 |
| Titel |
A Mulitivariate Statistical Model Describing the Compound Nature of Soil Moisture Drought |
| VerfasserIn |
Colin Manning, Martin Widmann, Emanuele Bevacqua, Douglas Maraun, Anne van Loon, Mathieu Vrac |
| Konferenz |
EGU General Assembly 2017
|
| Medientyp |
Artikel
|
| Sprache |
en
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 19 (2017) |
| Datensatznummer |
250151581
|
| Publikation (Nr.) |
 EGU/EGU2017-16217.pdf |
|
|
|
|
|
| Zusammenfassung |
| Soil moisture in Europe acts to partition incoming energy into sensible and
latent heat fluxes, thereby exerting a large influence on temperature
variability. Soil moisture is predominantly controlled by precipitation and
evapotranspiration. When these meteorological variables are accumulated over
different timescales, their joint multivariate distribution and dependence
structure can be used to provide information of soil moisture. We therefore
consider soil moisture drought as a compound event of meteorological drought
(deficits of precipitation) and heat waves, or more specifically, periods of
high Potential Evapotraspiration (PET). We present here a statistical model
of soil moisture based on Pair Copula Constructions (PCC) that can describe
the dependence amongst soil moisture and its contributing meteorological
variables.
\vskip 0.25cm
The model is designed in such a way that it can account for concurrences of
meteorological drought and heat waves and describe the dependence between
these conditions at a local level. The model is composed of four variables;
daily soil moisture ($h$); a short term and a long term accumulated precipitation variable ($Y_1$ and $Y_2$) that account for the propagation of meteorological drought to soil
moisture drought; and accumulated PET ($Y_3$), calculated using the Penman Monteith equation, which can represent the effect of a heat wave on soil conditions.
\vskip 0.25cm
Copula are multivariate distribution functions that allow one to model the
dependence structure of given variables separately from their marginal
behaviour. PCCs then allow in theory for the formulation of a multivariate
distribution of any dimension where the multivariate distribution is
decomposed into a product of marginal probability density functions and
two-dimensional copula, of which some are conditional. We apply PCC here in
such a way that allows us to provide estimates of $h$ and their uncertainty through conditioning on the $Y$ in the form
\vskip 0.25cm
\noindent\textbox{\hfill}\textbox{\hfil $h=h|y_1,y_2,y_3$\hfil}\textbox{\hfill (1)}
\vskip 0.25cm
Applying the model to various Fluxnet sites across Europe, we find the model
has good skill and can particularly capture periods of low soil moisture
well. We illustrate the relevance of the dependence structure of these $Y$ variables to soil moisture and show how it may be generalised to offer
information of soil moisture on a widespread scale where few observations of
soil moisture exist. We then present results from a validation study of a
selection of EURO CORDEX climate models where we demonstrate the skill of
these models in representing these dependencies and so offer insight into
the skill seen in the representation of soil moisture in these models.
\end{document} |
|
|
|
|
|
|