Importance analysis of local and global climate inputs for basin‐scale streamflow prediction

Abstract

Basin‐scale streamflow is influenced by numerous local and global climate inputs. In this paper, genetic programming (GP) is combined with “importance analysis” to identify the important global climate inputs and local meteorological variables needed for prediction of weekly streamflow at the basin scale. The analysis is carried out for the Mahanadi River in India using global climate inputs, namely, the El Niño–Southern Oscillation (ENSO) index and equatorial Indian Ocean Oscillation (EQUINOO) index; local meteorological inputs, including outgoing longwave radiation (OLR), total precipitable water (TPW), temperature anomaly (TA), and pressure anomaly (PA); and streamflow information from previous time steps. The rainfall information over the basin is intentionally not utilized so that the procedure may be applicable to basins with little or no rain gauge information and to achieve a longer prediction lead time. The Birnbaum importance measure is used to assess the importance of each input. Results of this study show that the relative importance of individual input variables is influenced by time lags. It is observed that among various local meteorological inputs, OLR and PA are more important than TA and TPW. Among large‐scale circulation indices, ENSO index is important for previous 5th to 7th week, whereas EQUINOO index is important for previous 3rd to 6th week. On the basis of their importance measures, 15 indices were selected from the initial group of 30 indices. The GP‐derived streamflow forecasting models could predict weekly streamflow with good accuracy (correlation coefficient r = 0.821) for such a complex system.