Accuracy of hydrologic forecasts

The analysis in this paper concentrates on the Colorado River Basin. Analyses of the correlations between forecasted and observed runoffs were made using Spearman rank-order coefficients, Product-Moment correlation coefficients, average forecast errors, and a derived statistical measure referred to as the Coefficient of Prediction. In addition, further statistical measures including the t-test were used. The analysis showed that even as early as January of each year worthwhile forecasts are being developed. The Coefficient of Prediction and the correlation coefficients compare favourably when used to judge which forecast stations had relatively higher forecast accuracy. The two statistical tests of correlation coefficients and Coefficients of Prediction are in agreement when analyzing the average forecast errors. These coefficients increase when the average forecast error decreases. The t-test proved to not be as good a measure to use when comparing different data sets.