H13R-06
eWaterCycle: Recent progress in a global operational hydrological forecasting model

Monday, 14 December 2015: 14:55
3011 (Moscone West)
Nick Van De Giesen, Delft University of Technology, Faculty of Civil Engineering and Geosciences, Delft, Netherlands, Edwin Sutanudjaja, Utrecht University, Utrecht, Netherlands, Marc FP Bierkens, Utrecht University, Department of Physcial Geography, Utrecht, Netherlands, Niels Drost, Netherlands eScience Center, Amsterdam, Netherlands and Rolf Hut, Delft University of Technology, Delft, Netherlands
Abstract:
Earlier this year, the eWaterCycle project launched its operational forecasting system (forecast.ewatercycle.org). The forecasts are ensemble based, and cover fourteen days. Near-real-time satellite data on soil moisture are assimilated in the forecasts. Presently, the model runs with a spatial resolution of 10km x 10km, and the plan is to move to 1km x 1km in the near future. The eWaterCycle forecast systems runs on a combination of a supercomputer and a cloud platform. Interactive visualization allows users to zoom in on any area of interest and select different variables. The project builds on close cooperation between hydrologists and computer scientists. What makes eWaterCycle relatively unique is that it was built with existing software, which is largely open source and uses existing standards. The Basic Model Interface (BMI) of the Community Surface Dynamics Modeling System (CSDMS) is an important tool that connects different modules. This allows for easy change and exchange of modules within the project. Only a few parts of the software needed to be re-engineerd for allowing it to run smoothly in a High-Performance Computing environment.

After a general introduction to the modeling framework, the presentation will focus on recent advances, especially with respect to quality control of runoff predictions. Different parts of the world show different predictive error. As the model does not use explicit calibration procedures, it is of interest to see where the model performs well and where it performs not so well. The next natural question is then why this is the case and how to move forward without ending up with ad hoc improvement measures.