Feature tracking in high-resolution regional climate data

• A hierarchical triangular mesh is created for a region by dividing an icosahedron.
• Regional data is regridded to the mesh, producing a compact representation.
• Detection of storm[HYPHEN]indicative features by search of anomalies from background field.
• Locally optimal tracking of features by passing a set of hierarchical rules.
• Global optimisation of the track set by merging and exchanging track sections.

In this paper, a suite of algorithms are presented which facilitate the identification and tracking of storm-indicative features, such as mean sea-level pressure minima, in high resolution regional climate data. The methods employ a hierarchical triangular mesh, which is tailored to the regional climate data by only subdividing triangles, from an initial icosahedron, within the domain of the data. The regional data is then regridded to this triangular mesh at each level of the grid, producing a compact representation of the data at numerous resolutions. Storm indicative features are detected by first subtracting the background field, represented by a low resolution version of the data, which occurs at a lower level in the mesh. Anomalies from this background field are detected, as feature objects, at a mesh level which corresponds to the spatial scale of the feature being detected and then refined to the highest mesh level. These feature objects are expanded to an outer contour and overlapping objects are merged. The centre points of these objects are tracked across timesteps by applying an optimisation scheme which uses five hierarchical rules. Objects are added to tracks based on the highest rule in the scheme they pass and, if two objects pass the same rule, the cost of adding the object to the track. An object exchange scheme ensures that adding an object to a track is locally optimal. An additional track optimisation phase is performed which exchanges segments between tracks and merges tracks to obtain a globally optimal track set. To validate the suite of algorithms they are applied to the ERA-Interim reanalysis dataset and compared to other storm-indicative feature tracking algorithms.