Streamlining a National Flood Assessment using a Flow Scheduler

Flooding is one of the costliest natural hazards in Aotearoa and climate change is increasing the frequency and severity of floods. We need a national-scale flood hazard and risk assessment under the current and future climate to plan for this. The Endeavour project “Mā te haumaru ō nga puna wai ō Rākaihautū ka ora mo ake tonu: Increasing flood resilience across Aotearoa” (http://niwa.co.nz/flood-resilience) aims to produce a consistent and automated method to generate flood maps nationwide. These flood maps will then be used for risk assessment, inform on societal vulnerability and identify solutions to reduce and adapt to flood risk.

A cylc workflow (https://cylc.github.io/) was built to generate these flood maps. Cylc was initially designed for weather forecasting and scheduling jobs by job dependencies and time increments. In this workflow, cylc is principally used as a central tool to link the different models and loop through all the computational domains (based on flood plains and their catchments) and scenarios. As the flood modelling needs to integrate climate science, rainfall statistics, hydrology, hydrodynamics, and geospatial data for each domain, a modular workflow was built. Each of these main tasks spins up a cylc sub-workflow (see Figure). Each subworkflow can also be run independently for testing and validation. This allows the high-level workflow to be simple, accessible and usable by a larger group of researchers. Through this modularisation, each sub-workflow can be designed specifically for its scientific domain. They then communicate smoothly through the main workflow. Each workflow is subdivided into tasks that are submitted independently to the workload manager (Slurm on NeSI). This functionality creates flexibility and allows the appropriate machine, computational environment and computing resources to be allocated for each job. For example, this workflow uses GPUs and parallel CPUs, as well as different computing languages such as C++, CUDA, Fortran90, Python, R, Julia and batch. 

The workflow has been used for the generation of the first set of flood maps. It iterated over 248 computational domains. For this first test, two flood maps corresponding to different scenarios have been generated for each domain. This will be increased next run with the inclusion of climate projection scenarios. This set-up, version controlled using git and deployed on NeSI, provides a robust method for collaborative code development, reproducibility and flexibility.

The author(s) wish to acknowledge the use of New Zealand eScience Infrastructure (NeSI) high performance computing facilities, consulting support and/or training services as part of this research.

ABOUT THE AUTHOR

Alice Harang is a hydrodynamic modeler part of the Natural Hazard group at NIWA. She has experience in computational fluid dynamic with modelling of small-scale mixing mechanism to large scale coastal flows or inundation scenarios. She grew her experience in flood modelling particularly through the ENDEAVOUR project “Mā te haumaru ō nga puna wai: Increasing flood resilience across Aotearoa” where she co-developed a workflow to create flood maps automatically across all New Zealand. She also was central in the creation of flood maps following TC Gabrielle for the recovery process, working with local councils, consultants, and university students. Finally, she is co-developing the 2D numerical model BG_Flood, a shallow water solver, producing to model inundation from compound sources such as tsunami, rain, storm surge, or river discharge.

-----

For more information about the eResearch NZ / eRangahau Aotearoa conference, visit:
https://eresearchnz.co.nz/