H53I-01 INVITED
The Value of Tracer Data in Catchment Modeling and Process Representation
The ability to make water quality predictions from hydrologic models has been challenging given that models must incorporate information about flowpaths, storages, and fluxes of water and solute. Many field studies are limited by observation capacity of subsurface processes in heterogeneous and dynamic systems, and resulting hydrological descriptions of catchments are often not easily translated into model structure. For water quality models, additional information is required to describe solute retention and transport processes, which increases the number of model parameters causing problems associated with calibration and parameter identification. Tracer methods have become increasingly common in hydrological studies since they provide spatially integrated information about catchment processes such as water flowpaths and sources. An important measure that tracers can provide information on is the water transit time through a catchment, which is relevant to water quality since the contact time in the subsurface largely controls stream water chemical composition. We will explore how tracers, such as stable isotopes of water, can be used to provide process insights, reductions in parameter uncertainty, and enhancement of model structure.
H53I-02
Understanding Flow Pathways, Mixing and Transit Times for Water Quality Modelling
Water quality modelling requires representation of the physical processes controlling the movement of solutes and particulates at an appropriate level of detail to address the objective of the model simulations. To understand and develop mitigation strategies for diffuse pollution at catchment scales, it is necessary for models to be able to represent the sources and age of water reaching rivers at different times. Experimental and modelling studies undertaken on several catchments in the north east of Scotland have used natural hydrochemical and isotopic tracers as a means of obtaining spatially integrated information about mixing processes. Methods for obtaining and integrating appropriate data are considered together with the implications of neglecting it. The tracer data have been incorporated in a conceptual hydrological model to study the sensitivity of the modelled tracer response to factors that may not affect runoff simulations but do affect mixing and transit times of the water. Results from the studies have shown how model structural and parameter uncertainties can lead to errors in the representation of: the flow pathways of water; the degree to which these flow pathways have mixed and the length of time for which water has been stored within the soil / groundwater system. It has been found to be difficult to eliminate structural uncertainty regarding the mechanisms of mixing, and parameter uncertainty regarding the role of groundwater. Simulations of nitrate pollution, resulting from the application of agricultural fertilisers, have been undertaken to demonstrate the sensitivity of water quality simulations to the potential errors in physical transport mechanisms, inherent in models that fail to account correctly for flow pathways, mixing and transit times.
H53I-03
Flow Data for Solute Transport Modeling from Tracer Experiments in a Stream Not Continuously Gaining Water
In-stream tracer experiments are a well-established method for determining flow data to be incorporated in solute transport modeling. For a gaining stream, this method is implemented to provide spatial flow data at scales of minutes and tens of meters without physical disturbance to the flow of water, the streambed, or biota. Of importance for solute transport modeling, solute inflow loading along the stream can be estimated with this spatial data. The tracer information can also be interpreted to characterize hyporheic exchange time-scales for a stream with hyporheic exchange flowpaths (HEFs) that are short relative to the distance over which the stream gains water. The interpretation of tracer data becomes uncertain for a stream that is not gaining water continuously over intended study reach. We demonstrate, with straight-forward mass-balances, uncertainties for solute loading which arise in the analysis of streams locally losing water while predominantly gaining water (and solutes) over a larger scale. With field data from Mineral Creek (Silverton, Colorado) we illustrate the further uncertainty distinguishing HEFs from (locally) losing segments of the stream. Comparison of bromide tracer with ambient sulfate concentrations suggests that subsurface inflows and outflows, concurrent with likely HEFs, occur in a hydrogeochemical setting of multiple, dispersed and mixed, sources of water along a 64 m sub-reach of the predominately gaining, but locally losing, stream. To compute stream-reach mass-balances (the simplest of water quality models) there is a need to quantitatively define the character and source of contaminants entering streams from ground-water pathways, as well as the potential for changes in water chemistry and contaminant concentrations along flow paths crossing the sediment-water interface. Identification of inflow solute mass requires quantifying water gain, loss, and hyporheic exchange in addition to concentration. http://smig.usgs.gov/SMIG/transtor_reader2.html
H53I-04 INVITED
Uncertainty Assessment for Surface Water Quality Models: The Challenge of Sparse Data
Water quality models are often used to aid stakeholders in making critical decisions about how to improve water quality. These decisions occur against a backdrop of process complexity and data scarcity. Basin scale water quality models generally simplify spatial complexity to some degree. This problem is especially relevant in water quality modeling since the sources of pollution as well as the hydrologic drivers vary spatially across the landscape. The spatial complexity problem also presents specific challenges for estimating model predictive uncertainty. Questions of how to integrate multiple sources of stream water and water quality data at multiple locations are likely to be even more daunting than they are for surface water hydrologic models. At this time, given the complexity of water quality models and the sparse data availability, true statistical techniques of integrating, multiple data sources and calculating uncertainty bounds are not reasonable approaches. For this reason, we have proposed uncertainty estimation methods not based on statistics but instead ones that fall in a class of methods that could be called "fit-to-purpose" methods. Within SWAT2005, both a statistical method for uncertainty analysis, ParaSol, and an evaluation method, SUNGLASSES, have been incorporated. The focus of calibration lies on capturing times series variability whereas the evaluation and uncertainty assessment of the model relies on predictions of sediment mass flux- the purpose motivating model development. These methods have been applied in the context of a water quality problem in the San Jacinto watershed in southern California. The analysis includes an extension of the uncertainty analysis into its economic implications. The economic impact of predictive uncertainty was compared to traditional margin of safety approaches. The economic implications of improved uncertainty assessment were shown to be most important under circumstances where water quality targets were close to being met. http://hwr.arizona.edu/tmeixner
H53I-05
Uncertainty Assessment in Watershed-Scale Water Quality Modeling and Management
Watershed-scale water quality models such as SWAT, WARMF or HSPF are widely used to support management decision-making. However, the uncertainty in model output is hard to determine explicitly. An uncertainty analysis is necessary to develop a safety factor, or in regulatory terms a Margin of Safety (MOS) for the Total Maximum Daily Load (TMDL). We have developed a framework to systematically assess the uncertainty in complex models, with a particular emphasis on watershed models used for decision-support. A key component of the framework is the Management Objectives Constrained Analysis of Uncertainty (MOCAU) method, which explicitly considers management objectives and observational uncertainty within the analysis. In this case study, we used the WARMF model and a specific catchment in an actual watershed (Santa Clara River) to demonstrate the applicability of the approach. A series of numerical experiments were conducted to investigate the performance of MOCAU. Although this method relies on a Monte Carlo approach, the use of management criteria, such as Non Attainment Frequency, Severity of Exceedance, Timing of Exceedance, are used to better constrain the uncertainty analysis. In addition to determining the necessary information for the MOS, a key result is the ability to design monitoring programs that use resources much more efficiently, by directing them towards those conditions that are most likely to reduce the uncertainty of management/regulatory decisions.
H53I-06
Propagating Water Quality Analysis Uncertainty Into Resource Management Decisions Through Probabilistic Modeling
Most probable number (MPN) and colony-forming-unit (CFU) are two estimates of fecal coliform bacteria concentration commonly used as measures of water quality in United States shellfish harvesting waters. The MPN is the maximum likelihood estimate (or MLE) of the true fecal coliform concentration based on counts of non-sterile tubes in serial dilution of a sample aliquot, indicating bacterial metabolic activity. The CFU is the MLE of the true fecal coliform concentration based on the number of bacteria colonies emerging on a growth plate after inoculation from a sample aliquot. Each estimating procedure has intrinsic variability and is subject to additional uncertainty arising from minor variations in experimental protocol. Several versions of each procedure (using different sized aliquots or different numbers of tubes, for example) are in common use, each with its own levels of probabilistic and experimental error and uncertainty. It has been observed empirically that the MPN procedure is more variable than the CFU procedure, and that MPN estimates are somewhat higher on average than CFU estimates, on split samples from the same water bodies. We construct a probabilistic model that provides a clear theoretical explanation for the observed variability in, and discrepancy between, MPN and CFU measurements. We then explore how this variability and uncertainty might propagate into shellfish harvesting area management decisions through a two-phased modeling strategy. First, we apply our probabilistic model in a simulation-based analysis of future water quality standard violation frequencies under alternative land use scenarios, such as those evaluated under guidelines of the total maximum daily load (TMDL) program. Second, we apply our model to water quality data from shellfish harvesting areas which at present are closed (either conditionally or permanently) to shellfishing, to determine if alternative laboratory analysis procedures might have led to different management decisions. Our research results indicate that the (often large) observed differences between MPN and CFU values for the same water body are well within the ranges predicted by our probabilistic model. Our research also indicates that the probability of violating current water quality guidelines at specified true fecal coliform concentrations depends on the laboratory procedure used. As a result, quality-based management decisions, such as opening or closing a shellfishing area, may also depend on the laboratory procedure used.
H53I-07 INVITED
How can models better aid the decision processes for the management of water resources?
Despite the long history of development and use of models in policy making, there is still a poor integration between modeling and the decision process. This is influenced by several factors, such as a lack of understanding of models from policy makers, limitations in the applicability and use of models, modeller's behaviour, and a lack of stakeholder involvement in the whole modelling process. But, most relevant of all, is the fact that models are perceived as non reliable tools for policy making, due to the uncertainty associated with them and the lack of confidence this uncertainty generates. Commonly, in fields like natural resource management, and in particular water management, models are build to predict, in space or time, the state of the system to be managed (e.g. real time flood forecasting). These models are then used by policy makers as a surrogate of a real system to inform their decisions. In this view, the efficacy of a model depends on how well it can approximate reality and how much confidence policy makers can have on model's results. However, even though predictive models can be used to convey scientific argumentation that can aid decision making, these models fall short in supporting the processes of negotiation, learning, and communication, which constitute the basis for policy making. In this presentation we explore and discuss the relationship between models used in water resource management and policy making: how models can be used and how uncertainty should be treated depending on the purpose models are supposed to serve. We identify four major modeling purposes that are important for understanding and managing complex environmental systems: prediction, exploratory analysis, communication and learning, and investigate the implications of the different purposes in dealing with uncertainty. In predictive models, the presence of uncertainty is understood as a critical constraint for decision making, and as such it ought to be eliminated or reduced as much as possible. On the other hand, in models for learning the presence of uncertainty become useful to identify the commonalities and differences in views, and can highlight point of conflicts, opening room for discussion and space for negotiation among different interest parties. Using these concepts, we present a set of strategies that can guide the development and use of models in support of the policy making process.