STREAM-FLOW CHARACTERIZATION AND FEATURE DETECTION USING A DISCRETE WAVELET TRANSFORM

An exploration of the wavelet transform as applied to daily river disc harge records demonstrates its strong potential for quantifying stream flow variability. Both periodic and non-periodic features are detecte d equally, and their locations in time preserved. Wavelet scalograms o ften reveal structures that are obscure in raw discharge data. Integra tion of transform magnitude vectors over time yields wavelet spectra t hat reflect the characteristic time-scales of a river's flow, which in turn are controlled by the hydroclimatic regime. For example, snowmel t rivers in Colorado possess maximum wavelet spectral energy at time-s cales on the order of 4 months owing to sustained high summer flows; H awaiian streams display high energies at time-scales of a few days, re flecting the domination of brief rainstorm events. Wavelet spectral an alyses of daily discharge records for 91 rivers in the US and on tropi cal islands indicate that this is a simple and robust way to character ize stream flow variability. Wavelet spectral shape is controlled by t he distribution of event time-scales, which in turn reflects the timin g, variability and often the mechanism of water delivery to the river. Five hydroclimatic regions, listed here in order of decreasing season ality and increasing pulsatory nature, are described from the wavelet spectral analysis: (a) western snowmelt, (b) north-eastern snowmelt, ( c) mid-central humid, (d) southwestern arid and (e) 'rainstorm island' . Spectral shape is qualitatively diagnostic for three of these region s. While more work is needed to establish the use of wavelets for hydr ograph analysis, our results suggest that river flows may be effective ly classified into distinct hydroclimatic categories using this approa ch. (C) 1998 John Wiley & Sons, Ltd.