An integrated, largely non-probabilistic, calibration-free approach
is proposed to identify, estimate, evaluate and attribute conceptual
components of a streamflow time series. We assess its gross functional aggregation
from the signal structure alone by consistently exploiting elementary
constraints. Starting from the separability concept of linear operator theory,
cross-connections are revealed of such a blind functional streamflow disaggregation
(FSD) to qualitative dynamics. The algorithm is initialized by
a first guess of regular behaviour using Singular–system Analysis (SSA). To
approach the regular/singular borderline of the data and to separate a fast
flow from total runoff, this (probabilistic) SSA mode is transformed into a
lower envelope to the series via iterative cubic spline interpolation (CSI). Repeated
CSI yields a hierarchy of lower envelopes that piles up part of a transient
component and converges into a slow one. A lower bound is constructed
as an instantaneous low flow (ILF), using the leading SSA eigenvector. We
demonstrate the method for highlands river stations, compare its results with
those from distributed hydrologic models, and discuss attributions to overland,
inter- and base-flows. For independent evaluation we resort to singularity
based multifractal analyses.
|