Assessing the utility of climate variability information in streamflow forecasting

The prediction of total and peak streamflows are essential for effective management of water resources systems. A data-driven approach, Model Tree (MT), is applied to predict daily streamflows for a tropical river basin in India. The Tapi River drains a total area of 65,225 km², wherein more than 20 million people are directly or indirectly dependent on it for their water and food requirements. The MT approach executes piece-wise linearization of a non-linear process for the input parameter space and develops linear regression models for each sub-space. The large-scale oceanic-atmospheric oscillations, such as El Niño-Southern Oscillation (ENSO), exert considerable influence on the hydroclimatic conditions across the globe. Based on the Oceanic Niño Index, the warm and cool phases of ENSO are identified as El Niño and La Niña, respectively. It is found that the El Niño and La Niña are associated with drier and wetter than normal conditions respectively across the Tapi basin. Hence, the hypothesis that incorporation of climate variability information would help in enhancing the predictive performance of the model is being tested. A daily-time step model for streamflow prediction is developed considering various hydrometerological inputs observed for the period 1975-2013 to predict streamflows at the catchment outlet. Additionally, two separate models, viz., El Niño- and La Niña-specific models, are developed considering the observed variables corresponding to these phases, and their skill of prediction with respect to the overall model is evaluated. The evaluation of the developed models is further carried out through a suite of statistical error and performance indices, and inferences are drawn.