Model-data fusion: Using observations to understand and reduce uncertainty in hydrological models

A. I.J.M. Van Dijk*

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

5 Citations (Scopus)

Abstract

Without corroborating observations, hydrological models should not be trusted. Hydrometric observations are the basis for any statistical or empirical hydrological model, and are widely used for 'calibration' - tweaking model parameter values to reduce the disagreement between model and observations. However there are other, less travelled paths to understand, quantify, and reduce hydrological model uncertainty using observations. A wealth of new observation types has become available from in situ sensor networks, airborne data collection and satellite remote sensing. With conventional hydrometric observation networks in decline, using these new observations is not merely of scientific interest but a bare necessity. This paper discusses opportunities and challenges to expand the application of model-data fusion techniques - using observations to make models behave better. Assessing the outcome of future boundary conditions (predictions or hypothetical scenarios) on water resources remains the most common hydrological model application. New observations can be used to improve model inputs (e.g. spatial parameter maps) or be considered in calibration to derive a more robust parameter set. Confronting the model with new observations helps to assess model uncertainty and build our confidence in it, though not usually without introducing some tough questions. (Is past performance actually relevant to uncertainty in future predictions? What about unknown or unquantifiable uncertainty components; are partial uncertainty estimates better than none? How does one determine whether a model is too poor to suit its intended purpose; what is the alternative? What is the uncertainty in the observations?) When looking into the past rather than the future, some of these questions disappear, while new ones emerge. Estimating past (and current) conditions is becoming increasingly important: as the basis of monitoring systems (of drought, flood, water resource generation, availability and use); to initialize forecasting systems; and to inform water resource accounting and situation analysis. The Australian Water Resources Assessment system (AWRA) is such an application, developed to support the Bureau of Meteorology's water information services. Having a predictive model is no longer an objective in this case. Rather, models take on a subservient role to reconcile and interpolate the available observations. In this case, observations are still used before the model is run (in model configuration and calibration), but now can also be applied during the model run (via data assimilation) and/or afterwards (e.g. via bias correction). Available techniques vary from the mathematically near-trivial to the very complicated, but assumptions about error in both model and observations are always critical. The generation of large volumes of relevant observations and derived products (e.g. from time series remote sensing) grows opportunities for model-data fusion, but equally grows the conceptual, mathematical and computational challenges. Several examples are discussed.

Original languageEnglish
Title of host publicationMODSIM 2011 - 19th International Congress on Modelling and Simulation - Sustaining Our Future
Subtitle of host publicationUnderstanding and Living with Uncertainty
Pages140-147
Number of pages8
Publication statusPublished - 2011
Externally publishedYes
Event19th International Congress on Modelling and Simulation - Sustaining Our Future: Understanding and Living with Uncertainty, MODSIM2011 - Perth, WA, Australia
Duration: 12 Dec 201116 Dec 2011

Publication series

NameMODSIM 2011 - 19th International Congress on Modelling and Simulation - Sustaining Our Future: Understanding and Living with Uncertainty

Conference

Conference19th International Congress on Modelling and Simulation - Sustaining Our Future: Understanding and Living with Uncertainty, MODSIM2011
Country/TerritoryAustralia
CityPerth, WA
Period12/12/1116/12/11

Fingerprint

Dive into the research topics of 'Model-data fusion: Using observations to understand and reduce uncertainty in hydrological models'. Together they form a unique fingerprint.

Cite this