West Africa, lies between latitudes 0 and 20° N and longitudes 20° W to 20° E (Fig. 1), and is home to approximately 456 million inhabitants (The United Nations – Department of Economic and Social Affairs, 2024). West Africa exhibits pronounced climatic variability on intra-seasonal to multidecadal timescales, with aridity increasing progressively inland. Three primary climatic zones are typically distinguished (Fink et al., 2017; Quenum et al., 2019; Ndao et al., 2020): (a) The Sahelian zone (∼12.5° N), a semi-arid belt that transitions into the Sahara desert, with highly irregular rainfall ranging from approximately 150 to 600 mm year⁻¹, typically peaking in August; (b) the Sudanian (savannah) zone (9 to 12.5° N), a sub-humid region influenced by the West African Monsoon, receiving around 200 to 1000 mm year⁻¹, and (c) the Guinean coastal zone (4 to 9° N), a humid tropical region characterized by a bimodal rainfall regime driven by the seasonal movement of the Inter-Tropical Discontinuity (ITD), with precipitation generally ranging from 1600 to 2000 mm year⁻¹.

The Upper Bandama (or Bandama Blanc) watershed, located in northern Côte d'Ivoire within the Sudanian climatic zone, was selected to evaluate the data quality. Covering approximately 14 500 km² between 8°40^′–10°20^′ N and 5°00^′–6°20^′ W, the watershed is of key importance due to its position upstream of a major hydroelectric dam (Kossou dam). Moreover, small agricultural dams are prevalent in the area, and its rural population relies heavily on rain-fed agriculture for livelihoods and food security, making it particularly vulnerable to hydrological variability (Moussa Ouedraogo, 2016). The Upper Bandama watershed therefore provides a valuable benchmark and represents a data-scarce context where existing datasets often have limited coverage and is currently the only validation dataset available to us for this assessment.

We carried out a literature review to identify, inventory, and assess the current state of existing datasets containing information on dams and reservoirs in West Africa. We consulted online scholarly literature datasets of peer-reviewed journal articles, theses and dissertations, books, conference papers, and technical reports. The online scholarly literature datasets used were Google Scholar (https://scholar.google.com/, last access: 30 September 2025) and SCOPUS (https://www.scopus.com/, last access: 30 September 2025). Search strings were defined using the keywords of this study such as dams, reservoirs, datasets, water management, and West Africa along with their synonyms, and related words to ensure comprehensive coverage of pertinent literature (Table 1). Based on this review, we selected datasets that are open-access with standardized file formats (CSV or XLS and SHP) for compilation (Fig. 2, Table 2). For each dataset, we used the most recent dataset version available at the time of the study and combined the versions when they are complementary. For instance, we combined two versions of the Global Lakes and Wetlands Database (GLWD) into a unified dataset by merging them: GLWD Level 1 (GLWD-L1) including large waterbodies (≥50 km² for lakes and ≥0.5 km³ for reservoirs), and GLWD Level 2 (GLWD-L2), which includes smaller waterbodies.

We extracted the data related to West Africa from each dataset and compared their entry (feature) types (dams and/or closed waterbodies) and counts, the proportional (percentage) representation of West African entries, and their attribute types and records. We then merged dam point and waterbody polygon entries separately into unified shapefiles using the Quantum Geographic Information System software (QGIS version 3.28.3) (Moyroud and Portet, 2018).

We overlaid the two merged files (point and polygon files) on high-resolution satellite basemaps from ESRI Imagery and Google Earth. This was to verify and refine the spatial accuracy of dam and reservoir geographic coordinates and to digitize missing dam points or reservoir polygons as appropriate. Dam points and reservoir polygons do not always have a strict one-to-one relationship in the compiled dataset as the selected source datasets differ in their feature types and purposes. Some datasets provide dam locations as points only, while others provide reservoir polygons without associated dam points. In addition, some large or complex reservoir systems may include more than one dam-related point, while some small dams may not have a clearly mapped reservoir polygon in the available products. Where possible, we linked dam points to reservoir polygons using spatial overlay and proximity analysis in QGIS. A dam point was associated with a reservoir polygon when it was located within, on the boundary of, or immediately downstream of the polygon. This spatial matching was supported by attribute information where available, including dam or reservoir name, country, river or basin, reservoir area, and source dataset. We manually inspected ambiguous cases using the spatial layers and, where necessary, satellite imagery or base maps. The satellite images also enabled us to visually identify natural waterbodies such as rivers and lakes that were subsequently removed manually from the data. We only retained man-made reservoirs with dams.

Several original datasets may share common source information or inherit records from earlier datasets. We accounted for this issue mainly through duplicate identification and data-source tracking. We identified duplicate records using a combination of spatial and attribute-based checks. Records with identical or nearly identical geographic coordinates were flagged as potential duplicates. We visually inspected in QGIS, overlapping or very closely located features. Additionally, available attributes such as dam or reservoir name, country, river or basin name, reservoir area, storage capacity, construction year, and source dataset were compared to confirm whether the records referred to the same infrastructure. When duplicate records were confirmed, we retained only one representative record in the compiled dataset. Attribute information from the duplicate sources was not simply discarded; rather, consistent information was retained and used to improve the completeness of the final record. When attribute values differed across sources, we retained the most complete and consistent record and identified the primary contributing dataset in the source field of the compiled dataset.

Finally, we assessed the differences between the newly compiled dataset and the original datasets (Sect. 2.2) at the West Africa scale by comparing the probability density distribution of attributes such as dam latitude, reservoir latitude, reservoir surface area, and reservoir volume. We estimated these distributions using Kernel Density Estimation (KDE), a widely used non-parametric method for probability density estimation (O'Brien et al., 2016). We defined the reservoir surface area classes based on the definition proposed by FAO (Table 3, Bernacsek, 1984). We classified minor and small reservoirs as “small-scale infrastructure” and medium, large, and major reservoirs as “large-scale infrastructure.” We further related dam size classes to the size class of their associated reservoirs. Accordingly, a small dam refers to a dam associated with a small reservoir.

The quality assessment of the dam and reservoir data at the watershed scale follows a structured, multi-step process designed to compare the compiled dataset with ground truth data. First, we consulted national institutions and agencies involved in studies related to dams and reservoirs in order to access their records. These institutions are the Geoscience & Environment Laboratory of Nangui Abrogoua University (UNA) in Côte d'Ivoire and the National Office of Technical Studies and Development (BNETD).

Next, we conducted field surveys during the dry season (December 2024) across the Upper Bandama watershed, which covers approximately 14 000 km². Using a structured questionnaire (Appendix A, Fig. A1), we carried out group interviews with local communities in each village that hosts a dam, covering ten villages and four districts (county) in total: Niakaramandougou, Dikodougou, Korhogo, M'Bengué. Interview questions were focused on dam and reservoir characteristics (e.g. year of construction, main uses, existence or not of maintenance plan, reservoir surface and volume, potential number of users). Participants included traditional authorities, local dam management bodies, user cooperatives, women's and youth associations, and smallholder farmers (Appendix B, Table B1). This participatory approach ensures that the data reflects both technical accuracy and socio-environmental realities (Quimby and Beresford, 2023). Additionally, we recorded ground truth geographic coordinates of dams and their reservoirs using a GPS-enabled handheld device. It is important to note that the survey focused on dams and reservoirs identified through consultations with institutions, local authorities, and community members, and that were accessible by vehicle or on foot and therefore verifiable during field visits.

Data from the field and institutional sources were combined to form a ground truth dataset for the Upper Bandama watershed. This field campaign data (ground truth data) was then compared with the compiled dataset (Sect. 2.3) to evaluate how accurately the compiled dataset represents the spatial representation and the characteristics of constructed dams and their reservoirs. The comparison involved the probability of density distribution of attributes including latitude of dams, year of completion, reservoir volume, reservoir surface area, average reservoir depth, dam height, and average discharge (outflow). In addition, we determined the descriptive statistics (mean, median, root mean square error (RMSE), and interquartile range (IQR)) for each attribute to quantify the differences between the compiled dataset (Sect. 2.3) and the field campaign data for the Upper Bandama watershed.

The International Commission on Large Dams (ICOLD) has maintained and regularly updated the World Register of Dams (WRD) since 1958. The WRD is a comprehensive inventory of over 62 000 large dams (≥15 m in height or ≥3×106 m³ in storage capacity) across 166 countries, of which two-thirds are georeferenced (ICOLD, 2024). The WRD has informed the development of several major global datasets, including the Global Lakes and Wetlands Database (GLWD) (Lehner and Döll, 2004). GLWD provides a global raster map with a 30 s resolution structured in three tiers: (1) 3067 large lakes (≥50 km²) and 654 large reservoirs (≥0.5 km³), (2) approximately 250 000 smaller waterbodies (≥0.1 km²), and (3) a classification of wetland type.

The Food and Agriculture Organization (FAO) incorporated data from the WRD and other sources, including national surveys and literature reviews, into the Global Information System on Water and Agriculture (AQUASTAT) (FAO, 2007, 2021). AQUASTAT provides detailed information on approximately 14 500 dams and their reservoirs, covering the geographic location, dam height, reservoir capacity, surface area, and primary purpose. This dataset contributed to the Global Reservoirs and Dams (GRanD) database, which supports integrated assessments of the environmental and socio-economic impacts of dams (Lehner et al., 2011). GRanD v1.1 contains 6862 records with a combined storage capacity of 6197 km³. In 2019, GRanD v1.3 added 458 additional reservoirs, bringing the total to 7320, and the earlier v1.2 which was not produced as a standalone product was integrated into the HydroLAKES v1.0 database (Lehner et al., 2011). These updates reflect the continued global growth of reservoir capacity, especially from projects completed between 2000 and 2016. Other relevant databases include the Future Hydropower Reservoirs and Dams (FHReD) database (Zarfl et al., 2015), which contains over 3700 planned or under-construction dams exceeding 1 MW in capacity. Mulligan et al. (2020) introduced the GlObal GeOreferenced Database of Dams (GOODD), which is containing over 38 000 dams identified through high-resolution satellite imagery, although with limited attribute information. The version 1.1 of the Global River Obstruction Database (GROD v1.1) by Yang et al. (2022) maps 30 549 artificial structures across 2.1×106 km of large rivers using Google Earth Engine. Each structure is categorized into one of six types of flow barriers. Wang et al. (2022) developed the Georeferenced global Dams And Reservoirs (GeoDAR) dataset using Google Maps and other inventories including ICOLD WRD, GRanD v1.3, and HydroLAKES v1.0. GeoDAR v1.1 includes, in all continents except Antarctica, over 24 000 dam points and 21 500 reservoir polygons linked to high-resolution water masks. Zhang and Gu (2023) relied on AQUASTAT, GRanD, and the World Resources Institute (WRI) database to build the Global Dam Tracker (GDAT), which includes over 35 000 geocoded dams, suitable for temporal analysis. A recent consolidation effort by Lehner et al. (2024) harmonized datasets by releasing the Global Dam Watch (GDW v1), integrating GRanD, GOODD, and FHReD, and complemented by GROD and the EC Joint Research Centre's Global Surface Water dataset. GDW v1 contains 41 145 dam and barrier locations, and 35 295 reservoir polygons, with a total storage of 7405 km³.

In parallel, recent initiatives focused on global monitoring of waterbody surface dynamics and morphology which contain data about reservoirs in all continents (except Antarctica). Khandelwal et al. (2022) developed the ReaLSAT dataset using machine learning applied to Earth Observation data and compiled surface area variations of over 681 000 waterbodies (≥0.1 km²) from 1984 to 2015. Donchyts et al. (2022) monitored 71 208 small to medium-sized reservoirs globally using multi-sensor satellite data. Khazaei et al. (2022) introduced a novel GLObal Bathymetric (GLOBathy) dataset, which estimates bathymetry for over 1.4 million waterbodies using a GIS-based framework. Bai et al. (2025) produced the GLRSED, combining data from HydroLAKES and OpenStreetMap to provide detailed spatial and physical attributes for 2.17 million entries (features). We provided a summary of existing global databases containing data for West African dams and reservoirs in the following table (Table 4).

The comparison among existing dam and reservoir datasets highlighted substantial differences, particularly in spatial resolution, entry types (dams and/or reservoirs), and attribute completeness. These differences reflect to the heterogeneity of data sources, reporting standards, geographical focus, the dynamic nature of dam and reservoir operations, and variation in data collection approach and dataset objectives (Zarfl et al., 2015; Zhang and Gu, 2023; Bai et al., 2025). Such discrepancies have critical implications for future mapping efforts and research in Earth system sciences, including hydrology, climatology, and land-surface processes. In particular, they can lead to substantial uncertainty in estimates of surface water storage, flow regulation, evaporation losses, and human water use, especially in data sparse regions such as West Africa, highlighting the urgent need for standardized, high-resolution, and systematically validated datasets.

The ground truthing assessment showed that only 32 % of the field observed dams (with their associated reservoirs) were contained in the compiled dataset for the Upper Bandama watershed corresponding to a gap of 68 %. This result support the findings in Zhang and Gu (2023) that revealed a low proportion of West African dams in existing datasets and emphasized that publicly available dam-related data for much of Africa (except South Africa) remain limited due to restricted access and scarce reporting from national institutions. Complementary findings by Cecchi et al. (2020) suggest that water infrastructure development in the region has largely focused on small scale systems, such as minor earth dams and farm reservoirs. These small scale structures are often constructed informally or at the community level (e.g., for community irrigation or livestock) without official documentation limiting their inventory (Brewitt and Colwyn, 2020; Hennig et al., 2023; Umukiza et al., 2023). The bias of the compiled dataset toward large scale dams and reservoirs is inherited from source datasets that primarily capture well-documented, officially reported, or remotely detectable large dams, which typically have long operational histories and broader geopolitical or hydrological significance (Biswas and Tortajada, 2001; Lehner et al., 2011; Schulz and Adams, 2019; Hennig et al., 2023; ICOLD, 2024). In addition, the use of moderate-resolution satellite imagery (e.g., 30 m) in most datasets limits the detection of small dams and their reservoirs (Pekel et al., 2016; Mulligan et al., 2020; Yang et al., 2022). Consequently, the combined limitations of documentation and detection technologies result in the underestimation of small dams and reservoirs in remote sensing-driven, curated inventory and hybrid datasets.

The spatial pattern derived from the compiled dataset revealed a pronounced concentration of dams and reservoirs between 8 and 14° N, corresponding to the Sudanian zone. This is consistent with previous studies (Cecchi et al., 2009b; Lèye et al., 2021; Fowé et al., 2023) and reflects long-term efforts to mitigate rainfall variability and sustain agricultural productivity in the savannah biome (Cecchi et al., 2009a; Lèye et al., 2021; Fowé et al., 2023; Peña-Angulo et al., 2025).

Similar to other regional compilations (Paredes-Beltran et al., 2021; Song et al., 2022) the compiled dataset for West Africa contains fewer missing entries and exhibits lower bias compared to the originate datasets. This advances the existing efforts by explicitly integrating both large and small scale reservoirs. The compiled dataset reduces spatial and attribute biases observed in individual datasets and improves the characterization of cumulative reservoir presence and distribution. This improvement is critical for regional hydrological modeling, water balance assessments, and climate land surface interactions analyses. In addition, a more balanced representation across latitudinal gradients further enhances the suitability of the compiled dataset for basin scale and regional analyses. By filling the gap between global scale inventories, this dataset provides a regionally optimized foundation for future mapping and model calibration. It also supports scenario-based assessments of dam and reservoir impacts in West Africa. However, the gaps between the compiled dataset and the ground truthing data may still have important implications for water management and Earth system science research.

The limited representation of smaller and recently constructed infrastructure, as well as gaps in associated attributes at the watershed scale, reflects structural limitations in existing dam and reservoir datasets when applied to the West African region. Further, numerous studies reported the existence of several thousand dams and reservoirs in West Africa (Cecchi et al., 2009b; Jeppe Kolding, 2016; Abobi and Wolff, 2020; Lèye et al., 2021). Yet, there is still no consensus on their total number which underscores a need for additional surveys ground-based or remotely to supplement the existing data.

Reservoirs play a crucial role in shaping hydrological processes by regulating runoff, energy balance, and water availability (Eslamian et al., 2018; Hou et al., 2022). Through their alteration of natural river flows, they exert important influence on surrounding ecosystems, agricultural productivity, and water management strategies (Kaup, 2015; Latrubesse et al., 2017). The compiled dataset accounts for 62 % of the total reservoir area and 59 % of the total storage volume reported from field campaign at watershed scale (Upper Bandama watershed). Such data gaps in dam and reservoir inventories can introduce considerable bias into hydrological and land surface analyses, resulting in misestimation of key parameters such as runoff, water storage capacity, and evapotranspiration, energy and ultimately affecting water availability estimates (Lehner et al., 2011; Helfer et al., 2012; Lehner and Grill, 2013; McManamay, 2014; Joseph et al., 2018). Dang et al. (2020) showed that models excluding reservoirs can reproduce observed streamflow by compensating through parameter adjustments, resulting in biased representations of key hydrological processes such as surface runoff, infiltration, and baseflow. These structural deficiencies may remain hidden during model calibration. However, they become evident when the models are applied to climate change impact assessments. In such cases, projections of low, mean, and high flows differ substantially from those produced by models that explicitly represent reservoirs. Such discrepancies highlight the importance of incorporating a better representation of dams and reservoirs to ensure physically realistic simulations and robust future water resource and fluxes assessments.

In semi-arid regions where communities depend heavily on reservoirs for water supply, inaccuracies in estimating storage capacity or spatial distribution can misguide water allocation decisions and undermine policy planning (Zogheib et al., 2018; Ekka et al., 2024). For instance, in the Volta Basin, mismanagement and incorrect estimation of available water resources have hindered equitable allocation among stakeholders (Youkhana and Laube, 2009). The competing demands for water between agricultural, domestic, and industrial uses have led to conflicts among different water users when actual water availability does not meet anticipated allocations (Youkhana and Laube, 2009). Therefore, in data-scarce regions such as West Africa, the use of these datasets requires balancing trade-offs between spatial resolution, completeness, accessibility, uncertainty in available water, and reliability.

Users of the compiled dataset should be cautious of not only the inherited errors from sources but also the potential errors stemming from the compilation process such as human errors and bias-trade-off decisions when resolving conflicting attribute values (Kopperud et al., 2019; Song et al., 2022; Wang et al., 2022; Ye et al., 2023). These issues can amplify uncertainty in available water estimates in model predictions and other downstream applications, underscoring the importance of transparent error reporting and continuous data validation.

Enhancing the quality and completeness of data on small sacle and recently constructed dam and reservoir in West Africa requires coordinated efforts that combine field validation, participatory approaches, and technological innovation. The ground-truthing of dam and reservoir data at watershed scale in this study revealed a much higher number of small infrastructures than recorded in the compiled data. This outcome highlights the importance of participatory mapping in identifying informal or community-built reservoirs that are often absent from official records. Expanding such participatory approaches to a regional level would improve data coverage and consistency (Rahman et al., 2025). This could be achieved by promoting open data policies across West African countries, ensuring the official publication of national inventories, and strengthening regional collaboration through basin organizations and water management authorities. Incorporating data-sharing requirements into donor-funded water projects would encourage regular updates and reduce data fragmentation. Moreover, establishing formal mechanisms for continuous dataset updates, supported by transparent exchange among national agencies and research institutions, would also help ensure that newly constructed or modified reservoirs are promptly included. Lastly, future research should focus on integrating field-based and participatory data with high-resolution satellite imagery, drone surveys, and machine-learning-based detection methods (Jing et al., 2021; Utama et al., 2024; Wang et al., 2025). Such a combination would enhance the detection and documentation of small and newly constructed dams and reservoirs, contributing to a more complete and up-to-date representation of water infrastructure across West Africa.

Despite the harmonization of dam and reservoir data in West Africa, the compiled dataset remains subject to several limitations. The quality assessment relied on field observations from a single watershed. Although this watershed scale comparison represents an important step toward validating existing datasets against locally observed information, the Upper Bandama watershed cannot capture the full diversity of environmental, hydrological, institutional, and data-availability conditions across West Africa. The performance of the compiled dataset may therefore differ in other climatic regions, river basins, and administrative contexts. Moreover, the field-reference dataset may itself be incomplete because some informal, inaccessible, recently constructed, or locally unreported dams could not be identified during the survey. These limitations introduce uncertainty into estimates of omission and dataset accuracy. Future assessments should include additional watersheds across the Guinean, Sudanian, and Sahelian climatic zones.

It is also important to note that the original datasets used in the compilation are not fully independent. Many of these datasets overlap, incorporate information from earlier inventories, or use similar remote-sensing products and mapping procedures. Consequently, agreement among multiple sources does not necessarily represent independent confirmation of a record. Additionally, the source field in the compiled dataset should be interpreted cautiously. It identifies the primary contributing source retained during harmonization rather than providing a complete record-level provenance history of all datasets that contributed information to a given dam or reservoir. Future versions could strengthen data traceability by recording all contributing sources, documenting attribute-level provenance, and assigning confidence scores based on source agreement, spatial accuracy, and attribute completeness.

Several attributes remain incomplete or uncertain. Reservoir surface area, storage capacity, watershed area, construction year, dam height, and other management-related characteristics are unavailable for some records. Missing values limit regional comparisons and may introduce bias because larger and formally managed infrastructure is generally better documented than small or informal structures. Differences in measurement methods, observation dates, spatial resolution, and definitions across the original datasets may also reduce the comparability of attribute values. Future work should establish standardized procedures for estimating missing reservoir areas and volumes, delineating upstream watersheds, and quantifying the uncertainty associated with derived attributes.