Skip to content

Cost mapping for friction surface

Here are a few notes on one of my projects on friction surface mapping

Goal

Have study area raster/statistics where each pixel represents the minimum travel time to market - Problem is generalisable to incorparating additional components, e.g. climate and seasonality to get perishable pressure of crop location

Background - What is it

These papers demonstrated how to do friction surface mapping on a global scale (in R)

  • Weiss, D., Nelson, A., Gibson, H. et al. A global map of travel time to cities to assess inequalities in accessibility in 2015. Nature 553, 333–336 (2018).
    - travel time to cities with more than 50,000 people

  • Weiss, D.J., Nelson, A., Vargas-Ruiz, C.A. et al.  Global maps of travel time to healthcare facilities. Nat Med (2020)

    • Least-cost-path algorithm developed by Dijkstra (1959) for use in graphs algorithm coded from Google Earth Engine and gDistance package

    • Data: roads, railways, navigable waterways, bodies of water, land cover types, elevation, slope angle, and national borders

      • roads location data (OSM, Google)
      • road speed data from OSM

Requirements - What do we need

  • data structure (image/raster) in which each pixel represents the cost per meter to traverse it

  • Impact factors in human travel

    • Traffic network infrastructure
    • Terrain specifics (where no infrastructure exists)

Tools - How to do it

When doing cost modelling, we're dealing with projected surfaces. When you accumulate costs over large projected surfaces, the pixel size represent different distances - but we need equal distance from any location!

  • e.g on a latitude, longitude map of the world, pixel at the equator is very different in size and shape than in Minnesota
  • The only way to do this is on a sphere
    • gdistance e.g. calculates great circle distances between every pair of location; this distance represents the shortest line between two points, taking into account the curvatur of the earth. It turns out to be the best predictor of genetic distance
    • with small scale (UTM) ARGIS is fine, otherwise there will be distortions

Google earth engine

  • image.cumulativeCost() to compute a cost map where every pixel contains the total cost of the lowest cost path to the nearest source location source

R package gdistance

Etten, Jacob van. "R package gdistance: distances and routes on geographical grids.” Journal of Statistical Software (2017)

the accCost() function

  • Superb handling of adjacency (transition matrix) and cumulative cost mapping to central point
  • geoCorrection function of the leastCost algorithm
  • corrects for distance distortions associated with data in a geographic coordinate system
  • corrects for ‘true’ local distance