How Do Discrete Global Grid Systems Actually Perform? A Systematic Benchmark Across Geometry, Computation and Relational Joins

Levente Juhász 

As geospatial datasets exceed the billion-row threshold, Discrete Global Grid Systems (DGGS) promise to replace expensive vector spatial joins with fast relational hash-joins on discrete cell identifiers. However, the real-world performance of different grid implementations and the upfront cost of converting vector geometries into grid indexes remains largely unquantified. This paper introduces dggs-bench, an opensource benchmarking framework that standardizes spatial operations across disparate grid libraries, and demonstrates it by evaluating four DGGS (H3, S2, rHEALPix, ISEA4H) and two legacy planar grids (XYZ Tiles, Geohash) across geometric fidelity, topological resilience, computational throughput, and relational performance experiments. Pre-computed DGGS equi-joins universally outperform ST_Intersects baselines by 13×–457×, with all grids completing ten-million-point joins in under 200 ms on a consumer workstation. On-the-fly polygon covering is not competitive at macro scales for any grid, though break-even query thresholds drop dramatically at coarser resolutions while maintaining ≥ 99% spatial accuracy. ISEA4H achieves competitive join performance and strict equal-area compliance but incurs covering costs up to 154× higher than H3, exposing a tooling maturity gap rather than a geometric limitation. Across all experiments, computational performance rank runs in partial reverse of geometric correctness. The benchmarking framework and all experimental data are openly available.