Introduction

Urban areas are characterized by highly complex spatial dynamics, with their sizes increasing dramatically. The expansion of a city beyond its periphery requires population growth to be spatially distributed. Population growth contributes to urban change through the absolute growth, which increases urban areas and changes the dynamics of the urban demography (Qiu et al. 2003; Alsharif et al. 2015). Urban growth is influenced by a number of factors, including geographic, demographic, economic, social, environmental, and cultural aspects. Therefore, modeling a dynamic system such as this poses analytical challenges (Kashem 2008; Abdullahi and Pradhan 2015).

Spatial analysis of urban growth dynamics becomes very important in urban studies. Several authors addressed issues of urban growth, discussing urban environment, urban development, urban change detection, modeling, and management (Cihlar 2000; Wang et al. 2003; Páez and Scott 2004; Zhu et al. 2006; Geymen and Baz 2008; Hedblom and Soderstrom 2008; Abdullahi et al. 2015; Alsharif and Pradhan 2014, 2015a; Alsharif et al. 2015; Dadras et al. 2015). Advances in satellite-based land surface monitoring contribute to the development of considerably more detailed urban maps, which offers planners an improved deeper understanding of urban growth dynamics and associated matters relating to territorial management (NASA 2001). Remote sensing (RS) and geographic information system (GIS) techniques are useful geospatial tools widely used for assessing natural resources and monitoring spatial changes. Land use/land cover (LULC) change dynamics can be analyzed through the use of time-series remotely sensed data, linked with socioeconomic or biophysical data in GIS platforms to better explore causal relationships (Moeller 2004; Reveshty 2011). The integration of RS and GIS enables researchers to analyze environmental changes, including land cover mapping, change detection, monitoring, and identification of land use attributes and change hotspots. With the advancement in technology, reduction in data cost, and availability of historical spatial-temporal data and high-resolution satellite images, GIS and RS techniques are now useful research tools in spatial change and modeling (Feng 2009; Bayes and Raquib 2012; Abdullahi et al. 2015; Alsharif and Pradhan 2014; Alsharif et al. 2015). Specific to the analysis of urban growth, Batty and Howes (2001) reported that remote sensing technology provides a unique perspective on growth and land use change processes. Datasets obtained through remote sensing are consistent over large areas and time periods and provide information at different geographic scales. Remote sensing-derived information is very useful in describing and modeling the urban development process. This leads to better understanding, management, and planning (Banister et al. 1997; Longley and Mesev 2000; Longley et al. 2001; Yikalo and Cabral 2010; Abdullahi et al. 2015; Alsharif and Pradhan 2014; Alsharif et al. 2015; Dadras et al. 2015). Remote sensing data can be used to understand the nature of urban landscape change through time. This understanding includes (1) urban growth rate, (2) spatial pattern of the growth, (3) differences between the observed and forecasted growth, (4) spatial or temporal variance in growth, and (5) if growth is sprawling or not.

Urban growth modeling is receiving increasing attention as an emerging research area. This is due to the recent dramatic increase in urban populations, resulting in considerable pressure on infrastructure services. Among all developed urban growth models, cellular automata (CA) models have better performance in simulating urban development than conventional mathematical models (Batty and Xie 1994; Alsharif and Pradhan 2015b; Alsharif et al. 2015). During the past 20 years, a new generation of models has been developed, based on the assumption that an understanding of the details can explain the whole—i.e., a bottom-up approach. One of these models is CA. CA has demonstrated success in capturing complexity with simple rules. One of the most important considerations to make CA more realistic is to find the transition rules which represent the real leading forces. CA is based on a defined neighborhood, where every entity (in a two-dimensional space, represented by a cell) is interacting with the surrounding cells only. Thus, CA has been considered most suitable for processes where the immediate surroundings have an influence on the cell, such as diffusion processes. This includes processes of ecological dynamics (Parker et al. 2003; Alsharif and Pradhan 2015b; Alsharif et al. 2015). Transition probability refers to a grid consisting of cells, where cell states are 1 and 0, a neighborhood within which transition rules can apply, as well as a temporal space or time-step interval (Torrens 2000).

Despite all the achievements in CA urban growth modeling, the selection of the CA transition rules remains a research topic (Batty 1998; Abutaleb et al. 2013; Ngie et al. 2013). CA models are usually designed based on individual preference and application requirements, with transition rules being defined in an ad hoc manner (Li and Yeh 2003; Alsharif and Pradhan 2015b). Furthermore, calibration of CA models remains a challenge. Most of the developed CA models require intensive computation to select the best parameter values for accurate modeling. CA models effectively predict the allocation of urban growth quantity. CA spatially quantifies change based on a pre-defined transition rule starting with the cells that have the highest probability.

Egypt’s cities are growing rapidly and continue growing. With population growth rate of 2.2–3 % per year over the past decades, every 8 months Egypt has approximately one million new citizens to be housed. Population growth and migration from rural to urban areas increase the rate of population expansion to approximately 4 % annually in urban centers (World Bank 2013). Egyptian urbanization levels approached 56 % in 2001, a 4.3 % increase from 1996 (Kok and Collinson 2006).

Many studies have quantified and recorded the rapid urban growth in The Greater Cairo Metropolitan Area (Taubenböck et al. 2009; Darwish et al. 2007; Rashed et al. 2005; Azaz 2001), but no attention has been paid to the behavior and the spatial distribution of this increase. This study aims to apply an integrated approach, incorporating GIS, RS, and modeling techniques, to quantify and analyze the pattern of urban changes within The Greater Cairo Metropolitan Area from 1990 to 2010. Additionally, this study projects the probable future areas in 2050 to enable the anticipation of planning policies that aim to preserve the natural characteristics of the study area.

Material and methods

Study area

The Greater Cairo Metropolitan Area is a typical mega-city, with approximately 20 million inhabitants living in three municipalities (Cairo, Giza, and Qaliyãbáya) as illustrated in Table 1. It is the largest metropolitan area both in Egypt and on the Africa continent. The region is situated between 29.72°–30.43° N and 30.717°–31.883° E with a total land area of approximately 1,226,771 ha (Fig. 1). The topography of the region is predominantly flat, bounded by hills to the east and west. The Greater Cairo Metropolitan Area comprises 20 % of the country’s population; the population density of The Greater Cairo Metropolitan Area is 2361 people per square kilometer.

Table 1 population distribution per governorate (World Bank Indicators 2013)
Full size table
Fig. 1
figure 1

Landsat false color composite image and DEM of the study area

Full size image

The tremendous social and economic inequalities within Egypt and The Greater Cairo Metropolitan Area are reflected in the urban spatial structure. A complex urban structure now exists, which bears the imprint of different phases of development. Modern urban architecture with a post-modern style is located adjacent to century-old buildings, and inner city slums border directly on high-rise buildings with mirror-glass windows (Raymond 1993).

Data used

  1. 1.

    Landsat satellite images for three different dates (176–39) were used in this study to produce LULC maps for the study area. The first one was captured in 1990 (TM), the second one in 2010 (ETM+) and the third one in 2014.

  2. 2.

    The road network layer was digitized from the 2010 image

  3. 3.

    SRTM 90-m DEM data was used in modeling the changes in the study area. The data was resampled to 30 m in ArcGIS to match the spatial resolution of the classified images used in the modeling phase.

  4. 4.

    Field surveys were conducted between April 23 and May 20, 2014, to collect ground truth data for the study area.

Image processing

Landsat images were downloaded and preprocessed such as layer staked, gap filling for the 2010 scene. An object-based classification rule set was then developed for image classifications for all of the selected images. Figure 2 illustrates the full methodology developed in the study.

Fig. 2
figure 2

Methodological flow chart

Full size image

Change analysis and modeling

Land change analysis and modeling were undertaken in IDRISI Land Change Modeler (LCM), in a stepwise manner comprising (1) change analysis from 1990 to 2010, (2) transition potential modeling, in which the potential of land transition was identified, and (3) change prediction. Land change analyses rely on image classification, and so it is imperative that this is accurate. Accuracy assessment to some extent guarantees the accuracy of the image classification. Based on the image classification, change analyses were carried out to quantify land changes and then deduce land transition. However, it is important to identify the dominant transitions that can be grouped and modeled. Three main transitions were considered in the modeling process: (1) sand and rock to urban, (2) sand and rock to agriculture, and (3) agriculture to urban. Transitions 1 and 2 represent the legal activities in the study area, which are new urban cities proclaimed by the government and land reclamation projects for agriculture, while the third represents the illegal transformation from agriculture to urban land use. Only these three transitions were modeled using a multi-layer perceptron (MLP) neural network, with four explanatory variables controlling these transitions, namely, distance to roads, distance of existing urban areas, DEM, and vicinity of change (calculated as continuous quantitative maps). All the produced transition probability maps were fed to LCM in IDRISI, and CA algorithms were chosen in the prediction process.

Model validation

Model validation was carried by running the model to predict land cover changes in 2014, and the resultant land cover was compared with a 2014 Landsat 8 classified image. Then, the same model (i.e., the ssame transition probability maps) was run for the year 2050.

Results and discussions

Accuracy assessment

Accuracy assessment is crucial for modeling and model performance. Table 2a–c shows the accuracy assessments for the three images used. The overall accuracy ranges from 94.17 to 96.52 %. This is one of the advantages of object-based classification that enables more control on the image by dividing it into segments with similar characteristics to be classified more accurately than with pixel-based algorithms. One can put stress on the urban class accuracy, which is particularly good with results ranging from 93.62 to 95.04 %, guaranteeing strong transition rules in the modeling phase, contributing to a predication more closely related to ground-based observations.

Table 2 Accuracy assessment for the classification results with overall Kappa
Full size table

Image classification

Figure 3 below illustrates the results for image classification for the years 1990 and 2010. The images were classified into five different classes which are dominant in the study area. It is clear that there are considerable differences in the land use land cover between the two dates. During these 20 years, the difference between the two studied dates includes new urban areas developed such as (1) 6th of October in the west, (2) Nasir city on the south, (3) Al-Salam, and (4) 10th of Ramadan cities in the east. This is simultaneous with the expansion of the existing areas in the center of the study region.

Fig. 3
figure 3

LULC for the study area in 1990 and 2010

Full size image

Table 3 shows the absolute changes in each class size. Urban areas increased by 65,460 ha (3273 ha per annum), while the agricultural class also increased by 31,734 ha. This is unexpected, but could be explained by the agriculture reclamation projects which took place during these 20 years.

Table 3 Class area and differences in hectares
Full size table

For understanding the dynamics of the study area, Table 3 is insufficient as it only portrays the absolute change, without any details regarding the provenance of the losses and gains. Table 4 and Fig. 4 illustrate in more detail the dynamics of these changes.

Table 4 Gains and losses in each class in hectares
Full size table
Fig. 4
figure 4

Change analysis of the study area

Full size image

Figure 4a reflects that there are two main classes for gains in the study area: urban and agricultural, respectively. The main transition in the area is from sand and rock class into these two gaining classes. This amounts to 44,339 ha transformed to urban areas in the form of the newly developed cities as mentioned above and 53,617 ha to agriculture areas in the form of the new reclaimed area west and east of the Greater Cairo Municipal Area. Agriculture is, however, also a losing class (Fig. 4c), which is transformed into urban areas (21,113 ha). The spatial distribution of these changes is better portrayed in Figure 5 which demonstrates the gains and losses in different classes.

Fig. 5
figure 5

Spatial distribution for the main changes in the area

Full size image

Modeling the urban change

Transition probability model

Figure 6 presents the results of the transition model for the three main changes occurring in the study area. Transition probability is a very important step in the modeling process, and therefore, the accuracy is crucial for model performance. The resultant accuracy of the transition is 83.7 %, which is considered as satisfying accuracy (Pontius 2000, 2002; Pontius and Malanson 2005).

Fig. 6
figure 6

Transition probability maps

Full size image

Model validation

Figure 7 demonstrates the spatial distribution of the model validation results with three categories: hits, false alarms, and misses. Hits are the correctly predicted cases, false alarms are model change-predicted cases which in reality did not occur, and misses are cases in which the model predicted no change but in reality transition occurred. Figure 7 reveals that the model is capable of detecting changes within the center of the study area, which includes the expansion of the existing areas. However, in the new development areas, it over-estimated the amount of change which occurred. Quantitative measure of the model accuracy was calculated at 80.7 % which is considered as a good performing model (Pontius 2000, 2002).

Fig. 7
figure 7

Model validation spatial results

Full size image

Change predications

Figure 8 demonstrates (a) gain and losses per class, (b) contribution to net change in urban areas, and (c) spatial distribution of the predicted changes up to the year 2050. It is clear that the area will suffer from continued urban expansion at the expense of the agriculture land. However, the agriculture areas will also be increasing due to the reclamation projects in the area. Overall, urban threats to the agricultural land have more severe impacts, especially in the inter-urban area (the agricultural area located between urban areas). Further threats are anticipated for the newly reclaimed areas due to their close proximity to the existing urban centers, unless the government mitigates these threats. Table 5 illustrates quantitatively the predicted change during these 40 years. Agricultural lands are projected to lose 44,544 ha to urban areas (1100 ha per annum). Assuming that agriculture reclamation projects will continue at the same rate of the past 20 years, agriculture will gain 84,668 ha.

Fig. 8
figure 8

Predicted changes up to year 2050

Full size image
Table 5 Predicted gain and loss (hectares)
Full size table

Finally, Fig. 9 presents the predicted LULC map of The Greater Cairo Metropolitan Area for the year 2050. The eastern part of the study area appears more suitable for new urban areas development, while the western region is predicted to be transformed into agriculture areas. This may be attributed to the slope of the area, with flatter topography in the west which is suitable for agriculture expansion, which is further supported due to the vicinity of the River Nile.

Fig. 9
figure 9

Predicted LULC in the year 2050

Full size image

Conclusion

This study assesses and models the trend of urban land cover changes in The Greater Cairo Metropolitan Area using an integrated approach including GIS, RS, and modeling tools. The area experienced extensive conversion to urban land cover over the last 20 years, eliminating approximately 10 % of the agriculture area. Results indicate that urban growth will continue to expand further in the future, with critical impacts on land resources, unless some preservation mechanism is implemented. Agricultural land is expected to experience both gains and losses if the reclamation projects are to continue. Transformation of vegetation to urban areas was noticed to occur in the peripheries of urban communities, which overloads the existing facilities in the area. New reclamation projects in the vicinity of Grater Cairo, such those along the Ismailia desert road and 6th of October city, are threatened by the urban expansion.

Cellular automata are demonstrated as successful in capturing complexity with simple rules. However, there are many uncertainties with the technique, and more research is required to improve the adaptation for an urban context. Future work will consider model accuracy enhancement and socioeconomic and demographic data leading parameters in urban expansion, allowing for long-term accurate simulation.