![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Extreme river flow dependence in Northern Scotland |
VerfasserIn |
M. Franco Villoria, M. Scott, T. Hoey, D. Fischbacher-Smith |
Konferenz |
EGU General Assembly 2012
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 14 (2012) |
Datensatznummer |
250066140
|
|
|
|
Zusammenfassung |
Various methods for the spatial analysis of hydrologic data have been developed recently.
Here we present results using the conditional probability approach proposed by Keef et al.
[Appl. Stat. (2009): 58,601-18] to investigate spatial interdependence in extreme river flows
in Scotland. This approach does not require the specification of a correlation function, being
mostly suitable for relatively small geographical areas. The work is motivated by
the Flood Risk Management Act (Scotland (2009)) which requires maps of flood
risk that take account of spatial dependence in extreme river flow. The method
is based on two conditional measures of spatial flood risk: firstly the conditional
probability PC(p) that a set of sites Y = (Y 1,...,Y d) within a region C of interest
exceed a flow threshold Qp at time t (or any lag of t), given that in the specified
conditioning site X > Qp; and, secondly the expected number of sites within C that will
exceed a flow Qp on average (given that X > Qp). The conditional probabilities are
estimated using the conditional distribution of Y |X = x (for large x), which can be
modeled using a semi-parametric approach (Heffernan and Tawn [Roy. Statist.
Soc. Ser. B (2004): 66,497-546]). Once the model is fitted, pseudo-samples can be
generated to estimate functionals of the joint tails of the distribution of (Y,X).
Conditional return level plots were directly compared to traditional return level plots thus
improving our understanding of the dependence structure of extreme river flow
events. Confidence intervals were calculated using block bootstrapping methods (100
replicates).
We report results from applying this approach to a set of four rivers (Dulnain, Lossie,
Ewe and Ness) in Northern Scotland. These sites were chosen based on data quality, spatial
location and catchment characteristics. The river Ness, being the largest (catchment size
1839.1km2) was chosen as the conditioning river. Both the Ewe (441.1km2) and Ness
catchments have predominantly impermeable bedrock, with the Ewe’s one being very wet.
The Lossie(216km2) and Dulnain (272.2km2) both contain significant areas of
glacial deposits. River flow in the Dulnain is usually affected by snowmelt. In all
cases, the conditional probability of each of the three rivers (Dulnain, Lossie, Ewe)
decreases as the event in the conditioning river (Ness) becomes more extreme. The
Ewe, despite being the furthest of the three sites from the Ness shows the strongest
dependence, with relatively high (>0.4) conditional probabilities even for very extreme
events (>0.995). Although the Lossie is closer geographically to the Ness than
the Ewe, it shows relatively low conditional probabilities and can be considered
independent of the Ness for very extreme events (> 0.990). The conditional probabilities
seem to reflect the different catchment characteristics and dominant precipitation
generating events, with the Ewe being more similar to the Ness than the other two
rivers. This interpretation suggests that the conditional method may yield improved
estimates of extreme events, but the approach is time consuming. An alternative
model that is easier to implement, using a spatial quantile regression, is currently
being investigated, which would also allow the introduction of further covariates,
essential as the effects of climate change are incorporated into estimation procedures. |
|
|
|
|
|