Fine-scale Landscape Variability of Cotonou City, Benin: Insights From Three Contrasted Urban Neighborhoods

Urbanization consist in a complex and deep human-driven environmental change that strongly impacts the ecology and evolution of living organisms, including pathogens, reservoir and vector species hence human health. Quantitative proxies of urban landscapes may be very useful to sum-up such a complexity and to guide fundamental and applied research as well as urban planning programs. Geographic Information Systems (GIS) provide landscape and uses metrics which can be investigated through multivariate analyses, thus providing pertinent synthetic landscape descriptors. As such, our study describes the ne-scale modelling of three urban neighborhoods of Cotonou city, Benin, using GIS, landscape metrics and Principal Component Analysis (PCA). Spatial variability between and within neighborhoods revealed different levels of variability, with elements differentiating the three areas from each other, while local neighbourhood-specic variations were also evidenced. We found that Cotonou landscapes are strongly inuenced by their history, the natural environment in which they develop as well as the urban planning trajectories. This case study shows that PCA-analyzed of GIS-based metrics may be very relevant to describe and understand the variability of urban landscapes at different scales, thus constituting a valuable tool for urban management of African cities.


Introduction
Cities are booming, especially in Africa where planning policies hardly keep up with unbridled urban growth (UN-WUP 2018). This may result in the emergence of large unsanitary areas lacking adequate infrastructures (Acuto 2018) and/or of marginal spaces often located at the cities' outskirts (Halligey 2020). In such environments, where formal (e.g. roads, formally subdivided neighborhoods with hard-built houses, etc.) and informal (e.g. slums with precarious houses illegally built on sites that are sometimes unsuitable for settlement) settings may coexist, social, economic and health inequalities are glaring (Linard et al. 2012).
Urbanization represents a complex, extreme human-driven environmental change that strongly impacts the ecology and evolution of living organisms, hence biological interaction networks, through deep modi cations of land use (Sun et al. 2020a), air quality (Liang and Gong 2020), hydrographic networks and available resources (reviewed in Rivkin et al. 2019). Some of the urban-dwelling species may be vectors or reservoirs of infectious agents; consequently, urban-associated eco-evolutionary changes may have important consequences in terms of human health, for example through increased human/animal interactions (Hassell et al. 2017; Dobigny and Morand, submitted).
The possible modeling and quanti cation of the landscape through Geographic Information Systems and associated mathematical treatments offer unique opportunities to rigorously characterize, and therefore to guide the arrangement of landscapes towards sustainable city building and management (Hu and Zhang 2020). Within the two last decades, these approaches have been undergoing signi cant conceptual and methodological development, together with the implementation of new tools that permit quicker processing of increasingly massive datasets (Dumas et al. 2008 (Turner et al. 2001) provide numerical data describing the composition and physiognomy of landscapes and are well suited for even deeply modi ed environments such as the urban ones (Turner et al. 2001; Rossi et al. 2018). However, such approaches have rarely, if ever, been used to study intra-urban landscape variability in Africa.
Cotonou is a coastal city subject to seasonal ooding, and displays a high human density (8,593 hbts/km 2 in 2013); its extant population reaches 1,23 million inhabitants, a value that is expected to reach >1,6 million in 2030 (INSAE 2018). It is characterized by an overall low household economic situation, rather poor infrastructures as well as inadequate management of solid waste. In many areas, these features altogether contribute to degraded socio-environmental conditions which may be associated with important infectious risks (e.g. cholera and other diarrheal diseases, malaria, leptospirosis, etc.). The habitat in which certain poor neighborhoods of Cotonou have been sprawling (e.g. seasonally oodable lowlands) may further exacerbates the precariousness of the population and reinforces inequalities in terms of environmental exposome, hence welfare and health. In addition to public urban policies, improvement of inhabitants' life conditions within such spaces could rely on locally-conducted management actions, thus requiring good knowledge of the context, including nescale landscape descriptions that allows one to guide and evaluate management strategies.
In this study, we provide a quantitative exploration of the landscape variability in Cotonou city, Benin, based on the extraction of synthetic landscape descriptors and their analysis through subsequent multivariate analysis. To do so, we rely on Principal Component Analysis (PCA) which allows for the simultaneous analysis of a large number of landscape metrics (Rossi and Dobigny 2019; Lemoine-Rodriguez et al. 2020), in order to model the urban landscape of three neighborhoods at a ne scale, something that has been rarely, if ever, performed in Africa. We then explore their individual characteristics and spatial variability with the aim of identifying synthetic landscape descriptors for each of these neighborhoods and testing whether they could be helpful to potential research and management programs in the future. We found that the multivariate analysis of such metrics are indeed well adapted to the study of a rapidly changing urban spaces such as those found in Cotonou, thus bringing additional tools for urban ecologists and epidemiologists as well as urban planners and decision makers dealing with city policies and development strategies in Africa.

Material And Methods
Cotonou city and studied neighborhoods Cotonou is located in the southern part of the Republic of Benin, between 6°20' and 6°24' North latitude and 2°20' and 2°30' East longitude. Before the colonial period, the current Cotonou area was probably home to a few isolated villages of To nou shermen (Ciavollela and Choplin 2018). Its location, favorable to the slave trade and then to the maritime trade of various commodities, is at the origin of the rst European settlements (Ciavollela and Choplin 2018). Its territory was ceded to France in 1868 (Brasseur-Marion 1953; Houngnikpo and Decalo 2013) and the construction of a wharf as early as 1900 in order to facilitate the transboarding of merchandises and people fueled the intensi cation of its role as an exchange hub (Janin 1964) and the development of the associated urban space. Though initially slow, Cotonou growth increased following the extension of the wharf in 1952 and the commissioning of a deep-water seaport in 1960 (Ciavolella and Choplin 2018). From there, its demography rapidly accelerated to reach 57,000 inhabitants in 1980, i.e. three times more than in 1945 (Capo 2008). Between 2002 and 2020, it grew from 665,000 to 1,190,000 inhabitants (INSAE 2018).
The city is compressed between Lake Nokoué at North, and by the Atlantic Ocean at South, thus undergoing spatial and demographic expansion to the East (towards Sémè-Kpodji, Porto Novo then Nigeria), to the West (towards Godomey, Cococodji, Cocotomey, Ouidah then Togo) and to the North on the West bank of Lake Nokoué (Abomey-Calavi). The sprawl of Cotonou thus participates to the formation of a wide urban area that occupies most of coastal Benin (INSAE 2017) and is part of the even larger, coastal so-called "Abidjan-Lagos corridor" that is in the process of giving rise to one of the largest African megacities expected to house up to 40 million inhabitants by 2050 (URBACOT 2017; Sun et al. 2020a). The core city has essentially sprawled into a low level sandy plain associated with a dense hydrographic network. Together with 1,200 mm of average annual rainfall, this results in repeated and sometimes long-lasting ooding events during most rainy seasons (Okou 1989).
In 2017, three model neighborhoods in Cotonou were investigated (Supplementary Information 1) as part of a broader program on small mammal-associated (essentially rodent-associated) zoonotic risks in urban and peri-urban settings of southern Benin. They were selected from previously established socioeconomic typologies (Dansou 2006) as well as our own observations to be representative of major neighborhoods pro les of the city. In Cotonou, ooding is an aggravating factor of households' precariousness and poor socio-environmental condition. Each year, they regularly force the poorest inhabitants to abandon their homes for several days when not weeks (PCUG3C 2012). The three neighborhoods studied here are also representative of the major ooding regimes that characterize the different areas of Cotonou.
Agla is a neighborhood in the process of rapid human densi cation and formalization. Some areas lack basic services (e.g. access to drinking water and electricity supply) and are home to economically poor households who inhabit very precarious housing structures. These zones correspond to large shallows that are humid all year round, but which sometimes over ow, from the beginning of the main rainy season onwards, due to the accumulation of rainwater. (NB: Agla has been the subject of many infrastructure developments since our work).
Ladji is a poor and densely populated informal settlement bordering and extending over Lake Nokoué (i.e., on-stilt cabans). Formal basic services are rare and housing range from hard-built to very precarious housing. Like almost all of the neighborhoods bordering the lake, it oods partially by over owing of the lake several weeks/months a year, especially at the end of the rainy season. Waste produced throughout the Cotonou city is partly dumped in vast unmanaged dumpsites, or sometimes used to ll-in the shallows and banks of the lake. Saint-Jean is an old colonial-type neighborhood, formally subdivided and equipped with drinking water and electricity networks; however, hygiene conditions remain poor in many homes. In this neighborhood, rainwater accumulates in large puddles that sometimes persist during a few days, but the area does not undergo lasting oods per se.
Mapping the landscape: land covers and associated social uses Land covers were de ned a priori on the basis of preliminary prospections in the eld and our own knowledge of Cotonou (Table 1 Cemented soils (Csoil): asphalt roads, paved roads, cemented terraces.
In addition to land uses, nine social uses associated with buildings were de ned from eld observations and further categorized according to the presence or absence of food stocks and/or foodstuffs, the presence or absence of humans at night as well as the presence of traded or manufactured non-food products ( Table 2). During the mapping process in the eld, each building (i.e., both hard-built and precarious houses) was systematically characterized for its social uses. Exceptionally, some buildings were characterized by two or more social uses (up to four uses for a single building), thus representing 7.4%, 3.3% and 11.6% of buildings of Agla, Ladji and Saint-Jean, respectively. In such cases, we relied on an a priori presumed attractiveness of social uses for small mammals (the focus of the wider program), particularly rodents, in order to retain only the most pertinent one. This led us to consider the following hierarchy, by decreasing attractiveness for rodents: presence of grain and other foodstocks = food shops > night dwelling > restaurant = bar > artisanal craft > non-food shops > services and o ces. As an example, a building showing both the "dwelling" and "presence of grain and other foodstocks" social uses was nally associated with "presence of grain and other foodstocks". In doing so, we were able to assign a unique social use to each building. Food storage and cooking space (FsCs): food storage, cooking and/or processing spaces, non inhabited at night (e.g. sh roasting sites, doughnut cooking, restaurants, food shops, butcher shops, grain and condiment mills, street kiosks/cafeterias, bars).
Manufacturing space (Manu): spaces dedicated for trade, manufacturing or storage of non-food materials (e.g. craft mechanical workshops and woodworking shops), hardware stores, printing works, sawmill).
Toilet (Toil): hard-built permanent buildings used as a public toilet.
All land cover and associated social uses were digitized on a 2016 Spot 7 (multispectral) satellite image base, 6m resolution. UTM projection to WGS 84 was implemented in QGIS v3.16.2 software. This digitizing process resulted in shape les (Supplementary Information 1) which were then converted to 0.5 m resolution raster les ( Supplementary Information 2) where each pixel is characterized by a speci c value corresponding to landscape unit.

Computation of landscape metrics
Rasters were processed under R (R Core Team 2020) in order to allow for the extraction of local minilandscapes following a moving window strategy (McGarigal et al. 2012). Mini-landscapes were represented by circular buffers which were preferred to square ones due to equidistance between the center and all edge points. In Cotonou, mapping of the above-mentioned landscape units was carried out on a perimeter covering an area whose edges were at least 100 meters away from the sampling sites (the most distant) -the latter being selected as part of a broader program on zoonotic risks in urban and periurban areas in southern Benin. On a square grid of 5 m mesh size, a series of 30m radius buffers (minilandscapes) were extracted from 20,398 points in Agla, 8,213 points in Ladji and 9,766 points in Saint-Jean. In order to avoid edge effects (i.e., the in uence of "incompleteness" of edge-located buffers on the metrics calculation), all buffers that were crossed by the edges of the neighborhoods right-of-way were removed.
Landscape metrics (Table 3) were computed using the R package ''landscapemetrics'' (Hesselbarth et al. 2019) on all individual buffers for the 10 landscape units and the 9 social uses associated with buildings described here above. The class-level landscape proportion (PLAND) was calculated for each landscape unit in order to quantify the composition of the landscape through the proportion of area covered by each land cover type. The Modi ed Simpson's Evenness Index (MSIEI) and the Edge Density Index (ED) were measured for both land covers and social uses in order to obtain synthetic indices of landscape diversity and complexity, respectively, thus re ecting of the intermingling of the different patches (Gerbeaud Maulin and Long 2008). Altogether, the landscape was thus described by 23 quantitative metrics. The Principal Component Analysis allows one to explore the relationships between a large set of descriptors (here, the metrics used as variables) as well as to identify and quantify the sources of landscape variability (see Rossi and Dobigny 2019, for an urban example).
A rst PCA analysis (PCA-Cot) was conducted at the scale of the Cotonou city by pooling the three studied neighborhoods, namely Agla, Ladji and Saint-Jean datasets. Three other analyses were conducted at the neighborhood scale, i.e. on each neighborhood considered independently (PCA-Agl for Agla; PCA-Lad for Ladji; and PCA-Jea for Saint-Jean).

Mapping of scores on principal components
The coordinates (scores) of each mini-landscape along the principal components (PC) were used to map the landscape characteristics as identi ed by the PCA (Supplementary Information 2). To do this, each pixel of the map was associated with its score along one of the different PCs. A colored gradient was used to re ect the range of scores, and allowed us to visualize the spatial variability of the PCA scores across the urban space (Rossi and Dobigny 2019). Such a spatial projection of the PCA scores was used to visualize the landscape variability within each of the three neighborhoods, either when analyzed together (PCA-Cot) or separately (PCA-Agl, PCA-Lad and PCA-Jea).

Results
At the Cotonou city scale (PCA-Cot) In PCA-Cot pooling all three nieghborhoods, PC1 (19.10%), PC2 (9.72%), PC3 (7.97%) and PC4 (7.09%) accounted for 48.88% of the total inertia. The distribution of the buffers along the scores of the PCA largely overlapped, especially towards the positive values of PC1 (Fig. 1a). The cloud of points that correspond to Saint-Jean appeared more compact than those representing the other two neighborhoods (Fig. 1a). Ladji cloud of points stretched towards the negative side of PC1 and PC2. The Agla cloud of points was also strongly stretched towards the negative values of PC1.
When looking at the contribution of variables to PCs (Fig. 1b), overlaps between neighborhoods appear associated with the hard-built houses (Hard) with multiple uses (Ed.U, Msiei.U) including dwelling (Dwel). Nevertheless, Saint-Jean partly differed from the other two neighborhoods with some pixels showing the preponderance of hard-built dwellings with multiple uses (Hard, Ed.U, Dwel, Msiei.U, Ed.C). In the same manner, many points from Agla was distinguished from Saint-Jean and Ladji by the importance of bushy shallows (Bush), while some areas of Ladji appeared distinct due the presence of Lake Nokoué (Nok) and some shallows.
Along PC3 ( Supplementary Information 3), shallow areas with a few hard-built houses (towards positive values of the axis) contrasted with bare soil (Bsoil) areas dotted with precarious roofed spaces (Psr) and houses (Preca) (towards negative values). Along PC4, the grassy-cover (Gras) often used as wild dumpsites (Wdum) and/or breeding-pens (Bpen) (positive values) was opposed to the lake and its banks (Nok, Preca, Msiei.U; negative values) ( Supplementary Information 3).
The projection on the map of the three neighborhoods of each pixel scores retrieved on PC1 and PC2 of the PCA-Cot is presented in Figure 2: several landscape elements could be clearly identi ed, such as the hard-built dwellings (represented by positive values on PC1), Lake Nokoué in Ladji, the shallows and wild dumps in Agla and Ladji (negative values on PC1), as well as the streets in Saint-Jean (negative values). Along PC2, bare soil, o ces and various services (positive values), precarious houses, grassy-cover used as wild dumps in Agla and Ladji, and Lake Nokoué in Ladji were represented by negative values (Fig. 2b).  (Fig. 3c).

Comparison of the city-wide and local PCAs
The PCA correlation circles that show the variables most involved in the PC1 and PC2 constructions of the inter-neighborhood ( Fig. 1; PCA-Cot) on the one hand, and the intra-neighborhood (PCA-Agl, PCA-Lad, and PCA-Jea on Fig. 3a, b, and c, respectively) analyses on the other hand, was quite similar. They point towards the same variables that stretched PC1 at both the Cotonou and the local (i.e., at each neighborhood) scales: "hard-built houses", "dwelling" and "complex and diverse" were opposed to "shallows", "Nok" and "bare soil" in both analyses. In the same manner, a few common variables (e.g. "bare soil," "precarious houses," and "Nok") characterized many of Ladji's points both globally (PCA-Cot) and locally (PCA-Lad) along PC2.

Discussion
Landscape variability Landscape metrics are important tools as they allow for the quanti cation and analysis of landscape variability, both between and within cities (Lechner et al. 2020). Examples of between-cities differences in urban landscapes are numerous (e.g. Ding 2021). Using multivariate analyses of metrics, we here provide a clear illustration of within-city variability that was detected between as well as within urban neighborhoods, thus demonstrating the usefulness of such approaches for ne-scale studies.
First, the three neighborhoods studied here contrasted sharply. The Saint-Jean neighborhood, with its many hard-built houses and formal organization, differed from Agla, which was dominated by the presence of vast areas of shallows that often sheltered a thick vegetation within which dwellings develop in a sometimes quite anarchic manner. The two previous neighborhoods differed from Ladji which is primarily characterized by the presence of Lake Nokoué and the associated constraints imposed to the riverside spatial and social organization. This rst level of variability was clearly detected by the multivariate analysis of metrics implemented from land and building-associated social uses. This remained true whatever the PCAs were conducted globally (PCA-Cot; i.e. with the three studied sites pooled) or locally (PCA-Agl, PCA-Lad and PCA-Jea; i.e. with the three sites investigated independently and compared a posteriori).
Second, we also revealed ner levels of landscape variability within each neighborhood. In Agla, our analysis succeeded in showing that shallows are often associated with wild dumpsites and are spatially dissociated from the spaces where habitations stand. More subtly, the hard-built dwellings in vicinity of shallows were opposed to the precarious houses that are rather located in bare soil areas (mostly representing streets and multi-family dwellings courtyards). Agla is currently undergoing very rapid changes. Indeed, this neighborhood was a shallow area and was only recently colonized by the inhabitants of Cotonou (<30 years). It is now one of the most densely populated areas of the city and continues to grow very rapidly with large areas of shallows gradually disappearing in favor of human settlements. Huge quantities of garbage from other neighborhoods of Cotonou city are brought in and dumped in the shallows to serve as ll (our own observations). Major works in progress (started after our eld surveys) aim at formalizing the development of the area, and at improving its sanitary situation, in particular by collecting and channeling part of the rainwater that used to be at the origin of the vast annual oods. Ladji is an outlying neighborhood located on the shore of Lake Nokoué. On the dry land, we could distinguish spaces occupied mainly by hard-built dwellings from those characterized by dwellings made of precarious material as well as grassy-covered wild dumpsites. This is indeed characteristic of this neighborhood, although a vast development project may accompany the construction of a major road and an interchange within this particularly unhealthy informal area. Finally, in Saint-Jean, the synthetic landscape indices e ciently retrieved a clear spatial segregation between hard-built dwellings, bare soil surfaces and cemented soils. These elements correspond perfectly to the formal and rectangular street grid (cemented soils or bare soil) of this central and relatively old neighborhood on the one hand, and the dwellings blocks that are characterized by a majority of de nitive material housings (i.e., hard-built houses) organized around large sandy courtyards (i.e., bare soil) on the other hand.

Impact of Cotonou history on the current landscape of its neighborhoods
The landscapes of Cotonou neighborhoods are strongly impacted by their history and the socioenvironmental context in which it took place. As an example, being of colonial origin, the neighborhood of Saint-Jean was one of the rst neighborhoods in Cotonou to bene t from major development work (e.g. design of formal streets, construction of hard-built houses, large sewage collector). So, the global structure of Saint Jean landscape is characterized by the well-aligned blocks of hard-built buildings, cemented streets and sandy alleys that regularly subdivide the area. The local name of Agla (Agla houn, bo wa! in the Fongbé language -"if you are bold, then come") refers in part to the many shallows that make human settlement di cult. Indeed, its hydromorphic soil and bowl topography make the area particularly prone to ooding today, especially since some adjacent neighborhoods (e.g. Houehiyo) have been lled in, developed and extensively cemented (Okou 1989). Originally, only a few indigent residents lived there. From the 1980s and especially from the 1990s, the area progressively densi ed against the shallows (URBACOT 2018), which is consistent with the outputs of our synthetic landscape proxies, i.e. the marked opposition of shallow spaces to spaces dominated by human settlements.
Similarly, the demographic growth of Cotonou over the last few decades and the urban sprawl that followed have inexorably caused the Cotonou city to move in the direction of Ladji ("where we land" in the To nou language) on the edge of Lake Nokoué. Initially occupied by xwla shermen's villages installed on the lake (on-stilts cabans) or along the banks, Ladji was overtaken by the expansion of the Cotonou city from the 1980s onwards, then rapidly densi ed thereafter (URBACOT 2018). The presence of many To nou populations accustomed to living on or near the lake (Ciavollela and Choplin 2018), the di culties of developing this informal and marshy space and the chronic poverty of often marginalized populations translate into clearly visible landscape signatures detected by our analysis, such as the lake, a mosaic of hard-built and precarious houses and the presence of several grassy-covered wild dumping sites on the dry land.
Urban landscape variability at various scales Each neighborhood displays speci c landscape features that were described above. But, interestingly, our comparative analysis of the global (PCA-Cot) and local (PCA-Agl, PCA-Lad, and PCA-Jea) PCAs also allowed us to detect landscape elements that were common to all three neighborhoods. For instance, the partial overlap of the Agla, Ladji and Saint-Jean clouds in PC1 as well as the sharing of variables strongly involved in the construction of the components of PCA (global and local) point towards quantitative landscapes elements that are shared across the city (e.g. hard-built houses, dwelling, complex and diverse). This further con rms that the method employed here is relevant to describe and to understand urban landscape variability at various scales.

Importance of urban landscape characterization
Cities are drastically transformed habitats that are strongly marked by human activities. As such, they are

Conclusion
All in all, our results thus illustrate how an approach based on PCA analysis of landscape metrics allows one to describe the structure and evolution of city landscapes at very local scale, even in the case of quickly changing urban areas. Consequently, such multivariate investigation of landscape metrics appear quite useful to detect and to describe both inter-neighborhood as well as ne-scale intra-neighborhood landscape variability, thus providing an interesting tool for the description and monitoring of urban organization. A major obstacle to the generalization of this approach lies in the tedious collection of eld data (i.e., eld mapping). Nevertheless, it can also be applied to data acquired in a more automated way (e.g. satellite images, drone-mediated acquisition), although this does not generally allow for the integration of certain data, such as usage data that were included here based on the Open Street Map toolkit (OSM 2017). The latter is freely available and allows easy-to-implement mapping directly from the ground; in addition, participative approach may make possible the acquisition of more and more numerous and the cartography of increasingly vast areas (https://www.openstreetmap.org/#map=17/6.36396/2.41592).

Declarations
Henri-Joël DOSSOU (HJD), Jean-Pierre ROSSI (JPR) and Gauthier DOBIGNY (GD) designed the study. HJD and Mariano Davy SOSSOU (MDS) conducted the mapping in the eld as well as the management of collected data. HJD and JPR performed the analyses. All authors contributed to the writing of the manuscript.