Flood susceptibility mapping in the Tongo Bassa watershed through GIS, remote sensing and frequency ratio model

Water resources management and related risks are increasingly becoming an important concern of contemporary society (Chomba et al. 2022; Ebodé 2022a; Ebodé et al. 2022; Nsangou et al. 2022). Among the wide range of existing hydrological risks, flooding appears to be one of the most devastating, considering the enormous damage it often causes (Samanta et al. 2018; Ekwueme 2022; Krisnayanti et al. 2022; Manzoor et al. 2022). Between the 1990s and 2010s, floods caused the deaths of more than 158,000 people around the world (Centre for Research on the Epidemiology of Disasters 2018). Economic losses of around $23 billion are also recorded due to floods each year (UN 2018). If we stick to climate forecasts and changes in land use and land cover (LULC) patterns, we will observe an increase in the number and territories exposed to these disasters (Vidhee & Amit 2020; Asinya & Alam 2021; Chen et al. 2023).

In the European Union, flood phenomena were responsible for one-third of the economic losses caused by natural disasters between 1980 and 2016 (Costache et al. 2020). In terms of the loss of human lives and destruction of property, floods are among the most widespread and severe natural hazards in this region (Costache et al. 2020). The amount of damage due to the California floods in 1997 was $2,000 million. That of Mississippi in 1995 was $7,499 million. Environmentally, Nsangou et al. (2022) reported the Indian Himalaya in 2013 witnessed the terrible event of the Kedarnath flash flood, which was very disastrous and tragic, resulting in a serious stage modification of some geomorphological units in the watershed. Finally, for the African continent alone, the costs could amount to several billions of euros for insurers (Nsangou et al. 2022). Thus, the human desire to build in flooding areas risk appears in flagrant contradiction with the current phenomenon of flooding risk increase, which threatens many human lives and infrastructure (socio-economic, heritage, etc.).

Compared to other parts of the world, sub-Saharan Africa seems more vulnerable to these disasters due to poverty, poor governance and low technological level (Ahouangan et al. 2010). In the case of Cameroon, a study published by Centre for Research on the Epidemiology of Disasters (2016) states that 367,276 people were affected by floods between 2007 and 2015. Douala City is the most affected area due to its flat terrain, its exponential demographic growth (which leads to an anarchic occupation of spaces, including those reserved for the water circulation) and its proximity to the Atlantic Ocean, which is the origin of tidal phenomena. Between 2000 and 2010, floods caused more than 100 deaths and significant material damage in Douala City.

The fight against flooding begins with detailed and precise mapping of risk areas (Tehrany et al. 2015; Ebodé 2022b). Geographic information system (GIS) and remote sensing (RS) techniques and data have largely contributed to the analysis of natural hazards (Jaafari et al. 2014; Moel et al. 2014). Many studies relating to flooding risk area mapping have been carried out using GIS (Strobl et al. 2012; Pradhan et al. 2014). Among the most popular approaches in natural hazard modelling, we have frequency ratio (FR) (Tehrany et al. 2015; Samanta et al. 2018), analytical hierarchy process (Stefanidis & Stathis 2013), fuzzy logic, logistic regression, artificial neural networks (Kia et al. 2012; Lohani et al. 2012) and weights of evidence (Dahal et al. 2008). Among all these approaches, the FR could be considered one of the simplest and most effective in flooding risk area mapping (Liao & Carin 2009). It is a relatively new tool widely used for risk areas mapping several other complex natural disasters, such as landslides (Rahmati et al. 2016). These different reasons led us to adopt this approach in this study.

The Tongo Bassa watershed (TBW) is the largest and most populated in Douala City. Even though it is one of the basins most affected by flooding, the Tongo Bassa has so far only been the subject of a very small number of relevant works on this issue (Elong et al. 2022; Sone et al. 2023). In these works, the FR model has never been used despite its effectiveness and simplicity. Also, the mapping of homes located in areas corresponding to each level of exposure to flooding (very low, low, medium, high and very high) has never been addressed in the few existing studies carried out in this basin, despite the importance it has for decision-making during development and for prioritizing interventions in the event of a hazard.

This study therefore aims to (1) validate the FR model in the TBW and the region; (2) produce a reliable map of flooding risk areas in the TBW with this model and (3) map homes included in each level of exposure (very low, low, medium, high and very strong). The third objective was not easily achieved due to the difficulty of accessing key information, allowing us to go further in the analysis than the few existing studies (distribution of houses in the basin). These data make it possible to propose additional, more concrete tools for development (maps of homes included in the flooding risk area and whose occupants should be relocated to appropriate sites). These data were extracted from Google Earth.

Where NIR is the surface ground reflectance in the near-infrared channel; R is the ground reflectance of the surface in the red channel and MWIR is the ground reflectance of the surface in the mid-infrared channel. The use of Google Earth, as well as the areas sampled from GPS, made it possible to identify with certainty impervious areas (built-up areas, savannahs, bare soils and crops), water bodies (large rivers, lakes and ponds) and forests. Before classification operation, the separability of the spectral signatures of the sampled objects to avoid interclass confusion was evaluated, by calculating the ‘transformed divergence’ index. The value of this index is between 0 and 2. A value > 1.8 indicates good separability between two given classes. The different classes used in this study show good separability between them, with indices >1.9.

The FR values indicate the types of correlation between different factors and flooding. A FR value greater than 1 indicates a strong correlation with flooding; on the other hand, a value less than 1 indicates a weak correlation with flooding. Table 2 shows the value of the FR of each of the classes of different variables.

To obtain the flooding risk area map, all variables are converted to raster format. The spatial resolution of each of the rasters was defined on a cell size of 10 × 10 m, and it is integrated into the ArcGIS database. This integrated database has been reclassified into five classes of flood sensitivity, namely very low, low, medium, high and very high.

As in the case of LULC patterns, the altitudes of the basin have been reclassified into four categories (–1 to 13 m; 14–25 m; 26–37 m and 38–57 m) (Figure 4). Only the –1 to 13 m class (20.89% of the total basin surface) is well correlated with flooding, with a significant FR of 4.49 (Table 2).

The slopes of the basin were also divided into four classes (0–3%; 4–7%; 8–14% and 15–42%) (Figure 5). Only the 4–7% slope class is poorly correlated with flooding. The other three classes are well correlated with flooding. Their FRs are greater than 1 (Table 2). The most important FR is that of the class of slopes 15–47%, which occupy 3.21% of the total area of the basin (Table 2).

The drainage densities of the basin are divided into four classes (44–53 km/km²; 54–63 km/km²; 64–72 km/km² and 73–82 km/km²; Figure 6). Two of them (54–63 km/km² and 73–82 km/km²) appear to be well correlated with flooding, with respective FRs of 1.12 and 4.62, and respective basin occupancy rates of 65.29 and 1.4% (Table 2).

Four classes (0–142 m; 143–283 m; 284–425 m and 426–567 m) also appear on the map of distances to watercourses (Figure 7). The 0–142 m class, whose basin occupancy rate is 53.71%, is the only one for which the FR is greater than 1 (Table 2).

The soils of the investigated basin are divided into only two classes (sandy-loamy and clayey) (Figure 8). Unlike sandy-loamy soils, clayey soils (54.86% of the basin) are well correlated with flooding, with a FR of 1.82 (Table 2).

High and very high flooding risk areas are where flooding is most likely to occur. They are characterized by significant imperviousness, low altitudes, weak slopes, significant proximity to watercourses and clayey soils.

It has already been shown in other studies around the world that low altitudes and slopes, proximity to watercourses and imperviousness present the most important links with flooding (Esteves 2013; Colmet-Daage et al. 2017; Sone et al. 2023).

To validate the FR model, it is important to calculate the success rate and accuracy of the predictions. The success rate was calculated using 200 historical flood points. The prediction accuracy was calculated using 31 historical flood points. Future flooding is most likely to occur in areas with high and very high flood risk (Samanta et al. 2018). The success and accuracy rates of the predictions are 89 and 96.78% (Table 3). Such prediction accuracy validates the FR model in the studied watershed and proves at the same time that it is suitable for flooding risk areas mapping in the studied region.

As is the case in this study, the FR model has already been used to reliably map flooding risk areas in other basins around the world (Liao & Carin 2009; Pradhan & Youssef 2011; Lee et al. 2012; Samanta et al. 2018).

The exposure levels which include the greatest number of houses are in the following order: high (25,712 houses or 38.5%), medium (20,509 houses or 30.72%), very high (16,268 houses or 24.42%), low (2,835 houses or 4.24%) and very low (1,418 or 2.12%) (Table 4). Most of the houses included in the flooding zone are located to the South and Southwest of the basin, regardless of the exposure level. More than half of the houses in the basin are located in high and very high-risk zones (41,980 houses or 62.92%), which are the exposure levels with the greatest impacts for which measures must be taken.

Most authors who have dealt with flooding in their work generally limit themselves to producing a map of risk areas using various methods (Chen et al. 2011; Lohani et al. 2012; Zou et al. 2013; Haghizadeh et al. 2017). The rate of houses included in each exposure level (very low, low, medium, high and very high) is a dimension of analysis that is missing in this work. However, this information can help to get an idea of the impact level of flooding on the populations of a given basin and the prioritization of interventions during different hazards. It can indeed happen that the zone at high risk of flooding has a large surface area in a basin, but that the population is mainly settled in the zones at medium and low risk. This reflects the low vulnerability of the population to flooding. On the other hand, if the majority of the population is settled in areas with a high and very high flooding risk, as is the case in this study, this implies high vulnerability and therefore, urgent decision-making. It is crucial to always cross information relating to the different levels of exposure with that of houses to get a clear idea of the current degree of vulnerability of the different parts of a given area and thus be able to prioritize interventions in the event of a hazard.

This study aimed to map flooding risk areas in the TBW using the GIS technique, field data (historical flood points), RS (Landsat image) data and the FR model. For this, six independent variables influencing floods were retained (LULC patterns, altitudes, slopes, drainage densities, distances from watercourses and soil types). At its end, it appears that areas with high and very high risk of flooding (those where flooding is most likely to occur) occupy more than half of the basin. Their respective basin occupancy rates are 33.86 and 27.33%. These areas are characterized by significant imperviousness, low altitudes, weak slopes, significant proximity to watercourses and clayey soils. Most of the houses in the basin (66.92%) are located in areas affected by these two levels of exposure (high and very high) (Figure 10). With respective success and prediction accuracy rates of 89 and 96.78%, the map of flood risk zones produced for the investigated basin deserves a certain amount of confidence to be placed in it. This tool can be validly used by the public authorities to fight against flooding in this basin. Although the results of this study are satisfactory overall, the inclusion of some key parameters, such as geology, would have made it possible to have significantly better results. Field studies are necessary to have a complete database on this basin and the region, which would make it possible to further refine the analyses in such studies.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.