H21K-01 INVITED
Reconciling Theory With Observations: Elements of A Diagnostic Approach To Model Evaluation
As the science of hydrology progresses, we must reconcile increasingly larger (more complex/detailed) models with larger and more information rich data sets. The classical approach, based in Bayesian and Likelihood theories, can work well for simple models, but must be refined and expanded to respond to this pressing challenge. Simply put, we need a robust and powerful Theory of Evaluation. This talk will argue that the way forward lies in developing improved (context relevant) methods for extracting information from data … and will require the active collaboration of process scientists, modelers and systems theorists. Such collaboration will enable us to build a "Diagnostic Approach" to model evaluation (one that helps in diagnosing and correcting model deficiencies), rooted in information theory and employing the notion of "Signature Indices" to measure theoretically relevant system process behaviors. An important aspect of the signature-based approach is that it addresses the issue of how much of the observable system complexity is resolvable by a specific model. The approach can easily be placed in the context of Bayesian inference to facilitate uncertainty analysis. Further, it can be readily applied to the problem of improving predictions in ungaged basins.
H21K-02 INVITED
New Approaches to Modelling an Ungauged Basin
In this paper, we address a number of typical problems in ungauged basins: first, currently available rainfall sources from satellites seem promising, but are biased and noisy; second, streamflow records, if available, often have no overlap with modern rainfall products and are unreliable; third, the question is how to incorporate innovative satellite-based data sources (for instance SEBAL evaporation maps) in model calibration? Experiences are presented from a truly ungauged sub-catchment of the Zambezi, the Luangwa river basin (~150 000 km2) in Zambia. The assumption is that all data, whether noisy, non-concomitant or biased, contains information that may constrain our model solution space. A set of old discharge records revealed: 1) a clear recession curve, 2) an estimate of the long-term water balance, and 3) the auto- regressive properties of the streamflow process. This information has been mapped into a number of innovative maximum likelihood estimators that are introduced step-wise in a Bayesian updating procedure for the development of a conceptual hydrological model. As a last step, the updated parameter distribution is introduced in a classical automatic Pareto optimization, where land-surface related parameters are distributed, justified by evaporation constraints based on the SEBAL evaporation maps. This information is included in the form of fuzzy measures for parameter acceptability. The results show that it is possible to constrain the parameter space considerably based on the combined information. It is shown that the likelihood estimators are not very sensitive to the quality of the rainfall product used for constraining model parameters, as long as the autoregressive properties of the rainfall are present in the time series. The use of the evaporation maps in the optimization process considerably reduces equifinality, preventing parameters to compensate for each other.
H21K-03 INVITED
Combining PUB and climate change scenarios for long-term water resource modeling in South Africa
The lack of historical observations of streamflow for watershed model calibration is a particular problem for many less developed countries where the resilience of society against disturbances of the hydrological cycle is very low and the value of reliable hydrological predictions is very high. However, gauging networks continue to decline in many of these countries and predictions in ungauged basins (PUB) are a necessary tool to develop strategies to achieve water security for people in such regions. In this talk, we focus on the Olifants basin in South Africa, and combine downscaled GCM climate change scenarios with PUB to assess the potential long-term change of the water resources in this region. We particularly focus on providing ensemble predictions of continuous streamflow in the current main source regions for freshwater, which are mainly ungauged.
H21K-04 INVITED
A national hydrological model for New Zealand
New Zealand is a fascinating laboratory for hydrological research. The land area of New Zealand is relatively small (269,000 km2), but within this area there are large differences in precipitation (300 to 12,000 mm/year), vegetation (rainforest, grassland, and desert), and geology (sandstone, pumice, and limestone). Snow can be an important component of the hydrological budget in the Southern Alps, and streamflow in many parts of New Zealand is affected by natural and managed lakes. River forms vary from steep mountain torrents to wide, braided, gravel beds. There are increasing demands for the available water resources and increasing vulnerability to floods, and a national hydrological model is needed for both water resource assessments and flood forecasting. This presentation discusses the use of research conducted as part of the Problem of Ungauged Basins (PUB) initiative to build a national hydrological model for New Zealand. The research has two main steps: (1) evaluate model simulations in experimental watersheds and use internal catchment observations of soil moisture, groundwater levels, and streamflow to identify appropriate model structure(s) and model parameters; and (2) evaluate the spatial patterns of nationwide model simulations and use hydrological classification systems to understand spatial differences in model performance. Nationwide hydrological datasets and modeling systems are already developed in New Zealand, and we invite the community to use this "virtual laboratory" for their own research.
H21K-05 INVITED
Progress with the PUB Initiative in Canada
Practicing hydrologists continually face the challenge of prediction in ungauged basins. They are well aware of the difficulties and risks inherent in making predictions and forecasts of the state of water resources. They are cognizant of the climate and landscape change that is forcing our community to address some of the long held assumptions in our methodologies – notably that of stationarity. Furthermore, as resources have become scarcer due to availability or quality limitations, decision makers' demands not only include reports of mere abundance or state but also change. Interactions among hydrological, biochemical and ecological processes now need to be understood and incorporated into new predictive tools. In Canada, progress has been slow but steady. Priorities were identified, including improving prediction in small basins, incorporating process algorithms into deterministic models, implementing new information generating methods, and expanding outreach of new knowledge and techniques. Individual successes are reflective of the needs of each segment of the community. Large utilities and operational forecast offices, with their larger infrastructure, have made progress incorporating new algorithms into deterministic models and applying advanced regionalization tools. The majority of consulting engineers remain constrained by time, budgets and access to data. They remain comfortable reducing uncertainty and building confidence with calibration and reproduction of past conditions. Conservative assumptions are a mainstay for reducing risk. Progress in reducing uncertainty for this segment is made by developing relationships and exchanging information so that practicing hydrologists are aware of the new tools and knowledge they need to ensure wise water management decisions.
H21K-06
Challenges of Predicting the Hydrologic Response of Ungauged Arctic Catchments
Few streams are gauged in the high latitudes and often those streams are gauged under unfavorable conditions in this harsh environment. This, coupled with a changing climate makes it even more challenging to use regional historical data to predict flow in ungauged basins. While it is anticipated that precipitation will increase in a warming scenario, this has not been universally found to be the case. Increasing variability in precipitation patterns and extreme events further increases the difficulty in using relatively short-term data to predict future conditions. Because of the long winters, the snowmelt runoff is guaranteed to be a significant event; however, in the Alaskan Arctic summer precipitation events have produced runoff peaks that are four times greater than maximum snowmelt peak events for some smaller headwater basins. Presently it is difficult to get hydrologic models to capture the wide range of responses that exist from droughts to floods for gauged basins. The need for good predictive models for ungauged basins in light of rapid resource development in this region of the world is crucial for good water-resource management. These models will need to deal more directly with uncertainty and risk evaluation, both due to the limited data and increasing variability in changing climate conditions. The methods to apply data and hydrologic characteristics from regional "index" watersheds will also be critical in improving the methods of simulating flow in ungauged basins.
H21K-07
The use of Hydrologic Regionalisation Techniques in the Australian Sustainable Yields Projects
The Australian Sustainable Yields Projects are currently being carried out by CSIRO. The aim of these
projects is to produce transparent, consistent and robust information on current and future water availability
in order to properly manage and share limited water resources in four regions across Australia: the Murray-
Darling Basin, Tasmania, Southwest Western Australia, and Northern Australia.
Deriving these water availabilities is made more difficult by the fact that most catchments in the study regions
are ungauged. As a result, a technique needs be derived to predict the hydrologic response of these
ungauged catchments. For the Australian Sustainable Yields Projects, a decision was made to calibrate
rainfall-runoff models to gauged catchments and predict the hydrologic response of nearby ungauged
catchments through hydrologic regionalisation techniques.
For the now completed Murray-Darling Basin Sustainable Yields Project, four rainfall-runoff models were
applied to 183 gauged catchments within the Murray-Darling Basin. These catchments cover less than 10
percent of the entire area of the Murray-Darling Basin. The hydrologic response of the remaining 90 percent
of the Basin therefore needed to be determined using hydrologic regionalisation techniques.
The rainfall-runoff models used were SIMHYD, Sacramento, IHACRES_Classic, and SMAR_G. Because of
the problems inherent in transferring individual parameter values from gauged to ungauged catchments
based on relationships with catchment attributes, the decision was made to transfer entire models from
gauged to ungauged catchments. A number of techniques for transferring models were investigated,
including the use of catchment similarity measures. However, the best results were obtained by simply
applying the model from the nearest gauged catchment to an ungauged catchment (as assessed by treating
each gauged catchment as ungauged in turn, predicting its hydrologic response and then comparing this to
the observed hydrologic response).
The results indicate that runoff in ungauged catchments can be estimated reasonably reliably using rainfall-
runoff models from a nearby gauged catchment. The Nash-Sutcliffe efficiency of monthly runoff for
simulations using optimised parameter values from a nearby gauged catchment is on average about 0.1
lower than that in the calibration results. The errors in mean annual runoff for ungauged catchments are less
than 50 percent in most of the catchments and less than 20 percent in more than half of the catchments.
http://www.csiro.au/partnerships/SYP.html
H21K-08
Use Of MODIS Leaf Area Index To Improve Rainfall-Runoff Modelling In Ungauged Catchments
This paper demonstrates the feasibility of using MODIS-LAI (MODerate resolution Imaging Spectrometer mounted on the polar-orbiting Terra satellite - leaf area index) and catchment evapotranspiration estimated using MODIS-LAI (ERS) to improve the modelling of runoff in ungauged catchments. Two applications of MODIS-LAI with the SIMHYD lumped conceptual daily rainfall-runoff model were explored. In the first application, SIMHYD was calibrated against both the observed streamflow and ERS. In the second application, SIMHYD was modified to use MODIS-LAI data directly. Data from 2001 to 2005 from 120 catchments in south-east Australia were used for the study, where runoff for each catchment was estimated using the optimized parameter values from the geographically closest calibration catchment. The results indicate that the SIMHYD calibration against both the observed streamflow and ERS produced better simulations of daily and monthly runoff compared to the SIMHYD calibration against only the observed streamflow data, despite the modelling results being assessed solely against the observed streamflow data. The runoff simulations were even better for the modified SIMHYD model, particularly in the simulation of daily runoff. The use of MODIS-LAI or other remotely sensed data directly is interesting because it accounts directly for the time varying changes in vegetation properties. It is likely that the use of other remotely sensed data (like soil moisture) and smarter modification of rainfall-runoff models to use remotely sensed data directly can further improve the prediction of runoff in ungauged catchments.