| Popis: |
The research addresses two aspects of uncertainty assessment in spatially distributed modeling: uncertainty analysis (UA), described as propagation of uncertainty from spatially distributed input factors on model outputs; and sensitivity analysis (SA), defined as assessment of relative importance of spatially distributed factors on the model output variance. An evaluation framework for spatially distributed models is proposed based on a combination of sequential simulation and the global, variance-based, Sobol SA method to quantify model output uncertainty together with the corresponding sensitivity measures. The framework is independent of model assumptions; it explores the whole space of input factors and provides measures of factors importance (first-order effects) and their interactions (higher-order effects). The Water Conservation Area 2A is a constructed subtropical wetland in the Florida Everglades that has been impacted by increased nutrient loads, hydrologic manipulation and incursions of invasive species. The topography of the area has been simulated at different resolutions considering regular and irregular grids for different data-density. The uncertainty increases from a threshold value of the fractal dimension from which the roughness of the surface is no longer preserved. The application of the proposed procedure allows for formal evaluation of model uncertainty which is essential if the value of the models used for estimation of TMDL is not to be undermined and if the main factors contributing to model uncertainty are to be identified. A hydro- logical model is used for the purpose of this illustration, but the procedure can be easily extended to spatially distributed water quality models. Furthermore, the procedure can evaluate propagation of uncertainty from alternative spatial representations of model input data or parameters (both continuous, like land elevation, and categorical, like land cover type). The proposed methodology has been applied for model quality control and for guiding field measurement campaigns by optimizing data collection in terms of cost-benefit analysis. |
