Historical photographs of waterways, harbours, and coastlines survive in archives worldwide, but reconstructing these lost waterscapes as interactive three-dimensional environments remains unsolved: archival photographs are sparse, single-angle, and captured with unknown camera parameters, while existing reconstruction methods require either multiple overlapping views or restrict output to planar facades. This paper proposes a pipeline that bridges the gap through three contributions. First, multi-angle hallucination: an AI video generation model synthesises a plausible camera orbit from a single archival photograph, and the resulting frames are submitted to a conventional Structure from Motion pipeline as a multi-view image set. A systematic literature search confirms that this technique has not been previously formalised in the academic record, although adjacent work exists in Gaussian Splatting augmentation and reconstruction–generation coupling. Second, a certainty-driven visualisation system maps reconstruction confidence continuously to rendering style: a per-vertex score derived from photogrammetric reprojection error, video-model temporal consistency, and archival evidence quality modulates opacity, geometric detail, edge character, and texture simultaneously through a multi-axis shader, advancing beyond the discrete colour-coded overlays used in prior heritage uncertainty visualisation. Third, a regional water profile system parameterises water appearance—sediment colour, turbidity, flow character, seasonal variation—as interchangeable ScriptableObjects rendered in a painterly aesthetic, demonstrated across four geographically distinct water bodies. The pipeline integrates GIS data ingestion, procedural building generation, and time-slice switching within Unity 6, and is demonstrated with Auckland archival data. Limitations including video-model geometric consistency, certainty-shader empirical validation, and water-profile manual calibration are assessed.
James Otto Danenberg (Thu,) studied this question.