The current graph construction assumes the input data lies on a regular rectilinear grid. With the mesh_layout work in #78 and CRS detection in #75, the infrastructure is becoming more flexible. However, there's no support yet for cases where the input data itself is irregular — e.g., sparse observational networks (ship reports, weather stations) or data on rotated/curvilinear grids where the point distribution is non-uniform.
This would involve:
- Detecting the spatial structure of input coordinates (regular, curvilinear, sparse)
- Generating graphs that adapt to the data geometry (e.g., Delaunay-based for irregular grids, density-aware k-NN for sparse data)
- Building multi-scale hierarchies from these adaptive base graphs
This relates to GSoC 2026 Project 1 (Flexible Graph Construction) — see GSoC ideas. I'm preparing a proposal for this and would appreciate any feedback on the approach from maintainers.
Related: #78 , #75 , #40
The current graph construction assumes the input data lies on a regular rectilinear grid. With the
mesh_layoutwork in #78 and CRS detection in #75, the infrastructure is becoming more flexible. However, there's no support yet for cases where the input data itself is irregular — e.g., sparse observational networks (ship reports, weather stations) or data on rotated/curvilinear grids where the point distribution is non-uniform.This would involve:
This relates to GSoC 2026 Project 1 (Flexible Graph Construction) — see GSoC ideas. I'm preparing a proposal for this and would appreciate any feedback on the approach from maintainers.
Related: #78 , #75 , #40