Evaluating Watershed Experiments through Recursive Residual Analysis

Abstract

The question of watershed response to changes in land use and land cover has received a great deal of attention in the hydrologic literature. One of the primary tools for quantifying such changes has been the paired watershed approach. In general this approach applies a linear regression analysis to relate water yield from a control watershed to a physiographically similar watershed in close proximity, and to predict changes in yield following treatment. Although much research on paired watershed experiments has focused on quantification of the impacts of land-use and land cover changes on water yield, little attention has been brought to evaluating explicitly the methodology used to quantify such effects. An alternative method is proposed for examining treatment impacts and their duration, through the application of a cumulative recursive residual test of model stability. The results from a paired watershed study of ponderosa pine watersheds in north-central Arizona are analyzed using both a traditional linear regression analysis and a recursive residuals approach. The results suggest that the linear regression approach may fail to account for the true complexity of watershed response to vegetation treatments, by underestimating the interactions of treatment impacts, shifts in climatic drivers, and revegetation rates.