Long-Lead Probabilistic Streamflow Forecasting Using Ensemble Streamflow Prediction (ESP) and Large-Scale Climate Signals

Understanding the ocean-atmospheric interactions that results in climate modes will not only improve weather and climate forecasting capabilities, but also could greatly enhance hydrological forecast skills. Therefore, identifying the effects of these phenomena on hydrological and meteorological parameters of different basins throughout the world is of great importance. On the other hand, Ensemble Streamflow Prediction (ESP) -part of the US National Weather Service River Forecasting System (NWSRFS)- is a well-known advanced hydrological forecasting technique that considers the forecast uncertainty in terms of occurrence probability and provides probabilistic forecast information rather than deterministic. This probabilistic approach, together with large-scale climate information could beneficially lead us to more accurate hydrological forecasts with longer forecasting lead times. In this research, as the first step, two of the most prominent known sources of interannual and interdecadal climate variability in the form of El Niño-Southern Oscillation (ENSO) and Pacific Decadal Oscillation (PDO) are analyzed to assess the influences of these large-scale climate phenomena on water supply in the Zayande-rood River Basin. For this purpose, any shifts in the mean and variance of the inflow volume of the Zayande-rood Dam in different climate conditions (resulted by different combination of ENSO and PDO phases) has been analyzed and compared with similar statistics in neutral condition to find out if the differences are statistically significant. Correlation analysis indicates that the inflow volume has a direct relation with Pacific Decadal Oscillation Index (PDO Index) and an inverse relation with Southern Oscillation Index (SOI). Also, it was found that any significant shift in the mean January through July inflow volume arises only when El Niño and La Niña events occurs respectively in the positive and negative phase of the PDO. To give some physical justification to the results, precipitation and temperature patterns in the above basin of Zayande-rood Dam are analyzed. As the second step, the hydrologic model is initialized based on meteorological and hydrological data of a year similar to the actual water year preceding the forecast year. Currently, forecast of ENSO condition with lead times of about 6 months up to a year are available. Having known the ENSO condition of the coming water year and the persistent phase of PDO, the climate condition of the coming water year is determined and then, the initialized hydrological model is driven to produce the ensemble members. The central tendency of the forecasts i.e. ensemble mean represents the forecasts hydrograph. Retrospective forecasts of the historic record are prepared to evaluate the technique, as well as applying probabilistic verification methods such as Skill Score.