Probabilistic approach to modeling under changing scenarios

Abstract

The complexity of the hydrologic system challenges the development of models. One issue faced at model development stage is the uncertainty involved when calibrating and validating the model. Model inputs and parameters can introduce large amount of uncertainties that can be propagated non-linearly to the model outputs. Additionally, several sets of parameters may also exist that acceptably represent the system (i.e., equifinality). As a result, converting model outputs into important environmental decisions become challenging. The main objective of this study was to define a framework that facilitates model development while evaluating uncertainty to assess the impacts of land management practices at watershed scale. A two-step probabilistic approach to model calibration and parametrization was implemented using global uncertainty and sensitivity analysis. The Agricultural Policy/Environmental eXtender (APEX) model was Winter Wheat) and two alternative land management scenarios (conventional tillage for grain, and conventional tillage for graze out) were proposed and simulated over the study area to observe their effects on the monthly N loads. Results suggested that changes in land use and land iii management did not affect the total water yield at watershed scale. However, the N loads showed a high variability ranging from 0.1 to 1.2 kg/ha during summer and fall seasons, with uncertainty of up to 77%. The 100% Pasture scenario was the most effective alternative in reducing nitrogen, the N loads in this scenario did not exceed 0.9 kg/ha in any season. This methodology demonstrated that the modeling process, coupled with the evaluation of uncertainty and equifinality, facilitate the adjustment of input/parameters and quantify the uncertainties in the model outputs. By considering all possible parameter combinations that represent the response of the system, the most likely ranges of hydrologic outcomes can be established under changing scenarios while accounting for the associated uncertainty.iv ACKNOWLEDGEMENTS