Building type classification in Mozambique using mobile phone data, high-res satellite images, night-time light data and digital surface model

According to the UN DESA report “World Population Prospects: The 2015 Revision”, The world population is expected to grow 33% by the year 2050. With the highest rate of population growth, Africa is expected to account for more than the half of the world’s population growth between 2015 and 2050. The study area presented in this paper is the Republic of Mozambique, an African country with 70% of its population of 28 million (2016) living and working in rural areas. The Real gross domestic product (GDP) of the country was 3.7% in 2017 shows it’s struggle of poor macroeconomic stability and investment of private sector.

High income countries often have extensive mapping resources and expertise to create reliable and accurate building maps and population databases, but across the low-income regions of the world, relevant data are either lacking or are of poor quality. For low-income regions of the world, accurate maps of human population distribution together with the knowledge of building types and its quantitative measures can play an essential part in planning for elections, calculating per-capita gross domestic product (GDP), poverty mapping, city planning, disaster management amongst countless other applications.

The rapid growth in availability of high resolution satellite imagery, computing power and expansion of geospatial analysis tools over the past decade are providing new opportunities to solve such problems. The use of high resolution images, geospatial data and road network together with state of the art machine learning technology can improve the understanding of human population distribution and building type estimation, which is necessary to predict the future infrastructure management for increasing population because it can be expanded to a bigger scale easily unlike the traditionally used method based on human visual interpretation and survey data collection.

In this paper, we proposed a methodology to classify the types of buildings in three classes; high- rise residential buildings, low-rise residential and non-residential buildings. We have used state- of-the-art machine learning algorithm on the combination of mobile phone sample data collected from a survey, high resolution satellite images, digital surface model and night time light data to extract the building footprints and classify the types of the buildings. The comparing results indicated that our methodology classified types of buildings efficiently with the accuracy of 84%.