Speaker
Description
Spatial planning increasingly relies on geospatial datasets, models, and decision-support systems to determine where attention, investigation, and resources should be concentrated. Yet data-intensive approaches can create a misleading impression of certainty when observations are incomplete, unevenly distributed, or collected using methods with different capacities to detect the phenomenon of interest. In such circumstances, an apparently empty spatial unit may represent genuine absence, inadequate observation, inappropriate survey methods, unavailable data, or simply an area that has never been investigated.
Submerged archaeological landscapes provide a particularly clear example of this problem. Late Quaternary sea-level change repeatedly exposed and inundated extensive areas of the continental shelf, creating former terrestrial and coastal landscapes that may preserve archaeological and palaeoenvironmental evidence. Their contemporary visibility, however, is conditioned by burial, erosion, sediment dynamics, water depth, survey design, and the capabilities of available sensors and samples. Consequently, the known underwater archaeological record cannot be treated as a representative inventory of what exists. Non-detection is not equivalent to absence, particularly where observation itself varies spatially and methodologically (Banning et al., 2006; Thomas & Darvill, 2022; Wallach, 2019).
This creates both an archaeological and a planning problem. Detailed investigation of an entire continental shelf is neither technically nor economically realistic. Researchers and planning systems therefore require a defensible way to determine where detailed assessment should begin. Conventional suitability or priority models frequently address this type of problem by combining indicators into weighted indices or by using observed occurrences as training data. Both approaches can conceal important epistemic distinctions. A favourable palaeolandscape is not an archaeological site; terrestrial occupation evidence does not demonstrate offshore presence; mapped preservation conditions are not equivalent to archaeological survival probability; and greater survey density does not make a landscape intrinsically more archaeologically relevant.
This contribution presents SCREEN, an observation-aware regional framework developed for the Iberian palaeo-shelf. Its central question is not “Where are archaeological sites likely to occur?” but rather: which broad sectors warrant progression to detailed submerged-archaeology assessment, given palaeogeographic opportunity, regional human-use evidence, preservation context, target-sensitive observation adequacy, and evidence confidence?
The conceptual proposition is that uncertainty should not merely appear as an error term attached to a final model score. Where incomplete evidence could materially alter a planning or assessment decision, uncertainty should affect the classification itself.
SCREEN separates landscape relevance from the adequacy of existing observations. The framework first represents Palaeogeographic Opportunity (P): whether a spatial unit was terrestrially available during a specified chronological window. This establishes environmental opportunity rather than archaeological probability. Regional Human-Use Support (H) then represents independently documented terrestrial occupation or use within predefined regional neighbourhoods. Such evidence remains contextual and cannot be translated into offshore archaeological presence.
P and H are combined through categorical, non-compensatory logic to derive Landscape Relevance (LR). Observation density is not allowed to increase LR. This prevents historically well-investigated areas from appearing inherently more relevant simply because more evidence has been collected there.
A separate component, Preservation-Process Evidence (R), describes mapped environmental states including burial or accumulation, consolidated or exposed conditions, erosion or reworking, mixed or complex settings, and unresolved evidence. These states are not interpreted as numerical probabilities of archaeological survival.
The observation question is then made explicit through Target (J). Three target categories are distinguished: surface or near-surface archaeological material; buried or depositional targets; and stratigraphic or palaeoenvironmental targets. This distinction is important because an observation method that is informative for one target may be inadequate for another.
Available survey and investigation information forms the Observation Vector (O), which is translated into Target-Sensitive Observation Adequacy (O*) according to the specified target and preservation context. Catalogue presence alone cannot create a strong observation state, while missing observations remain unresolved unless sufficient evidence demonstrates that the relevant footprint has been investigated using methods capable of addressing the specified target.
Finally, Evidence Confidence (K) qualifies the result. Confidence cannot increase the assessment priority. Instead, where uncertainty could change the classification materially, the framework retains an explicitly unresolved state.
The final output is Detailed-Assessment Need (AN). AN is not a site-probability estimate, risk score, legal status, or management recommendation. It is an upstream scientific screening state indicating the comparative remaining need to progress a sector to finer-resolution archaeological assessment. This distinction is central to the methodology: landscape relevance does not prove observation, survey intensity does not create relevance, and contemporary development pressure does not modify the archaeological evidence state. Development pressure, statutory designation, planned use, and institutional response are intentionally reserved for later planning translation. This boundary is already explicit in the Article 0 architecture.
SCREEN is implemented across the Iberian Peninsula and adjacent continental shelf using a common equal-area spatial lattice. The operational domain contains 18,319 unique hexagonal units, each covering exactly 4 km², represented in ETRS89-extended / LAEA Europe (EPSG:3035). The eligible analytical area covers approximately 62,526 km². The coarse regional lattice is deliberate: SCREEN is designed to select broad sectors for subsequent investigation rather than reproduce the fine-resolution archaeological-potential workflow applied within selected regions.
Assessment is performed independently across five chronological windows spanning approximately 55–6 ka cal BP and independently for the three target categories. There is therefore no single, target-free assessment map. A location may be relatively well characterised for one target while remaining unresolved for another. Each classification also retains an auditable reason code describing its landscape-relevance, preservation, target, observation, and confidence states.
The decision architecture is deterministic rather than machine-learned. No classifier is used because the research does not presently contain a defensible learning target, a representative investigated presence/absence sample, or unbiased survey labels. Artificial intelligence is therefore restricted to an optional explanation layer over frozen outputs. It may translate reason codes into accessible language, compare predefined scenarios, or identify which missing evidence prevents resolution, but it cannot alter classifications, infer archaeological sites, impute missing information, or recommend a planning response.
Interim implementation already demonstrates why target-sensitive uncertainty matters. Under the current preservation classification, the 18,319 analytical cells include substantial areas associated with burial or accumulation and mixed or complex preservation conditions, together with more than six thousand cells in which preservation evidence remains unresolved. A total of 17,266 cells retain the same primary preservation state under predefined 50%, 60%, and 70% dominance tests, indicating that the principal preservation classification is comparatively robust to the tested threshold specification.
Observation adequacy is substantially more target dependent. For buried or depositional targets, 5,421 cells currently contain limited observation support while 12,898 remain unresolved. For stratigraphic or palaeoenvironmental targets, 2,125 cells contain limited support while 16,194 remain unresolved. For surface or near-surface archaeological targets, all 18,319 cells remain unresolved under the accessible registered evidence. These states do not demonstrate archaeological absence. They reveal how strongly apparent knowledge depends on the question being asked and on whether the available observations are capable of answering it.
Final LR and AN results are intentionally withheld until the independent human-use evidence passes its remaining acquisition, calibration, deduplication, and quality-control gates. This withholding is itself methodologically important. Rather than completing the framework through proxy substitution or by treating incomplete information as neutral, SCREEN preserves insufficient evidence as a reason to postpone inference.
The framework illustrates a broader issue in model-mediated planning. Spatial models do not merely organise data; they determine which conditions become visible, which unknowns disappear, and which territories appear sufficiently understood to support subsequent decisions. Models may unintentionally privilege locations with abundant observations and treat poorly documented areas as neutral or low concern. When the geography of observation reflects previous research investment, infrastructure development, regulatory intervention, or institutional access, such outputs risk reproducing the history of evidence collection while presenting it as a property of space.
SCREEN addresses this problem by incorporating the observation process into the decision architecture and by allowing uncertainty to remain an explicit outcome. Its non-compensatory structure prevents strong evidence in one dimension from cancelling a decision-critical unknown in another. Its target-sensitive observation logic recognises that evidence adequacy cannot be evaluated independently of the question a planning process seeks to answer.
The broader planning proposition is therefore straightforward: unknown should not automatically become intermediate. Where missing or inadequate observation could materially alter a spatial decision, unresolved knowledge should remain visible and actionable.
For submerged archaeology, this provides a transparent basis for allocating scarce detailed-assessment effort without claiming knowledge that the evidence cannot support. More generally, the same principle is applicable to ecological monitoring, environmental assessment, mobility data, and other planning domains in which non-observation may be systematically confused with absence.
Rather than attempting to eliminate uncertainty through increasingly complex prediction, SCREEN makes the limits of evidence explicit, auditable, and target-specific. In data-intensive planning environments, this distinction can improve both analytical robustness and the transparency of the decisions that follow.
References
Bailey, G., Galanidou, N., Jöns, H., Peeters, H., & Mennenga, M. (Eds.). (2020). The archaeology of Europe’s drowned landscapes. Springer. https://doi.org/10.1007/978-3-030-37367-2
Banning, E. B., Hawkins, A. L., & Stewart, S. T. (2006). Detection functions for archaeological survey. American Antiquity, 71(4), 723–742. https://doi.org/10.2307/40035886
Flemming, N. C., Harff, J., Moura, D., Burgess, A., & Bailey, G. N. (Eds.). (2014). Land beneath the waves: Submerged landscapes and sea level change. European Marine Board.
Fraile-Jurado, P., Mejías-García, J. C., Roldán-Muñoz, M. E., & Borja-Barrera, C. (2025). Mapping subaerial exposure timing on the Iberian Peninsula continental shelf during the Last Glacial Cycle. Boletín de la Asociación de Geógrafos Españoles. https://doi.org/10.21138/bage.3425
McCarthy, J., Benjamin, J., Wiseman, C., Beaman, R., Kowlessar, J., McGowan, G., Baggaley, P., Jerbić, K., Fowler, M., & Neudorf, C. (2022). Beneath the Top End: A regional assessment of submerged archaeological potential in the Northern Territory, Australia. Australian Archaeology, 88(1), 65–83. https://doi.org/10.1080/03122417.2021.1960248
Mejías-García, J. C., & Fraile-Jurado, P. (2026). Spatial analysis of changing sea levels and its implications for understanding prehistoric human settlement patterns in the southwestern Iberian Peninsula. The Journal of Island and Coastal Archaeology, 21(1), 104–129. https://doi.org/10.1080/15564894.2025.2512155
Teschendorff, M., Benjamin, J., & O’Leary, M. (2025). Evaluating deep time: A model for the archaeological assessment of submerged landscapes along the Pilbara coast, Western Australia. Journal of Maritime Archaeology, 20, 419–454. https://doi.org/10.1007/s11457-025-09450-w
Thomas, R., & Darvill, T. (2022). What haven’t we found? Recognising the value of negative evidence in archaeology. Antiquity, 96(388), 955–967. https://doi.org/10.15184/aqy.2022.69
Wallach, E. (2019). Inference from absence: The case of archaeology. Palgrave Communications, 5, Article 94. https://doi.org/10.1057/s41599-019-0307-9
Ward, I., Larcombe, P., Firth, A., & Manders, M. (2014). Practical approaches to management of the marine prehistoric environment. Netherlands Journal of Geosciences, 93, 71–82. https://doi.org/10.1017/njg.2014.2