Cascade Dynamics and Drought Sensitivity of the Keban-Karakaya-Atatürk Reservoir System on the Euphrates River: A Satellite-Altimetry, Statistical, and Machine-Learning Assessment

Abstract

The Keban, Karakaya, and Atatürk dams constitute the primary cascade dam system—designed for hydroelectric power and irrigation—of the Southeastern Anatolia Project (GAP) on the Euphrates River. Since these three reservoirs are hydraulically connected in series, a water level anomaly occurring upstream can propagate downstream over a period spanning days or even months; nevertheless, this cascade system is rarely analyzed as an integrated whole. This study utilized multi-mission satellite radar altimetry WSE records obtained from the DAHITI database (Keban: 2016–2026, n = 289; Karakaya: 1993–2026, n = 672; Atatürk: 2002–2026, n = 535) to: characterize long-term and recent water surface elevation (WSE) trends (Mann-Kendall test, Theil-Sen slope); assess inter-reservoir interactions (monthly cross-correlation and lag analysis, 2016–2026); test—using the Mann-Whitney U test with Holm-Bonferroni correction—whether WSE values ​​differed significantly between drought periods (1999–2001, 2008–2009, 2020–2021) and wet periods (1993–1998, 2019–2020) identified in the literature; and develop an exploratory random forest classifier for low water level risk.

Raw Mann-Kendall trends were significant for all three reservoirs, but this significance was not maintained after applying the autocorrelation-resistant (Hamed-Rao) correction; Therefore, Theil-Sen slope values ​​based on paired time intervals (Keban +0.53 m/yr; Karakaya +0.28 m/yr; Atatürk +0.16 m/yr) are presented descriptively. Although zero-lag correlations were observed between successive reservoirs (Keban-Karakaya r = 0.72; Karakaya-Atatürk r = 0.61), only the relationship between the Keban-Karakaya pair remained significant after adjusting the effective sample size for autocorrelation. Water surface elevation (WSE) was significantly lower at Karakaya during each drought period (p < 10⁻⁶) and at Atatürk during the 2008–2009 period (p < 10⁻¹⁹); however, the Atatürk reservoir did not exhibit a significant decline during the 2020–2021 period (p = 0.79), a pattern consistent with the largest reservoir in this cascade system acting as a buffer. During the 2019–2020 wet season, water levels rose by 15.7 m at Keban, 13.1 m at Karakaya, and 8.8 m at Atatürk; these increases were consistent in terms of order of magnitude with the 4–16 m rise reported independently across the basin for the same event. Nevertheless, this signal associated with the wet season failed the robustness test based on monthly sampling and was considered merely a descriptive finding. In a random forest classification model adjusted for the leakage effect, the anomaly in the upstream reservoir emerged as the second most informative predictor variable—following the reservoir's own lagged state while turbine discharge estimates derived independently of electricity generation records revealed the same top-down lag structure. These results support treating the Keban-Karakaya-Atatürk system as an interconnected cascade system for the purposes of water level monitoring and early warning.