The 2024 water crisis in Bogotá, during which the El Niño phenomenon caused the city's reservoirs to reach critically low levels, revealed the necessity of reliable tools for forecasting water storage. This article presents a practical method for modeling the behavior of four important reservoirs in the city's water supply system: Chuza, Neusa, Sisga, and Tominé. The central hypothesis is that each reservoir's storage dynamics and the most suitable statistical model are governed by its geographical location. Chuza, which has a unimodal rainfall regime, was represented with a conventional SARIMA model. In contrast, the reservoirs of the Northern Aggregate, which are subject to a bimodal regime and strong human intervention for flood control, required a hybrid approach that combined Fourier series with ARIMA. The hybrid models produced low forecast errors: 5.37 \% for Sisga, 5.67 \% for Tominé, and 8.87 \% for Neusa. In contrast, the SARIMA model generated a 28.52 \% error for Chuza. Residual diagnostics confirmed that all models were correctly specified and produced residuals with no significant autocorrelation. Out-of-sample validation in 2024-2025 showed that the hybrid models kept the observations within their confidence intervals, even for Tominé, which was validated during the El Niño period. In contrast, the observed storage in Chuza fell below the intervals, reflecting an extraordinary climatic event rather than a model deficiency. These results suggest that high forecast errors do not necessarily imply incorrectly specified models but may instead reveal correctly specified models confronted with conditions outside their historical regimes. In conclusion, the geography of each basin defines its rainfall regime and the type of model best suited to its dynamics. Hybrid approaches, such as the one proposed, separate the climatic response from operational decisions. This offers water managers a useful tool to anticipate crises and plan with greater certainty.