Back to Blog
Working out distance along a network is a fairly simple process within most GIS packages but only knowing the distance rarely gives much more than surface level information to a GIS user. Whether the GIS user is measuring a water flowing through a river, or cars travelling along a road, distance is only ever one factor in determining the time it will take to travel through any given network.
Normally when studies conduct road network analysis in rural, low-income countries they use arbitrary theoretical speed values under the assumption that road users travel as quickly as is legally possible e.g. (Tanser et al 2006). This is obviously a highly flawed assumption and has potentially serious and significant ramifications. In developing countries where factors such as unpaved roads and increased slope significantly hamper travel times misleading data can result in poor allocation of resources.
Previous studies have largely been centred around urban areas, with emphasis on congestion (Sen et al, 2011) as well as improving road safety by investigating effective methods to reduce speed eg Bener et al, 2003 and Afukaar, 2003. There has been little research that attempts to map rural travel speeds and has left a hole in our understanding of access.
This blog post will focus predominantly on the creation of novel GIS techniques to improve travel time estimation within developing countries, and how this data can be used by researchers to improve network analysis in remote regions.
There are two main ways of measuring accessibility, one is a raster-based approach and is a more visual source, the other utilises vectors and would be of greater use to GIS users and policymakers.
Firstly, is a raster-based approach that uses data sources including, but not limited to; road cover type, land cover, slope, and elevation. The crucial part of this methodology will be assessing how to weight these variables. This could be done theoretically, though practical assessment of this would be useful as a form of grounding the data. A weighted sum can be used to prioritise these variables and produce a single raster grid. This would visually show more isolated communities.
The other approach would involve attaching cost measurements to existing road networks so that when analysed would give an accurate indication of travel time. Slope is one method of assessing cost – on an unpaved road a slight increase in slope will result in an increase in travel time, whilst a gentle decrease will result in a lower travel time. However, when the slope reaches a critical threshold travel times will rapidly begin to increase until it could be deemed that a road is impassable (see figure 1 in footer).
To gain an understanding of how these costs impact travel times they will need to be set against ‘base,’ times – that is typical travel speeds along an ‘optimal,’ route on both paved and unpaved surfaces. Once these variables are calculated then it becomes possible to make it more costly to travel along non-optimal routes.
Methodology and data
The methodology used by this report is designed to be easily expandable to incorporate new data sources as available and appropriate, as well as replicable to a theoretically global-scale coverage. Data has been sourced from the following locations;
Open Street Map
OSM data whilst containing significant artefacts that are inaccurate is worth persisting with not only because it is free (and therefore expandable without significant cost to the user) but is a network that improves in both coverage and quality with age. It also has natural features embedded which are critical to understanding how a road network might be affected, eg river crossings etc. This report has only used the road network
Shuttle Radar Topography Mission DEM
SRTM data is at 30m2 ground resolution generally with a 16m elevation positional accuracy. Whilst higher resolution DEM’s would be preferable there are no free products currently available that are available globally without significant artefacts (such as those associated with ASTER). A coarse DEM will not necessarily hinder this project – the slope values will be attached to differing sections of the road network which may well be greater than 30m which would make a higher resolution DEM have redundancy and require interpolation.
Global Land Cover classification from ESA’s MERIS sensor. 300m2 resolution is quite coarse and would preferably be higher resolution – but land cover type is the least important variable used in this report and lower resolution data will be easier to process
WorldPop 2015 population rasters
The area used as a sample in this investigation was quite rural, therefore the impact of larger populations is not expected to be significant in increasing travel times. Raster with a spatial resolution of 100m2 are available from WorldPop, indicating population density in any given location. This will give some indication of how many road users there are, this is a highly limited approach but will still benefit the overall methodology.
The landcover classification was reclassified as a cost depending on the surface type (see figure 3 - slideshow at footer) with a land cover deemed to be impassable receiving a maximum score of 20. The figures assigned are arbitrary and need to be validated in future research.
The SRTM DEM was converted into a Slope model in Arc by the angle between pixels (see figure 2). This was then categorised into 20 natural breaks using the Jenks clustering method. The steepest classification of slope was assigned a cost score of 20, with the score depreciating down the most gradual incline which received a score of 2.
The population data was manually assigned a classification out of 20 (where 0 represents no friction and 20 is impassable), as this was a rural area with only small villages it was determined that 4 would be the highest classification allocated, this is to allow for suitable scaling in areas with significantly higher population densities.
I determined that slope was a more significant cost to travelling along a network and accordingly weighted it more significantly than land cover classification by a factor of 8. These two variables were merged together to form a singular cost column using the field calculator (for more information on how I achieved this please contact me directly).
The output of this study is a road network that has embedded into it the cost of travelling derived from slope and land cover (see figure 4 and 5).
Whilst not embedded into the final product, the distance of each vector was multiplied by the cost to exaggerate the length of stretches of road to reflect the cost. Using the OD Cost Matrix function the distance to the nearest health clinics from villages was assessed, with both the original lengths and then with the exaggerated dataset. Out of the 368 villages that were connected to villages 11 were rerouted to alternative health clinics as a result of the introduction of a cost measure (see figure 6)
Understanding travel times in a network is crucial in a wide range of situations ranging from measuring access to healthcare facilities to learning how diseases can spread in transit. It is important that policymakers have advice driven by underlying data that best replicated real-world conditions.
This is only the first iteration of this methodology, with future refinements to improve the weighting of the variables.
This methodology has great potential for expansion - to a global scale (due to the nature of the data sources used) as well as to incorporate new variables such as a more reflective indication of road users so that the model is better tuned for urban environments and a better reflection of the seasonal variance of the quality of the road surface. This could be achieved by having a 'dry,' speed and a 'wet,' speed embedded within the same file, or alternatively as two altogether separate files.
The final step will to be actually calculate a speed along each section of the network - this will require a 'base' speed along a few different road types that are optimal for their category. The categories that I would suggest to be appropriate would be "unpaved_wet," unpaved_dry," "paved_wet," "paved_dry." Once an speed under ideal conditions is set per classification then the cost measure can be applied to give an indication of actual travel times.
Afukaar, F.K., 2003. Speed control in developing countries: issues, challenges and opportunities in reducing road traffic injuries. Injury control and safety promotion, 10(1-2), pp.77-81.
Bener, A., Abu-Zidan, F.M., Bensiali, A.K., Al-Mulla, A.A. and Jadaan, K.S., 2003. Strategy to improve road safety in developing countries. Saudi medical journal, 24(6), pp.603-608.
Park, D. and Rilett, L., 1998. Forecasting multiple-period freeway link travel times using modular neural networks. Transportation Research Record: Journal of the Transportation Research Board, (1617), pp.163-170.
Sen, R., Siriah, P. and Raman, B., 2011, June. Roadsoundsense: Acoustic sensing based road congestion monitoring in developing regions. In Sensor, Mesh and Ad Hoc Communications and Networks (SECON), 2011 8th Annual IEEE Communications Society Conference on (pp. 125-133). IEEE.
Tanser, F., Gijsbertsen, B. and Herbst, K., 2006. Modelling and understanding primary health care accessibility and utilization in rural South Africa: an exploration using a geographical information system. Social science & medicine, 63(3), pp.691-705.