Conceptualization and implementation of a global drought monitoring and forecasting system within the HydroSOS framework

Forecasting systems focusing on upcoming flood and drought events are essential to support various aspects such as disaster risk reduction, climate change mitigation, or long-term policy and planning. In particular, multiple model-based early warning systems have been developed to allow the simulation of future floods and droughts at different temporal-spatial scales. However, despite the successful development of many innovative and state-of-the-art modeling systems in the academic fields, their transition into an operational system is challenging, and it may take several years to set up appropriate technical requirements, especially into a new IT infrastructure. In this talk, we hence outline these challenges for the example of the ongoing project OUTLAST (operational, multi-sectoral global drought hazard forecasting system), where the main goal is to develop a modeling system that is ready for operational use. OUTLAST will provide model-based near real-time monitoring using recent updated ERA5 climate data and seasonal forecasting of drought globally across different sectors (water supply, riverine and non-agricultural land ecosystems, rainfed and irrigated agriculture). The system consists of a model chain of three models: (1) bias correction of global seasonal forecasting products SEAS5, (2) the global hydrological model WaterGAP, and (3) the global crop water model GCWM. The drought status in both monitoring and forecasting phase from OUTLAST will be provided globally for the next six months and be freely accessible via the HydroSOS portal, a Hydrological Status and Outlook System hosted by the World Meteorological Organization (WMO).

Highlights of OUTLAST are the ability to run the whole system within a cloud-ready automated workflow to ensure seamless integration into the HydroSOS framework. This includes the so-called “trigger” to automatically download the newly released climate data (ERA5 and SEAS5) from the source (ECWMF). To achieve this goal, each model and its dependencies in the model chain in OUTLAST are encapsulated in a "container" by the core developer in the research institution before being transferred to run in an IT infrastructure at an external government institution. The containers will then be orchestrated to enable the upscaling of the system based on computational requirements and the availability of hardware resources. This approach aims to (i) enable a seamless transition of OUTLAST into operation, (ii) avoid any conflict with the host operating system, and (iii) ensure a fast boot system in case one of the servers fails. We hope that the proposed infrastructure design can serve as a blueprint for other efforts to transfer scientific workflows into an operational environment.