 |
| Titel |
A non-linear transfer function time series model consistent with the Data Based Mechanistic framework |
| VerfasserIn |
Paul Smith, Keith Beven  , Wlodek Tych |
| Konferenz |
EGU General Assembly 2010
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 12 (2010) |
| Datensatznummer |
250035793
|
|
|
|
|
|
| Zusammenfassung |
| In hydrology a discrete time series of observations (y1,...,yT) can often be effectively
modelled from an input series (u1,...,uT) lagged by Ï time steps using a transfer function
model of the form yt + -
i=1naiyt-i = -
j=0mbjut-Ï-j + η where the noise η is
represented by an ARMA process. In many cases the input series is constructed as a function
of multiple observed series e.g. ut = ytÏrt when constructing an effective rainfall (ut) from
observed rainfall (rt) and river discharge (yt). Functions for constructing the input series can
often be rewritten as functions for the state dependency of b = (b0,...,bm). Increasingly
there is evidence that state dependant formulation for a = (a1,...,an) many also be
required to adequately represent some systems. We outline a method for representing the state
dependency of (a,b) as polynomial functions of observed values. An appropriate robust
estimation algorithm using an instrument variable technique is presented. It is shown that the
estimation algorithm can be constrained to maintain system stability so allowing
interpretation within the Data Based Mechanistic framework of Young (Environmetrics 5,
1994). |
|
|
|
|
|
|