Research Software Engineering for Climate Science Applications

Research Software Engineering for Climate Science Applications Jack Franklin, Tom Meltzer, Dominic Orchard, Nirav Vasant Shah, Joe Wallwork Institute of Computing for Climate Science, University of Cambridge, United Kingdom [email protected] Overview Climate change is one of the most serious crises faced by the world today. The understanding of climate science is critical to analyse the impact of climate change. In this context, research software engineering for climate science aims to empower researchers with software tools which are based on the principles of software sustainability and reproducible science. We discuss three applications related to climate science. - Image segmentation for Methane Plume detection using Satellite data - Domain decomposition for the development of a Discontinuous Galerkin method-based sea-ice model, NeXtSIM-DG, as part of Scale-Aware Sea Ice Project (SASIP) - Software modifications and Software engineering for evolving PlankTOM framework as part of TIMBER project Keywords: Climate science, Scientific software development, Software sustainability, Image segmentation, Sea-ice model, Biogeochemical model. Image segmentation using Zooniverse A B NeXtSIM-DG: next generation sea-ice model with a Discontinuous Galerkin (DG) method PlankTOM: a global marine biogeochemicalmodel Research software engineering contributions Acknowledgements While, researchers primarily aim to verify software implementation using established benchmark cases, research software engineering complements their efforts by adding features related to scientific results’ reproducibility and software sustainability. Some of the important aims of software engineering tasks include: - Facilitating wider usage by scientific community - Effective parallelization techniques - Implementing code maintenance practices - Improvement of software architecture - Ensuring consistent runs across different environments The objective of this work was to identify the methane plumes from satellite image data. Currently various ML tools and algorithms exist based on own proprietary training datasets. Besides, the problem of identifying methane plumes from satellite image is an image segmentation problem. In this context, it is important to develop a benchmark dataset for Machine Learning (ML) model training and evaluation. An important part of the creation of benchmark dataset is hand labelling of satellite images by experts. This effort requires creation of tools and software pipeline which supports the labelling by multiple users. To this end, Zooniverse [1], a people powered research platform, to support projects involving many volunteers. A feasibility study was performed to assess the suitability of Zooniverse functionalities for the image segmentation task described above. This involved: - Marking specific areas in the image - Labelling images by volunteers or subject experts - Creation of project in Zooniverse - Exporting data from Zooniverse - Importing data in a ML framework References: 1. Zooniverse: https://www.zooniverse.org We would like to acknowledge the financial support of Schmidt Sciences, LLC. . The core-idea behind NeXtSIM-DG [1, 2] is to explore the role of sea-ice mechanics and dynamics in determining the large-scale behaviour of Arctic sea-ice. NeXtSIM-DG uses Discontinuous Galerkin (DG) method for accurately capturing discontinuities such as fractures in sea-ice. . Sea-ice thickness calculations using NeXtSIM-DG The design of NeXtSIM-DG framework is modular and flexible for easy coupling with other models and easy addition of different routines. We develop Domain Decomposition tool [3] to split the domain into partitions for MPI parallelism using Recursive Coordinate Bisection (RCB) geometric partitioning algorithm [4]. Specifically, Domain Decomposition tool generates the decomposition metadata before running NeXtSIM-DG. Domain decomposition Halo exchange logic References: 1. NeXtSIM-DG repository: https://github.com/nextsimhub/nextsimdg 2. T. Richter, V. Dansereau, C. Lessig, and P. Minakowski. “A dynamical core based on a discontinuous Galerkin method for higher-order finite-element sea ice modelling.” Geoscientific Model Development, Volume 16 (2023) 3907-3926. 3. Domain decomposition in NeXtSIM-DG: https://github.com/nextsimhub/domain_decomp 4. M. Berger and S. Bokhari. "A partitioning strategy for nonuniform problems on multiprocessors.“ IEEE Trans. Computers, C-36 (1987) 570-580. We consider the modifications in the design of Planktom framework. Planktom model [1] is a global marine biogeochemical model based on the representation of twelve Plankton Functional Types: six phytoplankton, five zooplankton and bacteria and full cycles of Carbon, Oxygen, Phosphorus, Silicon and simplified cycles of Iron and Nitrogen. PlankTOM uses time dependent Advection-Diffusion Equation and is based on Nucleus for European Modelling of the Ocean (NEMO) [2], a numerical ocean model. Scope of TIMBER project. (Adapted from [3]) The Planktom framework requires modifications based on the principles of swappable module requirements, software sustainability for wider community usage, software maintainability and continuous development for integrating new routines. Specifically, the modifications include -Containerisation: For running simulations on different High Performance Computing (HPC) systems -Code refactoring: For easier maintenance and elimination of duplicate code -Documentation: For improving understanding and usage of software -Software design: Defining clear interfaces and low coupling between modules to allow replaceability of modules for specific applications References: 1. Description of the PlankTOM12.2 equations: https://greenoceandata.uea.ac.uk/model/PlankTOM_Manual.pdf 2. NEMO: a numerical ocean model: https://www.metoffice.gov.uk/blog/2024/nemo-anumerical-ocean-model 3. C. Heinze, T. Blenckner, H. Martins, D. Rusiecka, R. Döscher, M. Gehlen, N. Gruber, E. Holland, Ø. Hov, F. Joos, J.B.R. Matthews, R. Rødven, and S. Wilson. “The quiet crossing of ocean tipping points.” Proc. Natl. Acad. Sci., Volume 118 (2021).