Page Comparison

...

Users also need to be conscious of the impact that the change in resolution has on the travel time in cell where patients are reaching the road network.

More specifically, as roads segments get converted into a raster layer during the creation of the merged land cover, any cell containing a road segment will be considered as fully covered by a road in the merged land cover. As such, the travel speed over these cells will correspond to that of the road segment crossing it.

While the above process does not have much impact when working with high resolution grids (up to 100 meters), working with lower resolution grids does produce an impact because the travel time required to walk between the border of the cell and the road is not taken into account.

For example, consider a patient walking at a speed of 3 km/h to reach a road passing through the middle of the cell on which you can drive at a speed of 100 km/h:

1. When working at a resolution of 500 meters, the patient would have walked during 5 minutes to reach the road and then 9 seconds to exit it using the road, while in AccessMod this path will be modelled to only take 18 seconds to reach the road network and exit the cell (17 times faster). The error here is therefore equivalent to 4 minutes and 51 seconds
2. When working at a resolution of 1 kilometer, the patient would have walked during 10 minutes to reach the road and then be on the road during 18 seconds while in AccessMod this will be modelled to only take 36 seconds (16 times faster). The error in this case is equivalent to an underestimate of 9 minutes and 42 seconds.

Once the patient is on the road network in the resulting merged land cover layer then the above remarks no longer applies, since passing from one cell to another is done using the road itself. The error is therefore applied only once and not repeated along the full network. 

In conclusion, as the resolution of the dataset decreases, the extension of the catchment areas tends to be over-estimated and more patients are assumed to reach the corresponding facilities by “saving time” at the transition between the road network and the areas where they have to walk.

Unfortunately, it is not possible to adjust for the above errors. The recommendation is therefore to use the highest resolution possible taking the RAM consumption issue into account (see section 3.2.5).

In addition to the above, the rasterization of elongated water body polygons (and other barriers to movement stored as polygons) does not always generate continuous surfaces, unlike when rasterizing vector lines. Figure 5 below provides an illustration of such a problem. As we can see in Figure 5c, working with lower resolution (larger grid sizes) will create discontinuities, therefore generating artificial bridges in these areas.