Gangopadhyay, SubhrenduÌý1
1ÌýAMEC Hydrosphere
A non-parametric method for paleohydrologic reconstruction (NPP) of streamflows using tree-ring chronologies is developed. The method uses pattern matching of chronologies in EOF-space (Empirical Orthogonal Function) to identify a set of K-nearest neighbors (similar years) from a contemporary period (common period of tree-ring and streamflow data) to the paleohydrologic reconstruction years. Streamflow ensembles are subsequently generated using weighted resampling. This algorithm was developed and applied to reconstruct annual streamflow ensembles for the important Lees Ferry, AZ gage on the Colorado River Basin for the period 1400-1905. Results from the NPP algorithm was compared with existing parametric reconstructions for the Lees Ferry gage. Cross-validated streamflow reconstructions for the contemporary period compares very well with the observed flows and also with other published streamflow reconstructions. The non-parametric approach has several advantages over the parametric methods – the NPP is data-driven, and can generate streamflow ensembles with asymmetric confidence bounds. Furthermore, the non-parametric approach also proved to be very robust, and was able to generate ensemble streamflows using raw tree-ring width measurements directly from only sixteen individual trees in the Upper Colorado River Basin region. Since the NPP algorithm uses only tree-ring data, this approach can be used to reconstruct structured and even non-numerical data for use in water resources modeling.