![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Can streamflow data provide more certain TSS predictions? |
VerfasserIn |
Anna E. Sikorska, Andreas Scheidegger, Dario Del Giudice, Kazimierz Banasik, Jörg Rieckermann |
Konferenz |
EGU General Assembly 2014
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 16 (2014) |
Datensatznummer |
250086492
|
Publikation (Nr.) |
EGU/EGU2014-372.pdf |
|
|
|
Zusammenfassung |
Total suspended solids (TSS) is commonly accepted as a proxy of water quality due to
recognised impacts on receiving waters (physical, biological, ecological and ecotoxic).
Therefore, the accurate prediction of TSS loads and concentrations is important to assess the
risk of hazardous conditions in a stream, e.g. due to land-use changes or pollution
management strategies.
Unfortunately, TSS predictions are rather uncertain. First, erosion and sediment wash-off is
a complex interplay of many processes, which are not well identifiable. Second,
monitoring TSS is challenging. Thus, calibration data for TSS models are often
inaccurate and relatively short and do not cover the whole variability of TSS in a stream.
Consequently, although TSS models can reproduce observed events, their capacity to
predict unobserved events is usually very low. Third, when the focus is on TSS
concentrations, streamflow predictions are needed. This requires a hydrological
model, which introduces additional uncertainty. Interestingly, this uncertainty is often
overlooked.
A common approach to model TSS concentrations relies on a physically-based sediment
build-up/wash-off model (BWM), which has an integrated hydrological component with
precipitation as an input. Because streamflow is here an intermediate state, streamflow data
are not directly required for TSS calibration. However, it remains unexplored how
better i.e.more frequent or more accurate streamflow observations improve TSS
predictions.
In this work we therefore investigate the value of using streamflow data to better calibrate
TSS models. Specifically, we use two methods. First, we calibrate a TSS model, which
considers streamflow as an internal state, only on observed TSS concentrations. Second, we
use additional streamflow observations. To reliably calibrate the model, we adopt a
multi-objective Bayesian calibration approach, which considers input errors, model structure
deficits and observation errors as a sum of independent random noise and autocorrelated
error process (bias) (Reichert and Schuwirth, 2012). This seems very promising, because the
first term accounts for stochasticity, such as measurement error of TSS concentrations and
random effects in sediment production, whereas the second captures the remaining inability
to reproduce observed patterns. In addition, we also assess the predictive capability of the
models, analyse how the additional information of the streamflow data influence
different properties of both error terms and their contribution to the total predictive
uncertainty.
We illustrate the approach on a small catchment in Warsaw, Poland, where we monitored
precipitation, TSS concentrations, water levels and velocities during a dedicated field
campaign. To these data, we calibrated a hydrological model (HyMod) and a conceptual
BWM for TSS concentrations.
Our results show that i) using additionally streamflow data to calibrate the TSS model
substantially improves TSS predictions (assessed by data coverage and uncertainty band
sharpness), and ii) most of the prediction uncertainty is due to systematic errors and not
random noises. Furthermore, by formally describing systematic errors we are able to provide
more reliable uncertainty estimates than before. These findings are relevant for investigating
the frequency of exceeding hazardous thresholds for TSS concentrations in receiving
waters.
References:
P. Reichert and N. Schuwirth. 2012. Linking statistical bias description to multiobjective
model calibration. Water Resources Research, 48, W09543. |
|
|
|
|
|