Analysis of Medical Facility Placement in Ethiopia

ABT 181 - GIS and Spatial Modeling

Final Laboratory Project by Jim Mullins and Gregory McHugh


Introduction

Last Quarter we developed a method for determining the best placement for medical facilities in areas of Ethiopia where major diseases (malaria, onch, schistosomiasis, and yellow fever) were endemic. This was achieved by "overlaying" polygon and point coverages of the diseases to determine the cities which fell within those endemic areas that did not already have hospitals or health centers. Around each medical facility we placed a predetermined 10km "buffer" defining our service area. We also attempted to determine how many people, in theory , each medical facility would serve by overlaying population density onto our service areas. This quarter we wanted to address the questions left unanswered by our project last quarter. What if a natural barrier such as a river or a mountain range runs right through the middle of the "buffer" ? How would this affect the accessibility of our facilities ? Would this analysis lead us to put the medical facility somewhere else ? We did not bring into account the topography, landscape, or roads of our study regions to determine what the "cost" would be within each buffered area to access the medical facility.

(For reference see our Analysis of Medical Facility Placement from ABT 180, Winter Quarter, 1997,UC Davis. )




Check out a MPEG Aerial Fly-by of our Study Area !!



Aerial Fly-By Reference Maps




The city of METU is the
center and beginning
of the Aerial Fly-By

This shows the Fly-By Path from METU to X.


Objective / Hypothesis

Our objective with this project is to determine what the relative "cost" would be to access each medical facility from each point within each respective buffer. This will be determined from a combination of factors which include Euclidian distance from the city at the center of each buffer, proximity to rivers, proximity to roads, slope determined from a digital elevation model, and land use types. From these weighting factors we can calculate a "cost distance", that is the "cost" (not necessarily in monetary terms) to reach the facility from each location in the grid. In addition , we would like to again briefly return to our project from last quarter, by applying this analysis to the two proposed facility locations that were excluded from our previous analysis and how they compare with the four proposed sites.


Data Sources

We started with many vector coverages which we used from last quarters project, specifically the Ethiopian country boundary from ESRI's ArcWorld, the cities we digitized, and the medical facility buffers . We then added the roads, rivers, and lakes from ESRI's Digital Chart of the World for our study area. It turned out that the lakes weren't a factor because there were no lakes in any of our medical facility service buffers. We found a Digital Elevation Model for our study area at the Eros Data Center's G TOPO30 site on the WWW. Finally for our land use coverage we turned to good old Shields Library Map Room where we found a suitable hardcopy paper map from the National Atlas of Ethiopia which we could digitize.

It should be noted here that the data we used was for a particularly large scale and we would have preferred to have data at a much higher resolution. However we couldn't find such data either in Shields Library or on the Internet. In short the data wasn't good, but we believe our process was sound.


Importing Spatial Data

To import our spatial data for roads, rivers, and political boundaries we simply downloaded coverages from ESRI's Arc World and Digital Chart of the World. These coverages were already in a native ARC/Info format, on CD-ROMs in the Visualization Lab.

Greg downloaded our DEM from the USGS Eros Data Center G TOPO30 WWW site. He was then required to perform the following steps to import the DEM into a grid coverage which we could use in ARC/Info :

To import our landuse cover into digital form, Jim scanned it on a HP 4c flatbed scanner and "on-screen" digitized rather than using the digitizing table in order to achieve greater accuracy. The scanned image was saved as a ".tif " image and then used the ".tif " image as a background coverage in ARC Edit to create a vector coverage from the raster ".tif " image.

To accomplish this he first used the IMAGE command to bring up the ".tif " in the background environment in ARC Edit. Next he used CREATE to create a new coverage. Then he digitized the attributes and features of our scanned raster ".tif " image into a vector coverage format.

For creating the landuse polygon coverage, he first used the INTERSECTARCS ALL command to ensure that nodes were placed at all intersecting arcs. He then set the EDITFEATURE to ARC and ADDed the arcs using the SPLINE function to smooth the lines we digitized from the ".tif " image. Finally, when all arcs were added , he used the CLEAN command to create arc and polygon topology.


Imported Coverages of Proposed Service Areas

Landuse

Rivers and Lakes

Roads

Digital Elevation Model


Adding Attribute Information

The command we used to add attribute information was, not surprisingly, ADDITEM. For our roads, rivers, and cities vector coverages, we needed to assign an item which would be the same for all coverage features , which would then be used as the "value" item when converting them to raster coverages. The resulting grids would then have some value for the cells representing the vector feature and nodata everywhere else. We also had to add "weighting factor" items to the rivers, roads, and land use in order to perform our spatial analysis.


Editing and Processing Spatial Data

The first thing we had to do was to convert the ESRI data to the right coordinate system for our study area. For this process Jim developed a very helpful AML to convert the Digital Chart of the World coverages from their native projection of latitude and longitude in decimal degrees to the mercator projection system we used for our project. This was so helpful because the Digital Chart of the World comes in tiles which need to be combined to cover most study areas, including ours. The DEM from the USGS and the digitized landuse coverage were also converted to the mercator projection system from their native format of latitude and longitude decimal degrees using Jim's AML.

Once we had established topology for our land use, river , road, and city vector coverages , we then had to convert them into raster grid coverages using the POLYGRID, LINEGRID, and POINTGRID functions respectively in ARC/Info.

Once we had all of our necessary data converted into raster grid format, we then used the LATTICECLIP command to "cookie-cut" out the appropriate portions of the grids which fell within the 10km medical facility buffer. The purpose of this was to eliminate the processing of data which was uneccessary for our goal and to save time and disk space. We quickly learned a valuable lesson in this regard , after first traveling down the POLYGONSELECT road which led to huge files of a couple of hundred megabytes and major lapses in processing time. LATTICECLIP converts the output grid to the mapextent of the "cookie-cutter" while POLYGONSELECT keeps the mapextent the same as the source grids'.


Grids Developed for each Proposed Service Area

Landuse

Rivers and Lakes

Roads

Digital Elevation Model


Spatial Analysis

Once we had "cookie-cut" grid coverages of our six study sites, we began our spatial analysis; to determine the COSTDISTANCE to reach each proposed medical facility within each buffer. We began by performing the GRID function EUCDISTANCE to determine the distance from each grid cell in the coverage to the medical facility in the center of the coverage. Naturally, this distance measurement will be one of the most important weighting factors in our spatial analysis. Next, we used our DEM to calculate the SLOPE of our study sites. This function was followed by calculating EUCDISTANCE from all roads found in the 10 km buffered regions. The EUCDISTANCE function was used in this case because the proximity of each grid cell to a road is directly related to the cost associated with accessing the medical facility in this rural part of Ethiopia. Therefore, this distance value grid was used as a weighting factor for the final COSTDISTANCE calculation. Similarly, weighting factors were assigned to the land use and river grids. The river grid was weighted based upon whether the river was a natural barrier to "direct line access" to the medical facility at the center of the buffer. The land use was assigned weighting factors based upon the impedance each particular land use type created for the movement through each cell in a COSTDISTANCE function.

Before performing the COSTDISTANCE function, we created a cost grid representing all of these weighting factors. This grid consisted of the sum of the weighting factors assigned to all of the previously discussed variables, which we attempted to standardize so they would more accurately portray the trends in our data. This was very challenging due to the fact that we had no template to follow and were unsuccessful at finding any similar projects done previously in this manner. Initially, several variables (i.e. EUCDISTANCEs) skewed our data in such a way that all other weighted factors were smoothed out; their trends were virtually undetectable. Therefore, weighting factors were manipulated for each variable until a combination yielded results that supported the data that we had to work with.


Final COSTDISTANCE Grids for each Proposed Service Area

COSTDISTANCE Grids


Results

We were right !!!! Before even beginning with the spatial analysis we could determine that the two proposed sites which we threw out were indeed not suitable proposals by simply looking at the river, road, and land use vector coverages. Both sites had the "worst" land use in terms of COSTDISTANCE weighting. One site had no roads traversing the buffer at all and the other had many rivers running through it. After these observations , we decided to focus our spatial analysis on the four proposed sites from our project last quarter. After completing our spatial analysis we are now able to tell how easy or difficult it will be to access each facility from every point in the service buffer.


Conclusions

Last quarter we used vector based GIS to identify areas in Ethiopia that were in most need of medical facilities based on prevalence of endemic diseases, population density, etc. We asked the right questions, but without ample time, we were not able to ask all the right questions. As it was a vector based class, we did not have a chance to consider how topographical factors would affect our analysis. Upon completion of this quarter's project we have found that raster GIS is a invaluable compliment to vector GIS. We believe that we have developed a sound method for determining the placement of medical facilities, but would benefit greatly from discussions with medical professionals, geographical professionals, or other native people familiar with a region we are largely unfamiliar with. Perhaps there are localized variables to eastern, rural Ethiopia which we are unfamiliar with , which could potentially undermine our analysis.


Bibliography / Map Data Sources


Final Laboratory Project for :

Applied Biological Systems & Technology 181, WinterQtr.97, UC Davis.

Submitted By : Jim Mullins & Gregory McHugh


Thanks to : Prof. Wes Wallender , Lab Instructor Paul Grant , & T.A. Robert Zomer