-
Notifications
You must be signed in to change notification settings - Fork 0
Use in CRISP
NetworkModel_World is used on its own, but it is also a preparation step for the CRISP model. This page explains what the connection is.
CRISP downscales exogenously defined change in built-up land and population onto 1 km grid cells globally, up to 2100. The heart of it is a locational suitability: a logit probability, estimated by calibration, that says how likely a given grid cell is to receive additional built-up development compared to the other cells in its region.
That suitability has to be built from variables that describe the location. Some of those are simple to derive from a grid, such as slope or distance to water. The ones that describe how a location sits in the settlement system are not. A cell is attractive for development partly because of what can be reached from it, and reachability is a property of a network, not of a raster.
NetworkModel_World produces exactly that class of variable, globally and on the same 1 km grid:
- Travel time and network distance to the nearest settlement of each type, and to the nearest local, regional and national centre. See Accessibility. These are the direct measures of how peripheral a cell is.
- Potential accessibility per cell, the decay weighted population reachable within an hour, from the Travel distance proxy. This is a smoother measure than distance to the nearest place, and it distinguishes a cell near one small town from a cell in the middle of a dense urban field.
- Catchment membership, which says which centre a cell belongs to. See Catchments.
- Settlement hierarchy, which ranks the centres themselves. See Settlement hierarchies.
Because everything is computed on the same population grid in World Mollweide, these products line up cell for cell with the grids CRISP works on, without resampling.
The split is deliberate and follows from how expensive the network step is. Building a routable global network and running matrices over it takes days and a very large machine, but its result changes only when the OSM extracts or the settlement layer change. A CRISP scenario run, by contrast, is something you want to repeat often with different assumptions.
Keeping the network work in a separate model means the accessibility grids are computed once and then read as static input, and a scenario run never pays for routing. It also means the accessibility products can be validated, published and used on their own, independently of any projection.
The two models share more than data. Both are built in GeoDMS by Object Vision, both use the same GHSL population and settlement layers as their base, both work on 1 km grid cells in World Mollweide, and both are organised around per continent and per country iteration for the same memory reasons. Someone who can read one configuration can read the other.
The connection to CRISP is not the only one. The modelled link flows split by road class are produced for emission modelling work, the functional rural areas and rural accessibility grids were produced with the OECD in mind, and the settlement hierarchies are usable wherever a network based definition of central places is needed. See Outputs for what is available.