Integrating models on the web: applications for socio-environmental studies

Models are developed with the objective to understand a concept or system, to predict what the future states and trends may look like, and if possible to come up with appropriate management decisions and mitigation or adaptation strategies. Based on this, thousands of computer models are developed to explore certain aspects of the world. Complex problems like climate mitigation strategy analysis requires interdisciplinary assessment from many domains, including climate, hydrology, energy, economy, land use, behavioral sciences. The complex and interrelated nature of such real-world problems requires holistic system-of-systems thinking. However, it is not practical to construct a single model that could simulate such complex processes; integration of models and tools may be a solution. Besides, integration gives the chance of reusing existing models and it is also faster and less expensive than reengineering legacy systems.

So we need genuine techniques that enable us to integrate various models which are: (1) from different domain, (2) with different assumptions and semantics,  (3) they are likely to assume different spatial and temporal scale, (4) can be developed using different methodology, (5) tools and techniques, (6) can have different levels of complexity, and (7) they can be calibrated for different ranges of inputs and outputs, etc.  In this research we investigate how  technical, semantic, and dataset levels of interoperability can be addressed in linking range of heterogeneous models.

First we identified key design requirements for an integrated system that could link independently built heterogeneous models which could be hosted on different hardware and software platform.  Then we develop the design of the integrated system based on the principles of separation of concern and distributed computing. By using these principles we are able to treat  the different aspects of integration in a harmonious way.

To establish technical interoperability we used web services and service oriented architecture approach (SOA). For example, we wrap models developed using NetLogo, GAMS, and C++ as web services. And we used SOA to orchestrate the communication between these models. By using this approach we built an integrated system that link models that focus on climate, economy, energy, and households behavioral responses towards changes in the energy market. For example, by using this integrated system we are ables to simulate the effect of climate change policy scenarios on different sectors of the economy.

For semantic mediation of text-based input-output data and model attributes names we use Word Overlap or Token-base semantic matching algorithm. To consider conceptual relations between concepts we also used WordNet (https://wordnet.princeton.edu/) - a lexicon database that considers hierarchical semantic relations between concepts – in our semantic matching algorithm. For unit conversion in semantic mediation we used an openly available, QUDT ontology (http://www.qudt.org/ ).

We also investigated on how to provide generic user interface that enable to (1) run web service based models hosted anywhere on the internet, (2) to integrate web service based models on the graphical user interface level without the need for additional coding and debugging. The prototype of such interfaces are developed using our design approach – separation of concern & distributed computing.

We observed that technical integration of models can be provided in a generic way. However, providing a generic model integration framework is dependent on providing fully automated and intelligent semantic mediation. Standardization of model metadata, building semantic networks and ontologies on these standardized metadata, and employing artificial intelligence in semantic mediation could contribute in enabling fully automated semantic mediation.