Reconstructed streamflow using SST and tree-ring chronologies over the southeastern United States

- Using SST and tree rings with SSA improves reconstruction of high flow estimates.
- Multimodel combination on reconstructed streamflow provides better estimates.
- SSA provides a framework for separating periodic and non-periodic components.

SummaryA hybrid approach that considers both tree-ring chronologies and sea surface temperature (SST) data for reconstructing annual streamflow is proposed. The most common approach to reconstruct annual streamflow is to develop statistical regression relationships between principal components of tree rings and observed annual flow values and then extend the relationship to estimate annual streamflow values over the period for which tree-ring chronology is available. The primary limitation of this approach is in estimating high flow values since tree-ring growth reaches its potential limit during wet years. The proposed hybrid approach overcomes this limitation by using SST conditions from the tropical Pacific and tree-ring chronologies from the watershed for reconstructing annual streamflows. For this purpose, we considered eight virgin watersheds having long tree-ring chronologies over the southeastern United States. Given the role of El Nino Southern Oscillation (ENSO) in influencing the hydroclimatology of the southeastern United States, we estimated the periodic component of streamflow using Nino3.4 - an index representing ENSO - and the non-periodic component of streamflow using the non-periodic component of tree rings that represent interannual variability of moisture supply within the region. We employed Singular Spectrum Analysis (SSA) for extracting periodic and non-periodic components from tree-ring chronologies, Nino3.4 and streamflow data. The proposed tree ring and SST hybrid approach was compared with the traditional principal component regression (PCR) approach based on cross-validation. Results show that inclusion of SST provided better reconstructed flow values during high flow years but also resulted in overestimation of flow during low flow years. Combination of annual streamflow estimates from the two models - PCR and the hybrid approach - resulted in improved estimates of reconstructed annual streamflow for the selected eight watersheds. Potential applications for such improved reconstructed streamflow estimates is also discussed.