• Home
  • About

GIScience News Blog

News of Heidelberg University’s GIScience Research Group.

Feed on
Posts
Comments
« Usage of HELIOS for various applications
Colloquium on Assessment of Risk from Natural Hazards in Urban Areas »

Exploring OSM history: the example of health related amenities

May 16th, 2019 by Sven Lautenbach

Introduction
Exploring how OpenStreetMap data developed over time across different administrative untis might reveal interesting insights into the self organizing approach of the OSM communities and can potentially be used to derive intrinsic data quality indicators. It might even be possible to estimate the completeness of OSM for a specific key-value combination as done by Barrington-Leigh & Millard-Ball (2017) for the road network.

Here we want to investigate the development of health related amenities across countries. The focus of the post will be on exploration of the data to highlight a few interesting patterns. A scientific rigorous analysis is not the aim of the post but will follow in a dedicated scientific journal.

Data and methods
We queried the OSM history by using the OpenStreetMap History Data Analytics Platform - if you are unfamiliar with the ohsome platform (OSHDB and ohsome API) we encourage you to explore the related blog posts:

  • idea
  • general architecture
  • OSHDB open access journal article
  • blog series: “How to become oshome” first post,second post,third post, fourth post, fifth post

We queried ways and nodes but not relations for the key-value combinations provided later on, used national boundaries to group by and monthly time steps.

Results were stored in the postgresql database of the OSM history explorer (see the related blog post) and further analyzed by R & Rstudio using the packages from the tidyverse, ggplot2, forcats, stringr, RPostgreSQL and DBI.

Data exploration
Saturation type time series
If we look at France and Hungary it looks as if the number of hospitals has reached a peak which we might take as an indicator that the number of hospitals in both countries have been completely mapped (give or take a few). There have been some interesting ups and downs in Hungary.

Fitting a standard logistic saturation curve leads to reliable results. For France the estimated number of hospitals equals 2800 with a standard error of 11.3. For Hungary the estimated number of hospitals equals 333 with a standard error of 1.8.


The mapping of amenity=hospital for Haiti and Lesotho seems also to have leveled of. However the process looks different with steep jumps. For Haiti the jump corresponds to the disastrous earthquake in 2010 and the mapping activities by the HOT and OSM community afterwards. For Lesotho the steep increases can be less clearly linked to single events but could potentially be related to the Lesotho Mapathons.

In general one should be careful to interpret saturation curves too literaly. Have all hospitals in Haiti been mapped or did the mapping actitives stop leaving parts of the hospitals unmapped? If we would have looked at Lesotho in early 2014 we might have concluded that all hospitals have been mapped since the time series seemed to have leveled of. However, the strong increase afterwards shows that this would have been a wrong conclusion.

So even if the standard logistic curve fits the time series for Haiti quite well, we need to think carefully if the fitted asymptote (1296.56 , standrad error: 5.4 ) is a reliable estimate of the number of hospitals. For Lesotho it is clear that it would be better to use a function that combines two logistic curves.

Fitting a double logistic curve to the development clearly leads to a better fit which is also indicated by means of a comparision of AIC (Aikaike Information Criterion) values. The saturation asymptote is here estimated as 49.18 with a standard error of 0.21.

For Haiti it is worth to look at the spatial distribution of mapped hospitals. Are they mapped all over the country or are they concentrated in specific parts? If the later would be true we might take this as an indication that mapping might be incomplete - e.g. because mapping activities after the earthquake in 2010 would have concentrated at a specific part of the country.

As we can see from the map hospitals in Haiti are mapped all over the country. Not unexpectedly, hospitals are concentrated around the capital Port-au-Prince. The map gives at least no imediate indication of a geographical bias in the mapping activites.

Increasing trend
Other countries such as China or India show still an increase in mapped hospitals.

Fitting saturation curves here comes of course with a high attached uncertainty since not even the inflection point might be in the observations.

The expected number of hospitals in China is estimated as 7004.01 with a standard error of 305.09. For India the estimate is 7220.21 with a standard error of 136.93 - since the inflection point is part of the observations (from the perspective of the model fitting procedure) the estimated uncertainty of the parameter estimate (here the asymptote) is lower compared to China.

Stabilisation followed by a decrease
Both Belgium and Germany show first an increasing number of hospitals and a decay after a stabilisation period (which was longer in Belgium and relatively short for Germany). The decay might be related to either a real decrease in hospitals (which is at least for Germany the case) or due to a revision of the OSM data.

Fitting a monoton increasing function would not describe the behaviour too well in this situation. The functional relationship could for example be modelled by a Holling type IV relationship or by a double logistic function with a negative slope in the second part of the function.

Stabilization followed by a sharp decrease
Spain and Sweden show a confusing pattern: after an increase followed by a - shorter or longer - saturation phase the number of ways or nodes tagged by amenity=hospital show a sharp decrease.

We suspect that this might have been caused by a tag diversification and plot the count for amenity=clinic on top of the count of amenity=hospital. Tagging of amenity=clinic has begun much later in both countries and seems to explain the sharp decrease supporting the hypothesis that a tag diversification has been taken place.

In both cases it looks as if no saturation has been reached. If we would have looked at e.g. Spain in 2017 we presumably would have assmued wrongly that a saturation had been reached at around 2,200 hospitals. After the tag diversification it looks as if the number of hospitals is much lower (~1,000). Estimating saturation levels for clinics hospitals or their combination from the available information seems not unproblematic.

Mapping the distribution of amenity=hospital and amenity=clinic for January and November 2018 shows that the tag-diversivication was not presumably concentrated on specific parts of the country but seems to have affected the whole country in a similar way. This seems remarkable given the regional cultural diversity of Spain.

We could even go one step further and add a third health related amenity (doctors) on top of the stacked figure. Taken all three objects together it seems as if the mapping of health related amenities in Spain and Sweden has not leveled of so far.

Interesting observations
Germany and France
Both France and Germany show an ongoing mapping of ways and nodes with amenity=doctors while clinic and hospitals seemed to have been mapped more or less completely. Remarkable is also the much higher number of amenity=doctors in Germany (~25,000) compared to France (~8,000) which can not be explained by differences in population size (France ~67.12 Mio. Inh., Germany ~82.79 Mio. Inh.).

Thailand and Bolivia
Thailand and Bolivia are interesting since they show a strong increase in the mapping of hospitals followed by a steep descent, followed by an increase afterwards. Potentially, the strong increase could be triggered by a mass import and the drop by the removal of mass imported objects.

Concluding food for thoughts
The development of health related amenities captures interweaved phenomena: the tagging of real world phenomena, changes in tagging conventions, external events that trigger mapping activities (such as earthquakes or tsunamis) as well as mass imports. In addition, real world phenomena change over time: health related amenities might be created or be taken out of use (e.g. in Germany). To understand differences between countries it is also helpful to look at local tagging guidelines - amenity=clinic is for exmaple defined differently in Spain and Germany. Differences in the health systems are of course of importance as well. In Spain for example ambulant clinics are much more common than in Germany.

We invite you to have a look at the full detail of all countries here: Charts for all countries.

We will continue with OSM history analysis in further blog posts - so stay tuned.

And if you are studying geography at University of Heidelberg and are searching for a topic for your Bachelor or Master thesis in the domain of OSM history analysis feel welcome to approach Alexander Zipf or Sven Lautenbach.

Tags: Big Spatial Data, heigit, ohsome example, OSM History Analytics, OSM ohsome

Posted in OSM, Research, VGI Group

Comments are closed.

  • About

    GIScience News Blog
    News of Heidelberg University’s GIScience Research Group.
    There are 1,675 Posts and 0 Comments so far.

  • Meta

    • Log in
    • Entries RSS
    • Comments RSS
    • WordPress.org
  • Recent Posts

    • Assessing road criticality and loss of healthcare accessibility during floods: the case of Cyclone Idai, Mozambique 2019
    • New paper on the automatic characterization of surface activities from 4D point clouds
    • OSHDB Version 1.0 Has Arrived
    • Job Opening for Postdoc / Senior Researcher on OpenStreetMap Road Quality Analysis
    • Geography Awareness Week 14.-19.11.2022
  • Tags

    3D 3DGEO Big Spatial Data CAP4Access Citizen Science Conference crisis mapping Crowdsourcing data quality deep learning disaster DisasterMapping GeoNet.MRN GIScience heigit HOT humanitarian humanitarian mapping Humanitarian OpenStreetMap team intrinsic quality analysis landuse laser scanning Lidar machine-learning Mapathon MapSwipe MissingMaps Missing Maps ohsome ohsome example Open data openrouteservice OpenStreetMap OSM OSM History Analytics Public Health Quality quality analysis remote sensing routing social media spatial analysis Teaching VGI Workshop
  • Archives

    • January 2023
    • December 2022
    • November 2022
    • October 2022
    • September 2022
    • August 2022
    • July 2022
    • June 2022
    • May 2022
    • April 2022
    • March 2022
    • February 2022
    • January 2022
    • December 2021
    • November 2021
    • October 2021
    • September 2021
    • August 2021
    • July 2021
    • June 2021
    • May 2021
    • April 2021
    • March 2021
    • February 2021
    • January 2021
    • December 2020
    • November 2020
    • October 2020
    • September 2020
    • August 2020
    • July 2020
    • June 2020
    • May 2020
    • April 2020
    • March 2020
    • February 2020
    • January 2020
    • December 2019
    • November 2019
    • October 2019
    • September 2019
    • August 2019
    • July 2019
    • June 2019
    • May 2019
    • April 2019
    • March 2019
    • February 2019
    • January 2019
    • December 2018
    • November 2018
    • October 2018
    • September 2018
    • August 2018
    • July 2018
    • June 2018
    • May 2018
    • April 2018
    • March 2018
    • February 2018
    • January 2018
    • December 2017
    • November 2017
    • October 2017
    • September 2017
    • August 2017
    • July 2017
    • June 2017
    • May 2017
    • April 2017
    • March 2017
    • February 2017
    • January 2017
    • December 2016
    • November 2016
    • October 2016
    • September 2016
    • August 2016
    • July 2016
    • June 2016
    • May 2016
    • April 2016
    • March 2016
    • February 2016
    • January 2016
    • December 2015
    • November 2015
    • October 2015
    • September 2015
    • August 2015
    • July 2015
    • June 2015
    • May 2015
    • April 2015
    • March 2015
    • February 2015
    • January 2015
    • December 2014
    • November 2014
    • October 2014
    • September 2014
    • August 2014
    • July 2014
    • June 2014
    • May 2014
    • April 2014
    • March 2014
    • February 2014
    • January 2014
    • December 2013
    • November 2013
    • October 2013
    • September 2013
    • August 2013
    • July 2013
    • June 2013
    • May 2013
    • April 2013
  •  

    May 2019
    M T W T F S S
    « Apr   Jun »
     12345
    6789101112
    13141516171819
    20212223242526
    2728293031  
  • Recent Comments

    GIScience News Blog CC by-nc-sa Some Rights Reserved.

    Free WordPress Themes | Fresh WordPress Themes