Stochastic Streamflow Generation Incorporating Paleo-Reconstruction

The Colorado River Basin experienced the worst drought on record from 2000–2004. Though this drought was unprecedented in the observed streamflow record (1906–present) reconstructed streamflows dating back to 1490 generated from tree-ring chronologies have shown droughts of greater magnitude and duration. Yet, the decision to adopt the information from tree-rings is not without question. Alternate techniques to reconstruct streamflows based on tree-rings can display different magnitudes of past flow. Though the magnitudes are different these various reconstructions do show similar system state, i.e., wet or dry state. We present a technique to combine reconstructed streamflow state information with the observed records flow magnitude. This a achieved using a nonhomogeneous Markov chain model with kernel smoothing coupled with a K-nearest neighbor sampling algorithm. The technique is demonstrated for the Lees Ferry stream gauge on the Colorado River. The coupled models retain the ability to generate basic statistics similar to the observed record while also capturing the state properties of the reconstructed streamflows.