Reconstructing nineteenth-century Danube river water levels with transformer-based computer vision

We convert nineteenth-century Bavarian Danube gauge charts (1826–1894) into daily water-level series referenced to gauge zero through a novel semi-automated workflow combining light document pre-processing, dewarping, transformer-based line extraction, pixel-to-curve calibration, and targeted human checks. A curated ground-truth sample supported benchmarking and uncertainty quantification. Across three representative gauges (Neu-Ulm, Vilshofen, Passau), the pipeline attains high series-level accuracy (mean composite score 0.979) while reducing manual effort by roughly an order of magnitude relative to full manual digitisation. Outputs include versioned datasets with page-level provenance, confidence scores, and methodological descriptors to ensure transparency and reuse. The approach offers a replicable template for rescuing analogue hydrometric records and enabling long-term analyses of extremes, regulation impacts, and ecological context. Data are openly available under CC BY 4.0 (https://doi.org/10.5281/zenodo.17296750, Rehbein, 2025).