Hydrological models as web services: Experiences from the Environmental Virtual Observatory project

Data availability in environmental sciences is expanding at a rapid pace. From the constant stream of high-resolution satellite images to the local efforts of citizen scientists, there is an increasing need to process the growing stream of heterogeneous data and turn it into useful information for decision-making. Environmental models, ranging from simple rainfall - runoff relations to complex climate models, can be very useful tools to process data, identify patterns, and help predict the potential impact of management scenarios. Recent technological innovations in networking, computing and standardization may bring a new generation of interactive models plugged into virtual environments closer to the end-user. They are the driver of major funding initiatives such as the UK's Virtual Observatory program, and the U.S. National Science Foundation's Earth Cube. In this study we explore how hydrological models, being an important subset of environmental models, have to be adapted in order to function within a broader environment of web-services and user interactions. Historically, hydrological models have been developed for very different purposes. Typically they have a rigid model structure, requiring a very specific set of input data and parameters. As such, the process of implementing a model for a specific catchment requires careful collection and preparation of the input data, extensive calibration and subsequent validation. This procedure seems incompatible with a web-environment, where data availability is highly variable, heterogeneous and constantly changing in time, and where the requirements of end-users may be not necessarily align with the original intention of the model developer. We present prototypes of models that are web-enabled using the web standards of the Open Geospatial Consortium, and implemented in online decision-support systems. We identify issues related to (1) optimal use of available data; (2) the need for flexible and adaptive structures; (3) quantification and communication of uncertainties. Lastly, we present some road maps to address these issues and discuss them in the broader context of web-based data processing and "big data" science.