According to United Nations forecasts, the population of Africa will double over the next 40 years:from about 1 billion in 2010, Africa is expected to reach 2 billion inhabitants in 2045 (1). Much ofthis anticipated population growth is expected to be concentrated in urban areas, with almost 720million more people living in cities over the same period (2, 3). Such an increase in urbanpopulation combined with an overall decrease in average urban population density will lead tohigher urban expansion rates in Africa than other regions of the world. Some scenarios forecasturban land cover growth of more than 12-fold between the years 2000 and 2050 in sub-SaharanAfrica (4). These expected urban expansion rates will induce major changes in human populationdistribution, with important consequences on the environment, health and socio-economicdevelopment. In particular, urbanization in sub-Saharan Africa is characterized by an increasingproportion of poor people living in urban areas (5). In 15 years, the number of people living underslum conditions, in this region, was doubled and reach nearly 200 million people (70% of the urbanpopulation) in 2010 (6). Even if urban health and well-being have the potential to be much betterthan that of rural areas – thanks to improved access to healthcare and sanitation – this is usuallyonly true for the urban rich and it increases the gap between the urban rich and the urban poor (5).In addition, the poor often settle in marginal land, making them particularly vulnerable to naturaldisasters such as floods and landslides (7–9).
Knowing where people are and where they will habit in the future is essential to prepare for futuresocio-economic, environmental and health changes associated with urban growth and expansion(10). However, our knowledge of human population counts and distribution in Africa remainssurprisingly poor (11). Spatially detailed population databases for the present and the future arecrucial for any study that depends on a human population denominator, typically for calculatingpopulations at risk of human or natural disasters (12–14), for assessing vulnerabilities (15, 16) or toderive health or development indicators (17–20). A growing interest in the global mapping ofhuman populations emerged in the 1990s (21, 22) and led to the development of sophisticatedmethodologies to spatially disaggregate human population census data from large and irregularadministrative units to pixels of progressively finer size (17). The AfriPop project was initiated in2009 with the objective to provide spatially detailed and open access demographic datasets tosupport development and health applications in Africa (www.worldpop.org.uk) and is currently themost detailed and accurate population database available for Africa (23). However, the productionof human population distribution datasets for the future remains limited by the lack of spatial urbanexpansion forecasts that will drive major population distribution changes.
Spatial and temporal dynamics of cities are complex and controlled by various driving forces, suchas the regional economy, population movements, the policy environment and socio-culturalprocesses. The modelling of urban systems has occupied researchers for decades and anabundant literature has emerged focusing on industrialized countries (e.g. (24–31)). Only fewstudies analysed and simulated the urban expansion of African cities, including Accra (32, 33),Lagos (34, 35), Nairobi (36) and Tunis (37). Besides the models that focus on individual cities,some studies have tried to model urban expansion nationally (38, 39) or even globally (40, 41) atcoarse spatial resolutions (ranging from 1 to 10 km). A recent modelling approach was developedto simulate Africa-wide urban expansion at a moderate spatial resolution (100 m) and lead topromising results, though suffering from limited available training datasets (42).
Besides the prediction of human population distributions in the future, more detailed analyses ofurban areas can also greatly improve our knowledge of current population distributions in Africa, asthe main limitations of population databases generally concern the large urban areas. Because theAfripop disaggregation method only considers one urban class, the same population density isassigned to all urban pixels within an administrative unit, thus ignoring any intra-urban variations inpopulation density. This shortcoming has been highlighted as the main weakness of the AfriPopdatabase (43). Estimations of within-city variations in population density must be improvedin order to take urban heterogeneities into account and better help urban research anddecision making.
We have highlighted two limitations of population distribution datasets related to urban areas: (i)the lack of urban expansion forecasts and (ii) the homogenous distribution of people within cities.Both limitations can be significantly improved with remote sensing data and techniques. Satelliteremote sensing offers an effective solution for mapping settlements and monitoring urbanization ata range of spatial and temporal scales and allows the link to empirical observations with urbantheory. The built environment (i.e. non-vegetative, human constructed elements, includingbuildings, roads, etc.) is easily detectable using satellite imagery of different spatial resolutions (4,44–46) and it provides a more precise, consistent and comparable definition of an urban area thannotions such as population thresholds or administrative boundaries (45, 46). Here the urban extentis defined as the Morphological Urban Area (MUA), which is the densely populated area wheremost of the economic and/or cultural activities occur and all the important transportationconnections to other cities take place. It is usually defined as a contiguous area with highpopulation density or built-up (47). Moreover, remote sensing data have a great potential to mapintra-urban variations in population density. Population densities have long been estimated usingremote sensing data, particularly at large spatial scales using Landsat, SPOT and night-timeimageries (see (48) for a review). However, for applications at relatively small geographic extent,very high spatial resolution data ( < 5 m) are preferred (49) as they allow the counting of dwellingunits or the interpretation of residential land use types. To date, little research has explored thepotentialities of integrating data from various sensors for population mapping (50). One exceptionis the use of airborne Light Detection and Ranging (LiDAR) for deriving vertical informationcombined with optical remote sensors for urban studies (51).
Synthetic aperture radar (SAR) technology, being indifferent to weather and illumination conditions,ensures permanent urban monitoring (52) and demonstrates its suitability for operational urbanmapping using texture information extracted from SAR data (53). However, the relationshipbetween the radar signal and the built environment is complex, varying with the configuration of thesensor, the nature of the target, the environmental conditions and the employed data processingtechniques (54). Several researches have attempted to explore the joint-use of SAR and opticalsensors for urban delineation, characterization and monitoring (55, 56). In an overview of theliterature covering urban remote sensing and data fusion issues, Gamba et al. (53), cited anddiscussed the main studies that exploited the synergism of multiple sensor data in general andSAR and optical sensors in particular. Classification or delineation of African landscape usingsingle SAR or optical image, might produce inaccuracy due to the similarity of the man-madematerials and the natural environment or due to the heterogeneity of the build-up structures. Overcoming these shortcomings by fusing optical and SAR data was poorly studied and in theliterature we are finding only one attempt for classifying Eastern African land covers with moderatespatial resolution (57).
The advent, over the last few years, of a third generation of very high spatial resolution (<5 m) SAR(i.e. Radarsat-2, CosmoSkymed, TerraSAR-X) and optical satellite sensors (i.e. Quickbird, Ikonos,SPOT-5) stimulated the development of urban remote sensing still further (52). The data producedby these satellites facilitate improved detection of subtle urban changes (58) and rapid expandingagglomerations and ‘edge cities’ of many developing countries (59). It can also allow easierdiscrimination of the typology of urban landscapes especially in dense and heterogeneous cities.