Evaluation of preCICE (version 3.3.0) in an Earth System Model Regridding Benchmark

In Earth System Modeling (ESM), meshes of different models usually do not match, requiring data mapping algorithms implemented in coupling software. Valcke et al. recently introduced a benchmark to evaluate such algorithms and compared implementations in four specialized ESM couplers. In this paper, we assess preCICE, a general-purpose coupling library not limited to ESM, using this benchmark and compare our results to the original study. The generality of preCICE with its larger community offers potential benefits to ESM applications, but the software naturally lacks ESM-specific solutions. We describe necessary pre- and postprocessing steps to make the benchmark tangible for preCICE. Overall, preCICE achieves comparable results; using its radial basis function mapping yields significantly lower errors.

💡 Research Summary

The paper evaluates the performance of preCICE (version 3.3.0), a general‑purpose coupling library, on the Earth System Model (ESM) re‑gridding benchmark introduced by Valcke et al. (2022). In ESM, atmospheric and oceanic components typically use distinct meshes, which necessitates data mapping (re‑gridding) algorithms to exchange scalar or vector fields. The benchmark originally compared four specialized ESM couplers—SCRIP, YAC, ESMF, and XIOS—each offering nearest‑neighbour, first‑order, and second‑order mapping methods in both “conservative” and “non‑conservative” variants (according to the ESM definition of conservation).

The authors first formalise the mapping problem as a linear operation R = M Ψ_S, where M ∈ ℝ^{N_T×N_S} is the mapping matrix, and discuss the distinction between consistent (row‑sum = 1) and conservative (column‑sum = 1) mappings. preCICE implements three core mapping strategies: nearest‑neighbour (NN), nearest‑projection (NP), and a family of radial‑basis‑function (RBF) methods based on a partition‑of‑unity (POU) approach. The RBF method partitions the source mesh into small clusters, builds a local RBF interpolant for each cluster by solving a modest linear system, and then blends the interpolants for target points. This high‑order, mesh‑agnostic technique is the key novelty examined in the study.

Because preCICE treats meshes as pure point clouds, the authors had to devise a six‑step pre‑ and post‑processing pipeline. First, the original NetCDF meshes (provided in latitude/longitude) are converted to 3‑D Cartesian coordinates using the PROJ library, with duplicate points merged at a tolerance of 10⁻⁸. Ocean masks, originally stored as cell data, are averaged onto nodes, then Delaunay‑triangulated to generate the surface elements required by NP. For atmospheric meshes that contain hexagons and pentagons (which preCICE’s NP does not support directly), the authors employ a scaled‑consistent NP mapping as a surrogate for the conservative mask mapping used in the reference study. All meshes and associated data are written to VTK UnstructuredGrid files, which serve as the input format for preCICE’s artificial‑solver testing environment (ASTE).

The benchmark evaluates four synthetic test functions (sinusoid, harmonic, vortex, and Gulf‑Stream‑like) on several atmosphere–ocean mesh pairs (e.g., “torc” vs. “bggd”, “nogt” vs. “icos”). Accuracy metrics include mean, maximum, and RMS relative misfit, as well as L_min, L_max, and global conservation errors for integrals. Although the detailed numerical tables are not reproduced in the excerpt, the abstract and discussion state that preCICE achieves results comparable to the specialized couplers, and that the RBF mapping produces “significantly lower errors” across the test cases.

Beyond raw accuracy, the paper highlights preCICE’s broader ecosystem: a black‑box coupling philosophy, peer‑to‑peer MPI/TCP communication without a central server, support for explicit and implicit coupling schemes, and acceleration techniques (Aitken under‑relaxation, quasi‑Newton). These features make preCICE attractive for high‑performance computing environments and for coupling codes from disparate scientific domains (e.g., OpenFOAM, FEniCS). However, the study also notes limitations: the need for Cartesian coordinates, extra preprocessing for complex cell shapes, and the absence of ESM‑specific optimisations such as built‑in conservative cell‑based mappings.

In conclusion, the authors demonstrate that a general‑purpose library like preCICE can be successfully applied to a demanding ESM re‑gridding benchmark. The RBF mapping, in particular, offers superior accuracy without sacrificing the flexibility and sustainability provided by a large, active developer community. The work suggests that ESM practitioners could benefit from adopting preCICE, gaining easier integration with other simulation tools, better documentation, and long‑term software support, while still meeting the stringent accuracy requirements of climate modeling.