Premium
Multiobjective sensitivity analysis to understand the information content in streamflow observations for distributed watershed modeling
Author(s) -
Wagener Thorsten,
van Werkhoven Kathryn,
Reed Patrick,
Tang Yong
Publication year - 2009
Publication title -
water resources research
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.863
H-Index - 217
eISSN - 1944-7973
pISSN - 0043-1397
DOI - 10.1029/2008wr007347
Subject(s) - sensitivity (control systems) , hydrograph , watershed , streamflow , calibration , computer science , function (biology) , environmental science , hydrological modelling , surface runoff , event (particle physics) , hydrology (agriculture) , data mining , statistics , mathematics , machine learning , geography , geology , ecology , climatology , cartography , drainage basin , geotechnical engineering , electronic engineering , engineering , physics , quantum mechanics , evolutionary biology , biology
In a previous paper, van Werkhoven et al. (2008b) demonstrated that the information content of streamflow observations at a watershed outlet is a dynamic entity and is dependent on the spatiotemporal dynamics of the causal precipitation event. This result has important consequences for distributed hydrological model calibration strategies and for the design of observation networks. However, the conclusions drawn were based only on the analysis of the model parameter sensitivities to the hydrograph peak fit because of the use of the root‐mean‐square error objective function. An unanswered question is how will the previous result change if alternative objective functions are used? Here we extend the earlier analysis by adding low‐flow and water balance objective functions. We study their impact on how much information can be extracted during calibration overall and for specific model components (parameters) using a synthetic rainfall‐runoff event. Results suggest that both vertical (within a model cell) and spatial (across cells) sensitivities vary greatly with the objective function used. Timing‐related objective functions show sensitivity largely focused on the area close to the outlet, while a volume‐based objective function shows sensitivity distributed more evenly across the watershed. These results demonstrate the importance of using multiple evaluation metrics when assessing distributed model predictions. The resultant multiobjective sensitivity maps provide helpful tools for assessing the actual information provided by gauges in observation networks and motivate the need for a new generation of dynamic calibration strategies that would consider how the spatial parameter controls on the model response of interest vary in time.