The research team, led by Mustafa Utlu of Marmara University together with Mesut Şimşek and Abdulkadir Özkaya of Hatay Mustafa Kemal University and Eray Selçuk, employed a text mining technique known as Latent Dirichlet Allocation, or LDA, to sift through the corpus. LDA is an unsupervised machine learning algorithm, first formalized in 2003 by David Blei and colleagues, that treats every document as a probabilistic mixture of hidden themes and every theme as a probability distribution over words. Rather than imposing pre-existing categories on the literature, the algorithm lets the vocabulary of the papers themselves reveal the field’s latent structure. The team ran the analysis in Orange, an open-source data mining toolbox, and complemented the topic modeling with hierarchical clustering to group the resulting themes into broader thematic families.

Out of this computational sifting emerged fourteen distinct research topics that together describe the intellectual anatomy of modern geomorphology. At the top of the hierarchy sat three dominant clusters: fluvial geomorphology, the study of how rivers carve, transport and deposit sediment across landscapes; geomorphological analysis in general; and modeling, the increasingly mathematical simulation of surface processes. That rivers and quantitative modeling lead the field will surprise few geomorphologists, but the scale of the dominance, quantified across five decades and two databases, provides the first rigorous statistical confirmation of what had previously been an impressionistic observation.

More provocative are the findings at the bottom of the list. Coastal geomorphology, glacial geomorphology and mass movement research, which encompasses landslides and other slope failures, all ranked among the least represented categories in the corpus. Perhaps most striking of all, karst geomorphology, the study of landscapes sculpted by the dissolution of soluble rocks such as limestone, also landed among the least studied topics. This is a curious result, the authors note, because karst science has been pursued continuously and actively from the earliest decades of the record to the present day. Its low ranking suggests that the field’s output volume, rather than its vitality, has simply been outpaced by the explosive growth of other subdisciplines, a reminder that bibliometric rankings measure attention as much as importance.

The temporal dimension of the analysis tells a story that resonates far beyond academic bibliometrics. The researchers found that the topics gaining the most ground in recent decades coincide with two of the defining challenges of the twenty-first century: global climate change and the mounting toll of natural disasters. Concerns over rising temperatures and drought appear to have pulled research attention toward themes connected to hydrometeorological hazards, water scarcity and landscape instability. The authors are careful, however, to frame this relationship with scientific restraint. As a text mining study, their analysis treats climate change and natural disasters as co-occurring contextual factors rather than established causal drivers, an important distinction in an era when correlation in bibliometric data is easily mistaken for mechanism.

Technology, the study argues, has been the other great engine of change. The rise of high-resolution data streams has transformed what geomorphologists can see and measure. Orthomosaic imagery, geometrically corrected aerial photographs stitched into seamless maps, satellite data of ever finer resolution, and digital surface modeling techniques have collectively shifted the field from labor-intensive field mapping toward data-rich, computationally intensive science. The trajectory mirrors a broader transformation across the geosciences, in which light detection and ranging, or LiDAR, uncrewed aerial vehicles and satellite constellations have made centimeter-scale topographic measurement routine, enabling researchers to track riverbank migration, glacier retreat and landslide deformation with a precision unimaginable to researchers of the 1970s.

Methodologically, the study sits within a rapidly growing tradition of computational scientometrics. Topic modeling has become a standard instrument for mapping large literatures, and the authors ground their approach in a substantial methodological literature addressing questions such as how to choose the optimal number of topics in an LDA model, how to evaluate the semantic coherence of the resulting themes, and how to validate topic models against human interpretation. By combining LDA with hierarchical clustering, the team could move beyond a flat list of topics to a nested thematic structure, revealing how fine-grained research themes aggregate into the major branches of the discipline. The scale of the corpus, spanning both Web of Science and Scopus, matters as well, since the two databases differ in journal coverage and a single-database analysis risks a skewed picture of the field.

The findings arrive at a moment when geomorphology’s societal relevance has never been clearer. The authors emphasize that the discipline continues to contribute significantly to understanding hydrometeorological and geological disasters, accounting for both the morphological processes that shape terrain and the human impacts that destabilize it in the context of global climate change. Landslides, floods, coastal erosion and drought-driven land degradation are all, at their core, geomorphological problems, and the field’s growing engagement with hazard-related themes suggests that researchers are responding to a world in which those problems are intensifying. The study also implicitly documents the rise of anthropogenic geomorphology, the recognition that humans have become one of the most powerful geomorphic agents on the planet, moving earth at scales that rival rivers and glaciers.

For working scientists, the analysis doubles as a map of opportunity. The underrepresentation of coastal, glacial, mass movement and karst research, set against the backdrop of accelerating sea level rise, shrinking glaciers and increasing slope instability in a warming world, highlights potential mismatches between research effort and societal need. Whether these gaps reflect funding priorities, the difficulty of working in polar and coastal environments, or simply the gravitational pull of fluvial and modeling traditions, the study provides the quantitative baseline that research funders and journal editors have long lacked. It also demonstrates, at scale, how machine learning tools originally developed for information retrieval can be repurposed to audit the collective output of an entire scientific discipline.

What emerges from fifty years of compressed text is a portrait of a science in transition: anchored in its classical concern with rivers and landform analysis, increasingly computational, increasingly remote-sensing driven, and increasingly oriented toward the hazards of a changing climate. The authors’ machine had no stake in that story; it simply read thirty thousand papers and counted. Yet the pattern it found, of a discipline pivoting toward the most urgent environmental questions of its time while leaving some of its oldest and most distinctive subfields comparatively quiet, is one that geomorphologists, and the societies they serve, will want to read closely.

Subject of Research: Topic modeling analysis of fifty years of global geomorphology research trends

Article Title: Fifty years of geomorphology research: a topic modeling analysis of global trends (1975–2025)

Article References: Utlu, M., Şimşek, M., Özkaya, A., & Selçuk, E. (2026). Fifty years of geomorphology research: a topic modeling analysis of global trends (1975–2025). Earth Science Informatics, 19(11), Article 197. https://doi.org/10.1007/s12145-026-02253-0

Image Credits: AI Generated

DOI: 10.1007/s12145-026-02253-0

Keywords: geomorphology, topic modeling, LDA, text mining, bibliometrics, climate change, natural hazards, fluvial geomorphology, karst, remote sensing, Web of Science, Scopus