Virtual Roads for the BioEnergy Atlas

Virtual roads representing theoretical paths [roads] starting every 1km apart (across South Africa) from areas where no roads exists to the closest road segment in South Africa's National Geo-spatial Information (NGI) 2019 road layer (http://www.ngi.gov.za) -- for use in the SAEON BioEnergy Atlas feasibility model. A cost surface using a digital elevation model (SRTM v3) [~30m] was used as the basis, with main water features (rivers and dams) and a slope threshold of 15 degrees removed as obstacles from the surface. Start points (derived from a 1km x 1km raster grid, and end points (the NGI road layer) rasterised at a 100m x 100m resolution, both using the StatsSA BSU grid projection ['+proj=aea +lat_1=-22 +lat_2=-38 +lat_0=-30 +lon_0=25 +x_0=1400000 +y_0=1300000 +datum=WGS84 +units=m +no_defs]) were used as location inputs. Non-NA raster values from these location inputs were converted to points (the centroid of each Non-NA pixel), then reprojected to the same resolution as the cost surface (e.g. EPSG:4326 - WGS 84 – Geographic). A variable resolution window was used for modelling [based on 250 start points at a time, each linked to 100 closest road points [end points] to create an extent and perform the modelling operation in reasonably sized chunks]. The scikit-image python package was used to find the shortest path through each cost surface array cropped using the variable window extent, using a fully connected (i.e diagonal pixels) minimum cost path (MCU) algorithm. All outputs were merged to create a seamless Virtual roads layer with 1 051 696 individual paths covering South Africa.