Georeferenced diagnosis of drainage structures in the city of Bouaflé using geospatial techniques and machine learning

 Corresponding author: Houebagnon Saint-Jean Patrick Coulibaly Copyright © 2025 Author(s) retain the copyright of this article. This article is published under the terms of the Creative Commons Attribution License 4.0. Georeferenced diagnosis of drainage structures in the city of Bouaflé using geospatial techniques and machine learning Houebagnon Saint-Jean Patrick Coulibaly 1, *, Rock Armand Bouadou 2, Jean Claude Konin 3, Talnan Jean Honoré C 2 and Théophile Gnagne 2 1 Department of Space Sciences and Geomatics, Félix Houphouët-Boigny University, Côte d’Ivoire. 2 Faculty of Environmental Science and Management, Nangui Abrogoua University, Côte d’Ivoire. 3 Faculty of Governance and Sustainable Development, Bondoukou University, Côte d’Ivoire. World Journal of Advanced Research and Reviews, 2025, 28(01), 557-570 Publication history: Received on 30 August 2025; revised on 03 October 2025; accepted on 06 October 2025 Article DOI: https://doi.org/10.30574/wjarr.2025.28.1.3446 Abstract The traditional method of collecting and processing socio-economic, health and spatial identification data on urban water infrastructure is time-consuming and costly. To reduce costs, geospatial and machine learning (Random Forest) tools were used to establish the Bouaflé Drainage Master Plan. These tools made it possible to characterise and predict the condition of the hydraulic infrastructure. The thalwegs, extracted from altimetric data using connectivity algorithms, were cross-referenced with the road network to identify the collection points for individual structures. The theoretical points were verified in the field to draw up an inventory of the structures and assess their operating condition. Population density, elevations, and the distance of structures from roads and canals are among the variables included in the prediction model. Field data collection identified 102 crossing structures and 38 gutters over 13.69 km. The prediction model has a satisfactory accuracy of over 96%. The distance to the canals significantly impacts the accuracy of the model. Structures located far from the drainage network often fail due to poor hydraulic connectivity. The high impact of population density creates significant anthropogenic pressure on infrastructure. The algorithmic approach reduced the diagnostic phase from three months to one and a half, while also identifying problems and enabling solutions to be targeted more effectively. The trained model could be applied in similar contexts, even in the absence of data on the condition of structures. Keywords: Random Forest; Hydraulic structure; Bouaflé; Geospatial 1. Introduction Ensuring the proper functioning of urban drainage systems is a challenge in most cities in developing countries. As a result, the effects of climate change tend to exacerbate frequent flooding, as noted by (Andrés-Doménech et al., 2021 [2]). Urban areas are rapidly expanding beyond their traditional centres, creating sprawling peri-urban areas that often lack infrastructure and services (Angel et al., 2011 [1]). At the same time, urban drainage networks are not developing or modernizing at the same pace, and given the effects of climate change, the risk of flooding is increasing (Wu et al., 2018 [20]). Urbanization has progressively increased the extent of impervious surfaces (Andrés-Doménech et al., 2021; Lapiński & Wiater 2018; Limthongsakul et al., 2017 [2], [10], [11]), such as roads, car parks and roofs, and a reduction in forest areas and other forms of open space that absorb rainwater. Changes in the water balance are mainly caused by an increase in impervious surfaces. This leads to significant changes in both the quality and quantity of runoff (Paul & Meyer 2001; Muller et al., 2020 [16], [14]). World Journal of Advanced Research and Reviews, 2025, 28(01), 557-570 558 Urbanisation in developing countries poses significant challenges in terms of planning and managing drainage networks (Mohammed et al., 2019 [12]). It is essential to understand the condition and performance of drainage infrastructure to ensure sustainable urban development, particularly in the context of rapid urbanisation and climate change (Francisco et al., 2022 [8]). Several factors need to be considered when assessing the effectiveness of an urban drainage system (Molzahn & Burke, 1986 [13]). To plan repairs and enhance the performance of drainage structures, it is important to assess their condition and the factors contributing to their deterioration. In Côte d'Ivoire, these assessments are rarely conducted comprehensively and systematically at city level. Most studies focus on the design and installation of structures, with limited consideration given to monitoring and maintenance, both of which are essential for ensuring longevity. The traditional method of gathering, processing and identifying urban water infrastructure spatially is not only labour-intensive, but also costly. To reduce costs, the Bouaflé Drainage Master Plan was formulated using geospatial and machine learning tools (Random Forest). Bouaflé, a town located in central-western Côte d'Ivoire, faces significant challenges in stormwater management. Drainage systems, which are essential for flood prevention, are often inadequate. This may be due to congestion or poor design. In the context of increasing rainfall, it is crucial to assess these structures in order to prioritize the interventions of the relevant services. This involves conducting a spatial assessment of the hydraulic infrastructure of the city of Bouaflé using geospatial data and machine learning models, specifically the Random Forest method. 2. Materials and methods 2.1. Study area Bouaflé is the capital of the Marahoué region. This crossroads town is an important part of the department of the same name in terms of its enormous economic, social and cultural potential. The town of Bouaflé is located approximately 294 kilometres from Abidjan (the economic capital of Côte d'Ivoire) and 59 kilometres from Yamoussoukro (the political capital of Côte d'Ivoire). It is situated between longitudes 5°48' and 5°42' west and latitudes 6°57' and 7°3' north (Figure 1). The urban area of Bouaflé covers an area of 45 km². Bouaflé is traversed by the Marahoué River. According to the National Institute of Statistics, Bouaflé has a population of 109,186. The urban area of Bouaflé consists of 19 neighbourhoods : Agbanou, Bellou, Biaka-boda, Bromacote, Centre-administratif, Datecoumana, Dehita, Dioulabougou, Dioulabougou-Extension, Hermankono, Koblata, Koko, Lopoifla, Marahoue, Millionnaire, N'gatakro, Port-Bouet, Residentiel Est, Solibra. Figure 1 Neighbourhood in the town of Bouaflé World Journal of Advanced Research and Reviews, 2025, 28(01), 557-570 559 The town of Bouaflé is characterized by generally flat terrain. The municipality is located in a basin along the banks of the Marahoué River. The altitude of the urban area of Bouaflé varie between 156 and 257 metres above sea level. According to the soil classification system developed by FAO et al (1999) [6], the study area is dominated by ferric Acrisols and ferric Cambisols (Af). Acrisols are highly acidic clay soils with low cation exchange capacity and low base saturation. In the town of Bouaflé, ferric cambisols (Bf) are also found, which are thick soils with a silty-sandy texture. These soils are well suited to agriculture and are ideal for cocoa cultivation. Average annual rainfall is around 1,200 millimetres. The town of Bouaflé is characterised by four distinct rainy seasons: two dry seasons and two rainy seasons, a long dry season from November to March, a long rainy season from April to June, a short dry season from July to August, and a short rainy season from September to October. Average temperatures in Bouaflé range from 24.7°C (minimum) to 27.9°C (maximum) according to SODEXAM. The town is drained by the Marahoué River and some of its tributaries. The town has a dense hydrographic network. These natural waterways are characterised by the presence of two main rivers, the Frako and the Glopeny. The river system reflects the rainfall pattern. 2.2. Data and materials 2.2.1. Geospatial data This type of data consists of: • Road network (roads, tracks) represented as polylines, obtained from the CCT (Centre for Cartography and Remote Sensing); • Open Street Map vector maps; • Mosaic of satellite and aerial images, from Google Earth; • Sentinel-2 images (level 2A, resolution 10–20 m), acquired from the Copernicus programme and used to calculate spectral indices; • SRTM Digital Elevation Model (DEM) with 30 m resolution, available for download via the USGS Earth Explorer platform. 2.2.2. Population data The high-resolution population data, comes from the WorldPop database for 2023 hosted by the University of Southampton. This database provides high-resolution annual estimates (from 100 m to 1 km depending on the product). They are derived from : • National and regional population censuses; • United Nations population projections (CIESIN (Centre for International Earth Science Information Network), (2018) [5]; • UNFPA (United Nations Population Fund); • Auxiliary geospatial data (land use, road networks, infrastructure, altimetry, satellite imagery, OpenStreetMap, Microsoft Building Footprints, Microsoft Roads Dectior). For this study, WorldPop population data with a resolution of 100 metres was used. 2.2.3. Data on drainage structures A georeferenced inventory of drainage structures was conducted in 2023 in the urban area of Bouaflé using GPS during field surveys. 2.2.4. Data collection and processing tools A series of software programs and tools were used to collect and process the data required for this study. These included: • Qgis for geoprocessing and map production; • Rstudio, for statistical analysis and model training; • GPS for collecting geographical coordinates; • Qfield for collecting data on hydraulic structures. 2.3. Methodology The analyses were carried out using three main geospatial layers; World Journal of Advanced Research and Reviews, 2025, 28(01), 557-570 560 • Road network (roads, tracks, paths) represented as polylines ; • Hydrographic network of thalwegs extracted from an SRTM Digital Elevation Model (DEM); • Sentinel 2 satellite image. 2.3.1. Extraction of potential sites for the placement of individual drainage structures It is carried out in two main stages: first, extracting the thalwegs and watercourses from the study area, and second, cross-referencing them with the road network data. Extraction of thalwegs from the DEM • Pre-processing of the DEM The raw DEM often contains artificial depressions or ‘sinks’ that interrupt the flow network. These depressions must be filled in to ensure hydrological continuity (Crave, 1995 [4]), (Wang et Liu, 2006 [19]). The operation known as sink filling consists of creating a modified DEM, Zf (x,y) such that: Zf (x,y) ≥ Zf (x,y), ∀ (x,y). • Extraction of the hydrographic network and delineation of Sub-Catchment Areas For the study, the D8 connectivity algorithm was applied to extract the drainage network. The direction 8 method was introduced by O’Callaghan et Mark (1984) [15]. Algorithm: Min [(Z (x, y) -z (xi, yi)] Où xi= (x-1, x, x+1) ; yi= (y-1, y, y+1) ; Equation (1) Z (x, y): Elevation of the central pixel. z (xi, yi): Elevations of the 8 neighbours. Identification of road–thaleweg crossing points The aim is to identify road–thalweg intersection points to determine potential locations for hydraulic crossing structures (culverts, drains). To achieve this, a 10 m buffer zone was generated around the road network polylines. Intersections were then detected using the following equation: let X represent the buffered polygon of the road network and Y the thalweg network. The intersections are given by : P = X ∩ Y. Where P represents the probable crossing points, i.e., the potential locations for culverts or pipes. 2.3.2. Collection of drainage structures (pipes, culverts, channels) Hydraulic interpretation and field validation The road–thalweg crossing points identified through topographical analysis are ideal locations for the installation of drainage structures. Intеrseсtiоns betweеn rоads and thаlwegs represent zоnes where runоff frоm adjaсеnt slоpes intеrsects with the rоadway. This intersection creates an increased risk of erosion or overflow. The relevance of these locations has been validated by field observations, revealing recurrent water stagnation during or after rainfall events. This cоmprеhensive methоdоlоgy, which merges hydraulic assessments with еmpirical field validаtiоn, reinfоrces thе reliаbility оf site selectiоn fоr the implеmentatiоn оf suitable drainage sоlutiоns. 2.3.3. Preparation of predictor variables The Sentinel-2 imagery underwent atmоspheriс cоrrectiоn and was subsequently crоpped tо fоcus оn thе study regiоn. Bоth raster аnd vectоr data layers wеre reprоjected tо cоnfоrm tо the WGS 84 / UTM Zоne 29N cооrdinаte system. Pоpulatiоn and elеvatiоn datasets wеre aligned with the rеsоlutiоn оf the Sеntinel-2 images (10 m) thrоugh bilinеar resampling techniquеs. The integrity оf the rоad and сanаl netwоrks was cоnfirmed thrоugh tоpоlоgical verificatiоn in QGIS 3.30. Calculation of spectral indices Two indices were calculated from Sentinel-2 bands : World Journal of Advanced Research and Reviews, 2025, 28(01), 557-570 561 • SAVI : Soil Adjusted Vegetation Index SAVI is a vegetation index that corrects for the influence of bare soil on vegetation measurements, particularly useful in areas with sparse vegetation cover (urban areas) (Huete A. R. 1988 [9]). This index is used to quantify vegetation cover around hydraulic structures. SAVI = NIR−RED NIR+RED+L ⋅(1 + L) Equation (2) Where NIR = B8, RED =B4 et L= 0.5 to limit the influence of the soil. Vаlues apprоaching 0 suggest minimal оr аbsеnt vеgetatiоn, whеreas values neаring 1 indicаte rоbust and thriving vegetatiоn. • NDBI: Normalized Difference Built-up Index The NDBI index was calculated to quantify the degree of urbanization around each hydraulic structure. It reflects the extent of soil sealing and its impact on the local hydrological regime Zha, Y., Gao, J., & Ni, S. (2003) [21]. NDBI = SWIR−NIR SWIR+NIR ⋅(1 + L) Equation (3) Avec SWIR = B11 et NIR =B8 Population density Pоpulatiоn dаtа, disаggrеgated fоr this analysis, werе sourced frоm WоrldPоp (Tаtеm, 2017 [18]). WоrldPоp emplоys spatiаl mоdeling techniques that leverage satellite imagery in conjunction with census infоrmatiоn. Fоr the purpоses оf this reseаrсh, the WоrldPоp 2023 datasеt was utilized, and pоpulаtiоn density values clоsest tо еaсh structure were extracted through spatial join. This metric асted as a prохy fоr assessing pоpulаtiоn pressurе within thе mоdel proposed by Stevens et al. (2022) [17]. Elevation Altitudе dаtа were retrieved frоm the SRTM Digital Elevation Model (DEM), which has a resolution of 30 meters, by cоnduсting spоt sаmpling at each struсture's lоcatiоn (Fаrr et al., 2007 [7]). Distances to roads and canals The shortest Euclidean distances from each structure to the surrounding road and canal networks were computed using QGIS's Distance Matrix functionality. Integration into the geospatial analysis database All derived variables, including SAVI, NDBI, population density, elevation and distances to roads and canals, were integrated into the attribute table associated with drainage structures. This comprehensive database will serve as the foundation for subsequent statistical and predictive analyses. 2.3.4. Predictive modelling of hydraulic structure condition Theoretical points identified through GIS and topographical analyses were verified in the field to create a comprehensive inventory of urban drainage structures. During these visits, the location of each structure was determined, and its operating condition was assessed according to standardised criteria (good or damaged). This phase allowed us to validate the theoretical data and collect detailed information on the existing infrastructure. A Random Forest machine learning model was created to forecast the operational status of drainage systems based on explanatory variables. 2.3.4.1. Target variable: The observed condition of the structures, as determined by field data collection, was considered a dependent variable with two qualitative categories: 'good' and 'damaged'. World Journal of Advanced Research and Reviews, 2025, 28(01), 557-570 562 Explanatory variables Several explanatory variables were included in the model to represent the condition of structures. • Population density, which is an indicator of the pressure placed on infrastructure by human activity; • Elevation, derived from the digital terrain model, was used to estimate the vulnerability of structures based on topography; • Distance to roads and canals to quantify hydraulic connectivity and accessibility of structures; • The SAVI spectral index indicates the presence of vegetation, which can signify obstruction, abnormal humidity or lack of maintenance; • The NDBI spectral index, which measures urban pressure and impermeability (i.e. the level of stress and risk of damage to structures). Thus, each structure is represented by a vector of predictors. The Random Forest (RF) algorithm (Breiman 2001 [3]) was applied to predict the condition of drainage structures based on environmental and socio-spatial explanatory variables. Construction des arbres Random Forest (RF) is based on the concept of decision trees, which are formed from the results of numerous independently generated trees. This strategy avoids overfitting, improves accuracy and offers greater robustness (Zhang et al., 2025 [22]). The condition of drainage structures (either 'good' or 'damaged') was predicted using the RF algorithm, which involves building a set of decision trees on bootstrap samples using a random subset of variables at each node. The final prediction for each structure is obtained by majority vote of the trees. The model's performance was evaluated using a confusion matrix and variable importance, which allowed the most influential factors affecting the condition of the structures to be identified. The condition of drainage structures (Y = Good or Damaged) was predicted using the RF algorithm, which is based on the aggregation of T decision trees built on bootstrap samples. For each tree t, a prediction 𝑓𝑡 (𝑋) is produced based on the explanatory variables X = (X₁, ..., X_p). 𝑓𝑡 (𝑋)=∑𝑐𝑚 𝑀 𝑚=1 1{𝑥 ∈ 𝑅𝑚} Equation (4) Where 𝑹𝒎 is the region corresponding to leaf M et m, cm is the majority class in that leaf. The final prediction of the forest is obtained by a majority vote. 𝑌 =mode{𝑓1 (𝑋),𝑓2 (𝑋),…,𝑓𝑇 (𝑋)} Equation (5) The model's performance was evaluated using a confusion matrix and variable importance. This enabled the identification of the factors that have the greatest influence on the condition of structures. Model evaluation We evaluated the model's overall performance using the out-of-bag (OOB) error and the confusion matrix. This allowed us to estimate the model's accuracy, as well as its sensitivity and specificity. Thanks to Random Forest's internal crossvalidation principle, these metrics provide a robust evaluation without the need for an external test set. Additionally, the importance of each variable was measured using Mean Decrease Accuracy (MDA), reflecting its contribution to the model's accuracy. MDA is based on the idea that if a variable is informative, permuting its values at random will significantly deteriorate predictive performance. Mathematically, the MDA for variable i is defined as follows: 𝑀𝐷𝐴ᵢ = (1/𝑁)×∑[𝑂𝑂𝐵ⱼ − 𝑂𝑂𝐵ⱼ,ᵢᵖᵉʳᵐ]𝑝𝑜𝑢𝑟𝑗 = 1à𝑁 Equation (6) • MDA₍ᵢ₎: The importance of the variable i, according to the Mean Decrease in Accuracy. • N: Number of observations • OOBⱼ: Prediction error for observation j in the out-of-bag (OOB) sample. World Journal of Advanced Research and Reviews, 2025, 28(01), 557-570 563 • OOBⱼ, ᵢᵖᵉʳᵐ: Error obtained in observation j after permutation of variable i • ∑: Sum of all observation samples. 3. Results and discussion 3.2. Watercourses and thalweg in the urban area of Bouaflé Figure 2 illustrates the dense hydrographic network in the area. The Marahoué River and the two main tributaries, the Frako and the Glopeny, flow through the city of Bouaflé. This puts the population of Bouaflé at significant risk of flooding. Therefore, it is necessary to establish an effective drainage system in Bouaflé. Figure 2 Watercourses and thalweg in the urban area of Bouaflé. 3.3. Potential sites for positioning individual drainage structures Field data collection identified 312 potential drainage structures. These locations could accommodate culverts, drains or bridges. The high number of potential sites indicates the density of thalwegs in the urban area of Bouaflé (see Figure 3). World Journal of Advanced Research and Reviews, 2025, 28(01), 557-570 564 Figure 3 Potential drainage structures in the urban area of Bouaflé. 3.4. Identifying potential drainage structures in the field. A systematic field survey (Figure 4) was carried out to identify existing unique drainage structures. This survey identified 102 crossing structures and 38 gutters along a 13.69 km stretch of land. The characteristics and geographical locations of these structures were also recorded. Figure 4 Drainage structure in the urban area of Bouaflé. World Journal of Advanced Research and Reviews, 2025, 28(01), 557-570 565 3.5. Results relating to explanatory variables. 3.5.1. Spectral indices SAVI: Soil Adjusted Vegetation Index Typical SAVI values for the urban area of Bouaflé are low. They range from -0.02 to 0.25 (see Figure 5). There is existing but sparse vegetation cover. This is a protective rather than a risk factor. Figure 5 Soil Adjusted Vegetation Index of Bouaflé urban area NDBI: Normalized Difference Built-up Index The prominent position of the NDBI in our model highlights the significant effect that urbanisation has on the deterioration of hydraulic infrastructure. This highlights the importance of urban planning in the sustainable management of such structures. In Bouaflé, most structures are in areas with low permeability and NDBI values below 0.15 (see Figure 6).