One of the key features of Delft-FEWS is the ability to integrate external models from third party suppliers via the GeneralAdapter concept. Generally, an additional model adapter facilitates the communication between Delft-FEWS and the model by converting data formats to and from the native model format.  Running the model adapter introduces a dependency, can impact the performance, and increases the workflow runtime. 

Eight Canadian agencies use the Raven Hydrological Modelling Framework in Delft-FEWS for flood and reservoir inflow forecasting. The extremely short runtimes of this semi-distributed model are especially useful when making ensemble forecasts and in allowing interaction with the forecaster. Use of the Raven framework in Delft-FEWS was previously enabled by a Python-based model adapter, run from the GeneralAdapter. 

Supported by a strong community of Delft-FEWS users that use Raven in their forecasting systems, the Raven lead developer, Prof. James Craig from the University of Waterloo, recently removed the need for a separate model adapter. The new functionality allows Raven models to directly read a Delft-FEWS runinfo file, provide state and parameter updating, and link model forcings directly to Delft-FEWS inputs.

A beta version of the new Raven software was published this month. Initial tests show a major performance increase, with workflows that run up to 25x faster. Additional benefits include:

• A more robust integration thanks to enabling id-mapping
• All state variables and model parameters are accessible
• Reduced dependencies
• A smaller and simpler Delft-FEWS configuration

With this improved integration of Raven models, Delft-FEWS users can generate more forecasts in less time and expand their forecasting systems with more robust functionality that is easier to maintain and expand to new watersheds.

The beta version is currently available to a small group of users. The new release will be available on the Raven website by or before January 2022.