Organised Editing/Activities/UBOS Pilot Census Preparation Mapping

From OpenStreetMap Wiki
Jump to navigation Jump to search


With Uganda’s national census scheduled to take place in 2024, the country’s national statistics agency - the Uganda Bureau of Statistics (UBOS) - is responsible for planning and implementing a multi-district pilot census activity to determine their methodologies’ practicality, effectiveness and feasibility. UBOS is currently planning to implement a household survey as part of the pilot census in four districts across Uganda - Kapelebyong, Butambala, Amuria and Kitagwenda. The purpose of the pilot is to experiment with and improve data collection tools, methodologies and workflows used in previous censuses to ensure the final approach can gather, analyze and visualize household data more efficiently and effectively. In more detail, the census pilot aims to do the following:

  • Estimate the amount of resources (people and time) that will be required to implement the national data collection activities
  • Assess whether or not new enumeration areas will be needed to speed up the field data collection process
  • Determine the most effective data collection methodology (sampling versus full range coverage) to deploy for the national census.

In early 2020, HOT received a request from the Manager of Geo information Services at UBOS for support in acquiring building footprints dataset of four districts in Uganda; these datasets would provide a blueprint for parts of the country that have not been previously mapped and help guide the initial planning of the pilot census activity to take place in mid-2020. Providing some additional context, in mid-2019, Microsoft released country-wide building footprint datasets for Uganda and Tanzania in an effort to enable (open) mapping to support disaster and humanitarian responses. These datasets represented some of the first openly available country-wide building datasets in Africa.

With these new datasets and AI-specific tools (RapID) being available for the first time in Africa and UBOS’ specific request for their national pilot census activities, HOT began developing a working methodology that would harness the potential of Facebook’s AI platform combined with Microsoft's buildings to generate an updated building footprint layer that could be leveraged for an innovative approach to national census planning.


Contact information Name: Kiggudde Deogratius
Hastag used #MapwithAI #Uganda #UBOScensus
Length of activity February - May 2020


The team responsible for implementing this activity include HOT staff in Uganda

OSM Username Role
kiggudde Validation
Allan Mbabani Validator
Shamillah Validator
Beza Mapper
Micheal Yani Mapper
Bakole Mapper

Tools and data sources

Tools and their use

  • JOSM is used to map buildings structures that were not identified by the AI tool, remove structures that no longer exist and align the new Maxar imagery to the Bing imagery. The JOSM tool is also going to be used to validate the mapped data through using the todo list where each mapped feature will be manually inspected by the mapper before uploading to OpenStreetMap. JOSM will also be used to improve the shape of the buildings traced by the AI to reflect the actual footprint of the building structure. This will therefore improve the overall accuracy and representation of the data on OpenStreetMap.
  • RapiD is used to map building features identified by AI and uploaded to OpenStreetMap.
  • MapwithAI plugin is used to assist with mapping building features to be identified by AI in OpenStreetMap. This plugin will be used as a substitute for mappers that prefer to use JOSM to carry out their mapping activities
  • QGIS is used by mappers to split the AI buildings into manageable proportions along village administrative boundaries.

Data sources

  • Imagery to be used: Bing, Maxar
  • Microsoft AI  buildings: These building features will be used to assist with identifying where building structures exist.

Measuring our Success

HOT plans to map approximately 100,000 buildings across the three districts of Amuria, Butambala and Kitagwenda. As part of the broader activity, HOT will also provide training to GIS staff at UBOS on how to download the newly created building footprints from OpenStreetMap as well as how RapID and MapwithAI tools can be used for similar activities. From early discussions, the building footprint data layer will be used by UBOS staff to create map products that will support and inform the planning of the pilot census exercise. This, in itself, represents a successful indicator for HOT as a government institution is using data from OSM to support national census activities.


Through the duration of this activity, three training sessions will be carried out with mappers on importing/mapping building footprints across the AOI. The first training, which was held in February 2020, provided mappers an overview of the activity workflow and a demonstration on how different tools would be used to add Microsoft-generated buildings to OSM. During the second training in March 2020, mappers received refresher training on the tools and prescribed workflow; feedback was gathered from mappers on how they were each progressing and in what ways the existing methodology for adding new buildings to OSM could be improved. The third and final session will focus on wrapping up the activity and providing mappers an opportunity to discuss and rate the methodology used to determine how this approach can be modified and improved in future instances without jeopardizing the process or data quality.

Remote mapping


Before remote mapping can begin, specific data preparation activities will be carried out. First, download the Microsoft AI buildings and then split the buildings along village-level administrative boundaries. Once this is done, mappers can be assigned a particular set of villages, where s/he will be adding buildings features.

For this activity, 7 remote mapping tasks were planned to be set up on HOT’s Tasking Manager to assist with monitoring mapping progress across the districts. Depending on the size of the district, some of them will be split along sub-county level administrative boundaries, to enable faster mapping and reduce ‘mapping fatigue’ whereby mappers’ mapping rate slows after time due to working on one task for a significant amount of time. These remote mapping tasks were tailored for intermediate and/or advanced mappers only. With that said, strict supervision and monitoring protocols will be applied throughout the activity to ensure high quality mapping.

Once remote mapping tasks are set up, mappers are instructed to navigate the village task assigned to them and download it to his/her JOSM application. Once downloaded in JOSM, the mapper will add the (identified) village boundary to allow him/her to be sure of the mapping limits. Then, the mapper will add the Microsoft AI buildings for that specific village into JOSM as a new layer.

After loading Microsoft AI buildings into JOSM, the mapper must activate the Bing imagery layer. Since Microsoft AI buildings were largely generated using Bing imagery, it is important to use the same imagery to ensure building structures are properly aligned. From this point on, the mapper must assess each building feature individually using the todolist plugin in JOSM to ensure buildings are aligned with Bing imagery. Once this process is complete, the mapper will be required to activate the Maxar imagery service in JOSM. Maxar, in most areas in Uganda, is newer than Bing imagery (2015-2018 for Maxar versus 2011-2013 for Bing). From this point, the mapper will map new and existing buildings structures as well as remove building structures that no longer exist by aligning the Maxar imagery with Bing imagery and then begin mapping with the appropriate tags - building=yes, source=Microsoft/BuildingFootprints.

After completing this process, the mapper will apply a second assessment using the todolist plugin for all features added to the layer. The mapper will then run the validation process to identify overlapping, fix wrongly tagged or poorly traced building features. If any errors are identified, the mapper will work to resolve these issues before uploading them to OSM and moving to the next task.

Validating the map

For this activity, validation is carried out by two experienced validators/trainers. Validators will be tasked with assessing each village task individually for potential errors. For simple errors, the validator will edit and improve the features directly. If numerous errors are found, the village task will be in-validated and the responsible mapper will be asked to re-map the village task. Validators will be required to validate both building features across the activity AOI.


The building footprints mapped in OpenStreetMap will be summarized on the Organised Editing Guidelines page on the OSM Wiki once mapping activities are completed. Upon completion, the data will be shared with UBOS and a series of training sessions will be scheduled with the national statistics agency to ensure their adequate access and use of the data in OSM.