-
Notifications
You must be signed in to change notification settings - Fork 0
Prior variable preparation
Many variables are prepared before ingestion in the model, or (partially) within the model configuration. This occurs in the /SourceData and /Preprocessing containers. We discuss the prior preparation of variables in this section. The model also entails some variables that respond to concurrent model results. These variables are computed internally and discussed in the section dynamic variable computation.
Population grids are originally obtained from GHS-POP. These grids describe population at the 1 square kilometer resolution at 5 year intervals between 1975 and 2020. However, these grids have been modified to maintain consistency with model assumptions. The GHS-POP data are consistent with CIESIN population assumptions. The current reference model is based on UN WPP counts and projections from 1950 to 2100. To ensure that totals from the input historical population grids match the input population projections, Marcello Schiavina has provided rescaled GHS-POP grids that match the historical WPP counts per country. The resulting input grids describe floating-point population estimates. The reference model produces integer population counts. To make sure that this discrepancy does not lead to inconsistencies in historical vs projected model results, all historical population grids are transformed into integer population grids. This is done in /sourcedata/Population/PerYear/[YEAR]/Population in the same way as we do in the population allocation. Thus, we
- round down the population grids to integer numbers; then
- sum the amount of people that are lost by rounding down (a number n per functional area);
- then give the n cells with the highest loss of fraction population (ie, the floating population - rounded down population) an additional inhabitant.
The terrain roughness index (TRI; see Riley et al, 1999) measures the local height differences of a geography. The calibration exercise executed for this paper included it as an important factor to explain the presence of built-up land. The model is based on TRI grids that have been derived from a GLO 30 arcseconds, global digital elevation model. Because of the size of the input DEM files, computing this TRI grid globally is computationally intensive. This computation has been done as part of the model development by computing TRIs per continent, and then 'mozaiking' these continental results together into one global TRI grid. The global TRI grids are part of the model sourcedata; the underlying DEM are not. The code to calculate TRI grids is available through /sourcedata/Physical/TerrainRoughnessIndex/TRI_generation/Generate.
Elevation grids were, similar to the TRI grid, also created internally from an underlying high-resolution DEM from GLO. Similarly, the resulting elevation grid is part of the standard model sourcedata. The code to generate this elevation grid is available through /sourcedata/Physical/Elevation/Mean_1k.
A number of static model inputs have to be created for a study area before the model can be run. In all cases, these are data that are most efficiently transferred as shapefiles, but, prior to the modelling, still need to be transformed into grids describing the wanted variable. Another advantage of this process is that shapefiles are much more easily reprojected to another geographic projection system than grids. The variables that need to be transformed to grids are described here.
The distance to the nearest coastline is derived internally from a shapefile describing coastlines. This is done internally, prior to the model can be run. As the source shapefile is smaller than the resulting grids, only the shapefile is provided, and users will have to generate the distance to coastline grid (and other, similar inputs) themselves for the study area they want to model. The process to create the distance to coastline grid is available through /sourcedata/Physical/Coast/Location2Coast_sqr_domain_export, or /MakeUnlinkedData/CreateFiles
The distance to all water bodies, and large inland water bodies, are derived from a shapefile describing surface water. Any waterbody that is more than 0.5 square kilometers in size is considered large here. The grids describing distance to the closest water body and the closest large water body are available through /sourcedata/Physical/Water/Location2AnyInlandWater_sqr_domain_export and /sourcedata/Physical/Water/Location2LargeInlandWater_sqr_domain_export, respectively; or alternatively through /MakeUnlinkedData/CreateFiles
Road shapefiles from the GRIP project are included in the sourcedata. They are split into a set with main throughfares (categories 1, 2 and 3 describing highways, primary roads and secondary roads, respectively) and local roads (categories 4 and 5 describing tertiary roads and local roads). Both sets were converted to variables indicating the closest distance to either main roads, or local roads. They were used in the calibration and are significantly associated with the presence of built-up land. These variables need to be converted to distance grids through /sourcedata/Socioeconomic/Infrastructure/Location2Roads123_sqr_domain_export (main thoroughfares) and /sourcedata/Socioeconomic/Infrastructure/Location2Roads45_sqr_domain_export (local roads).