Recent applications using tree-ring data have been effective in providing resource managers with an understanding of past hydroclimatic variability. Multicentury tree-ring records of hydroclimatic variability are now being considered in assessments of the regional risk of multiyear droughts and the expected occurrence interval of extreme drought. Critical questions raised by the water managers when introducing tree-ring reconstructions into the decision process have focused on describing and quantifying uncertainty in the annual stream flow reconstructions. Potential sources of uncertainty in the annual stream flow reconstructions are: 1) measurement errors in both the instrumental data and the tree-ring data, and 2) model calibration errors due to the less than perfect relationship between tree-ring data and hydroclimatic variables. Previous efforts describing and quantifying uncertainty in tree-ring reconstructions of past environmental conditions have focused on statistical analyses of a regression model that maximize the variance and characterize uncertainty in terms of model fit to calibration data. These approaches assume the regression model that maximizes the variance explained in the calibration data set represents the most likely relationship in the reconstruction period. In this study we explore the use of a suite of regression models to reconstruct past hydroclimatic variability as an alternative to using a single “best” model”.

Our ensemble tree-ring based reconstructions draw on the ensemble methodology used in operational weather and climate forecasting prediction. Our application of an ensemble of regression models in tree-ring reconstructions of past hydroclimatic variability allows for a more realistic representation of uncertainty in the development and application of regression model relationships between tree-ring indices and hydroclimatic variability. The ensemble approach allows for the possibility that a suite of models based on a variety of calibration and independent verification periods can provide a more accurate representation of the past. Although individual regression model ensemble members may explain less of the variance, the regression model ensemble mean should yield a more robust reconstruction while providing a more complete representation of uncertainty. Application of this approach to reconstruct streamflow in the South Platte suggests that use of a single “best” model may be underestimating hydrologic extremes. In addition, the ensemble approach indicates an increase in uncertainty (as measured by spread of the ensemble members) in the reconstruction of hydrologic extremes relative to other parts of the record.