Identifying Patterns in Long Term Streamflow Variability and Predictability in the Upper Colorado River Basin using a Nonlinear Dynamics Approach

Abstract:

Long term shifts and oscillations in hydrologic regimes can influence streamflow dynamics and predictability. Combining wavelet analysis and nonlinear embedding previous research demonstrated decadal streamflow epochs with statistically significant differences in streamflow predictability at the Lees Ferry Gauge using tree ring reconstructions dating back to the 1400s. Here we expand on this work to explore spatial patterns and controls on predictability across the Upper Colorado River Basin (CRB). We use nonlinear embedding to unfold the system’s dynamics through the construction of a phase space. Lyaponav exponents that measure the rate of divergence of trajectories in the phase space are used to identify periods of high predictability. We evaluate the sensitivity of the resulting streamflow predictability epochs to spatial location and length of record within the basin. First, we compare the derived predictability from different tree ring reconstructions at the Lees Ferry Gauge, in order to understand the sensitivity of the behavior to tree ring reconstructions (both length and source). The Lees Ferry Gauge is heavily studied and there are four different tree ring reconstructions available which range in starting dates from 762, 1116, 1416, and 1490. Second, we compare how predictability varies across gauges in the Upper CRB. Tree ring reconstructions are available for over 30 gauges in Upper Colorado ranging in start date from 762 to 1615. Differences in the identified high predictability epochs highlight the sensitivity of predictability to locations within the basin and the period of record available for study.