Long-Range Forecasting of Colorado Streamflows Based on Hydrologic, Atmospheric, and Oceanic Data

Abstract

Climatic fluctuations have profound effects on water resources variability in the western United States. The research reported herein centers on streamflow predictability at the medium- and long-range scales in rivers that originate in Colorado. Specifically, we want to improve forecasting seasonal and yearly streamflows based on atmospheric-oceanic forcing factors, such as geopotential height, wind, and sea surface temperature, as well as hydrologic factors, such as snow water equivalent. The approach followed in the study involves searching for potential predictors, applying principal component analysis (PCA) and multiple linear regression (MLR) for forecasting at individual sites, canonical correlation analysis (CCA) for forecasting at multiple sites, and testing the forecasts using various performance measures. The analysis includes comparisons of forecasts by using various combinations of possible predictors, such as hydrologic, atmospheric, and oceanic variables. The study brought into relevance the significant benefits of using atmospheric, oceanic, and hydrological predictors for long-range streamflow forecasting. It has been shown that forecasts based on PCA applied to individual sites give very good results for both seasonal and yearly timescales. We also found that although PCA has been applied on a site-by-site basis, the forecasts approximated well the historical cross correlations, although some underestimation was noted for two sites. Furthermore, the forecasts based on CCA were less efficient than those based on PCA.