(Scaling up): miscalculation of population covered by health centres

rdoctors commented 5 years ago

Current Behavior

I'm using AccessMod version 5.3.2 to estimate how many health centres would be necessary to cover x% of the population in Malawi.

I'm not using any existing health centres, so I would assume that the start coverage would be very close to 0% (as commented by Steeve). In my results, however, I find that the initial coverage is of 17%. I've tried varying the parameters with no success.

In addition to this, when running the scaling up analysis for a population coverage smaller than 17% (such as 5% or 10%), I get no results of the type "table_result_scaling_upanalysis*", as AccessMod understands that the population is being covered, and thus no additional health facilities are needed.

Expected Behavior

As mentioned, I would expect to find an initial population coverage close to 0%, but the simulation starts at 17%, only adding thus new facilities for population coverages that are higher than 17%.

Possible Solution

Steps to Reproduce

The data that I'm using can be found here: https://www.dropbox.com/sh/6qmqdrvwypvi7nz/AAB_SroOw1_X4fJA37ELavrca?dl=0

In the simulations for the Scaling up analysis I'm using the following parameters:

"Select population layer" equal to "Select residual population layer"
Option for the output layer: Start with empty layer
Type of analysis: anisotropic
Direction of travel: toward facilities
Maximum travel time: 120min
Number of new health facilities to locate: nothing added
Maximum processing time: nothing added

Detailed Description

SteeveEbener commented 5 years ago

Hi Fred,

Just a small adjustment to Rebeca's message for having looked at the issue over the past days, without finding any issues in the data: The expected coverage after the first iteration should be around 0.0006% (population covered = 10,000, total population in the study area=15,735,870) and not 0.

Thanks in advance for your help on this.

Steeve

fxi commented 5 years ago

@rdoctors thanks for reporting this.

When starting with an empty set of facilities with no population on barrier or outside the current project extent, the population covered should indeed be 0% at the beginning of the scaling up.

We will look at this as soon as possible. I let @nicolasray organize the follow-up and set a priority.

Thanks again.

fxi commented 5 years ago

During unrelated tests, I've just seen that your population resolution (~908 ~908 [m]) did not match the resolution of the project (1000 1000 [m]), other parameters seems compatible.

	Resolution ( m )	Projection ( proj 4 )
Project before importation	1000 x 1000	+proj=utm +no_defs +zone=36 +a=6378137 +rf=298.257223563 +towgs84=0.000,0.000,0.000 +to_meter=1
Project after importation	1000 x 1000	+proj=utm +no_defs +zone=36 +a=6378137 +rf=298.257223563 +towgs84=0.000,0.000,0.000 +to_meter=1
Imported dataset	908.8702 x 908.8702	+proj=utm +zone=36 +datum=WGS84 +units=m +no_defs +ellps=WGS84 +towgs84=0,0,0

Then, it appears that the population was not corrected:

@nicolasray It could be a starting point for investigating this, no ?

SteeveEbener commented 5 years ago

Hi Fred. I missed that one during my tests.

This brings us back to issue #200 => It would be nice to raise the priority level of this issue.

Thanks in advance and have a nice weekend.

syzeyna commented 5 years ago

Hello,

Could you indicate which capacity column (in the shapefile of the facilities) you used during the geographic coverage that generated the residual population?

Zeyna

rdoctors commented 5 years ago

Hello Zeyna,

I'm not using the geographic coverage tool (the residual population is thus equal to the total population in the scaling up analysis). I am, however, using a capacity of 10,000 for the new facilities to be added in the scaling up analysis.

Rebeca

syzeyna commented 5 years ago

Hi @rdoctors can you share the administrative boundary of your study area please. It will help for the correction of the population

rdoctors commented 5 years ago

Hi @rdoctors can you share the administrative boundary of your study area please. It will help for the correction of the population

Hi @syzeyna, could you give me a contact where I could send it to you? I'm asking you this because this data is confidential and I was asked not to share it (although @SteeveEbener already had a look at it).

just an update, I have the census limits, not the administrative boundary. I'll send them to you.

SteeveEbener commented 5 years ago

@syzeyna,Would the country boundaries be enough as Rebecca's study area is covering all of it?

syzeyna commented 5 years ago

@rdoctors , Hi Rebecca, I did the test with the population data that I re-sampled at 1000m resolution like the DEM and made the correction (population on barriers). I have the following result: 0.077% for the first iteration.

Best regards

rdoctors commented 5 years ago

@rdoctors , Hi Rebecca, I did the test with the population data that I re-sampled at 1000m resolution like the DEM and made the correction (population on barriers). I have the following result: 0.077% for the first iteration.

Best regards

@syzeyna thank you very much for your reply and help on this. Could you let me know which resampling method you used and how you made this correction?

syzeyna commented 5 years ago

Hi @rdoctors , I did the resample with the nearest neighbour method. In your case, once you have resampled your population layer at 1000m, you must adjust it so that the population in your layer ~ 908m is equal to the population in your layer at 1000m.

After the adjustment, you can make the correction in Accessmod using the module: Correct for population on barriers (in Toolbox). For that you will need your landcover merge and the administrative limit (at the most precise level possible)

Zeyna

SteeveEbener commented 5 years ago

Hi @syzeyna

Was the resolution problem specific to this dataset or to a problem in AccessMod? Did you do some test with other datasets presenting a different resolution (not only population but also landcover) to see if the problem persists?

Is the difference in resolution the only thing that explains the significant difference in results or is there other problems in AccessMod that needs to be addressed?

Thanks in advance for your answer on the above.

@nicolasray: this is a critical issue that needs to be addressed rapidly if it is not dataset specific => I would like once again to suggest for issue #200 to pass from "enhancement" to a high level of priority. Many thanks in advance for your consideration in this regards.

unige-geohealth / accessmod