BEGIN:VCALENDAR VERSION:2.0 PRODID:-//pretalx//talks.osgeo.org//foss4g-2026//speaker//GWMUZA BEGIN:VTIMEZONE TZID:JST BEGIN:STANDARD DTSTART:20000101T000000 RRULE:FREQ=YEARLY;BYMONTH=1 TZNAME:JST TZOFFSETFROM:+0900 TZOFFSETTO:+0900 END:STANDARD END:VTIMEZONE BEGIN:VEVENT UID:pretalx-foss4g-2026-CY3TZS@talks.osgeo.org DTSTART;TZID=JST:20260903T150000 DTEND;TZID=JST:20260903T153000 DESCRIPTION:Terrain traversability mapping is an important geospatial task in off-road mobility assessment\, route planning\, environmental analysis\ , and spatial decision support. In recent years\, AI and machine learning methods have increasingly been used to estimate terrain-related indicators from heterogeneous spatial inputs\, offering a scalable alternative to pu rely rule-based GIS procedures. However\, model performance in this domain is often summarized through aggregate regression metrics alone\, while th e spatial structure of prediction error and the influence of incomplete ge ospatial reference data remain underexplored. This paper addresses that ga p by focusing not only on predictive accuracy\, but also on uncertainty in AI-based terrain traversability modelling. Rather than asking only whethe r a model predicts well on average\, we ask where it fails\, why those fai lures occur\, and how limitations of source geodata propagate into mapped analytical outputs.\n\nThe study investigates the estimation of a terrain passability coefficient (IOP)\, expressed as a continuous value in the ran ge from 0 to 1 and interpreted as the difficulty of traversing a given ter rain unit. The modelling workflow is built around regular 100 m × 100 m g rid cells\, which serve as primary spatial units for analysis. Within each cell\, vector-based topographic and thematic data are transformed into a structured set of quantitative attributes describing surface features\, li near features\, point objects\, and terrain morphology. This conversion fr om a discrete vector data model to a continuous feature space enables the use of machine learning methods for prediction of terrain passability. The final dataset contains 236\,617 records and 115 non-empty explanatory var iables selected from a larger attribute structure affected by missing valu es. A compact artificial neural network is used for regression-based estim ation of IOP\, with 20% of the observations reserved for model testing.\n\ nThe architecture of the model is intentionally simple. It consists of an input layer\, one hidden layer with 57 neurons\, and a single-neuron outpu t layer. The hidden layer uses the ReLU activation function and L2 regular ization\, while model training is guided by mean absolute error as the los s function and a decreasing learning rate. This design choice is deliberat e: the goal of the study is not to claim novelty through increasingly comp lex network architectures\, but to examine how geospatial data quality con strains the reliability of AI-based prediction. In this sense\, the model is treated as an analytical instrument for exposing data-driven uncertaint y rather than as an end in itself.\n\nAt the aggregate level\, the model a chieves good performance. On the test set\, the results reach BIAS = 0.002 4\, MAE = 0.0097\, and RMSE = 0.015\, with training and validation behavio ur suggesting neither severe overfitting nor underfitting. These values in dicate that machine learning can successfully approximate the terrain pass ability coefficient from structured geospatial attributes. Yet the study s hows that such global statistics do not fully capture model behaviour in s pace. The distribution of errors is not random\, and the most important di screpancies emerge in value ranges and terrain contexts that are weakly re presented or incompletely encoded in the source data. In particular\, low IOP values tend to be underestimated\, while the upper extreme of the coef ficient range is also less stable. This means that the model may overestim ate traversability in some difficult areas and underestimate it in highly passable terrain\, despite apparently good overall metrics.\n\nThe most si gnificant contribution of the paper lies in the spatial interpretation of these discrepancies. Local analysis reveals that some of the strongest dev iations are not primarily caused by the neural model itself\, but by incom pleteness in the geospatial reference base used to construct training and validation labels. Two examples are especially illustrative. A very large discrepancy was identified over a mining area\, where the lack of sufficie ntly detailed source information caused the feature representation of the terrain cell to differ substantially from actual conditions. Another impor tant discrepancy appeared in wetland terrain\, where missing swamp-related information in VMAP led to a clear mismatch between predicted and referen ce passability. These cases show that a model may appear statistically rob ust while still producing spatially misleading outputs in areas where key terrain categories are absent\, weakly represented\, or overly simplified in the underlying geodata. From the perspective of GeoAI\, uncertainty is therefore not only an algorithmic issue\; it is also inherited from the th ematic completeness and semantic fidelity of the input spatial database.\n \nA second major source of uncertainty concerns representativeness. The tr aining data originate from a specific study area and do not include explic it temporal labels\, even though terrain traversability is inherently sens itive to seasonal and environmental variability. A model that performs wel l in one region may not preserve the same level of accuracy when transferr ed to another region or to similar terrain observed under different weathe r or seasonal conditions. This issue is particularly important for travers ability analysis\, where the same land cover or soil-related feature may b ehave differently across time due to moisture\, freeze-thaw effects\, vege tation changes\, or local environmental context. The paper therefore argue s that evaluation of geospatial machine learning models should explicitly consider both thematic incompleteness and spatiotemporal representativenes s\, rather than treating prediction error as a purely technical property o f model architecture.\n\nThe contribution of the paper is threefold. First \, it presents a practical AI-based workflow for terrain traversability pr ediction from structured geospatial attributes derived from open spatial a nalysis procedures. Second\, it demonstrates that spatially explicit error analysis provides insights that remain invisible in global regression sta tistics alone. Third\, it frames incomplete and unevenly representative ge ospatial reference data as a primary source of uncertainty in traversabili ty prediction. By shifting attention from model novelty to data-driven rel iability\, the paper contributes a realistic and methodologically transpar ent perspective to GeoAI research in the open geospatial domain. The broad er implication is that future work on terrain traversability prediction sh ould invest not only in better models\, but also in richer\, more complete \, and more temporally explicit geospatial reference datasets. DTSTAMP:20260717T220449Z LOCATION:Cosmos2 SUMMARY:Machine Learning for Terrain Traversability Mapping: Accuracy\, Unc ertainty\, and Data Limitations - Marek Wyszyński\, Krzysztof Pokonieczny URL:https://talks.osgeo.org/foss4g-2026/talk/CY3TZS/ END:VEVENT END:VCALENDAR