News & Views
Get More Out of Water Resources Engineering Models
by Venu Kandiah, PhD, P.E.
Senior Water Resources Engineer
The effort required to develop water resources engineering models can be significant. But once developed, in many instances the models are only evaluated for a limited, or a single set of conditions. The results and predictions obtained from these evaluations are the ones used in design, presented in reports and conveyed to decision makers and the public. This approach might serve the immediate needs of a project; however, for most modeling efforts a limited amount of additional effort can provide more knowledge and understanding at a low marginal cost to that required to develop the model itself.
Advances in the hydrological sciences and in the areas of climate change, which are providing new insight into the foundational data used in the models, and advances in computational modeling and computational resources that increase overall reliability and capacity of modeling methods provide additional reasons to get more out of models. In many instances, for decision makers and the general public, a single outcome is often all that is associated with the overall modeling effort. Modelers should aim to deliver more, and in the process can increase the value and insight that modeling brings to the decision making processes and engineering project, and increase the profile of the modeling practice in general.
The traditional approach of reporting model outcomes based on a limited set of runs need to be reconsidered for a number of reasons. For example, flood extents might be evaluated using a “design storm”, or stormwater facilities sized using a continuous simulation covering a long term period for which rainfall records and other records needed for model development and calibration are available. Reported results are often assumed to be accurate, and fail to properly convey the uncertainties associated with them. Beyond the limitations of the model itself, there is inherent uncertainty in the parameters and boundary conditions used: inaccuracies exist in model data inputs.
Hydrological models used in the engineering community are underpinned by the idea of stationarity, the concept that hydrological systems operate within a window that can be established from past records. Research indicates that hydrological trends are changing. Climate change, and changes associated with development, are impacting rainfall, evapotranspiration and stream flows. Means and other statistical measures for these parameters, obtained from historical records, and assumed to be good descriptors for these parameters might no longer be valid in light of observed increased variability and extremes in hydrological events.
To ensure that users of model results are fully aware of the limitations in model results, modelers should attempt to better convey the variability and possible wide range of outcomes given uncertainties. This can be achieved by running a large number of model evaluations, and conducting more rigorous sensitivity analysis exercises. In the past limitations in computational power and computing acted as a constraint and provided a justification to limit model runs. Advances in computing (better numerical methods and algorithms), desktop computer technology (increased processing power and speeds, the advent of multi-core machines and graphical processing units) and computational models allow a greater number and more accurate models to be run in shorter amounts of time by modelers . Multiple evaluations can be done, ranging from a few scenarios using scenario-builder tools, to running an ensemble of models to numerous evaluations for parameter estimation.
Mauger G.S., et al., “State of Knowledge: Climate Change in Puget Sound”, Report prepared for the Puget Sound Partnership and the National Oceanic and Atmospheric Administration, Climate Impacts Group, University of Washington (2015).
Milly P.C.D., et al., “Stationarity Is Dead: Whither Water Management?”, Science, Vol 319 (2008).
Peel M.C., and Bloschl, G., “Hydrological modeling in a changing world.”, Progress in Physical Geography, Vol 35(2), 2011.