Socio-spatial Modelling of Village Territory Boundaries in North and Centre Benin

Socio-spatial Modelling of Village Territory Boundaries in North and Centre Benin | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Socio-spatial Modelling of Village Territory Boundaries in North and Centre Benin Omoto Aurelle Christelle Sedegnan, Comlan Hervé Sossou, Raffaele Gaetano, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7242782/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Accurate delineation of village boundaries is essential for analysing agricultural and socio-economic dynamics, yet such spatial data are often unavailable in many African countries. Ground surveys can provide this information, but they are time-consuming and costly. This study evaluates spatial models for the automatic delineation of village territories using widely available data: village coordinates, population census data, and agroecological zones. We applied five fixed-shape models (10 km square; buffers of 2.5 km, 5 km, 10 km radius; Voronoi) and two population-weighted models (buffer and Weighted Nearest Neighbour, WNN) to a dataset of 1,059 villages in northern and central Benin. This geospatial database was compiled from public sources and refined through cleaning and validation processes. Model outputs were assessed using participatory mapping data from 62 villages across four agroecological zones. Evaluation relied on four geometric indicators (territory area, user and producer accuracies, F1-score) and two landscape-based indicators (cropland fraction and NDVI). Population-weighted models outperformed fixed models on geometric criteria, with the WNN model achieving the highest F1-score (54.1% vs. 46.1% for the 5 km buffer). Landscape indicators revealed substantial ecoclimatic regional variation but limited model discrimination, suggesting similar landscapes across neighbouring villages. Integrating agroecological zoning notably improved model accuracy at the regional level. Population-weighted models demonstrated adequate precision for applications such as linking household surveys with satellite data. However, their performance declined near national borders or large natural features. The proposed methodology is scalable and reproducible across African regions where detailed administrative boundaries are lacking. spatial model buffer Voronoi population settlements Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 Figure 13 Figure 14 Figure 15 Figure 16 Figure 17 Figure 18 Highlights Spatial models for automatic delineation of village territory boundary are compared. Models are evaluated against participatory maps of 62 village boundaries in Benin. Models geometric and landscape performances differ across agroecological zones. Population-weighted models, especially Voronoi-based, outperform standard approaches. 1. Introduction The use of Earth observation data in scientific research and land-use planning has grown substantially in recent years. Satellite images provide access to direct information, at different spatial scales, on land cover and land conversion, and the conditions of natural and cultivated vegetation. These data are used in many applications, such as monitoring the state and dynamics of natural resources, agricultural season, human settlements, and infrastructures. Alongside these developments, an expanding body of research is exploring the potential of Earth observation data to derive indirect socio-economic indicators. The underlying hypothesis—that landscape patterns reflect human practices—has been well established since the seminal work People and Pixels (National Research Council, 1998 ), which emphasizes the importance of linking satellite imagery with social science insights. New images (THRS) combined with artificial intelligence (AI) and cloud computing are now enabling significant progress in this field (Kugler et al., 2019 ), but this research is hampered by the definition of geographical objects of interest to human activities. Indeed, most socio-economic surveys in rural areas are carried out at the level of households or village communities, which are not objects directly visible on satellite images. While access to the territory occupied by a household seems difficult to map on a large scale (Entwisle et al., 1998 ), the mapping of village territories seems to be more accessible. This information also exists in many countries, but little on the African continent where cadastral data are scarce and often incomplete. Thus, there is a real challenge for the development of remote sensing applications in the socio-economics of rural areas in Africa, which is the delineation of village territories (not to be confused with the delineation of village settlements which concerns only the built-up part, and which has been the subject of numerous publications; e.g Liu & Liu, 2022 ). At national or continental scales, conducting ground-based village delineation through GPS receiver surveys or participatory mapping is logistically and financially infeasible. In response, the scientific community has developed spatial models to approximate village territories. These models are generally guided either by pragmatic constraints (simplicity of implementation, spatial uncertainty in village location, etc.) or by assumptions about the spatial organization of rural livelihoods—namely, that the productive environment supplying food and cash crops to a village operates at a characteristic spatial scale, itself shaped by rural settlement dynamics and the degree of lumpiness in the natural environment (Grace et al., 2019 ). The simplest and most applied approach is the use of square polygons centred on the geolocated village centroid. This method has gained popularity in recent years with the increasing availability of large-scale household survey data, such as the USAID-funded Demographic and Health Surveys (DHS) data and the World Bank Living Standards Measurement Study (LSMS). In rural areas, the geographical precision of these surveyed villages is +/- 5 km (+/- 10 km for 1% of the surveyed villages). Consequently, many studies using machine learning techniques to extract satellite-derived features rely on 10 km by 10 km square polygon to associate spatial patterns with socio-economic indicators (e.g. Zhao et al., 2019 ; Browne et al., 2021 ; Jarry et al., 2021 ; Tang et al., 2024 ). A more traditional but equally simple alternative is the use of circular buffer zones centred on the village with a fixed radius regardless of the village size (e.g. Entwisle et al., 1998 ; Malan et al., 2024 ). This model is based on the ring cultivation concept (Prudencio, 1993 ), with values generally less than or equal to 3 km, which is the distance accessible on foot to cultivate land (the radius varies according to the study region, Prudencio ( 1993 ). Other studies have introduced more sophisticated versions, using variable radii determined by village size (Wilebore & Coomes, 2016 ), or socio-demographic characteristics such as average householdsize and number of landowners (Entwisle et al., 2005 ; Watmough et al., 2013 ). However, the latter approaches require high-resolution village-level survey data, which remain unavailable or outdated in many African countries. Beyond simple geometric models, several studies have implemented landscape partitioning by assigning each location in space to nearest village’s territory. Basically, each location accounts to a Voronoi tessellation (also called Thiessen polygonation) of space based on village centroids. In this case, each village boundary is equidistant from the location of each adjacent village (Muller & Zeller, 2002 ). This method for determining village boundaries has been applied in different contexts (e.g. Malan et al., 2024 ; Brewer et al., 2022 ; Muller & Zeller, 2002 ; Hu et al., 2022 ; Wilebore & Coomes, 2016 ; Watmough et al., 2013 ). To further account for village-specific characteristics, some studies have employed weighted Voronoi polygons, where the Euclidean distance raster is adjusted by a weight factor—typically based on population size. In Sierra Leone, for example, Wilebore & Coomes ( 2016 ) and Malan et al.(2024) adjusted the size of the Voronoi polygons based on village population estimates derived from dedicated surveys conducted around a national park. In Burkina Faso, Turner et al. ( 2021 ) created population-weighted influence zones by assigning each pixel a score based on the ratio of village population to squared distance from the village and attributing each pixel to the village with the highest score. Finally, some approaches incorporate environmental constraints. In a mountainous province of northern Vietnam, Castella et al.(2005) generated initial Voronoi polygons and manually adjusted village boundaries based on topographic features such as ridgelines and watershed divides. In this study, we test the ability of various spatial models to automatically delineate village boundaries in areas where rural village boundaries are generally not surveyed; these models must rely on existing and available data, such as the location of settlements and national population surveys, that exist in many countries, to be deployable at a regional or national scale. To conduct this study, we choose the North and Centre Benin that displays a large diversity of habitats and agricultural systems. To assess the relevance of the spatial models for village boundaries delineation and analyse the validity of those models in different agroclimatic regions, we calculated a set of geometric and landscape indicators and tested whether these variables differed statistically from those derived from village boundaries surveyed on the ground through a participatory approach. 2. Methods 2.1. General approach In Benin, land tenure is rarely formalized, and access to natural resources is generally regulated through customary use rights. In this context, several key principles form the basis of our approach: Land is fully exploited by humans for various purposes, including agriculture, livestock rearing, gathering, and hunting; Access to these resources is constrained by distance and is closely tied to the spatial distribution of human populations concentrated in distinct localities; Villages and their hamlets share and exploit the same territory; Urban populations have limited involvement in land-based activities and are not associated with clearly defined territorial spaces. Based on these principles, we formulated several hypotheses regarding the delineation of village territories. First, their boundaries are influenced by the relative location of settlements, the surrounding agroecological context, and the size of the population. Second, village territories may include forests and rangelands, as these areas provide essential ecosystem services to local communities. Finally, in the case of neighbouring villages, village territory boundaries may overlap, either through intermingled land parcels or shared access to natural resources. To evaluate these hypotheses and the corresponding village territory spatial models, the first step involved constructing a geospatial database of villages at the regional scale, including population data for each locality. In Benin, as in many countries in the region, such a database is not readily available and had to be reconstructed from multiple data sources. This reconstruction was carried out in the first step of the workflow (Fig. 1 .) by combining the national geographic database of localities, demographic census data, built-up data from the GHSL-S product, and agroecological zone maps. In the second step, seven spatial models of village territories - each based on distinct assumptions - were applied to the village geospatial database. The resulting village boundaries were then evaluated based on geometric and landscape criteria, using field data collected from a sample of 62 surveyed villages. Spatial analyses were conducted using QGIS and the R programming environment. 2.2. Study area The study area covers about 98000 km² and five departments of the North and Centre Benin: Alibori, Borgou, Atacora, Donga and Collines. These departments intersect five of the eight agro-ecological zones (AEZ) of the country (Fig. 2 .). The AEZ were defined by cross-referencing agri-environmental data (climate, topography, soil, agronomic potential and constraints) with administrative boundaries (MDRAC/PNUD, 1995 ): The far North area of Benin (Zone_I) contains most of the forest reserve known as the W National Park. There is rice and vegetable crops, near the Niger River, and pastoral activities. The climate is of the Sudano-Sahelian type. The cotton-growing zone of northern Benin (Zone_II), whose name is essentially based on its specialization in cotton cultivation. This area is influenced by the continental trade winds with a Sudanian-type climate. The South Borgou food zone (Zone_III) is essentially characterized by a very high availability of agricultural land with a dominance of food crops (maize, sorghum, yam, cassava, rice, groundnuts, cowpeas, etc.), which is a major asset for food security. It is characterized by a humid Sudanese climate. The West-Atacora area (Zone_IV) benefits from the presence of the Atacora mountain range, which gives it a particular climate where temperatures are cooler. It is a cotton-food diversification zone. This area is home to an agro-sylvo-pastoral integration system. The Centre Cotton Zone (Zone_V) is the largest and most suitable for agriculture. It is home to "agricultural colonizers" who come mostly from Zone_IV. It is an area of cotton-food-cashew diversification. In the framework of the European OBSYDYA project ( https://www.obsydya.org/ ), six sites representative of agroecological zones were chosen by the project research team (Fig. 2 .). Zone_I, in the far north, was excluded due to security issues not allowing fieldwork. 2.3. Data 2.3.1. Localities and demographic databases The localities geospatial database was produced by the National Geographic Institute (IGN) of Benin in 2018. For North and Centre Benin, 10 948 localities out of the country's 23 739 are referenced, with the following attributes: name of the locality, type (city, village, district, hamlet, Fulani camp), administrative status (chief town of the district or village), name of the commune, name of the district. Other fields exist, such as population size, but they are incomplete and have not been used. In total, 33 cities, 2015 villages, 6440 hamlets, 1172 districts and 1288 Fulani camps are counted in our study area. The General Population Census Database (RGPH) was produced in 2013 by the National Institute of Statistics and Demography of Benin (INStaD). This non-georeferenced dataset includes 3 769 localities (encompassing villages, neighbourhoods, and other settlements) across the national territory, of which 1 400 are in the North and Centre regions. According to the 2013 census, the total national population was 10 008 749, with 4 114581 inhabitants residing in the North and Centre Benin. To estimate the population for the year 2018—corresponding to the reference year of the IGN database and the Global Human Settlement Layer (GHSL) raster product—a compound annual growth rate of 2.7% was applied to the 2013 figures ( https://instad.bj/statistiques/indicateurs-recents/43-population ). 2.3.2. Geospatial raster products The Global Human Settlement Layer Built-up Surface (GHSL) product provides the spatial distribution of built-up areas, expressed in square meters (Pesaresi & Politis, 2023 ). For the 2018 reference year (GHS-BUILT-S R2023A), the dataset is derived from Sentinel-2 imagery and is available at a spatial resolution of 10 meters. The data can be accessed via the Copernicus Human Settlement platform ( https://human-settlement.emergency.copernicus.eu/download.php ). The land cover map produced as part of the OBSYDYA Benin project in 2023 for North and Centre Benin, is the most recent and accurate map of the area. It is composed of 6 classes (Fig. 3. a), including 2 classes dedicated to agriculture (Cropland, Tree crops). It is produced from Sentinel2 image time series using the IOTA2 (Infrastructure for Land Use by Automatic Processing Incorporating Orfeo Toolbox Applications) processing chain (Inglada et al., 2017 ). The mean NDVI was derived from the MODIS MOD13Q1 V6 product (Terra Vegetation Indices, 16-day composite, 250 m resolution) over the period 2016–2018 (Fig. 3. b). This three-year window was selected to minimize noise caused by outlier conditions in particular years. The product was accessed via the NASA’s AppEEARS platform ( https://appeears.earthdatacloud.nasa.gov ). 2.3.3. Village survey data The models tested were validated by field surveys carried out in 62 villages spread over the OBSYDYA project sites, in four ZAEs (Fig. 4., Table 1 ). The participatory mapping approach used in this study draws on previous work on village territory delineation ( Boissiere et al., 2019 ;Wulan et al., 2020 ). The methodology can be summarized as follows: Initial small-group discussions with a minimum of three key informants were conducted to gather information on village boundaries, land tenure, areas of conflict or dispute, road access, and market locations; Village meetings involving 10–15 adult participants (men and women) were organized to refine the collected information; During these village meetings, a participatory mapping exercise is conducted (Fig. 5.). A spatial map centred on the village was used to provide an overview of the village and its surrounding localities (5–10 km). The background image consisted of a true colour Google satellite image, onto which key geographic features were overlaid, including settlement locations (main village, hamlets), roads, paths and tracks, schools, health centres, rivers and streams, and water bodies (ponds and lakes). In addition, a Qfield project with the image of the village, the localities and the infrastructures, was configured on tablets to support the village boundaries mapping. Focus group discussions were conducted using an interview guide. The main characteristics of the 62 villages are given in Table 1 . There is a great variability in the surface area of the villages surveyed, with small village areas in Zone_IV and large villages in Zone_V. Table 1 Summary of the surveyed villages by AEZ. Number of villages Mean (std) village territory area Built-up (GHSL) 1 Estimated population 2 Zone_II 16 31.9 km² (22.4) 14.88% 5454 Zone_III 26 72.5 km² (35.4) 10.69% 2804 Zone_IV 12 18.2 km² (12.4) 9.37% 2746 Zone_V 8 103 km² (61.4) 16.03% 7091 TOTAL 62 55.4 km² (44) 12.21% 4030 1 the % built is calculated within a 500 m radial buffer around the village centroid. 2 Population estimation using regression model between population (RGPH, 2018) and % of built-up area (GHSL,10m, 2018) 2.4. Methods The overall approach followed a two-phase process (Fig. 6 .): 1. the creation of a population/village geospatial database in North and Centre Benin, including population data and associated agroecological zone, and 2. the application of different spatial models for delineating village territories, and their evaluation using ground-truth data collected during village surveys. Spatial analyses were performed in QGIS and R software. 2.4.1. Creation of the population/village geospatial database This study focused exclusively on villages, defined as administrative units with a population between 1 000 and 10 000 inhabitants, in accordance with Law No. 2013-05 of 27 May 2013 on the local administrative units in the Republic of Benin (Article 65). The 10 948 localities recorded in the IGN geospatial database lack demographic data or contain only partial population information. The Global Human Settlement Layer (GHSL) was therefore used as a proxy to estimate village populations across northern and central Benin. This proxy was calibrated using data from the 2013 national population census which includes 1400 non-georeferenced localities. Although limited in coverage, this dataset supported both the calibration and selection of villages. A population growth rate of 2.7% per year was applied to update the 2013 figures to 2018, aligning with the production year of the GHSL raster data. The creation of the spatial village DB involved five main steps: Villages from the national census database (BD RGPH) were manually matched to those in the IGN geospatial database based on village names. This manual process was necessary due to inconsistencies in the spelling of locality names across the two datasets. As a result, a sample of 490 villages, evenly distributed across northern and central Benin, was compiled.; The percentage of built-up area was calculated for each sampled locality using a 500-meter radial buffer. Several buffer sizes were tested, and the 500-meter radius was found to be the most appropriate, as it consistently encompassed the core built-up area of the villages; An exponential regression model was fitted between the percentage of built-up area and the estimated 2018 population of the sampled localities, with separate models constructed for each agroecological zone; The calibrated models were applied to 1059 localities from the IGN geospatial database to estimate population size; Finally, overlapping or duplicated localities were removed, and only those with estimated populations between 1 000 and 10 000 inhabitants were retained—consistent with the definition of a village. A specific protocol was developed to resolve localities overlaps; neighbourhood and hamlet entries overlapping with a village, village overlapping with district capitals, and nearby villages sharing the same base name were removed (see examples in Appendix A.). 2.4.2. Modelling the boundaries of village territories The spatial models On the cleaned geospatial village database, seven spatial models were tested, each corresponding to different assumptions on the land access and resources exploitation by rural population : a single buffer model around village centroid, with an anisotropic assumption of land use and based on a maximum walking distance (2.5 km) or bicycle or motorcycle (5–10 km) to reach the fields ( Thenkabail & Nolte, 1996 ; Turner et al., 2021 ) or other natural resources; a population-weighted buffer model based on the assumption of an average area farmed per capita; a 10 km x 10 km square model centred of the village, which was a model widely used in studies using LSMS data which are geographically anonymized with a ± 10 km uncertainty in rural areas (Grace et al., 2019 ); Voronoi polygons, which were based on the assumption that land was spatially divided among neighbouring villages according to proximity. This model approximated equal access to surrounding land, constrained by village density and distribution ( Crawford, 2002 ; Castella et al., 2005 ); and finally, population-weighted Voronoi polygons, that we called Weighted Nearest Neighbour (WNN), where each village’s influence was adjusted based on its population size relative to neighbouring villages, thereby assigning more land to larger settlements. Considering the population-weighted buffer, the size of the buffer radius was proportional to the square root of the village population. This proportion was calculated based on an average value of the area farmed per inhabitant and per ZAE with limits of 1000 and 10000 inhabitants (see Appendix B for details). The resulting average radius calculated over North and Centre Benin is 4.7 km, which was of the same order of magnitude as our field surveys (4.2 km on average; Appendix B.). As for the population-weighted Voronoi, we adopted an approach close to the one used in Turner et al., 2021 , except for the way in which the score was computed. In our case, for each location we apply a coefficient to the Euclidean distances, d , to village centroids, by dividing each of them by a given power w of the number of inhabitants ( pop ): $$\:{d}^{{\prime\:}}\:=\frac{d}{po{p}^{w}}$$ eventually assigning locations in space to the village territory accounting for the minimum modified distance d’ . The choice of the power parameter w allowed to “tune” the effect of taking into account population in the Voronoi tessellation process. Note that for a value of w equal to zero, the result falls back to regular Voronoi polygons. Several values of w were tested on our dataset (w from 0.25 to 1, in steps of 0.25). The value w = 0.25 yielded the best visual results and was therefore used in subsequent analyses. The spatial models evaluation The indicators used to evaluate the tested models were of two types: geometric and landscape indicators. The four geometric indicators, based on village boundaries, included the village territory area, the producer accuracy, the user accuracy, and the F1-scores. The village territory area was defined as the total surface area of the polygon delineating the village territories, whether from surveyed or modelled village data. The Producer Accuracy (PA) quantified the proportion of the intersected area between the modelled village polygon and the surveyed village polygon relative to the total area of the surveyed polygon. It reflected omission errors, indicating the extent to which the spatial model fails to identify areas that should be included. The User Accuracy (UA) represented the proportion of the overlapping area between the modelled village polygon and the surveyed village polygon relative to the total area of the modelled village polygon. It reflected commission error, indicating the extent to which the spatial model incorrectly includes areas not part of the actual village. The F1-scores was a composite metric that evaluates classification performance as the harmonic mean of producer and user accuracies, balancing omission and commission errors. For landscape indicators, we selected variables reflecting land use and vegetation conditions: 1. the proportion of cropland, derived from the OBSYDYA land cover map, and 2. the mean MODIS NDVI (averaged over 2016–2018). Both indicators are calculated within the delineated village territories. 3. Results 3.1. The village geospatial database The relationships between the percentages of built-up area (on a 500 m radial buffer) and the locality population were calculated for both the entire North and Centre Benin region, and individually for each of the six agroecological zones (Fig. 8 .). At the regional scale, the relationship followed a linear model, with an r² value of 0.64 and a root mean square error (RMSE) of 894 inhabitants, based on 490 localities ( Appendix C). At the agroecological zone scale, the exponential regressions revealed different thresholds of built-up percentages to define a village (Fig. 8 .). For Zone I, the data were not sufficient to fit the regression model so an average of the proportions of the other five zones was used (5%). Zones II and III exhibited similar thresholds of built-up proportion (5%), while Zone IV showed a lower proportion (4%), and Zone V a higher one (7%). These differences reflected contrasting patterns of land organization, with settlement structures in Central Benin (Zone V) tending to be more clustered compared to those in the North region. The regression models were subsequently applied to estimate the population of localities classified as “hamlets” and “villages” in the IGN database, while “cities” were excluded given that their inhabitants are generally not engaged in agricultural activities. Localities with estimated populations below 1 000 inhabitants were excluded from the analysis, while those exceeding 10 000 inhabitants were capped at 10 000. This filtering resulted in a georeferenced database of 1 059 localities, which primarily corresponded to villages, with a minority representing hamlets (Fig. 9 .). These 1 059 localities were hereafter referred to as villages. 3.2. Testing and evaluation of village land delineation models Geometric evaluation At the scale of the study area, the average modelled village area closest to that of the surveyed villages (55.4 km²) was achieved using a 5 km radial buffer (78.5 km²; Appendix D.). However, fixed-size models did not adequately account for the variability in village areas across different agroecological zones ( Appendix D.). For their part, the variable-size models exhibited significant difficulties in accurately reproducing the village areas in Zone_V (Fig. 10 .), consistently resulting in substantial overestimates (Fig. 11 .). Outside of Zone_V, Voronoi-based models exhibited the strongest correlations ( Appendix E.) with observed village land areas (r² = 0.53 for Voronoi and r² = 0.51 for WNN), whereas the weighted buffer model showed poor performance in reproducing village areas (r² = 0.02). F1-scores across the entire dataset (Fig. 12 .) indicated that the highest classification performance was achieved by the Voronoi and WNN models, with scores of 53.7% and 54.1%, respectively. These were followed by the weighted buffer and 5 km radial buffer models, with F1-scores of 47.0% and 46.0%. The 10 km buffer model demonstrated the lowest performance, with an F1-score of just 26.0%. However, these overall scores concealed substantial variation across agroecological zones (AEZs) and error types. When disaggregated by AEZ ( Appendix F.), the spatial models achieved high F1-scores in Zones III and II (52.0% and 43.5%, respectively), and substantially lower scores in Zones IV and V (36.8% and 33.8%). The Voronoi and WNN models consistently delivered strong performance and were closely aligned, with Voronoi polygons slightly outperforming in Zones II and III, and WNN polygons having an advantage in Zones IV and V. The strong performance of these two models can be attributed to their consistently high user accuracy (> 51%) and adequate producer accuracy (> 65%) across the full dataset, as shown in Fig. 12 . and detailed in Appendix F. (F1-scores by AEZ). Thematic landscape evaluation Cropland coverage indicated a marked landscape variability across agroecological zones (Fig. 14 .), ranging from very high (> 92% in Zone II, North), to moderate (60–70% in Zones III and IV), and low (35% in Zone V). Similarly, plant productivity differed by zone (Fig. 16 .), with NDVI lower values in the North (0.44 in Zone II), intermediate values in Zones III and IV (0.51), and higher productivity in the Centre (0.61 in Zone V). These patterns aligned with the known climatic gradient and population density variations in North and Central Benin. At the scale of the study area, there was a strong correlation between measured and modelled values of both cultivated fraction and NDVI, irrespective of the spatial model used (r² >0.9; Fig. 15 ., Fig. 17 .). This high correlation generally indicated good local homogeneity of landscapes surrounding the villages. At the scale of the agroecological zones, results were more variable. Both the proportion of cultivated land (Fig. 14 .) and NDVI (Fig. 16 .) values were accurately reproduced by all models in Zone III, and are well approximated in Zones II and IV (NDVI error < 0.02; cropland error < 5%), reflecting local landscape homogeneity. Conversely, in Zone V, none of the models accurately captured the percentage of cropland, consistently underestimating it by 8–13%, with the poorest performance observed for the Voronoi-based models. This suggested greater local heterogeneity of landscapes within Zone V. 3.3. Synthesis of the results and village territories map at the North and Central Benin scale In terms of geometric evaluation—aligned with our objective of delineating village territories for the estimation of socio-economic indicators at the village level—the weighted models yielded the most reliable results. Among these, the WNN model achieved the highest F1-scores and provided a robust estimate of village land area, with user accuracy slightly surpassing that of the Voronoi model. The selection of the WNN model based on geometric criteria was further supported by the landscape evaluation, which revealed minimal discrepancies between model outputs and ground observations, apart from Zone V (Gbanlin site; Fig. 18 .c) where the modelled percentage of cropland was consistently underestimated by approximately 10%. 4. Discussions 4.1. Realistic spatial models, despite strong assumptions The spatial models evaluated in this study rely on several key assumptions related to spatial land use patterns, differentiation between urban and rural populations, and the maximum distance between village centres and cultivated fields. The results from the seven tested models allowed for a more nuanced evaluation of these assumptions. Regarding the maximum distance between cultivated areas and village centres (i.e., the radial buffer size), the 5 km buffer model yielded the best performance in terms of F1-scores, along with strong landscape similarity metrics. This average distance aligns well with values reported in the literature (Turner et al., 2021 ; Thenkabail & Nolte, 1996 ) and results in an estimated average village area of approximately 78 km², which is reasonably close to the ground survey value of 55 km². However, this model does not account for the geometric variability of village territories across different agroecological zones, limiting its representational accuracy in diverse contexts. The 10 km × 10 km square buffer model, which is frequently used in studies linking socio-economic data with remote sensing, performs poorly in our evaluation, both in terms of F1-scores and in estimating realistic village territory areas. In contrast, we did not observe significant differences between observed and modelled landscape indicators. This contrasts with the findings of (Grace et al., 2019 ), who reported that such geographic approximations can significantly affect the calculation of patterns in LSMS-based studies. These discrepancies suggest that the impact of spatial approximation may vary depending on the specific context and characteristics of the dataset. Concerning the hypothesis that village territory size varies with population (population-weighted vs. fixed spatial models), our results showed that population-weighted models generally performed better, offering improved geometric accuracy and better alignment with landscape attributes. This supports the relevance of incorporating demographic data into spatial modelling approaches for village territory delineation. Finally, when comparing population-weighted models (weighted buffers vs. weighted Voronoi), the WNN model emerged as the best overall compromise, combining strong geometric performance with good landscape alignment. Nevertheless, Voronoi-based models encounter difficulties reproducing village boundaries in locally heterogeneous regions such as Zone V. These limitations are likely due to edge effects and the sensitivity of the Voronoi method to local spatial discontinuities (Okabe et al., 2000 ). The quality of spatial modelling achieved in this study is comparable to that reported in previous research conducted in West Africa. Notably, these earlier studies typically report only the correlation between modelled territory areas and measured (or census-derived) ground-truth data. For instance, Turner et al. ( 2021 ) reported a R² of 0.61 between influence zone estimates - calculated using population-weighted buffers - and measured village areas for 24 villages located in two provinces in Northern Burkina Faso. Similarly, Wilebore & Coomes ( 2016 ) evaluated different Voronoi-based approaches in a community land of Sierra Leone. Unweighted Voronoi polygons showed a weak correlation with census areas (R² = 0.03), whereas weighted Voronoi polygons yielded a substantially higher correlation (R² = 0.46) across 98 villages. In comparison, the weighted nearest neighbour (WNN) model used in our study produced R² values ranging from 0.36 to 0.59 at the scale of agroecological zones (excluding Zone_V) based on data from 54 villages, which aligns with the range of performances reported in the literature studies. Despite the numerous approximations and assumptions inherent in our approach—as well as limitations related to the databases used, including village selection and the spatial and temporal coherence of the datasets—the results indicate that the weighted models provide a reasonable approximation of village territories. In particular, the WNN model shows promising performance, with a user accuracy of 56%. This level of accuracy is considered satisfactory given that many important environmental and socio-economic factors are not accounted for in the spatial models due to the lack of data at national or sub-national scales. Such factors include the presence of marginal lands, migratory pressures, and land tenure systems involving large-scale farms that may extend well beyond the land immediately surrounding villages ( Turner et al., 2021 ; Prudencio, 1993 & Brewer et al., 2022 ). These environmental and socio-economic conditions are known to influence the use of natural resources, agricultural practices, and ultimately the spatial extent of land exploitation. To further improve model accuracy, it would be necessary to constrain the Voronoi-based models using exclusion zones representing areas that are inaccessible or unsuitable for exploitation. Additionally, access to spatial data on villages in neighbouring countries would help reduce edge effects and improve the delineation of village territories near national borders. 4.2. Village territories with significant regional variations The combined analysis of field and satellite datasets reveals substantial heterogeneity between the agroecological zones in terms of village territory area, settlement's structure, land area used per inhabitant, proportion of land allocated to annual crops and cropland productivity. This variability poses a significant challenge for spatial modelling, with the geometric performance of the tested models varying markedly from one zone to another. Zone V is particularly illustrative of these limitations, as none of the models accurately reconstruct village territory boundaries in that area. At the scale of the agroecological zones, landscape indicators derived from modelled village boundaries align well with those obtained from ground-surveyed data, regardless of the spatial model used. This consistency suggests a local landscape homogeneity and limited variation in land use between villages within the OBSYDYA sites. However, Zone V again stands out as an exception. Here, the Voronoi and WNN models fail to accurately estimate both the proportion of cropland and mean NDVI. This is likely due to the location of the sampled villages in Gbanlin site, which are situated near the Wari-Maro and Monts Kouffé forest reserves (Fig. 18 .c). The presence of these reserves may constrain and induce anisotropic development of village territories—a spatial pattern that is not accounted for by the models currently tested (Fig. 18 .c). In conclusion, two key points should be kept in mind: 1. the importance of accounting for local specificities – through zoning – when modelling village territories, in line with Grace et al., ( 2019 ), who showed that the impact of geographical approximations on patterns calculations varies by region; and 2. the limitations of theoretical modelling approaches in areas with a high local heterogeneity of land use. These insights underscore the critical role of high-quality of zoning and land use maps in the spatial modelling process. 5. Conclusions The boundaries of village territory serve as a spatial framework for analysing the interactions between populations and their environments. They enable a better understanding of socio-economic disparities and local development dynamics. Unfortunately, such data are largely absent across many African countries, including Benin. This study proposes an automatic, reproducible method for delineating village territory using widely available data, including village geographic databases and population census data. Seven spatial models were tested. While simple approaches—such as fixed-distance buffers and Voronoi polygons—are easy to implement at large scales, more refined models (e.g., population weighted models) offer improved accuracy but require additional survey data or post-processing, limiting their scalability. The incorporation of agroecological zoning was shown to highly improve model relevance at the regional level. Although some models showed sufficient accuracy for use in remote sensing applications—such as linking household survey data with geospatial indicators—they remain approximations. These spatial models cannot replace field-based delineation and should be applied cautiously, especially in land-use planning and resource management contexts where precise spatial data and local knowledge are essential. Declarations Conflict of interest The authors declare that they have no conflicts of interest. Funding This work was funded by the European Commission, project OBSYDYA FOOD/2020/417–846. Author Contribution All authors contributed to the study conception and design. Material preparation, data collection were performed by O.A.C. S., C.H.S., N.A.K., L.G., N.R.A.A., A.B. Data Analysis were performed by O.A.C. S., C.H.S., R.G., N.A.K., A.B. The first draft of the manuscript was written by O.A.C. S., C.H.S., R.G., A.B. and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript. Acknowledgement The authors wish to thank Benin National Research Institute (INRAB) for its support in data collection, and in particular Olou Denis, Codjo Victor, Yarou Koto Jean, and Bio Nikki Bruno, who participated in the field data collection. References Boissiere, M., Duchelle, A. E., Atmadja, S., & Simonet, G. (2019). Technical guidelines for participatory village mapping exercise. CIFOR-ICRAF . https://doi.org/10.17528/cifor/007282 Brewer, T. D., Andrew, N., Gruber, B., & Kool, J. (2022). Large-sample-size assessment of socioeconomic predictors of community-level resource management occurrence. Conservation Biology , 36 (2), e13800. https://doi.org/10.1111/cobi.13800 Browne, C., Matteson, D. S., McBride, L., Hu, L., Liu, Y., Sun, Y., Wen, J., & Barrett, C. B. (2021). Multivariate random forest prediction of poverty and malnutrition prevalence. PLOS ONE , 16 (9), e0255519. https://doi.org/10.1371/journal.pone.0255519 Castella, J.-C., Manh, P. H., Kam, S. P., Villano, L., & Tronche, N. R. (2005). Analysis of village accessibility and its impact on land use dynamics in a mountainous province of northern Vietnam. Applied Geography , 25 (4), 308–326. https://doi.org/10.1016/j.apgeog.2005.07.003 Crawford, T. W. (2002). Spatial Modeling of Village Functional Territories to Support Population-Environment Linkages. In S. J. Walsh & K. A. Crews-Meyer (Eds.), Linking People, Place, and Policy: A GIScience Approach (pp. 91–111). Springer US. https://doi.org/10.1007/978-1-4615-0985-1_5 Entwisle, B., Walsh, S., J., Rindfuss, R. R., & Chamratrithirong, A. (1998). Land-Use/Land-Cover and Population Dynamics, Nang Rong, Thailand. In People and Pixels: Linking Remote Sensing and Social Science (pp. 121–144). The National Academies Press. https://doi.org/10.17226/5963 Entwisle, B., Walsh, S. J., Rindfuss, R. R., & VanWey, L. K. (2005). Population and Upland Crop Production in Nang Rong, Thailand. Population and Environment , 26 (6), 449–470. https://doi.org/10.1007/s11111-005-0007-9 Grace, K., Nagle, N. N., Burgert-Brucker, C. R., Rutzick, S., Van Riper, D. C., Dontamsetti, T., & Croft, T. (2019). Integrating Environmental Context into DHS Analysis While Protecting Participant Confidentiality: A New Remote Sensing Method. Population and Development Review , 45 (1), 197–218. https://doi.org/10.1111/padr.12222 Hu, S., Ge, Y., Liu, M., Ren, Z., & Zhang, X. (2022). Village-level poverty identification using machine learning, high-resolution images, and geospatial data. International Journal of Applied Earth Observation and Geoinformation , 107 , 102694. https://doi.org/10.1016/j.jag.2022.102694 Inglada, J., Vincent, A., Arias, M., Tardy, B., Morin, D., & Rodes, I. (2017). Operational High Resolution Land Cover Map Production at the Country Scale Using Satellite Image Time Series. Remote Sensing , 9 (1), Article 1. https://doi.org/10.3390/rs9010095 Jarry, R., Chaumont, M., Berti-Équille, L., & Subsol, G. (2021). Assessment of CNN-based Methods for Poverty Estimation from Satellite Images. Lecture Notes in Computer Science , 12667 , 550–565. https://doi.org/10.1007/978-3-030-68787-8_40 Kugler, T. A., Grace, K., Wrathall, D. J., de Sherbinin, A., Van Riper, D., Aubrecht, C., Comer, D., Adamo, S. B., Cervone, G., Engstrom, R., Hultquist, C., Gaughan, A. E., Linard, C., Moran, E., Stevens, F., Tatem, A. J., Tellman, B., & Van Den Hoek, J. (2019). People and Pixels 20 years later: The current data landscape and research trends blending population and environmental data. Population and Environment , 41 (2), 209–234. https://doi.org/10.1007/s11111-019-00326-5 Liu, L., & Liu, Z. (2022). Delineation of traditional village boundaries: The case of Haishangqiao village in the Yiluo River Basin, China. PLoS ONE , 17 (12), e0279042. https://doi.org/10.1371/journal.pone.0279042 Malan, M., Carmenta, R., Gsottbauer, E., Hofman, P., Kontoleon, A., Swinfield, T., & Voors, M. (2024). Evaluating the impacts of a large-scale voluntary REDD+ project in Sierra Leone. Nature Sustainability , 1–10. https://doi.org/10.1038/s41893-023-01256-9 MDRAC/PNUD. (1995). Les huit (8) zones agroécologiques du Bénin . Muller, D., & Zeller, M. (2002). Land use dynamics in the central highlands of Vietnam: A spatial model combining village survey data with satellite imagery interpretation. Agricultural Economics , 27 (3), 333–354. https://doi.org/10.1111/j.1574-0862.2002.tb00124.x National Research Council. (1998). People and Pixels: Linking Remote Sensing and Social Science . National Academies Press. https://doi.org/10.17226/5963 Okabe, A., Boots, B., Sugihara, K., & Chiu, S. N. (2000). Frontmatter. In Spatial Tessellations (pp. i–xv). John Wiley & Sons, Ltd. https://doi.org/10.1002/9780470317013.fmatter Pesaresi, M., & Politis, P. (2023). GHS-BUILT-S R2023A - GHS built-up surface grid, derived from Sentinel2 composite and Landsat, multitemporal (1975-2030) . https://doi.org/10.2905/9F06F36F-4B11-47EC-ABB0-4F8B7B1D72EA Prudencio, C. Y. (1993). Ring management of soils and crops in the west African semi-arid tropics: The case of the mossi farming system in Burkina Faso. Agriculture, Ecosystems & Environment , 47 (3), 237–264. https://doi.org/10.1016/0167-8809(93)90125-9 Tang, J., Zhao, X., Zhang, F., Qiu, A., & Tao, K. (2024). Poverty Estimation Using a ConvLSTM-Based Model With Multisource Remote Sensing Data: A Case Study in Nigeria. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 17 , 3516–3529. https://doi.org/10.1109/JSTARS.2024.3353754 Thenkabail, P. S., & Nolte, C. (1996). Capabilities of Landsat-5 Thematic Mapper (TM) data in regional mapping and characterization of inland valley agroecosystems in West Africa. International Journal of Remote Sensing (United Kingdom) , 17 (8). https://agris.fao.org/search/en/providers/122469/records/64775c90bc45d9ecdbc2453b Turner, M. D., Eggen, M., Teague, M. S., & Ayantunde, A. A. (2021). Variation in land endowments among villages in West Africa: Implications for land management. Land Use Policy , 111 , 105768. https://doi.org/10.1016/j.landusepol.2021.105768 Watmough, G. R., Atkinson, P. M., & Hutton, C. W. (2013). Predicting socioeconomic conditions from satellite sensor data in rural developing countries: A case study using female literacy in Assam, India. Applied Geography , 44 , 192–200. https://doi.org/10.1016/j.apgeog.2013.07.023 Wilebore, B., & Coomes, D. (2016). Combining spatial data with survey data improves predictions of boundaries between settlements. Applied Geography , 77 , 1–7. https://doi.org/10.1016/j.apgeog.2016.09.007 Wulan, T. R., Prihanto, Y., & Hidayatullah, T. (2020). Cartometric mapping methods of village boundary in Tegineneng sub-district, Pasawaran Regency, Lampung Province, Indonesia . 40th Asian Conference on Remote Sensing, ACRS 2019: Progress of Remote Sensing Technology for Smart Future. Scopus. Zhao, X., Yu, B., Liu, Y., Chen, Z., Li, Q., Wang, C., & Wu, J. (2019). Estimation of Poverty Using Random Forest Regression with Multi-Source Data: A Case Study in Bangladesh. Remote Sensing , 11 (4), Article 4. https://doi.org/10.3390/rs11040375 Additional Declarations No competing interests reported. Supplementary Files Appendix.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7242782","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":501109835,"identity":"d6ab8baf-d563-4a9c-83f4-630f14753c8c","order_by":0,"name":"Omoto Aurelle Christelle Sedegnan","email":"data:image/png;base64,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","orcid":"","institution":"Centre de Coopération Internationale en Recherche Agronomique pour le Développement","correspondingAuthor":true,"prefix":"","firstName":"Omoto","middleName":"Aurelle Christelle","lastName":"Sedegnan","suffix":""},{"id":501109837,"identity":"585c3f8a-9583-4742-b444-7af57a261354","order_by":1,"name":"Comlan Hervé Sossou","email":"","orcid":"","institution":"Institut National des Recherches Agricoles du Bénin","correspondingAuthor":false,"prefix":"","firstName":"Comlan","middleName":"Hervé","lastName":"Sossou","suffix":""},{"id":501109839,"identity":"8816ab66-222e-4ddf-995e-ee22246062b8","order_by":2,"name":"Raffaele Gaetano","email":"","orcid":"","institution":"Centre de Coopération Internationale en Recherche Agronomique pour le Développement","correspondingAuthor":false,"prefix":"","firstName":"Raffaele","middleName":"","lastName":"Gaetano","suffix":""},{"id":501109841,"identity":"353e97ca-303e-4bec-a00c-521ef55f6511","order_by":3,"name":"Naboua Abdelkader Kouhoundji","email":"","orcid":"","institution":"Université d'Abomey-Calavi","correspondingAuthor":false,"prefix":"","firstName":"Naboua","middleName":"Abdelkader","lastName":"Kouhoundji","suffix":""},{"id":501109842,"identity":"233843a0-a210-4135-9217-a4420e829d03","order_by":4,"name":"Laurent Gazull","email":"","orcid":"","institution":"Centre de Coopération Internationale en Recherche Agronomique pour le Développement","correspondingAuthor":false,"prefix":"","firstName":"Laurent","middleName":"","lastName":"Gazull","suffix":""},{"id":501109843,"identity":"8c2abe4f-23c7-4857-ba76-51b63aabf0b6","order_by":5,"name":"Nestor René Ahoyo Adjovi","email":"","orcid":"","institution":"Institut National des Recherches Agricoles du Bénin","correspondingAuthor":false,"prefix":"","firstName":"Nestor","middleName":"René Ahoyo","lastName":"Adjovi","suffix":""},{"id":501109844,"identity":"8c022c74-71b8-4eeb-985f-8f63e7f50f54","order_by":6,"name":"Agnès Bégué","email":"","orcid":"","institution":"Centre de Coopération Internationale en Recherche Agronomique pour le Développement","correspondingAuthor":false,"prefix":"","firstName":"Agnès","middleName":"","lastName":"Bégué","suffix":""}],"badges":[],"createdAt":"2025-07-29 11:38:10","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7242782/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7242782/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":89696382,"identity":"6116ab5f-f25d-4747-acc0-5da79dceacbc","added_by":"auto","created_at":"2025-08-22 18:12:23","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":74896,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart, in two steps.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/4788ec362b3cd1b6194e5abc.png"},{"id":89696788,"identity":"fb4938f4-e3a6-4ad3-8c9e-0a896233bcbe","added_by":"auto","created_at":"2025-08-22 18:20:23","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":146941,"visible":true,"origin":"","legend":"\u003cp\u003eAdministrative boundaries of the departments and agro-ecological zones (AEZ) of North and Centre Benin. The red squares correspond to the study sites of the OBSYDYA project (www.obsydya.org).\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/32ad04ebb47f02d09732708f.png"},{"id":89696790,"identity":"7920c994-e187-4313-a8e5-790e4206b518","added_by":"auto","created_at":"2025-08-22 18:20:23","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":417395,"visible":true,"origin":"","legend":"\u003cp\u003e(a) Land cover map in North and Centre Benin (source: www.obsydya.org); (b) MODIS NDVI, averaged over the 2016-2018 period. The red squares correspond to the study sites.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/2f8c8249b8d29e0226851104.png"},{"id":89697456,"identity":"db92d247-be69-41f1-b3c2-87916407fdc8","added_by":"auto","created_at":"2025-08-22 18:28:23","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":138955,"visible":true,"origin":"","legend":"\u003cp\u003ea) Location and mapping of the 62 village territories, obtained during the participatory workshops conducted in 2024-2025; b) Zoom on Ouenou and Parakou sites.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/8a4c3e591a353981c9e0a75e.png"},{"id":89696400,"identity":"235c4680-acfc-42b2-bf72-9406145b79cf","added_by":"auto","created_at":"2025-08-22 18:12:24","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":790239,"visible":true,"origin":"","legend":"\u003cp\u003eSpatial map (a) used during the participatory mapping exercise (b) for the delimitation of the village land of Sero, on the site of Awanla (Credit: A. Sedegnan 2025).\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/20753ca660743a5e2374d88d.png"},{"id":89696390,"identity":"0e7f2b17-d415-4bc2-a5dc-b25cf1552303","added_by":"auto","created_at":"2025-08-22 18:12:23","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":98955,"visible":true,"origin":"","legend":"\u003cp\u003eCreation of the geospatial database.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/d433242552a8508b0cf9f63c.png"},{"id":89696381,"identity":"045fd720-5cb4-4211-983f-49fb7916b3b2","added_by":"auto","created_at":"2025-08-22 18:12:23","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":100828,"visible":true,"origin":"","legend":"\u003cp\u003eEvaluation of spatial models for delineating village territories.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/2150233d140a61e1d0195350.png"},{"id":89696797,"identity":"93db79fb-ef97-4d1e-bc53-7da9bc70b288","added_by":"auto","created_at":"2025-08-22 18:20:24","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":184310,"visible":true,"origin":"","legend":"\u003cp\u003eCorrelation between the percentage of built-up area (GHSL 2018) and population data (2018-extrapolated RGPH data), per agroecological zone.\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/f12ba67d8d9ed048363bf10e.png"},{"id":89696392,"identity":"c134cf93-c2db-4a3b-9b96-d21eab66ef76","added_by":"auto","created_at":"2025-08-22 18:12:23","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":331351,"visible":true,"origin":"","legend":"\u003cp\u003eMap of the village database for North and Centre Benin. Green, yellow, blue and red filled dots represent IGN Benin localities retained as “villages” in this study. Grey dots indicate IGN Benin localities excluded during database processing (see text for details).\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/b91521173aaff93def53d701.png"},{"id":89696791,"identity":"c1682e3f-4262-4374-865b-33a8f860175c","added_by":"auto","created_at":"2025-08-22 18:20:23","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":34312,"visible":true,"origin":"","legend":"\u003cp\u003eHeatmap of R² values representing the correlation between village areas from ground survey data and estimated from variable-size spatial models across agroecological zones (AEZ).\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/c6aae20416184849f55dffd1.png"},{"id":89696387,"identity":"3f39c5bd-40ed-43d5-9de3-d13267c1ee19","added_by":"auto","created_at":"2025-08-22 18:12:23","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":59806,"visible":true,"origin":"","legend":"\u003cp\u003eBoxplots of village areas obtained from ground survey data (green bars) and from variable-size spatial models across agroecological zones (AEZs), including a combined \"All Zones\" category on the right. \u003cem\u003eBoxes represent the interquartile range (25\u003c/em\u003e\u003csup\u003e\u003cem\u003eth\u003c/em\u003e\u003c/sup\u003e\u003cem\u003e–75\u003c/em\u003e\u003csup\u003e\u003cem\u003eth\u003c/em\u003e\u003c/sup\u003e\u003cem\u003e percentiles), with the median as a horizontal line. Black dots indicate mean F1-scores. Whiskers extend to 1.5× interquartile range (outliers are not represented).\u003c/em\u003e\u003c/p\u003e","description":"","filename":"11.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/e5ef1b062904fdceeb0406cc.png"},{"id":89696419,"identity":"d6865a5c-27f9-4ead-bdbd-b5d9a244a52f","added_by":"auto","created_at":"2025-08-22 18:12:24","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":72834,"visible":true,"origin":"","legend":"\u003cp\u003eBoxplots of village territories F1-scores obtained for village spatial models across agroecological zones (AEZs), including a combined \"All Zones\" category on the right.\u003c/p\u003e","description":"","filename":"12.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/6d932f09ec755242c99ec70b.png"},{"id":89696798,"identity":"274d2b49-a1ce-4753-89d8-a0775f284551","added_by":"auto","created_at":"2025-08-22 18:20:24","extension":"png","order_by":13,"title":"Figure 13","display":"","copyAsset":false,"role":"figure","size":24539,"visible":true,"origin":"","legend":"\u003cp\u003eUser (left side) and producer (right side) accuracies (%) of the village spatial models, for the North and Center Benin (all zones).\u003c/p\u003e","description":"","filename":"13.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/8b800a11cf2f9eece8601e06.png"},{"id":89696792,"identity":"344d91e4-738e-42d9-96be-e86b8b31201e","added_by":"auto","created_at":"2025-08-22 18:20:23","extension":"png","order_by":14,"title":"Figure 14","display":"","copyAsset":false,"role":"figure","size":64931,"visible":true,"origin":"","legend":"\u003cp\u003eBoxplots of the percentage of cropland within village territories, obtained from ground survey data (green bars) and from spatial models across agroecological zones (AEZs), including a combined \"All Zones\" category on the right.\u003c/p\u003e","description":"","filename":"14.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/77f0fc5f469a5cbe34cb2082.png"},{"id":89696395,"identity":"75fd38ca-5f6f-4ff3-9e91-6e314fedbc63","added_by":"auto","created_at":"2025-08-22 18:12:24","extension":"png","order_by":15,"title":"Figure 15","display":"","copyAsset":false,"role":"figure","size":48451,"visible":true,"origin":"","legend":"\u003cp\u003eHeatmap of R² representing the correlation between the percentage of cropland within village territories, obtained from ground survey data and estimated from spatial models across agroecological zones (AEZs).\u003c/p\u003e","description":"","filename":"15.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/df6796a25ed6e97c0154fab7.png"},{"id":89696413,"identity":"27f74320-232a-455f-aebc-3ce06716985e","added_by":"auto","created_at":"2025-08-22 18:12:24","extension":"png","order_by":16,"title":"Figure 16","display":"","copyAsset":false,"role":"figure","size":67965,"visible":true,"origin":"","legend":"\u003cp\u003eBoxplots of the mean NDVI within village boundaries, obtained from ground survey data (green bars) and from spatial models across agroecological zones (AEZs), including a combined \"All Zones\" category on the right.\u003c/p\u003e","description":"","filename":"16.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/fa9c25cc47568515e987b2c0.png"},{"id":89697460,"identity":"f3b116f2-5618-4acf-8bd8-0ba55ccc2365","added_by":"auto","created_at":"2025-08-22 18:28:24","extension":"png","order_by":17,"title":"Figure 17","display":"","copyAsset":false,"role":"figure","size":50128,"visible":true,"origin":"","legend":"\u003cp\u003eHeatmap of R² representing the correlation between the mean NDVI within village territories, obtained from ground survey data and estimated from spatial models across agroecological zones (AEZs).\u003c/p\u003e","description":"","filename":"17.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/9d58ae28306e7b5e671e21d7.png"},{"id":89696405,"identity":"22144207-38bf-4787-9285-c9bce4e6381f","added_by":"auto","created_at":"2025-08-22 18:12:24","extension":"png","order_by":18,"title":"Figure 18","display":"","copyAsset":false,"role":"figure","size":334732,"visible":true,"origin":"","legend":"\u003cp\u003eMap of the modeled village territory boundaries, based on WNN model, for the North and Center Benin (a), with zooms for the Ouenou (b) and Gbanlin (c) sites.\u003c/p\u003e","description":"","filename":"18.png","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/ead43d71f2cacf16cfa357d3.png"},{"id":89698019,"identity":"570efdae-092c-4a6a-9b03-bf4e8bc5e271","added_by":"auto","created_at":"2025-08-22 18:44:26","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3737572,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/db47ba48-91d5-4d9c-8fbe-bbbe225a4ff3.pdf"},{"id":89696380,"identity":"c9236d73-25e9-483f-a405-9021ef4c68f0","added_by":"auto","created_at":"2025-08-22 18:12:23","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":2132314,"visible":true,"origin":"","legend":"","description":"","filename":"Appendix.docx","url":"https://assets-eu.researchsquare.com/files/rs-7242782/v1/90102c8a667c0f05f2d73eb3.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Socio-spatial Modelling of Village Territory Boundaries in North and Centre Benin","fulltext":[{"header":"Highlights","content":"\u003cp\u003eSpatial models for automatic delineation of village territory boundary are compared.\u003c/p\u003e\u003cp\u003eModels are evaluated against participatory maps of 62 village boundaries in Benin.\u003c/p\u003e\u003cp\u003eModels geometric and landscape performances differ across agroecological zones.\u003c/p\u003e\u003cp\u003ePopulation-weighted models, especially Voronoi-based, outperform standard approaches.\u003c/p\u003e"},{"header":"1. Introduction","content":"\u003cp\u003eThe use of Earth observation data in scientific research and land-use planning has grown substantially in recent years. Satellite images provide access to direct information, at different spatial scales, on land cover and land conversion, and the conditions of natural and cultivated vegetation. These data are used in many applications, such as monitoring the state and dynamics of natural resources, agricultural season, human settlements, and infrastructures. Alongside these developments, an expanding body of research is exploring the potential of Earth observation data to derive indirect socio-economic indicators. The underlying hypothesis\u0026mdash;that landscape patterns reflect human practices\u0026mdash;has been well established since the seminal work \u003cem\u003ePeople and Pixels\u003c/em\u003e (National Research Council, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e1998\u003c/span\u003e), which emphasizes the importance of linking satellite imagery with social science insights. New images (THRS) combined with artificial intelligence (AI) and cloud computing are now enabling significant progress in this field (Kugler et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), but this research is hampered by the definition of geographical objects of interest to human activities. Indeed, most socio-economic surveys in rural areas are carried out at the level of households or village communities, which are not objects directly visible on satellite images. While access to the territory occupied by a household seems difficult to map on a large scale (Entwisle et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e1998\u003c/span\u003e), the mapping of village territories seems to be more accessible. This information also exists in many countries, but little on the African continent where cadastral data are scarce and often incomplete. Thus, there is a real challenge for the development of remote sensing applications in the socio-economics of rural areas in Africa, which is the delineation of village territories (not to be confused with the delineation of village settlements which concerns only the built-up part, and which has been the subject of numerous publications; e.g Liu \u0026amp; Liu, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2022\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eAt national or continental scales, conducting ground-based village delineation through GPS receiver surveys or participatory mapping is logistically and financially infeasible. In response, the scientific community has developed spatial models to approximate village territories. These models are generally guided either by pragmatic constraints (simplicity of implementation, spatial uncertainty in village location, etc.) or by assumptions about the spatial organization of rural livelihoods\u0026mdash;namely, that the productive environment supplying food and cash crops to a village operates at a characteristic spatial scale, itself shaped by rural settlement dynamics and the degree of lumpiness in the natural environment (Grace et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2019\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe simplest and most applied approach is the use of square polygons centred on the geolocated village centroid. This method has gained popularity in recent years with the increasing availability of large-scale household survey data, such as the USAID-funded Demographic and Health Surveys (DHS) data and the World Bank Living Standards Measurement Study (LSMS). In rural areas, the geographical precision of these surveyed villages is +/- 5 km (+/- 10 km for 1% of the surveyed villages). Consequently, many studies using machine learning techniques to extract satellite-derived features rely on 10 km by 10 km square polygon to associate spatial patterns with socio-economic indicators (e.g. Zhao et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Browne et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Jarry et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Tang et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2024\u003c/span\u003e ).\u003c/p\u003e\u003cp\u003eA more traditional but equally simple alternative is the use of circular buffer zones centred on the village with a fixed radius regardless of the village size (e.g. Entwisle et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e1998\u003c/span\u003e ; Malan et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2024\u003c/span\u003e ). This model is based on the ring cultivation concept (Prudencio, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e1993\u003c/span\u003e), with values generally less than or equal to 3 km, which is the distance accessible on foot to cultivate land (the radius varies according to the study region, Prudencio (\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e1993\u003c/span\u003e). Other studies have introduced more sophisticated versions, using variable radii determined by village size (Wilebore \u0026amp; Coomes, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2016\u003c/span\u003e), or socio-demographic characteristics such as average householdsize and number of landowners (Entwisle et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2005\u003c/span\u003e ; Watmough et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2013\u003c/span\u003e). However, the latter approaches require high-resolution village-level survey data, which remain unavailable or outdated in many African countries.\u003c/p\u003e\u003cp\u003eBeyond simple geometric models, several studies have implemented landscape partitioning by assigning each location in space to nearest village\u0026rsquo;s territory. Basically, each location accounts to a Voronoi tessellation (also called Thiessen polygonation) of space based on village centroids. In this case, each village boundary is equidistant from the location of each adjacent village (Muller \u0026amp; Zeller, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2002\u003c/span\u003e). This method for determining village boundaries has been applied in different contexts (e.g. Malan et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2024\u003c/span\u003e ; Brewer et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Muller \u0026amp; Zeller, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2002\u003c/span\u003e ; Hu et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2022\u003c/span\u003e ; Wilebore \u0026amp; Coomes, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2016\u003c/span\u003e ; Watmough et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2013\u003c/span\u003e ).\u003c/p\u003e\u003cp\u003eTo further account for village-specific characteristics, some studies have employed weighted Voronoi polygons, where the Euclidean distance raster is adjusted by a weight factor\u0026mdash;typically based on population size. In Sierra Leone, for example, Wilebore \u0026amp; Coomes (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2016\u003c/span\u003e) and Malan et al.(2024) adjusted the size of the Voronoi polygons based on village population estimates derived from dedicated surveys conducted around a national park. In Burkina Faso, Turner et al. (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) created population-weighted influence zones by assigning each pixel a score based on the ratio of village population to squared distance from the village and attributing each pixel to the village with the highest score.\u003c/p\u003e\u003cp\u003eFinally, some approaches incorporate environmental constraints. In a mountainous province of northern Vietnam, Castella et al.(2005) generated initial Voronoi polygons and manually adjusted village boundaries based on topographic features such as ridgelines and watershed divides.\u003c/p\u003e\u003cp\u003eIn this study, we test the ability of various spatial models to automatically delineate village boundaries in areas where rural village boundaries are generally not surveyed; these models must rely on existing and available data, such as the location of settlements and national population surveys, that exist in many countries, to be deployable at a regional or national scale. To conduct this study, we choose the North and Centre Benin that displays a large diversity of habitats and agricultural systems. To assess the relevance of the spatial models for village boundaries delineation and analyse the validity of those models in different agroclimatic regions, we calculated a set of geometric and landscape indicators and tested whether these variables differed statistically from those derived from village boundaries surveyed on the ground through a participatory approach.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\n \u003ch2\u003e2.1. General approach\u003c/h2\u003e\n \u003cp\u003eIn Benin, land tenure is rarely formalized, and access to natural resources is generally regulated through customary use rights. In this context, several key principles form the basis of our approach:\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003e\n \u003cp\u003eLand is fully exploited by humans for various purposes, including agriculture, livestock rearing, gathering, and hunting;\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eAccess to these resources is constrained by distance and is closely tied to the spatial distribution of human populations concentrated in distinct localities;\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eVillages and their hamlets share and exploit the same territory;\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eUrban populations have limited involvement in land-based activities and are not associated with clearly defined territorial spaces.\u003c/p\u003e\n \u003c/li\u003e\n \u003c/ul\u003e\n \u003cp\u003eBased on these principles, we formulated several hypotheses regarding the delineation of village territories. First, their boundaries are influenced by the relative location of settlements, the surrounding agroecological context, and the size of the population. Second, village territories may include forests and rangelands, as these areas provide essential ecosystem services to local communities. Finally, in the case of neighbouring villages, village territory boundaries may overlap, either through intermingled land parcels or shared access to natural resources.\u003c/p\u003e\n \u003cp\u003eTo evaluate these hypotheses and the corresponding village territory spatial models, the first step involved constructing a geospatial database of villages at the regional scale, including population data for each locality. In Benin, as in many countries in the region, such a database is not readily available and had to be reconstructed from multiple data sources. This reconstruction was carried out in the first step of the workflow (Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e.) by combining the national geographic database of localities, demographic census data, built-up data from the GHSL-S product, and agroecological zone maps. In the second step, seven spatial models of village territories - each based on distinct assumptions - were applied to the village geospatial database. The resulting village boundaries were then evaluated based on geometric and landscape criteria, using field data collected from a sample of 62 surveyed villages.\u003c/p\u003e\n \u003cp\u003eSpatial analyses were conducted using QGIS and the R programming environment.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\n \u003ch2\u003e2.2. Study area\u003c/h2\u003e\n \u003cp\u003eThe study area covers about 98000 km\u0026sup2; and five departments of the North and Centre Benin: Alibori, Borgou, Atacora, Donga and Collines. These departments intersect five of the eight agro-ecological zones (AEZ) of the country (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e.).\u003c/p\u003e\n \u003cp\u003eThe AEZ were defined by cross-referencing agri-environmental data (climate, topography, soil, agronomic potential and constraints) with administrative boundaries (MDRAC/PNUD, \u003cspan class=\"CitationRef\"\u003e1995\u003c/span\u003e):\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003e\n \u003cp\u003eThe far North area of Benin (Zone_I) contains most of the forest reserve known as the W National Park. There is rice and vegetable crops, near the Niger River, and pastoral activities. The climate is of the Sudano-Sahelian type.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eThe cotton-growing zone of northern Benin (Zone_II), whose name is essentially based on its specialization in cotton cultivation. This area is influenced by the continental trade winds with a Sudanian-type climate.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eThe South Borgou food zone (Zone_III) is essentially characterized by a very high availability of agricultural land with a dominance of food crops (maize, sorghum, yam, cassava, rice, groundnuts, cowpeas, etc.), which is a major asset for food security. It is characterized by a humid Sudanese climate.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eThe West-Atacora area (Zone_IV) benefits from the presence of the Atacora mountain range, which gives it a particular climate where temperatures are cooler. It is a cotton-food diversification zone. This area is home to an agro-sylvo-pastoral integration system.\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eThe Centre Cotton Zone (Zone_V) is the largest and most suitable for agriculture. It is home to \u0026quot;agricultural colonizers\u0026quot; who come mostly from Zone_IV. It is an area of cotton-food-cashew diversification.\u003c/p\u003e\n \u003c/li\u003e\n \u003c/ul\u003e\n \u003cp\u003eIn the framework of the European OBSYDYA project (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.obsydya.org/\u003c/span\u003e\u003c/span\u003e), six sites representative of agroecological zones were chosen by the project research team (Fig. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e.). Zone_I, in the far north, was excluded due to security issues not allowing fieldwork.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\n \u003ch2\u003e2.3. Data\u003c/h2\u003e\n \u003cdiv id=\"Sec6\" class=\"Section3\"\u003e\n \u003ch2\u003e2.3.1. Localities and demographic databases\u003c/h2\u003e\n \u003cp\u003eThe localities geospatial database was produced by the National Geographic Institute (IGN) of Benin in 2018. For North and Centre Benin, 10 948 localities out of the country\u0026apos;s 23 739 are referenced, with the following attributes: name of the locality, type (city, village, district, hamlet, Fulani camp), administrative status (chief town of the district or village), name of the commune, name of the district. Other fields exist, such as population size, but they are incomplete and have not been used. In total, 33 cities, 2015 villages, 6440 hamlets, 1172 districts and 1288 Fulani camps are counted in our study area.\u003c/p\u003e\n \u003cp\u003eThe General Population Census Database (RGPH) was produced in 2013 by the National Institute of Statistics and Demography of Benin (INStaD). This non-georeferenced dataset includes 3 769 localities (encompassing villages, neighbourhoods, and other settlements) across the national territory, of which 1 400 are in the North and Centre regions.\u003c/p\u003e\n \u003cp\u003eAccording to the 2013 census, the total national population was 10 008 749, with 4 114581 inhabitants residing in the North and Centre Benin. To estimate the population for the year 2018\u0026mdash;corresponding to the reference year of the IGN database and the Global Human Settlement Layer (GHSL) raster product\u0026mdash;a compound annual growth rate of 2.7% was applied to the 2013 figures (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://instad.bj/statistiques/indicateurs-recents/43-population\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec7\" class=\"Section3\"\u003e\n \u003ch2\u003e2.3.2. Geospatial raster products\u003c/h2\u003e\n \u003cp\u003eThe Global Human Settlement Layer Built-up Surface (GHSL) product provides the spatial distribution of built-up areas, expressed in square meters (Pesaresi \u0026amp; Politis, \u003cspan class=\"CitationRef\"\u003e2023\u003c/span\u003e). For the 2018 reference year (GHS-BUILT-S R2023A), the dataset is derived from Sentinel-2 imagery and is available at a spatial resolution of 10 meters. The data can be accessed via the Copernicus Human Settlement platform (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://human-settlement.emergency.copernicus.eu/download.php\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eThe land cover map produced as part of the OBSYDYA Benin project in 2023 for North and Centre Benin, is the most recent and accurate map of the area. It is composed of 6 classes (Fig. 3. a), including 2 classes dedicated to agriculture (Cropland, Tree crops). It is produced from Sentinel2 image time series using the IOTA2 (Infrastructure for Land Use by Automatic Processing Incorporating Orfeo Toolbox Applications) processing chain (Inglada et al., \u003cspan class=\"CitationRef\"\u003e2017\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eThe mean NDVI was derived from the MODIS MOD13Q1 V6 product (Terra Vegetation Indices, 16-day composite, 250 m resolution) over the period 2016\u0026ndash;2018 (Fig.\u0026nbsp;3. b). This three-year window was selected to minimize noise caused by outlier conditions in particular years. The product was accessed via the NASA\u0026rsquo;s AppEEARS platform (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://appeears.earthdatacloud.nasa.gov\u003c/span\u003e\u003c/span\u003e).\u003c/p\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec8\" class=\"Section3\"\u003e\n \u003ch2\u003e2.3.3. Village survey data\u003c/h2\u003e\n \u003cp\u003eThe models tested were validated by field surveys carried out in 62 villages spread over the OBSYDYA project sites, in four ZAEs (Fig. 4., Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eThe participatory mapping approach used in this study draws on previous work on village territory delineation ( Boissiere et al., \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e ;Wulan et al., \u003cspan class=\"CitationRef\"\u003e2020\u003c/span\u003e ). The methodology can be summarized as follows:\u003c/p\u003e\n \u003cul\u003e\n \u003cli\u003e\n \u003cp\u003eInitial small-group discussions with a minimum of three key informants were conducted to gather information on village boundaries, land tenure, areas of conflict or dispute, road access, and market locations;\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eVillage meetings involving 10\u0026ndash;15 adult participants (men and women) were organized to refine the collected information;\u003c/p\u003e\n \u003c/li\u003e\n \u003cli\u003e\n \u003cp\u003eDuring these village meetings, a participatory mapping exercise is conducted (Fig. 5.). A spatial map centred on the village was used to provide an overview of the village and its surrounding localities (5\u0026ndash;10 km). The background image consisted of a true colour Google satellite image, onto which key geographic features were overlaid, including settlement locations (main village, hamlets), roads, paths and tracks, schools, health centres, rivers and streams, and water bodies (ponds and lakes).\u003c/p\u003e\n \u003c/li\u003e\n \u003c/ul\u003e\n \u003cp\u003eIn addition, a Qfield project with the image of the village, the localities and the infrastructures, was configured on tablets to support the village boundaries mapping. Focus group discussions were conducted using an interview guide.\u003c/p\u003e\n \u003cp\u003eThe main characteristics of the 62 villages are given in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e. There is a great variability in the surface area of the villages surveyed, with small village areas in Zone_IV and large villages in Zone_V.\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eSummary of the surveyed villages by AEZ.\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eNumber of villages\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMean (std) village territory area\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eBuilt-up (GHSL)\u003csup\u003e1\u003c/sup\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eEstimated population\u003csup\u003e2\u003c/sup\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eZone_II\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e16\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e31.9 km\u0026sup2; (22.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14.88%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5454\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eZone_III\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e26\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e72.5 km\u0026sup2; (35.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10.69%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2804\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eZone_IV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e18.2 km\u0026sup2; (12.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9.37%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2746\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eZone_V\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e103 km\u0026sup2; (61.4)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e16.03%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7091\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eTOTAL\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e55.4 km\u0026sup2; (44)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12.21%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4030\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003c/div\u003e\n \u003cp\u003e\u003csup\u003e\u003cem\u003e1\u003c/em\u003e\u003c/sup\u003e \u003cem\u003ethe % built is calculated within a 500 m radial buffer around the village centroid.\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003e\u003csup\u003e\u003cem\u003e2\u003c/em\u003e\u003c/sup\u003e \u003cem\u003ePopulation estimation using regression model between population (RGPH, 2018) and % of built-up area (GHSL,10m, 2018)\u003c/em\u003e\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\n \u003ch2\u003e2.4. Methods\u003c/h2\u003e\n \u003cp\u003eThe overall approach followed a two-phase process (Fig. \u003cspan class=\"InternalRef\"\u003e6\u003c/span\u003e.): 1. the creation of a population/village geospatial database in North and Centre Benin, including population data and associated agroecological zone, and 2. the application of different spatial models for delineating village territories, and their evaluation using ground-truth data collected during village surveys.\u003c/p\u003e\n \u003cp\u003eSpatial analyses were performed in QGIS and R software.\u003c/p\u003e\n \u003cdiv id=\"Sec10\" class=\"Section3\"\u003e\n \u003ch2\u003e2.4.1. Creation of the population/village geospatial database\u003c/h2\u003e\n \u003cp\u003eThis study focused exclusively on villages, defined as administrative units with a population between 1 000 and 10 000 inhabitants, in accordance with Law No. 2013-05 of 27 May 2013 on the local administrative units in the Republic of Benin (Article 65).\u003c/p\u003e\n \u003cp\u003eThe 10 948 localities recorded in the IGN geospatial database lack demographic data or contain only partial population information. The Global Human Settlement Layer (GHSL) was therefore used as a proxy to estimate village populations across northern and central Benin. This proxy was calibrated using data from the 2013 national population census which includes 1400 non-georeferenced localities. Although limited in coverage, this dataset supported both the calibration and selection of villages. A population growth rate of 2.7% per year was applied to update the 2013 figures to 2018, aligning with the production year of the GHSL raster data.\u003c/p\u003e\n \u003cp\u003eThe creation of the spatial village DB involved five main steps:\u003c/p\u003e\n \u003col\u003e\n \u003cli\u003eVillages from the national census database (BD RGPH) were manually matched to those in the IGN geospatial database based on village names. This manual process was necessary due to inconsistencies in the spelling of locality names across the two datasets. As a result, a sample of 490 villages, evenly distributed across northern and central Benin, was compiled.;\u003c/li\u003e\n \u003cli\u003eThe percentage of built-up area was calculated for each sampled locality using a 500-meter radial buffer. Several buffer sizes were tested, and the 500-meter radius was found to be the most appropriate, as it consistently encompassed the core built-up area of the villages;\u003c/li\u003e\n \u003cli\u003eAn exponential regression model was fitted between the percentage of built-up area and the estimated 2018 population of the sampled localities, with separate models constructed for each agroecological zone;\u003c/li\u003e\n \u003cli\u003eThe calibrated models were applied to 1059 localities from the IGN geospatial database to estimate population size;\u003c/li\u003e\n \u003cli\u003eFinally, overlapping or duplicated localities were removed, and only those with estimated populations between 1 000 and 10 000 inhabitants were retained\u0026mdash;consistent with the definition of a village. A specific protocol was developed to resolve localities overlaps; neighbourhood and hamlet entries overlapping with a village, village overlapping with district capitals, and nearby villages sharing the same base name were removed (see examples in Appendix A.).\u003c/li\u003e\n \u003c/ol\u003e\n \u003c/div\u003e\n \u003cdiv id=\"Sec11\" class=\"Section3\"\u003e\n \u003ch2\u003e2.4.2. Modelling the boundaries of village territories\u003c/h2\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003e\u003cstrong\u003eThe spatial models\u003c/strong\u003e\u003c/p\u003e\n \u003c/div\u003e\n \u003cp\u003eOn the cleaned geospatial village database, seven spatial models were tested, each corresponding to different assumptions on the land access and resources exploitation by rural population : a single buffer model around village centroid, with an anisotropic assumption of land use and based on a maximum walking distance (2.5 km) or bicycle or motorcycle (5\u0026ndash;10 km) to reach the fields ( Thenkabail \u0026amp; Nolte, \u003cspan class=\"CitationRef\"\u003e1996\u003c/span\u003e ; Turner et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e ) or other natural resources; a population-weighted buffer model based on the assumption of an average area farmed per capita; a 10 km x 10 km square model centred of the village, which was a model widely used in studies using LSMS data which are geographically anonymized with a\u0026thinsp;\u0026plusmn;\u0026thinsp;10 km uncertainty in rural areas (Grace et al., \u003cspan class=\"CitationRef\"\u003e2019\u003c/span\u003e); Voronoi polygons, which were based on the assumption that land was spatially divided among neighbouring villages according to proximity. This model approximated equal access to surrounding land, constrained by village density and distribution ( Crawford, \u003cspan class=\"CitationRef\"\u003e2002\u003c/span\u003e ; Castella et al., \u003cspan class=\"CitationRef\"\u003e2005\u003c/span\u003e ); and finally, population-weighted Voronoi polygons, that we called Weighted Nearest Neighbour (WNN), where each village\u0026rsquo;s influence was adjusted based on its population size relative to neighbouring villages, thereby assigning more land to larger settlements.\u003c/p\u003e\n \u003cp\u003eConsidering the population-weighted buffer, the size of the buffer radius was proportional to the square root of the village population. This proportion was calculated based on an average value of the area farmed per inhabitant and per ZAE with limits of 1000 and 10000 inhabitants (see \u003cspan class=\"InternalRef\"\u003eAppendix\u003c/span\u003e B for details). The resulting average radius calculated over North and Centre Benin is 4.7 km, which was of the same order of magnitude as our field surveys (4.2 km on average; \u003cspan class=\"InternalRef\"\u003eAppendix\u003c/span\u003e B.).\u003c/p\u003e\n \u003cp\u003eAs for the population-weighted Voronoi, we adopted an approach close to the one used in Turner et al., \u003cspan class=\"CitationRef\"\u003e2021\u003c/span\u003e, except for the way in which the score was computed. In our case, for each location we apply a coefficient to the Euclidean distances, \u003cem\u003ed\u003c/em\u003e, to village centroids, by dividing each of them by a given power \u003cem\u003ew\u003c/em\u003e of the number of inhabitants (\u003cem\u003epop\u003c/em\u003e):\u003c/p\u003e\n \u003cdiv id=\"Equa\" class=\"Equation\"\u003e\n \u003cdiv class=\"mathdisplay\" id=\"FileID_Equa\" name=\"EquationSource\"\u003e$$\\:{d}^{{\\prime\\:}}\\:=\\frac{d}{po{p}^{w}}$$\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003eeventually assigning locations in space to the village territory accounting for the minimum modified distance \u003cem\u003ed\u0026rsquo;\u003c/em\u003e. The choice of the power parameter w allowed to \u0026ldquo;tune\u0026rdquo; the effect of taking into account population in the Voronoi tessellation process. Note that for a value of w equal to zero, the result falls back to regular Voronoi polygons. Several values of w were tested on our dataset (w from 0.25 to 1, in steps of 0.25). The value w\u0026thinsp;=\u0026thinsp;0.25 yielded the best visual results and was therefore used in subsequent analyses.\u003c/p\u003e\n \u003cdiv class=\"BlockQuote\"\u003e\n \u003cp\u003e\u003cstrong\u003eThe spatial models evaluation\u003c/strong\u003e\u003c/p\u003e\n \u003c/div\u003e\n \u003cp\u003eThe indicators used to evaluate the tested models were of two types: geometric and landscape indicators.\u003c/p\u003e\n \u003cp\u003eThe four geometric indicators, based on village boundaries, included the village territory area, the producer accuracy, the user accuracy, and the F1-scores. The village territory area was defined as the total surface area of the polygon delineating the village territories, whether from surveyed or modelled village data. The Producer Accuracy (PA) quantified the proportion of the intersected area between the modelled village polygon and the surveyed village polygon relative to the total area of the surveyed polygon. It reflected omission errors, indicating the extent to which the spatial model fails to identify areas that should be included. The User Accuracy (UA) represented the proportion of the overlapping area between the modelled village polygon and the surveyed village polygon relative to the total area of the modelled village polygon. It reflected commission error, indicating the extent to which the spatial model incorrectly includes areas not part of the actual village. The F1-scores was a composite metric that evaluates classification performance as the harmonic mean of producer and user accuracies, balancing omission and commission errors.\u003c/p\u003e\n \u003cp\u003eFor landscape indicators, we selected variables reflecting land use and vegetation conditions: 1. the proportion of cropland, derived from the OBSYDYA land cover map, and 2. the mean MODIS NDVI (averaged over 2016\u0026ndash;2018). Both indicators are calculated within the delineated village territories.\u003c/p\u003e\n \u003c/div\u003e\n\u003c/div\u003e"},{"header":"3. Results","content":"\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\u003ch2\u003e3.1. The village geospatial database\u003c/h2\u003e\u003cp\u003eThe relationships between the percentages of built-up area (on a 500 m radial buffer) and the locality population were calculated for both the entire North and Centre Benin region, and individually for each of the six agroecological zones (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e8\u003c/span\u003e.). At the regional scale, the relationship followed a linear model, with an r\u0026sup2; value of 0.64 and a root mean square error (RMSE) of 894 inhabitants, based on 490 localities (\u003cspan refid=\"Sec20\" class=\"InternalRef\"\u003eAppendix\u003c/span\u003e C). At the agroecological zone scale, the exponential regressions revealed different thresholds of built-up percentages to define a village (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e8\u003c/span\u003e.). For Zone I, the data were not sufficient to fit the regression model so an average of the proportions of the other five zones was used (5%). Zones II and III exhibited similar thresholds of built-up proportion (5%), while Zone IV showed a lower proportion (4%), and Zone V a higher one (7%). These differences reflected contrasting patterns of land organization, with settlement structures in Central Benin (Zone V) tending to be more clustered compared to those in the North region.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThe regression models were subsequently applied to estimate the population of localities classified as \u0026ldquo;hamlets\u0026rdquo; and \u0026ldquo;villages\u0026rdquo; in the IGN database, while \u0026ldquo;cities\u0026rdquo; were excluded given that their inhabitants are generally not engaged in agricultural activities. Localities with estimated populations below 1 000 inhabitants were excluded from the analysis, while those exceeding 10 000 inhabitants were capped at 10 000. This filtering resulted in a georeferenced database of 1 059 localities, which primarily corresponded to villages, with a minority representing hamlets (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e9\u003c/span\u003e.). These 1 059 localities were hereafter referred to as villages.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003e3.2. Testing and evaluation of village land delineation models\u003c/h2\u003e\u003cp\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003e\u003cb\u003eGeometric evaluation\u003c/b\u003e\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eAt the scale of the study area, the average modelled village area closest to that of the surveyed villages (55.4 km\u0026sup2;) was achieved using a 5 km radial buffer (78.5 km\u0026sup2;; \u003cspan refid=\"Sec20\" class=\"InternalRef\"\u003eAppendix\u003c/span\u003e D.). However, fixed-size models did not adequately account for the variability in village areas across different agroecological zones (\u003cspan refid=\"Sec20\" class=\"InternalRef\"\u003eAppendix\u003c/span\u003e D.). For their part, the variable-size models exhibited significant difficulties in accurately reproducing the village areas in Zone_V (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e10\u003c/span\u003e.), consistently resulting in substantial overestimates (Fig.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e11\u003c/span\u003e.). Outside of Zone_V, Voronoi-based models exhibited the strongest correlations (\u003cspan refid=\"Sec20\" class=\"InternalRef\"\u003eAppendix\u003c/span\u003e E.) with observed village land areas (r\u0026sup2; = 0.53 for Voronoi and r\u0026sup2; = 0.51 for WNN), whereas the weighted buffer model showed poor performance in reproducing village areas (r\u0026sup2; = 0.02).\u003c/p\u003e\u003cp\u003eF1-scores across the entire dataset (Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e12\u003c/span\u003e.) indicated that the highest classification performance was achieved by the Voronoi and WNN models, with scores of 53.7% and 54.1%, respectively. These were followed by the weighted buffer and 5 km radial buffer models, with F1-scores of 47.0% and 46.0%. The 10 km buffer model demonstrated the lowest performance, with an F1-score of just 26.0%. However, these overall scores concealed substantial variation across agroecological zones (AEZs) and error types.\u003c/p\u003e\u003cp\u003eWhen disaggregated by AEZ (\u003cspan refid=\"Sec20\" class=\"InternalRef\"\u003eAppendix\u003c/span\u003e F.), the spatial models achieved high F1-scores in Zones III and II (52.0% and 43.5%, respectively), and substantially lower scores in Zones IV and V (36.8% and 33.8%). The Voronoi and WNN models consistently delivered strong performance and were closely aligned, with Voronoi polygons slightly outperforming in Zones II and III, and WNN polygons having an advantage in Zones IV and V. The strong performance of these two models can be attributed to their consistently high user accuracy (\u0026gt;\u0026thinsp;51%) and adequate producer accuracy (\u0026gt;\u0026thinsp;65%) across the full dataset, as shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig9\" class=\"InternalRef\"\u003e12\u003c/span\u003e. and detailed in \u003cspan refid=\"Sec20\" class=\"InternalRef\"\u003eAppendix\u003c/span\u003e F. (F1-scores by AEZ).\u003c/p\u003e\u003cdiv class=\"BlockQuote\"\u003e\u003cp\u003e\u003cb\u003eThematic landscape evaluation\u003c/b\u003e\u003c/p\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003eCropland coverage indicated a marked landscape variability across agroecological zones (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e14\u003c/span\u003e.), ranging from very high (\u0026gt;\u0026thinsp;92% in Zone II, North), to moderate (60\u0026ndash;70% in Zones III and IV), and low (35% in Zone V). Similarly, plant productivity differed by zone (Fig.\u0026nbsp;\u003cspan refid=\"Fig13\" class=\"InternalRef\"\u003e16\u003c/span\u003e.), with NDVI lower values in the North (0.44 in Zone II), intermediate values in Zones III and IV (0.51), and higher productivity in the Centre (0.61 in Zone V). These patterns aligned with the known climatic gradient and population density variations in North and Central Benin.\u003c/p\u003e\u003cp\u003eAt the scale of the study area, there was a strong correlation between measured and modelled values of both cultivated fraction and NDVI, irrespective of the spatial model used (r\u0026sup2; \u0026gt;0.9; Fig.\u0026nbsp;\u003cspan refid=\"Fig12\" class=\"InternalRef\"\u003e15\u003c/span\u003e., Fig.\u0026nbsp;\u003cspan refid=\"Fig14\" class=\"InternalRef\"\u003e17\u003c/span\u003e.). This high correlation generally indicated good local homogeneity of landscapes surrounding the villages.\u003c/p\u003e\u003cp\u003eAt the scale of the agroecological zones, results were more variable. Both the proportion of cultivated land (Fig.\u0026nbsp;\u003cspan refid=\"Fig11\" class=\"InternalRef\"\u003e14\u003c/span\u003e.) and NDVI (Fig.\u0026nbsp;\u003cspan refid=\"Fig13\" class=\"InternalRef\"\u003e16\u003c/span\u003e.) values were accurately reproduced by all models in Zone III, and are well approximated in Zones II and IV (NDVI error\u0026thinsp;\u0026lt;\u0026thinsp;0.02; cropland error\u0026thinsp;\u0026lt;\u0026thinsp;5%), reflecting local landscape homogeneity. Conversely, in Zone V, none of the models accurately captured the percentage of cropland, consistently underestimating it by 8\u0026ndash;13%, with the poorest performance observed for the Voronoi-based models. This suggested greater local heterogeneity of landscapes within Zone V.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003e3.3. Synthesis of the results and village territories map at the North and Central Benin scale\u003c/h2\u003e\u003cp\u003eIn terms of geometric evaluation\u0026mdash;aligned with our objective of delineating village territories for the estimation of socio-economic indicators at the village level\u0026mdash;the weighted models yielded the most reliable results. Among these, the WNN model achieved the highest F1-scores and provided a robust estimate of village land area, with user accuracy slightly surpassing that of the Voronoi model. The selection of the WNN model based on geometric criteria was further supported by the landscape evaluation, which revealed minimal discrepancies between model outputs and ground observations, apart from Zone V (Gbanlin site; Fig.\u0026nbsp;\u003cspan refid=\"Fig15\" class=\"InternalRef\"\u003e18\u003c/span\u003e.c) where the modelled percentage of cropland was consistently underestimated by approximately 10%.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"4. Discussions","content":"\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003e4.1. Realistic spatial models, despite strong assumptions\u003c/h2\u003e\u003cp\u003eThe spatial models evaluated in this study rely on several key assumptions related to spatial land use patterns, differentiation between urban and rural populations, and the maximum distance between village centres and cultivated fields. The results from the seven tested models allowed for a more nuanced evaluation of these assumptions.\u003c/p\u003e\u003cp\u003eRegarding the maximum distance between cultivated areas and village centres (i.e., the radial buffer size), the 5 km buffer model yielded the best performance in terms of F1-scores, along with strong landscape similarity metrics. This average distance aligns well with values reported in the literature (Turner et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Thenkabail \u0026amp; Nolte, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e1996\u003c/span\u003e) and results in an estimated average village area of approximately 78 km\u0026sup2;, which is reasonably close to the ground survey value of 55 km\u0026sup2;. However, this model does not account for the geometric variability of village territories across different agroecological zones, limiting its representational accuracy in diverse contexts.\u003c/p\u003e\u003cp\u003eThe 10 km \u0026times; 10 km square buffer model, which is frequently used in studies linking socio-economic data with remote sensing, performs poorly in our evaluation, both in terms of F1-scores and in estimating realistic village territory areas. In contrast, we did not observe significant differences between observed and modelled landscape indicators. This contrasts with the findings of (Grace et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), who reported that such geographic approximations can significantly affect the calculation of patterns in LSMS-based studies. These discrepancies suggest that the impact of spatial approximation may vary depending on the specific context and characteristics of the dataset.\u003c/p\u003e\u003cp\u003eConcerning the hypothesis that village territory size varies with population (population-weighted vs. fixed spatial models), our results showed that population-weighted models generally performed better, offering improved geometric accuracy and better alignment with landscape attributes. This supports the relevance of incorporating demographic data into spatial modelling approaches for village territory delineation.\u003c/p\u003e\u003cp\u003eFinally, when comparing population-weighted models (weighted buffers vs. weighted Voronoi), the WNN model emerged as the best overall compromise, combining strong geometric performance with good landscape alignment. Nevertheless, Voronoi-based models encounter difficulties reproducing village boundaries in locally heterogeneous regions such as Zone V. These limitations are likely due to edge effects and the sensitivity of the Voronoi method to local spatial discontinuities (Okabe et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2000\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe quality of spatial modelling achieved in this study is comparable to that reported in previous research conducted in West Africa. Notably, these earlier studies typically report only the correlation between modelled territory areas and measured (or census-derived) ground-truth data. For instance, Turner et al. (\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2021\u003c/span\u003e) reported a R\u0026sup2; of 0.61 between influence zone estimates - calculated using population-weighted buffers - and measured village areas for 24 villages located in two provinces in Northern Burkina Faso. Similarly, Wilebore \u0026amp; Coomes (\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2016\u003c/span\u003e) evaluated different Voronoi-based approaches in a community land of Sierra Leone. Unweighted Voronoi polygons showed a weak correlation with census areas (R\u0026sup2; = 0.03), whereas weighted Voronoi polygons yielded a substantially higher correlation (R\u0026sup2; = 0.46) across 98 villages. In comparison, the weighted nearest neighbour (WNN) model used in our study produced R\u0026sup2; values ranging from 0.36 to 0.59 at the scale of agroecological zones (excluding Zone_V) based on data from 54 villages, which aligns with the range of performances reported in the literature studies.\u003c/p\u003e\u003cp\u003eDespite the numerous approximations and assumptions inherent in our approach\u0026mdash;as well as limitations related to the databases used, including village selection and the spatial and temporal coherence of the datasets\u0026mdash;the results indicate that the weighted models provide a reasonable approximation of village territories. In particular, the WNN model shows promising performance, with a user accuracy of 56%. This level of accuracy is considered satisfactory given that many important environmental and socio-economic factors are not accounted for in the spatial models due to the lack of data at national or sub-national scales. Such factors include the presence of marginal lands, migratory pressures, and land tenure systems involving large-scale farms that may extend well beyond the land immediately surrounding villages ( Turner et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Prudencio, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e1993\u003c/span\u003e \u0026amp; Brewer et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2022\u003c/span\u003e ). These environmental and socio-economic conditions are known to influence the use of natural resources, agricultural practices, and ultimately the spatial extent of land exploitation. To further improve model accuracy, it would be necessary to constrain the Voronoi-based models using exclusion zones representing areas that are inaccessible or unsuitable for exploitation. Additionally, access to spatial data on villages in neighbouring countries would help reduce edge effects and improve the delineation of village territories near national borders.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003e4.2. Village territories with significant regional variations\u003c/h2\u003e\u003cp\u003eThe combined analysis of field and satellite datasets reveals substantial heterogeneity between the agroecological zones in terms of village territory area, settlement's structure, land area used per inhabitant, proportion of land allocated to annual crops and cropland productivity. This variability poses a significant challenge for spatial modelling, with the geometric performance of the tested models varying markedly from one zone to another. Zone V is particularly illustrative of these limitations, as none of the models accurately reconstruct village territory boundaries in that area.\u003c/p\u003e\u003cp\u003eAt the scale of the agroecological zones, landscape indicators derived from modelled village boundaries align well with those obtained from ground-surveyed data, regardless of the spatial model used. This consistency suggests a local landscape homogeneity and limited variation in land use between villages within the OBSYDYA sites. However, Zone V again stands out as an exception. Here, the Voronoi and WNN models fail to accurately estimate both the proportion of cropland and mean NDVI. This is likely due to the location of the sampled villages in Gbanlin site, which are situated near the Wari-Maro and Monts Kouff\u0026eacute; forest reserves (Fig.\u0026nbsp;\u003cspan refid=\"Fig15\" class=\"InternalRef\"\u003e18\u003c/span\u003e.c). The presence of these reserves may constrain and induce anisotropic development of village territories\u0026mdash;a spatial pattern that is not accounted for by the models currently tested (Fig.\u0026nbsp;\u003cspan refid=\"Fig15\" class=\"InternalRef\"\u003e18\u003c/span\u003e.c).\u003c/p\u003e\u003cp\u003eIn conclusion, two key points should be kept in mind: 1. the importance of accounting for local specificities \u0026ndash; through zoning \u0026ndash; when modelling village territories, in line with Grace et al., (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), who showed that the impact of geographical approximations on patterns calculations varies by region; and 2. the limitations of theoretical modelling approaches in areas with a high local heterogeneity of land use. These insights underscore the critical role of high-quality of zoning and land use maps in the spatial modelling process.\u003c/p\u003e\u003c/div\u003e"},{"header":"5. Conclusions","content":"\u003cp\u003eThe boundaries of village territory serve as a spatial framework for analysing the interactions between populations and their environments. They enable a better understanding of socio-economic disparities and local development dynamics. Unfortunately, such data are largely absent across many African countries, including Benin. This study proposes an automatic, reproducible method for delineating village territory using widely available data, including village geographic databases and population census data.\u003c/p\u003e\u003cp\u003eSeven spatial models were tested. While simple approaches\u0026mdash;such as fixed-distance buffers and Voronoi polygons\u0026mdash;are easy to implement at large scales, more refined models (e.g., population weighted models) offer improved accuracy but require additional survey data or post-processing, limiting their scalability. The incorporation of agroecological zoning was shown to highly improve model relevance at the regional level.\u003c/p\u003e\u003cp\u003eAlthough some models showed sufficient accuracy for use in remote sensing applications\u0026mdash;such as linking household survey data with geospatial indicators\u0026mdash;they remain approximations. These spatial models cannot replace field-based delineation and should be applied cautiously, especially in land-use planning and resource management contexts where precise spatial data and local knowledge are essential.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eConflict of interest\u003c/h2\u003e\u003cp\u003eThe authors declare that they have no conflicts of interest.\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eThis work was funded by the European Commission, project OBSYDYA FOOD/2020/417\u0026ndash;846.\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eAll authors contributed to the study conception and design. Material preparation, data collection were performed by O.A.C. S., C.H.S., N.A.K., L.G., N.R.A.A., A.B. Data Analysis were performed by O.A.C. S., C.H.S., R.G., N.A.K., A.B. The first draft of the manuscript was written by O.A.C. S., C.H.S., R.G., A.B. and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgement\u003c/h2\u003e\u003cp\u003eThe authors wish to thank Benin National Research Institute (INRAB) for its support in data collection, and in particular Olou Denis, Codjo Victor, Yarou Koto Jean, and Bio Nikki Bruno, who participated in the field data collection.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eBoissiere, M., Duchelle, A. E., Atmadja, S., \u0026amp; Simonet, G. (2019). Technical guidelines for participatory village mapping exercise. \u003cem\u003eCIFOR-ICRAF\u003c/em\u003e. https://doi.org/10.17528/cifor/007282\u003c/li\u003e\n\u003cli\u003eBrewer, T. D., Andrew, N., Gruber, B., \u0026amp; Kool, J. (2022). Large-sample-size assessment of socioeconomic predictors of community-level resource management occurrence. \u003cem\u003eConservation Biology\u003c/em\u003e, \u003cem\u003e36\u003c/em\u003e(2), e13800. https://doi.org/10.1111/cobi.13800\u003c/li\u003e\n\u003cli\u003eBrowne, C., Matteson, D. S., McBride, L., Hu, L., Liu, Y., Sun, Y., Wen, J., \u0026amp; Barrett, C. B. (2021). Multivariate random forest prediction of poverty and malnutrition prevalence. \u003cem\u003ePLOS ONE\u003c/em\u003e, \u003cem\u003e16\u003c/em\u003e(9), e0255519. https://doi.org/10.1371/journal.pone.0255519\u003c/li\u003e\n\u003cli\u003eCastella, J.-C., Manh, P. H., Kam, S. P., Villano, L., \u0026amp; Tronche, N. R. (2005). Analysis of village accessibility and its impact on land use dynamics in a mountainous province of northern Vietnam. \u003cem\u003eApplied Geography\u003c/em\u003e, \u003cem\u003e25\u003c/em\u003e(4), 308\u0026ndash;326. https://doi.org/10.1016/j.apgeog.2005.07.003\u003c/li\u003e\n\u003cli\u003eCrawford, T. W. (2002). Spatial Modeling of Village Functional Territories to Support Population-Environment Linkages. In S. J. Walsh \u0026amp; K. A. Crews-Meyer (Eds.), \u003cem\u003eLinking People, Place, and Policy: A GIScience Approach\u003c/em\u003e (pp. 91\u0026ndash;111). Springer US. https://doi.org/10.1007/978-1-4615-0985-1_5\u003c/li\u003e\n\u003cli\u003eEntwisle, B., Walsh, S., J., Rindfuss, R. R., \u0026amp; Chamratrithirong, A. (1998). Land-Use/Land-Cover and Population Dynamics, Nang Rong, Thailand. In \u003cem\u003ePeople and Pixels: Linking Remote Sensing and Social Science\u003c/em\u003e (pp. 121\u0026ndash;144). The National Academies Press. https://doi.org/10.17226/5963\u003c/li\u003e\n\u003cli\u003eEntwisle, B., Walsh, S. J., Rindfuss, R. R., \u0026amp; VanWey, L. K. (2005). Population and Upland Crop Production in Nang Rong, Thailand. \u003cem\u003ePopulation and Environment\u003c/em\u003e, \u003cem\u003e26\u003c/em\u003e(6), 449\u0026ndash;470. https://doi.org/10.1007/s11111-005-0007-9\u003c/li\u003e\n\u003cli\u003eGrace, K., Nagle, N. N., Burgert-Brucker, C. R., Rutzick, S., Van Riper, D. C., Dontamsetti, T., \u0026amp; Croft, T. (2019). Integrating Environmental Context into DHS Analysis While Protecting Participant Confidentiality: A New Remote Sensing Method. \u003cem\u003ePopulation and Development Review\u003c/em\u003e, \u003cem\u003e45\u003c/em\u003e(1), 197\u0026ndash;218. https://doi.org/10.1111/padr.12222\u003c/li\u003e\n\u003cli\u003eHu, S., Ge, Y., Liu, M., Ren, Z., \u0026amp; Zhang, X. (2022). Village-level poverty identification using machine learning, high-resolution images, and geospatial data. \u003cem\u003eInternational Journal of Applied Earth Observation and Geoinformation\u003c/em\u003e, \u003cem\u003e107\u003c/em\u003e, 102694. https://doi.org/10.1016/j.jag.2022.102694\u003c/li\u003e\n\u003cli\u003eInglada, J., Vincent, A., Arias, M., Tardy, B., Morin, D., \u0026amp; Rodes, I. (2017). Operational High Resolution Land Cover Map Production at the Country Scale Using Satellite Image Time Series. \u003cem\u003eRemote Sensing\u003c/em\u003e, \u003cem\u003e9\u003c/em\u003e(1), Article 1. https://doi.org/10.3390/rs9010095\u003c/li\u003e\n\u003cli\u003eJarry, R., Chaumont, M., Berti-\u0026Eacute;quille, L., \u0026amp; Subsol, G. (2021). Assessment of CNN-based Methods for Poverty Estimation from Satellite Images. \u003cem\u003eLecture Notes in Computer Science\u003c/em\u003e, \u003cem\u003e12667\u003c/em\u003e, 550\u0026ndash;565. https://doi.org/10.1007/978-3-030-68787-8_40\u003c/li\u003e\n\u003cli\u003eKugler, T. A., Grace, K., Wrathall, D. J., de Sherbinin, A., Van Riper, D., Aubrecht, C., Comer, D., Adamo, S. B., Cervone, G., Engstrom, R., Hultquist, C., Gaughan, A. E., Linard, C., Moran, E., Stevens, F., Tatem, A. J., Tellman, B., \u0026amp; Van Den Hoek, J. (2019). People and Pixels 20 years later: The current data landscape and research trends blending population and environmental data. \u003cem\u003ePopulation and Environment\u003c/em\u003e, \u003cem\u003e41\u003c/em\u003e(2), 209\u0026ndash;234. https://doi.org/10.1007/s11111-019-00326-5\u003c/li\u003e\n\u003cli\u003eLiu, L., \u0026amp; Liu, Z. (2022). Delineation of traditional village boundaries: The case of Haishangqiao village in the Yiluo River Basin, China. \u003cem\u003ePLoS ONE\u003c/em\u003e, \u003cem\u003e17\u003c/em\u003e(12), e0279042. https://doi.org/10.1371/journal.pone.0279042\u003c/li\u003e\n\u003cli\u003eMalan, M., Carmenta, R., Gsottbauer, E., Hofman, P., Kontoleon, A., Swinfield, T., \u0026amp; Voors, M. (2024). Evaluating the impacts of a large-scale voluntary REDD+ project in Sierra Leone. \u003cem\u003eNature Sustainability\u003c/em\u003e, 1\u0026ndash;10. https://doi.org/10.1038/s41893-023-01256-9\u003c/li\u003e\n\u003cli\u003eMDRAC/PNUD. (1995). \u003cem\u003eLes huit (8) zones agro\u0026eacute;cologiques du B\u0026eacute;nin\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eMuller, D., \u0026amp; Zeller, M. (2002). Land use dynamics in the central highlands of Vietnam: A spatial model combining village survey data with satellite imagery interpretation. \u003cem\u003eAgricultural Economics\u003c/em\u003e, \u003cem\u003e27\u003c/em\u003e(3), 333\u0026ndash;354. https://doi.org/10.1111/j.1574-0862.2002.tb00124.x\u003c/li\u003e\n\u003cli\u003eNational Research Council. (1998). \u003cem\u003ePeople and Pixels: Linking Remote Sensing and Social Science\u003c/em\u003e. National Academies Press. https://doi.org/10.17226/5963\u003c/li\u003e\n\u003cli\u003eOkabe, A., Boots, B., Sugihara, K., \u0026amp; Chiu, S. N. (2000). Frontmatter. In \u003cem\u003eSpatial Tessellations\u003c/em\u003e (pp. i\u0026ndash;xv). John Wiley \u0026amp; Sons, Ltd. https://doi.org/10.1002/9780470317013.fmatter\u003c/li\u003e\n\u003cli\u003ePesaresi, M., \u0026amp; Politis, P. (2023). \u003cem\u003eGHS-BUILT-S R2023A - GHS built-up surface grid, derived from Sentinel2 composite and Landsat, multitemporal (1975-2030)\u003c/em\u003e. https://doi.org/10.2905/9F06F36F-4B11-47EC-ABB0-4F8B7B1D72EA\u003c/li\u003e\n\u003cli\u003ePrudencio, C. Y. (1993). Ring management of soils and crops in the west African semi-arid tropics: The case of the mossi farming system in Burkina Faso. \u003cem\u003eAgriculture, Ecosystems \u0026amp; Environment\u003c/em\u003e, \u003cem\u003e47\u003c/em\u003e(3), 237\u0026ndash;264. https://doi.org/10.1016/0167-8809(93)90125-9\u003c/li\u003e\n\u003cli\u003eTang, J., Zhao, X., Zhang, F., Qiu, A., \u0026amp; Tao, K. (2024). Poverty Estimation Using a ConvLSTM-Based Model With Multisource Remote Sensing Data: A Case Study in Nigeria. \u003cem\u003eIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing\u003c/em\u003e, \u003cem\u003e17\u003c/em\u003e, 3516\u0026ndash;3529. https://doi.org/10.1109/JSTARS.2024.3353754\u003c/li\u003e\n\u003cli\u003eThenkabail, P. S., \u0026amp; Nolte, C. (1996). Capabilities of Landsat-5 Thematic Mapper (TM) data in regional mapping and characterization of inland valley agroecosystems in West Africa. \u003cem\u003eInternational Journal of Remote Sensing (United Kingdom)\u003c/em\u003e, \u003cem\u003e17\u003c/em\u003e(8). https://agris.fao.org/search/en/providers/122469/records/64775c90bc45d9ecdbc2453b\u003c/li\u003e\n\u003cli\u003eTurner, M. D., Eggen, M., Teague, M. S., \u0026amp; Ayantunde, A. A. (2021). Variation in land endowments among villages in West Africa: Implications for land management. \u003cem\u003eLand Use Policy\u003c/em\u003e, \u003cem\u003e111\u003c/em\u003e, 105768. https://doi.org/10.1016/j.landusepol.2021.105768\u003c/li\u003e\n\u003cli\u003eWatmough, G. R., Atkinson, P. M., \u0026amp; Hutton, C. W. (2013). Predicting socioeconomic conditions from satellite sensor data in rural developing countries: A case study using female literacy in Assam, India. \u003cem\u003eApplied Geography\u003c/em\u003e, \u003cem\u003e44\u003c/em\u003e, 192\u0026ndash;200. https://doi.org/10.1016/j.apgeog.2013.07.023\u003c/li\u003e\n\u003cli\u003eWilebore, B., \u0026amp; Coomes, D. (2016). Combining spatial data with survey data improves predictions of boundaries between settlements. \u003cem\u003eApplied Geography\u003c/em\u003e, \u003cem\u003e77\u003c/em\u003e, 1\u0026ndash;7. https://doi.org/10.1016/j.apgeog.2016.09.007\u003c/li\u003e\n\u003cli\u003eWulan, T. R., Prihanto, Y., \u0026amp; Hidayatullah, T. (2020). \u003cem\u003eCartometric mapping methods of village boundary in Tegineneng sub-district, Pasawaran Regency, Lampung Province, Indonesia\u003c/em\u003e. 40th Asian Conference on Remote Sensing, ACRS 2019: Progress of Remote Sensing Technology for Smart Future. Scopus.\u003c/li\u003e\n\u003cli\u003eZhao, X., Yu, B., Liu, Y., Chen, Z., Li, Q., Wang, C., \u0026amp; Wu, J. (2019). Estimation of Poverty Using Random Forest Regression with Multi-Source Data: A Case Study in Bangladesh. \u003cem\u003eRemote Sensing\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e(4), Article 4. https://doi.org/10.3390/rs11040375\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"spatial model, buffer, Voronoi, population, settlements","lastPublishedDoi":"10.21203/rs.3.rs-7242782/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7242782/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAccurate delineation of village boundaries is essential for analysing agricultural and socio-economic dynamics, yet such spatial data are often unavailable in many African countries. Ground surveys can provide this information, but they are time-consuming and costly. This study evaluates spatial models for the automatic delineation of village territories using widely available data: village coordinates, population census data, and agroecological zones. We applied five fixed-shape models (10 km square; buffers of 2.5 km, 5 km, 10 km radius; Voronoi) and two population-weighted models (buffer and Weighted Nearest Neighbour, WNN) to a dataset of 1,059 villages in northern and central Benin. This geospatial database was compiled from public sources and refined through cleaning and validation processes. Model outputs were assessed using participatory mapping data from 62 villages across four agroecological zones. Evaluation relied on four geometric indicators (territory area, user and producer accuracies, F1-score) and two landscape-based indicators (cropland fraction and NDVI). Population-weighted models outperformed fixed models on geometric criteria, with the WNN model achieving the highest F1-score (54.1% vs. 46.1% for the 5 km buffer). Landscape indicators revealed substantial ecoclimatic regional variation but limited model discrimination, suggesting similar landscapes across neighbouring villages. Integrating agroecological zoning notably improved model accuracy at the regional level. Population-weighted models demonstrated adequate precision for applications such as linking household surveys with satellite data. However, their performance declined near national borders or large natural features. The proposed methodology is scalable and reproducible across African regions where detailed administrative boundaries are lacking.\u003c/p\u003e","manuscriptTitle":"Socio-spatial Modelling of Village Territory Boundaries in North and Centre Benin","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-08-22 18:12:18","doi":"10.21203/rs.3.rs-7242782/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"204ce09f-68b1-4c58-9cd4-edd313bcf25c","owner":[],"postedDate":"August 22nd, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-03-18T02:38:38+00:00","versionOfRecord":[],"versionCreatedAt":"2025-08-22 18:12:18","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7242782","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7242782","identity":"rs-7242782","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}