Assessment of hydrodynamic simulation results for eco-hydraulic and eco-hydrological applications: a spatial semivariance approach

摘要

Output from a three-dimensional numerical flow model (SS1IM) is used in conjunction with high-resolution topographic and velocity data to assess such models for eco-hydraulic applications in river channel design and habitat appraisal. A new methodology for the comparison between field measurement and model output is detailed. This involves a comparison between conventional goodness-of-fit approaches applied to a spatially structured (riffle and pool) sample of model and field data, and a 'relaxation' method based upon the spatial semivariance of model/field departures. Conventional assessment indicates that the model predicts point-by-point velocity characteristics on a 0.45 m mesh to within ±0.1 m s~(-1) over 80% of the channel area at low flow, and 50% of the area at high in-bank flow. When a relative criterion of model fit is used, however, the model appears to perform less well: 60-70% of channel area has predicted velocities that depart from observed velocities by more than 10%. Regression analysis of observed and predicted velocities gives more cause for optimism, but all of these conventional indicators of goodness of fit neglect important spatial characteristics of model performance. Spatial semivariance is a means of supplementing model appraisal in this respect. In particular, using the relaxation approach, results are greatly improved: at a high in-bank flow, the model results match field measurements to within 0.1 m s~(-1) for more than 95% of the total channel area, provided that model and field comparisons are allowed within a radius of approximately 1 m from the original point of measurement. It is suggested that this revised form of model assessment is of particular relevance to eco-hydraulic applications, where some degree of spatial and temporal dynamism (or uncertainty) is a characteristic. The approach may also be generalized to other environmental science modelling applications where the spatial attributes of model fits are of interest.