Home Page

The low quality of the data that was available played a huge role in the effectiveness of the protocol. Ideal data should contain multiple bands with at least 2.5m or smaller spatial resolution. This is especially true for our object-based protocol, as the lack of bands made the identification of houses extremely difficult. Both rule based and example based object-based image analysis attempted within ENVI were unable to distinguish houses from other objects in the study area as seen in the figures above. While it is possible to obtain other data, all that is available is lacking in either spectral or spatial resolution. A formal request to companies that sell higher quality data was also considered, but due to time and monetary constraints acquisition was not possible for this project. However, the object-based protocol should still be in theory the most effective method for counting population within informal settlements assuming that ideal data is acquired. The use of other programs to make up for non-ideal data were also considered. Feature Analyst from ArcGIS has been determined to have superior ability to distinguish objects than ENVI (Tsai et al. 2012). Unfortunately, the version ArcMap that we had access to did not have the licensing associated with Feature Analyst.

Area sampling by manual count was the second method by which we attempted to estimate the informal population within the study area. Even though the method is relatively simple when compared to object-based analysis, there are still a couple of things that can still be improved to further increase its accuracy. Manual counts are more effective with better spectral and spatial resolution, as the likelihood of miscounting houses is reduced. For our analysis only thirty sampling areas were chosen due to limited time and manpower. In statistics, thirty is considered to be statistically powerful (Brown et al. 2001). However, it would be possible to outsource time and labour to the general public to produce an even more powerful sampling. The act of manually counting houses can be done by anyone even if they do not have specialist GIS knowledge.

The protocols outlined in this paper are not designed to be used independently without the application of other techniques. The object-based protocol utilizes algorithms to distinguish houses within the aerial photography, but determining the number of houses within the study area is not sufficient to estimate the total population. Survey data taken from the Cape Town website on the average persons per house ratio has to be used in conjunction with object-based analysis to obtain the total population within the Cape Town informal settlements. Similarly, the area sampling by manual count protocol requires that each house to be counted manually within the blocks before it can be extrapolated to the whole study area. Again, the census data from Cape Town website for average persons per house ratio was used to obtain the total informal population. It should be noted that population growth models were not included in the protocols because of the lack of historical Cape Town data, as well as the lack of manpower to perform trial and error on different available models. Population growth model models also fail to consider immigration, which is one of the major reasons for the growth of informal settlements. Pixel analysis is also not considered, as the primary strength for pixel analysis is to compute land cover and land use and there are no existing papers that use this method to calculate population estimates.

Ideally, the provided data should have a high enough spatial and spectral resolution to use an object-based protocol to estimate population. In our case, where object-based methods cannot be used effectively, it is possible to use an area sampling by manual count method instead. Since object-based methods were unable to give us any meaningful results, we are unable to compare our two protocols and examine any potential tradeoffs in accuracy. It is important to note that area sampling by manual count can be time consuming when applied to a larger area, but this limitation can potentially be solved by outsourcing the work to members of the public. With this in mind, future studies should attempt to evaluate the tradeoffs between object-based methods and area sampling by manual count when estimating the populations of informal settlements.

Brown, Vincent, Guy Jacquier, Denis Coulombier, Serge Balandine, François Belanger, and Dominique Legros. “Rapid Assessment of Population Size by Area Sampling in Disaster Situations.” Disasters 25, no. 2 (June 2001).

Tsai, Yu Hsin, Douglas Stow, and John Weeks. “Comparison of Object-Based Image Analysis Approaches to Mapping New Buildings in Accra, Ghana Using Multi-Temporal QuickBird Satellite Imagery.” International Journal of Remote Sensing 3, no. 12 (December 16, 2011): 2707–26.