Recent Developments and Applications of The Ostrich Calibration Toolkit

Calibration is a critical step in developing a successful hydrologic model when observation data is available adjust physical process representation. One of the largest sources of uncertainty in a hydrologic model comes from error in estimating parameters that are input into the model. The model’s input parameter set is adjusted during calibration to reduce or minimize the model error with respect to the observed data. Manual calibration is a time-consuming and laborious task in which parameters are adjusted to improve model performance, and a new model simulation is initiated with the adjusted parameters. This process is repeated many times until the model achieves sufficient performance metric values, and the model is deemed successful. The manual calibration process can take days or weeks, depending on the number of parameters and the complexity of the model. In addition to the substantial time investment, manual calibration likely does not result in parameters sets that realize the full optimization of model results and is unrealistic for many of the distributed parameter hydrologic models used today.

Many parameter optimization tools already exist that address the issues with manual calibration. Optimization algorithms can be scripted into a programming language software and run automatically to find an optimal parameter set. The tools can be customized for each hydrologic model by making modifications to the tool to read the different input or outputs specific to that model. In addition to optimizing parameter sets, the tools can be used to provide parameter uncertainty estimates. This is especially useful when working in a probabilistic framework, which is often required to quantify a system’s risk and reliability. Implementation of optimization techniques save substantial time and budget from hydrologic model development.

This work highlights recent improvements to and applications of the Optimization Software Toolkit for Research Involving Computational Heuristics (OSTRICH) toolkit. Maintained as a collaboration between the Bureau of Reclamation (Reclamation), University of Waterloo, and University of Buffalo, OSTRICH is intended to be an approachable, flexible, and scalable optimization toolkit. OSTRICH utilizes an input structure that allows rapid development and debugging of optimization workflows. Additionally, OSTRICH provides multiple optimization algorithms to match the objective function topology as well as for rapid transition among the algorithms. Several recent deployments of OSTRICH within Reclamation will be highlighted as use cases.