 |
| Titel |
Optimizing weather radar observations using an adaptive multiquadric surface fitting algorithm |
| VerfasserIn |
Brecht Martens, Pieter Cabus, Inge De Jongh, Niko Verhoest |
| Konferenz |
EGU General Assembly 2013
|
| Medientyp |
Artikel
|
| Sprache |
Englisch
|
| Digitales Dokument |
PDF |
| Erschienen |
In: GRA - Volume 15 (2013) |
| Datensatznummer |
250073542
|
|
|
|
|
|
| Zusammenfassung |
| Abstract
Real time forecasting of river flow is an essential tool in operational water management. Such
real time modelling systems require well calibrated models which can make use of spatially
distributed rainfall observations. Weather radars provide spatial data, however, since radar
measurements are sensitive to a large range of error sources, often a discrepancy between
radar observations and ground-based measurements, which are mostly considered as
ground truth, can be observed. Through merging ground observations with the radar
product, often referred to as data merging, one may force the radar observations to
better correspond to the ground-based measurements, without losing the spatial
information.
In this paper, radar images and ground-based measurements of rainfall are merged based
on interpolated gauge-adjustment factors (Moore et al., 1998; Cole and Moore, 2008) or
scaling factors. Using the following equation, scaling factors (C(xα)) are calculated at each
position xα where a gauge measurement (Ig(xα)) is available:
Ig(xα)+-ε
C (xα) = Ir(xα)+ ε
(1)
where Ir(xα) is the radar-based observation in the pixel overlapping the rain gauge and ε
is a constant making sure the scaling factor can be calculated when Ir(xα) is zero. These
scaling factors are interpolated on the radar grid, resulting in a unique scaling factor for each
pixel. Multiquadric surface fitting is used as an interpolation algorithm (Hardy,
1971):
C*(x0) = aTv + a0
(2)
where C*(x0) is the prediction at location x0, the vector a (Nx1, with N the number of
ground-based measurements used) and the constant a0 parameters describing the surface and
v an Nx1 vector containing the (Euclidian) distance between each point xα used in the
interpolation and the point x0. The parameters describing the surface are derived by forcing
the surface to be an exact interpolator and impose that the sum of the parameters in a
should be zero. However, often, the surface is allowed to pass near the observations
(i.e. the observed scaling factors C(xα)) on a distance aαK by introducing an
offset parameter K, which results in slightly different equations to calculate a and
a0.
The described technique is currently being used by the Flemish Environmental Agency in
an online forecasting system of river discharges within Flanders (Belgium). However,
rescaling the radar data using the described algorithm is not always giving rise
to an improved weather radar product. Probably one of the main reasons is the
parameters K and ε which are implemented as constants. It can be expected that,
among others, depending on the characteristics of the rainfall, different values for the
parameters should be used. Adaptation of the parameter values is achieved by an online
calibration of K and ε at each time step (every 15 minutes), using validated rain gauge
measurements as ground truth. Results demonstrate that rescaling radar images using
optimized values for K and ε at each time step lead to a significant improvement of the
rainfall estimation, which in turn will result in higher quality discharge predictions.
Moreover, it is shown that calibrated values for K and ε can be obtained in near-real
time.
References
Cole, S. J., and Moore, R. J. (2008). Hydrological modelling using raingauge- and
radar-based estimators of areal rainfall. Journal of Hydrology, 358(3-4), 159-181.
Hardy, R.L., (1971) Multiquadric equations of topography and other irregular surfaces,
Journal of Geophysical Research, 76(8): 1905-1915.
Moore, R. J., Watson, B. C., Jones, D. A. and Black, K. B. (1989). London weather radar
local calibration study. Technical report, Institute of Hydrology. |
|
|
|
|
|
|