HR: 0800h
AN: H51B-0347 [Abstracts]
TI: Input Uncertainty and its Implications on Parameter Assessment in Hydrologic and Hydroclimatic
Modelling Studies
AU: * Chowdhury, S
EM: shahadat@civeng.unsw.edu.au
AF: School of Civil Engineering, University of New South Wales, Sydney, NSW 2052
Australia
AU: Sharma, A
EM: a.sharma@unsw.edu.au
AF: School of Civil Engineering, University of New South Wales, Sydney, NSW 2052
Australia
AB:
Hydrological model inputs are often derived from measurements at point locations taken at discrete time steps. The nature of
uncertainty associated with such inputs is thus a function of the quality and number of measurements available in time. A
change in these characteristics (such as a change in the number of rain-gauge inputs used to derive spatially averaged
rainfall) results in inhomogeneity in the associated distributional profile. Ignoring such uncertainty can lead to models
that aim to simulate based on the observed input variable instead of the true measurement, resulting in a biased
representation of the underlying system dynamics as well as an increase in both bias and the predictive uncertainty in
simulations. This is especially true of cases where the nature of uncertainty likely in the future is significantly different
to that in the past. Possible examples include situations where the accuracy of the catchment averaged rainfall has
increased substantially due to an increase in the rain-gauge density, or accuracy of climatic observations (such as sea
surface temperatures) increased due to the use of more accurate remote sensing technologies.
We introduce here a method to ascertain the true value of parameters in the presence of additive uncertainty in model inputs.
This method, known as SIMulation EXtrapolation (SIMEX, [Cook, 1994]) operates on the basis of an empirical relationship
between parameters and the level of additive input noise (or uncertainty). The method starts with generating a series of
alternate realisations of model inputs by artificially adding white noise in increasing multiples of the known error
variance. The alternate realisations lead to alternate sets of parameters that are increasingly biased with respect to the
truth due to the increased variability in the inputs. Once several such realisations have been drawn, one is able to
formulate an empirical relationship between the parameter values and the level of additive noise present. SIMEX is based on
theory that the trend in alternate parameters can be extrapolated back to the notional error free zone. We illustrate the
utility of SIMEX in a synthetic rainfall-runoff modelling scenario and an application to study the dependence of uncertain
distributed sea surface temperature anomalies with an indicator of the El Nino Southern Oscillation, the Southern Oscillation
Index (SOI).
The errors in rainfall data and its affect is explored using Sacramento rainfall runoff model. The rainfall uncertainty is
assumed to be multiplicative and temporally invariant. The model used to relate the sea surface temperature anomalies (SSTA)
to the SOI is assumed to be of a linear form. The nature of uncertainty in the SSTA is additive and varies with time. The
SIMEX framework allows assessment of the relationship between the error free inputs and response.
Cook, J.R., Stefanski, L. A., Simulation-Extrapolation Estimation in Parametric Measurement Error Models, Journal of the
American Statistical Association, 89 (428), 1314-1328, 1994.
DE: 1805 Computational hydrology
DE: 1816 Estimation and forecasting
DE: 1846 Model calibration (3333)
DE: 1854 Precipitation (3354)
DE: 1873 Uncertainty assessment (3275)
SC: Hydrology [H]
MN: Fall Meeting 2005