Linking hydrological simulation models with workflow and optimisation software

Abstract

The literature is rich in discussion of methods, platforms and standards that hope to ease the process of integrating models in workflows or decision support systems. The reality for practitioners is that integrating models is still complex and time consuming in most cases. In addition, the literature contains few descriptions of the practicalities of integrating models in the hydrology domain. This paper presents two case studies to explore opportunities for reducing the complexity associated with model integration. Even though the case studies were different, significant commonalities exist in the linkage requirements for both case studies. The paper finds that meeting these requirements required significant change to the existing model architecture. The first case study involves linking the hydrological modeling system, Source IMS, with the Delft Flood Early Warning System (FEWS). The FEWS system is being trialed in the Australian Water Resources Assessment system (AWRA) that the Water Information Research and Development Alliance (WIRADA) is developing to support water assessment and water accounts products for the Bureau of Meteorology. The second case study involves linking the Source IMS with Parameter ESTimation (PEST) software for automated calibration of model parameters. These two case studies identified four key requirements that are considered common to how workflow tools would need to integrate with Source IMS. Specifically, 1. Run and retrieve results without a user's interaction. 2. Model independence - both FEWS and PEST remain independent of the model. 3. Manipulate parameters and boundary conditions; there was no need to modify the structure or configuration of the model. 4. Manage project load times and overall simulation times. The architecture of Source IMS changed significantly to support mechanisms for simulation stepping, distributed execution, stay-alive functionality and parameter passing. Adopting The Open Modelling Interface (OpenMI) would have addressed these requirements. However, the start up costs of implementing the OpenMI made this solution unachievable in the short term with the available resourcing. This paper discusses a pragmatic solution that addressed the key integration requirements while reducing the cost of moving to OpenMI in the future. The pragmatic solution involved changes such as:  Refactoring persistence functions (load and save), data model parameterisation and simulation methods so they were brokered through the application layer interfaces;  Decorating the application layer interfaces with Windows Communication Foundation (WCF) to enable inter-program, cross-process and cross-computer communication; and  Developing a command-line client and server application for providing application layer services to external programs. Practitioners and software developers solving model integration problems desire to leverage standards such as OpenMI. This paper contributes to our understanding of the precursor steps required to implement some of these standards. The precursor to integrating hydrological models in Source IMS was a reengineering of the software architecture. The reengineered architecture supported requirements of both case studies due to commonalities of PEST and FEWS. This paper encourages practitioners to consider major architectural decisions that might simplify or complicate the integration with other products.