Reconstructing Multi-Decadal Daily Water Storage Changes in Scottish Standing Waters: A Classification-Based Remote Sensing Framework

Global water storage faces a crisis driven by climate warming, with significant declines observed in 53% of large water bodies (Yao et al., 2023). Crucially, recent analyses reveal that surface water dynamics are predominantly driven by seasonal variability (Li et al., 2025). However, as current assessments are biased toward large lakes, the high-frequency storage dynamics of small systems remain unquantified due to the spatiotemporal limitations of current satellite observations (Cooley et al., 2021). Scotland offers an ideal case study to address this observational gap: it hosts ~25,500 water bodies, of which > 90% are small (<0.1 km²) and poorly monitored (Taylor, 2021).

Currently, Scotland is undergoing a fundamental hydro-climatic transition, indicated by a pronounced intensification of seasonality, with substantially wetter winters but markedly drier summers (Lowe et al., 2018), challenging the reliability of these water resources. Recent extremes, such as Loch Ness recording its lowest levels since 1990 in May 2023, highlight the vulnerability of existing storage capacity (SEPA Water Scarcity Report, 2023). To effectively manage these emerging risks, a comprehensive understanding of storage dynamics is essential. Yet, a multi-decadal, daily-resolution dataset of water storage changes remains absent. Consequently, this study aims to bridge this gap by reconstructing continuous storage dynamics from 1980 to the present.

To account for heterogeneous basin morphology and anthropogenic regulation, we develop a scalable, typology-based framework that categorizes water bodies into three representative classes: (1) shallow/responsive basins (e.g., Loch Leven), where surface area is highly sensitive to water level changes; (2) deep, morphologically constrained basins (e.g., Loch Ness), where storage variability is primarily volumetric; and (3) regulated reservoirs (e.g., Loch Katrine), which exhibit non-natural level fluctuations due to abstraction. Targeting these calibration sites, we integrate Sentinel-1 (SAR) and Sentinel-2 (optical) imagery (2017-2024) with daily in-situ water level observations from SEPA to derive class-specific area-level relationships and validate model performance across contrasting hydrological regimes.

To extend storage reconstructions beyond the satellite era, we employ a machine learning approach driven by long-term meteorological reanalysis data. Models trained on the high-resolution dynamics of the Sentinel era are applied retrospectively to reconstruct daily water storage changes dating back to 1980. By including a dedicated class for regulated systems (Loch Katrine), this framework incorporates features to distinguish human-driven storage patterns from natural climatic responses. The resulting dataset provides the first multi-decadal quantification of Scottish water storage, enabling the identification of historical low water extremes and attribution of their climatic and anthropogenic drivers. This work provides a critical baseline for assessing hydrological resilience and water security in temperate regions under increasing climate variability.