| Summary: | Diverse non-linear methods are applied for medium to long-term hydrological forecasting in Colombia, involving the hydrological persistence and the influence of macro-climatic phenomena such as El Nino/Southern Oscillation (ENSO), the Quasi-Biennial Oscillation (QBO), the North Atlantic Oscillation (NAO), etc. We applied such methods to forecast mean river discharges of the Nare River at the central Andes of Colombia, whose inflows are used in the largest regulation reservoir of the Colombian hydropower system. Non-linear methods include multiple linnear regression (MLR), Multivariate Adaptive Regression Splines (MARS), Neural Networks, Historical Analogues, Non-Parametric Joint Distributions, and the Wavelet Transform is used in the context of a new prediction method that we introduced based upon decomposition on spectral bands (PREBEW). The methods are calibrated and verified to estimate monthly prediction errors during «blind prediction» exercises. The most important timescales that explain Colombian hydroclimatology are identified, and the degree of importance of each independent variable is established, depending upon the month of the year being forecasted, and the sensitivity of predictions with respect of the initial month. Our results indicate important predictive skills when comparing the illustrated methods with more traditional linear and markovian predictive models. Se implementan diversos métodos no lineales para predicción hidrológica de mediano y largo plazo en Colombia, involucrando la persistencia hidrológica y la influencia de fenómenos macro-climáticos como el ENSO, la QBO, la NAO, etc. Los métodos se ilustran aplicados al caso de la predicción de los caudales medios mensuales del río Nare, en los Andes centrales de Colombia, que alimenta el embalse de regulación más grande del país para la generación de energía eléctrica. Se usan los métodos de regresión lineal múltiple (RLM), MARS (“Multivariate Adaptive Regression Splines”), Redes Neuronales, método de Análogos Históricos, métodos no ...
|
|---|