Using large scale climatic patterns for improving long lead time streamflow forecasts for Gunnison and San Juan River Basins

• Main Authors: Kalra, Ajay, Miller, William Paul, Lamb, Kenneth W., Ahmad, Sajjad, Piechota, Thomas C.
• Format: Article in Journal/Newspaper
• Language: English
• Published: Digital Scholarship@UNLV 2012
• Subjects: Climate variability; Colorado – Gunnison River Watershed; Forecasting; Long-range weather forecasting; North America – Colorado River Watershed; Ocean-atmosphere interaction; Oscillations; Streamflow – Forecasting; Support vector machine; Water resource management; Water-supply – Management; United States – San Juan River Watershed; Climate; Environmental Engineering; Environmental Sciences; Fresh Water Studies; Meteorology; San Juan; North Atlantic; North Atlantic oscillation
• Online Access: https://digitalscholarship.unlv.edu/fac_articles/87

In a water-stressed region, such as the western United States, it is essential to have long lead times for streamflow forecasts used in reservoir operations and water resources management. Current water supply forecasts provide a 3-month to 6-month lead time, depending on the time of year. However, there is a growing demand from stakeholders to have forecasts that run lead times of 1 year or more. In this study, a data-driven model, the support vector machine (SVM) based on the statistical learning theory, was used to predict annual streamflow volume with a 1-year lead time. Annual average oceanic–atmospheric indices consisting of the Pacific decadal oscillation, North Atlantic oscillation (NAO), Atlantic multidecadal oscillation, El Niño southern oscillation (ENSO), and a new sea surface temperature (SST) data set for the ‘Hondo’ region for the period of 1906–2006 were used to generate annual streamflow volumes for multiple sites in the Gunnison River Basin and San Juan River Basin, both located in the Upper Colorado River Basin. Based on the performance measures, the model showed very good forecasts, and the forecasts were in good agreement with measured streamflow volumes. Inclusion of SST information from the Hondo region improved the model’s forecasting ability; in addition, the combination of NAO and Hondo region SST data resulted in the best streamflow forecasts for a 1-year lead time. The results of the SVM model were found to be better than the feed-forward, back propagation artificial neural network and multiple linear regression. The results from this study have the potential of providing useful information for the planning and management of water resources within these basins.