Methods

Average discharge of the Colorado River at Lees Ferry for the water year (October-September) was reconstructed using a two-stage regression procedure. Tree-growth at each site was first converted into an estimate of discharge by stepwise regression of discharge using tree-ring width indices, from the current year and lagged one year, as predictors. Squared terms on the tree-ring predictors were also included in the regression to allow for possible curvature in relationships between tree-growth and discharge. In the second step, the gage reconstruction was generated by averaging an appropriate set of single site reconstructions. Final estimates of discharge were interpolated from a piecewise-linear smoothed scatterplot of the observed discharge values and the discharge estimates averaged over the individual tree-ring sites. The procedure was repeated for subsets of tree-ring chronologies with different periods of common time coverage to build a “most-skillful” reconstruction, starting in the early 1400s, and a “longest” reconstruction, starting in the early 1100s. Details of the reconstruction method can be found here.

Note: The statistics listed in the table represent average accuracy, while the reconstruction method yields error bars that vary in width over time -- generally wider for wet years than for dry years. The listed statistic R2, the decimal proportion of variance explained by the reconstruction in the calibration period, is computed directly from the reconstruction residuals. For an explanation of these statistics, see this document.