Ocean-atmospheric Anomalies and Simulated Snow Cover Data Improve Statistical Forecasts of Seasonal Discharge in Central Asia

Abstract:

Forecasts of water availability in the main growing season of crops are crucial for irrigated agriculture in water-deficient regions. In Central Asia snowmelt dominates water discharge during vegetation period but is subject to high inter-annual fluctuations induced by high variability of precipitation during the cold season. To date statistical forecasting models for seasonal river discharge in Central Asia followed two main tracks. One accounts teleconnections of the seasonal precipitation in the region caused by global ocean-atmosphere anomalies, such as the El Nino Southern Oscillation (ENSO) and the North Atlantic Oscillation (NAO). The second strand is to regress discharge volume on the dynamics of seasonal snow cover area.

Here we combine regression-based forecasting of seasonal discharge with global ocean-atmosphere oscillation indices with simulated snow water equivalent (SWE) from the NOAA National Operational Hydrologic Remote Sensing Center. We regionalized multiple river basins in Central Asia into distinct groups using catchment-averaged monthly precipitation dynamics. For each catchment group, early winter hydrological outlook uses a similar set of global climate indices for predicting mean discharge from April to September. With decreasing forecast lead time, these predictors are gradually replaced with the monthly SWE estimates.

This approach provides robust seasonal forecasts with high predictive capacity while allowing greater control for overfitting. We evaluated the resulted statistical models and their predictive uncertainty with leave-one-out cross validation. The adjusted R-squared values of the resulted statistical models for forecasts at the beginning of vegetation season were above 0.7, whilst normalized RMSE values remained below 20% on average. The results reveal a strong overall influence of the NAO and ENSO anomalies on the variability in annual discharge, with substantial spatial and temporal variation conditioned by the mean catchment altitude. Overall, our approach provides a novel way to improve the forecasting of seasonal discharge volumes in snowmelt dominated watershed in arid and semi-arid regions.