Network data model

[tomalrussell]

As in #21, define a spatial network graph G = {V,E} as a set of nodes V and edges E where each edge connects two nodes.

• see notes below for objects and attributes
• some opinions about file formats and how to store metadata, conventions to follow for ease of use:
  • nodes and edges of a single network can be stored in one GeoPackage with layers named nodes and edges (a railway network could be in rail.gpkg).
  • if nodes have associated polygon geometries, store them in an node_areas layer in the same GeoPackage, with node_id to link them (GeoPackage modelling guidelines have related guidance)
  • sources can be stored in a CSV or Excel sheet (taking care to have one table per sheet, consistent column naming)
  • asset_type definitions can be captured in two tables: one for asset_types (name, title) and one table for asset_type_attributes, linked to asset type by asset_type column.

Nodes

A node:

• has a Point geometry
• has a node_id String/UUID which should be unique within an analysis project
• has an asset_type String which groups it with other assets sharing custom attributes
• may have a source String which refers to a third-party dataset
• may have a source_id which links it to an entry in a third-party dataset
• may have an associated area Polygon that represents a site, building, or service area
• may have conventional attributes to support analysis:
  • unit_cost: rehabilitation cost (units must be defined, may be made consistent across a project, e.g. million $USD in 2020)
  • cost_per_km2: rehabilitation cost if partial rehabilitation is possible per unit area

Edges

An edge:

• has a LineString geometry
• has an edge_id
• has from_id and to_id referring to its start and end nodes
• may have a length_km length in kilometres (which might otherwise be calculated from the geometry)
• may be directed (True/False), if flows on the graph should only be allowed to traverse the edge in one direction
• may have conventional attributes to support analysis:
  • unit_cost: rehabilitation cost if any damage is supposed to require full replacement
  • cost_per_km: rehabilitation cost per length damaged if partial repairs are possible

Asset types

An asset type:

• has an asset_type short name or id, for use in network data to refer back to it
• has a title human readable name that can be more descriptive
• may have a description with more details or notes
• has a list of attributes
  • name used as attribute key or column name, conventionally all in lower_snake_case
  • title human readable name that can be more descriptive
  • description human-readable description (spell out abbreviations)
  • unit units
  • dtype data type: boolean, integer, string, float, categorical
  • categories: list of options if dtype is categorical

Data sources

A data source:

• has a source short name or id, for use in network data to refer back to it
• has a title for a longer descriptive name or title
• may have a description for an abstract or longer description or notes
• may have a url link
• may have a citation recommended citation
• may have a contact.name
• may have a contact.email
• may have a release_date year or standard timestamp

For design ideas and similar approaches (particularly for data source/provenance schema):

• Spatio-Temporal Asset Catalogues STAC
• Data Catalog Vocabulary DCAT-2