interpretation (Kim & Joe, 2024), or using rule-based techniques that generate explanations based on frequent patterns of certain features, making complex predictions more understandable (Moradi & Samwald, 2021). Despite all of these advances and diversity, xAI still faces significant challenges. The main one is that explanations need to be both accurate and accessible, which often requires a balance between interpretability and model precision, as excessive simplification can compromise prediction quality (Botari et al., 2020;Ghassemi et al., 2021). Another obstacle is the need to tailor explanations to different stakeholders, such as programmers, end-users, and regulators, each with specific requirements for clarity and detail (Yen et al., 2023). In some cases, the generated explanations can be ambiguous or prone to misinterpretation, especially for users without technical training (Betco, Ribeiro, et al., 2026). GeoAI should therefore be understood neither as a replacement for traditional geographical methods nor simply as an extension of GIS through more powerful algorithms. Its significance lies in the changing relationship between data, analytical specification, geographical theory, and inference. While conventional approaches often require researchers to define a greater proportion of the relationships, rules, or features under investigation, GeoAI allows some of these representations to be learned from increasingly complex geospatial datasets. This shift increases rather than diminishes the importance of geographical reasoning. Spatial context, scale, dependence, heterogeneity, transferability, and the social conditions under which data are produced remain central to determining whether computationally detected patterns constitute meaningful geographical knowledge. The future of GeoAI therefore depends not only on advances in artificial intelligence, but also on its capacity to remain theoretically grounded, spatially explicit, methodologically transparent, and critically aware of the territories and societies to which it is applied.