PhD defence Mr Getachew Feleke Belete
Department of Geo-Information Processing
Title of defence
Integrating models on the web: application for socio-environmental studies
Most socio-environmental problems are wicked problems that are difficult to solve because of incomplete, contradictory, and changing requirements and information about the systems. Engaging stakeholders in the problem solving process is one essential way to address such problems. Modeling has been also recognized as an important tool for such problems. The challenge then is to make potentially quite complex and sophisticated models accessible for stakeholders. We see model integration as an option to explore the complexity continuum and improve our understanding. The assumption is that it makes more sense to use existing well developed and tested models as building blocks rather than build the whole system each time from scratch. This requires transparency and flexibility of the model integration process, allowing stakeholders to play a more significant role in deciding what modules are to be linked, how they should be treated, and what scenarios should be analyzed. By switching various modules on and off they can then also test the overall system sensitivity, both parametrically and structurally. We find that one way to remove technical and accessibility constraints and facilitate stakeholder participation is by presenting the models as web applications, making them available through standard web browsers. We performed this research with the objective to develop a methodology and software design that enables to transform independently developed models into web-based interoperable components and further link them into integrated models that can represent complex socio-environmental systems. The research was conducted in five blocks that are described as follows:
(1) Develop a methodology that facilitates the model integration process from requirements identification to system testing.
We did a literature review to improve our understanding of the model integration process, and to design better strategies of integrated modeling. Integration of models is an iterative and incremental process that requires higher level of software engineering and semantic processing than conventional modeling. We classified the model integration processes into five phases: pre-integration assessment, preparation of models for integration, orchestration of models during simulation, data interoperability, and testing. We highlighted key strategies, features, standards, and practices that can be employed during integration of models across disciplines, and suggested techniques that can improve discovery, reusability and ease of use of integrated modeling frameworks.
(2) Design a web based model integration framework that links independently developed multidisciplinary models into an integrated system.
Simulating complex socio-environmental systems may require linking models from different disciplines and developed using different assumptions, semantics, and tools. The models can also be located remotely and can be hosted on different platforms. We developed the design and prototype of the Distributed Model Integration Framework (DMIF) that can address these requirements by using distributed computing and service-oriented software development approaches. We used web services to wrap legacy models so that they can be accessed with a regular web browser and to convert them into interoperable components. In this approach the users do not need to install any additional software and can explore modules as stand-alone components, or build chains of modules by connecting outputs from one module to inputs of other modules. Data sets and user scenarios can be also packaged as web services and made available for integration. We also investigated runtime integration of models, i.e. an integration method in which users can access and integrate models 'on the fly' using a graphical user interface. Prototypes of generic interfaces for runtime access and integration of web service based models are developed.
(3) Method to reconcile implicit semantic relations in integration of models.
Integration of models requires establishing data exchange between participating component models. Semantic mediation is a mechanism to ensure the consistency of exchanged data by checking that the contextual meaning is understood, correctly mapped and, if necessary, translated. In this research we developed semantic matching algorithms for input-output text data and attribute names of models and also a dynamic unit conversion feature. We used freely available lexical database called WordNet and ontology called QUDT as a backend for some of the semantic mediation functions. The inclusion of the lexical database enabled the semantic matching algorithm to consider hierarchical semantic relationships between such concepts as synonym, hypernym, and hyponym. Similarly, the ontology provided dynamic unit conversion between SI, Derived, SI-Derived and None-SI units. The semantic matching algorithms were used in a case study of matching text-based data and also for searching of components within the Community of Surface Dynamics Modelers (CSDMS) model repository. The results indicated that the algorithms can effectively automate some of the semantic mediation tasks in integrating models. However, users can interactively improve results by setting the margins for ‘valid’ matching outputs.
(4) Sensitivity analysis to investigate how integration results are affected by time steps, numeric integration methods, and functional responses assumed in the component models.
To understand how integration arrangements can impact the overall results we used a demo case study where we split the classical predator-prey model into two separate component models. Then, we integrated the two components and conducted multiple runs for different time steps, numeric integration methods, and mathematical expressions used in components. The results indicated that integration of models with different time steps can be (1) highly sensitive to the size of the time steps chosen; (2) quite sensitive to the choice of the component where the bigger time step is assumed, and (3) relatively less sensitive to the difference between the time steps in component models. We then used different combinations of Euler and Runge-Kutta numeric integration methods in components. As expected, usage of Runge-Kutta method in even one of the coupled models can improve overall accuracy, but usage of different methods in component models is not symmetrical. We also found that the results are asymmetrical to the type of trophic function assumed in component models. Overall, sensitivity analysis is clearly very essential for our understanding of the overall system performance.
(5) A case study of integrated modeling of complex scenarios, such as climate change mitigation actions.
The case study was performed in the context of the EU FP7 project COMPLEX. The high-level system requirement for integrated modeling was to simulate the effect of United Nations Framework Convention on Climate Change (UNFCCC) policy scenarios in different sectors of the economy. We did pre-integration assessment and we identified that an integrated climate-energy-economy system can be constructed by linking three models from the COMPLEX project model repository: the Global Change Assessment Model - GCAM, a Computable General Equilibrium economic model – EXIOMOD, and an agent-based energy market model - NIROO. The models were developed using C++, GAMS, and NetLogo programming. We converted them into interoperable components by wrapping them with web services. By using the integrated system we simulated two selected scenarios. The first one is the business as usual scenario that simulates the situation when there are no climate change policy interventions, and the second one is a policy-based scenario in which UNFCCC policy targets set by different regions and countries are fed into the integrated system. The results indicate that by linking the three models we were able to simulate the scenarios in a more detailed way, which could be hardly accomplished using stand-alone model components.
Getachew Feleke Belete was born on 4 July 1976 in Bahir Dar, Ethiopia. He attended primary education in King Sertse Dengel elementary school and junior high school in Fasilo junior secondary school. Then, he attended high school in Tana Haike Comprehensive High School and he received Ethiopian School Leaving Certificate Examination (ESLCE) in 1993. From 1993 – 1997 he studied his bachelor’s degree at Addis Ababa University Faculty of Science and he graduated with BSc degree in Physics. In 2002 he joined HiLCoE School of Computer Science and Technology and in 2004 he received Postgraduate Diploma in Computer Science. From 2006-2008 he did his masters study at Addis Ababa University Department of Computer Science and he graduated with MSc degree in Computer Science.
Getachew started his career by teaching Physics in high school and then in a college. In 1999 while he was teaching Physics in Defence Engineering College he took some courses on computer application programs, which latter resulted in changing his profession to computing. He has worked as software developer for Ministry of Defence, Addis Ababa University, and United Bank. In 2012 he got short term research opportunity at Aarhus University in Denmark and he conducted research on Domain-Specific Programming Languages. In March 2013, he joined University of Twente, Faculty of ITC to pursue his PhD. His research was financially supported by European Union FP7 project called COMPLEX and the focus of the research was on integration of various simulation models. During his PhD, he has presented the research findings in various international conferences and in project workshops. He also published his research findings in highly reputable journals.
|Event starts:||Thursday 23 February 2017 at 12:30|
|Venue:||UT Waaier 4|
|City where event takes place:||Enschede|