Raster data model

[tomalrussell]

So far we've considered joining single raster datasets to a vector dataset.

In practice, it will be useful to have metadata about each raster, and collections of rasters. This might be represented and stored in a conventional way. It could enable useful programmatic manipulation of vector-raster intersection results, where there's a natural compact way of storing results..

asset_id | <raster_id> | <raster_id>
123      | depth=12    | depth=1.1

..but we also then want to use metadata associated with a given raster ID to calculate aggregate statistics (expected damages across multiple return periods; min/max exposure across models or scenarios).

Raster

• 2D array of data
• e.g. contents of a single TIFF from Aqueduct floods
• affine transform (typically recorded in file metadata)
• data type
• other metadata: RCP 4.5, GCM HadGEM2-ES, RP 1000, Fluvial, 2080

Raster Dataset

• set of rasters
• e.g. all the coastal flooding TIFFs from Aqueduct floods
• share a transform
• share a metadata spec (same keys, maybe enums/types of allowed values)
• share a source, citation, description

Related work on metadata/schema in this domain:

• https://github.com/OasisLMF/OpenDataStandards
• http://docs.riskdatalibrary.org/overview.html

[tomalrussell]

Consider also NetCDF-style rasters, where x-axis and y-axis coordinates are provided explicitly (e.g. as arrays of longitude and latitude values), and other dimensions may be encoded within the file (e.g. return period, as in STORM tropical cyclones [data])