Towards achieving Sustainability of coastal environments: Urban Growth analysis and prediction of Lagos, State Nigeria

The most extensive urban growths in the next 30 years are expected to occur in developing countries. Lagos, Nigeria Africa’s second most populous megacityis a prime example. To achieve more sustainable and resilient cities, there is a need for modeling the urban growth patterns of major cities and analyzing their implications. In this study, the urban growth of Lagos state was modeled using the Multi-Layer Perceptron (MLP) neural network for the transition modeling and the Markov Chain analysis for the change prediction, achieving a model accuracy of 81.8%. An innovative visual validation of the model results using the ArcGIS was combined with kappa correlation statistics. The results show that by 2031, built-up areas will be the most spatially extensive LULC class in the study area with percentage coverage of 34.1% as opposed to 9% in 1986. The coverage of bare areas is also expected to increase by 53% between 2016 and 2031. Conversely, 24.9% and 68.3% loss of forestlands and wetlands respectively, are expected between 2016 and 2031. In view of the 11th goal of SDGs which focuses on achieving sustainable cities and communities, the objectives of African Union’s Agenda 2063, and based on the urban growth trends observed, the study recommends a prioritization of vertical expansion as opposed to the current horizontal urban growth trends in the study area.


Introduction
A century ago, only 20% of the global population resided in urban areas, but this is projected to reach 70% by the year 2050 (WHO 2010). Globally, the most extensive urban growths between 2014 and 2050 will occur in developing countries, and the leading country is Nigeria (United Nations 2014). It currently experiences the 3 rd highest urban growth rate in the world and urban dwellers are projected to reach 212 million by 2050 (United Nations 2014)-a double-fold increase from slightly over 100 million in 2019 (Worldometers 2019).
Coastal cities are of greater concern due to their significant roles as hubs of international commerce. Hence, coastal cities in many parts of the world experience a relatively higher population and urban growth rates than the national average trends (Wong et al. 2014). Three of the megacities (with population > 10 million) to emerge in Africa by 2030, -Lagos, Luanda, and Dar es salam -, are coastal (UNDESA 2018). The population increase experienced by these coastal cities, coupled with climate change-related events aggravate coastal challenges like seawater intrusion and groundwater contamination (Ezekiel et al. 2016, Idowu et al. 2017, coastal flash floods (Ajibade et al. 2013), urban sprawls (Li et al. 2016), land subsidence (Showstack 2014) amongst others. These challenges necessitate deliberate and proactive management plans for sustainable development. The Sustainable Development Goals (SDGs) 11 emphasizes the need for proactive actions towards ensuring "Sustainable Cities and Communities by making cities safe, inclusive, resilient, and sustainable" (UN 2017). One of the pivotal steps towards achieving this target is the generation of empirical knowledge on spatial changes in land use land cover and the observation of the transformation of these cities (McPhearson et al. 2016).
The number of land cover predictive models has increased rapidly in the last two decades because they have been found valuable for urban planning and environmental management (Kumar et al. 2015, Giri 2012. The predictive models aid a better understanding of the complex systems and invariably support land use planning and policy decisions (Megahed et al. 2015). Verburg et al., (2004) highlights the six important concepts in land use modeling-Level of analysis; Level of integration; Spatial interaction and neighborhood driving forces; Cross-scale dynamics; Temporal dynamics; and driving forces. The characteristics of some of the models are also reported in Verburg, et al., (2004) and Table 1 highlights recently used LULC change prediction models and their applications. This paper combined the modeling capabilities of the Land Change Modeler (LCM) in IDRISI with the spatial capabilities of the ArcGIS software to assess the urban growth trend in the study area.
The Land Change Modeler (LCM) is a component of an ensemble software platform -IDRISI (Eastman 2012). IDRISI Selva (version 17), possesses faster processes and additional tools in comparison to previous versions of the software (Clark Labs 2012). This paper 1) explores the effectiveness of a loose coupling of LCM with the ArcGIS for predicting the urban growth change of coastal cities and links the results with sustainable development goals, and 2) identifies some current challenges inherent in the use of the LCM prediction model and areas requiring improvements.
Integrating the LCM and ArcGIS platforms has two advantages; 1) the LCM can also be utilized as an extension to ArcGIS and 2) ArcGIS has the capability to summarize LULC areas based on the number of pixels, generate tables and export to excel which is not currently achievable with the LCM in IDRISI.

Study area
Lagos is a coastal megacity located slightly above the equator in southwest Nigeria ( Fig. 1) and is one of three megacities in Africa alongside Cairo and Kinshasa (UNDESA 2018). It is currently the second most populous city on the continent (after Cairo) with a population growth rate of 6 -8% per annum which is more than twice the national average of 2.6% per annum (UNDESA 2018, The World Bank 2019). It is a low lying coastal plain, characterized by a flat topography with notable geographical features such as; the mainland, lagoon, creeks, wetlands, seaports, several small Islands, and reclaimed lands such as the Banana Island and Eco Atlantic City. Although it is the most populous state in Nigeria, its 3,577km 2 land area is only about 0.4% of the country's landmass (MPPUD 2016).
Urbanization expressed as the outward expansion of built-up areas and the conversion of primary agricultural and forestlands into industrial and residential uses (Opoko et al., 2014), and land reclamation activities are visibly evident in Lagos Nigeria (Idowu & Home 2015).

Methods
An object-based image analysis and post-classification comparison were performed on the multitemporal datasets for the three years -1986, 2001, and 2016-to obtain the LULC maps of each year (Idowu et al. 2020). The maps contain six LULC classes-bare areas, built-up areas, forestlands, shrublands, waterbodies, and wetlands, according to the CORINE classification nomenclature as described in an earlier study (Idowu et al. 2020). The TIFF formats of the 1986 and 2001 LULC maps were subsequently imported into IDRISI where they were first converted into RST before uploading on the Land Change Modeler. The comparative LULC changes were analyzed and subsequently, the transition potentials and sub-models were generated, forming the basis for creating the predicted 2016 LULC by the model. The model's accuracy was assessed by validating the predicted 2016 with the actual 2016 LULC map. Once validated, the model was used to predict urban growth by the year 2031. Fig. 2 gives an overview of the steps taken in the modelling process.

Change analysis
The changes being investigated include gains and losses in each LULC class, the net changes in each class, and the contributors to the changes. At this stage, the analyzed changes in the LULC between the 1986 and 2001 maps helped identify transitions from one land cover type to another. The process was crucial in identifying the dominant transitions contributing to urbanization which formed the basis for the resulting transition sub-models created. It was observed in previous studies that one of the major drivers of land-use change in Lagos, Nigeria is accessibility (Braimoh & Onishi 2007).
Hence, ancillary data -DEM and road layers -were entered to account for the impact of topography and road development on possible urban growth. These data act as structures and factors for guiding the creation of the transition sub-models.

Transition potential modeling and driving forces determination
The transition potential modeling helps to determine the location of the change, and it results in the generation of transition potential maps (Eastman 2012

Change prediction
The predicted change in each transition was modeled using the Markov chain analysis.
Markov Chain is a random procedure that measures the expected transitions to the predicted date based on the projections of the transition potentials (Mishra et al. 2014). The two basic models of change-the hard and soft prediction models were obtained. The hard prediction yields LULC maps with discrete values and similar legends to the input maps while the soft produces a vulnerability map with continuous pixel values of probability of change from 0 to 1.

Model validation/accuracy assessment
The two broad types of validation are visual and statistical validations. The VALIDATE module in the LCM, a visual validation process was used to analyze the degree of agreement between the reference (actual) and the predicted 2016 maps by generating a map of correctness and error.
Statistical validation is based on Kappa correlation statistics. These statistics include Kno, Klocation, and Kstandard, where Kno reflects the overall accuracy of the simulation run; Klocation measures the level of agreement of location and Kstandard is the ratio of the proportion assigned correctly to the proportion which is correct by chance (Tewolde & Cabral 2011, Megahed et al. 2015. Some predictive studies employed the statistical approach only (Feng et al. 2016, Hamdy et al. 2016, while some utilized the visual validation method only (Mahmoud et al. 2016). Few studies adopted both approaches but were not elaborate with the detailed steps taken (Megahed et al. 2015). This study combined both and went further by using the ArcGIS to analyze the map of correctness and error to determine the Hits, Null success (persistence), Misses, and False alarms. Hits and persistence represent the correctness of the predicted map while the misses and false alarms represent the discrepancy between the predicted map and the actual map (Megahed et al. 2015). Once the visual and statistical validations were found satisfactory, the model was used to predict the LULC map of 2031.

Results
The results of the hard prediction which yields discrete values like the LULC input maps show that from 2016 to 2031, the LULC is likely to undergo drastic changes towards massive urbanization and deforestation (Table 2). Increases in bare areas which may be due to land reclamation activities and built-up areas are direct indicators of urbanization. On the other hand, the losses in forestland and gradual increase in shrublands -areas dominated by low lying plants with few scattered trees such as agricultural lands and developing areas-are pointers to deforestation.
Over 30% (>28,000ha) and 50% (>17,000ha) increases in built-up and bare areas, respectively are expected between 2016 and 2031 (Table 2). There may be wetland losses of almost 70% (>600ha), while further deforestation (over 35,000ha loss) may occur in the same period. The waterbodies in the study area mainly comprise the lagoon and the creeks found in different parts of Lagos state. The progressive losses (although <1%) over the years can be attributed to the land reclamation activities and widespread floating informal settlements sprawled on these water bodies. The LULC in 2016 and the predicted LULC map by the year 2031 are represented in Fig. 3.
The prediction results indicate that by 2031, built-up areas will likely have surpassed forestlands and waterbodies as the most extensive LULC coverage in the study area at 121,918.5ha (about 34%).
To put that in perspective, built-up area coverage was only 9% (32,200ha) of the total land cover in 1986. In contrast, massive deforestation (>35,000ha loss) may occur in the same period with a decrease from 143,848.8ha (40%) to 108,086.9 (30%). When Fig. 3 is compared with actual administrative maps, the projected urban growth trend are evident in areas such as the Badagry axis; areas along the Lagos -Badagry express road including Agbara; Ikorodu / Ibese axis and areas along the Ikorodu -Epe road; areas along Lekki -Epe express road; and the Lekki lagoon's environs.
Therefore, as with the changes observed between 1986 and 2016, the model predicts that between 2016 and 2031 there will be net gains in shrublands, built-up and bare areas, massive losses of forestlands and wetlands, and slight losses in waterbody areas (Fig. 4). The most significant gains and losses between 1986 and 2016 were built-up areas and forestlands, respectively (Fig. 4a) and this trend is likely to continue (Fig. 4b) but the rapid losses in forestlands to other LULC classes will contribute to the emergence of built-up areas as the most spatially extensive land cover type by 2031 (Table 2). Built-up and bare area coverage may give a combined coverage of 126,986.6ha which is about 35.5% of the entire LULC of the study area. Fig. 5 shows the contributions to the transitions into each LULC class with Fig. 5a showing that the losses of forestlands will largely be to built-up areas, followed by shrublands and bare areas. These gains in built-up areas, shrublands, and bare areas at the expense of the forestland are clear indicators of rapid urbanization and it is also reflected in Fig. 5c where the main contributor to built-up areas are forestlands, shrublands, and bare areas. Also, the possible losses in waterbodies are due to an increase in built-up and bare areas (Fig. 5d).

Transition potential sub-model and model validation
An overall Cramer's V of 0.53 and 0.55 were obtained for 1986 -2001 and 2001 -2016, respectively. Cramer's V of over 0.4 is considered highly suitable (Eastman 2012

Discussion
Urban growth modeling in Lagos state is only just evolving. The earliest study on urban growth modeling in the study area was focused on the viability of cellular automata (CA) for simulating urban growth (Barredo et al. 2004). The more recent urban growth modeling and prediction study by Eyoh et al., (2012)  Finally, while carrying out the modeling exercise, the major challenge observed in the IDRSSIbased LCM is its limited compatibility, as it only accepts TIFF raster files, therefore, the interaction between the stand-alone software and other software platforms is not yet seamless and needs improvement. Overall, the high model accuracy obtained in this study demonstrates the effectiveness of the loose coupling of LCM and ArcGIS while the findings shows the strong need for sustainable land use management to forestall further deforestation, losses of wetlands, and the effective management of the health and livelihoods of the burgeoning population.

Conclusion and recommendations
This study focused on the urban growth prediction in the study area by the year 2031. Between 1986 and 2016, there were net increases in bare areas, built-up areas, and shrublands, and a general decline in forestlands, waterbodies, and wetlands. The urban growth modeling shows that these trends are likely to continue into 2031 except drastic and proactive measures are taken. According to the analyses, bare and built-up areas will increase by 1754.8ha (53%) and 28,805.3ha (31%), respectively between 2016 and 2031. On the other hand, a wetland loss of 608.1ha (68.3% loss) should be expected. These do not only have socioeconomic implications but also ecological. Also, by achieving high model accuracy (>80%) and kappa statistics (>75%), the study demonstrates the effectiveness of loosely coupling the LCM (using the Multi-Layer Perceptron (MLP) neural network for the transition modeling and the Markov Chain analysis for the change prediction) with the ArcGIS for LULC Map statistics estimation. However, the quality and reliability of the outputs of current models can still be improved by introducing more input data. Nevertheless, since this study provides insights on the trends in urbanization, it will contribute to the achievement of goal 11 of SDGs which focuses on achieving sustainable cities and communities. At the continental level, the achievement of African Union's Agenda 2063-a 50year development plan geared towards creating a dependent and prosperous Africa (African Union Commission 2015)-will be hinged on timely and accurate predictive information data. Hence, this study's findings can be incorporated while making infrastructural plans for the next decade as it provides pointers to where possible expansions will be concentrated. Finally, to curtail the potentially massive deforestation (>35,000ha), which is largely at the expense of the gains in built-up areas, it is recommended that vertical expansion in the form of high-rise green buildings be prioritized as against the horizontal urban growth and urban sprawl trend currently being experienced.

Acknowledgments
The authors acknowledge the Regional Centre for Mapping of Resources for Development (RCMRD) for providing access to the software packages and data used for this study. All thanks to Oluwseun Somefun, Niyi Glory Bass, and Margaret Abiola for their immense assistance all through the data collection and ground-truthing phase of the project. Finally, the authors appreciate the staff and management of Jomo Kenyatta University of Agriculture and Technology (JKUAT) and RCMRD for providing the platforms to present and improve the work.