The study was carried out in the central and western parts of the Volta basin ( Figure 1 ). These two regions have two alternating seasons, one rainy (3 to 5 months) and the other dry (7 to 9 months) [18] , and three climatic zones that can be distinguished on the basis of annual rainfall and temperature regime. The Centre region, located in the Sudano-Sahelian zone, is characterised by an average annual temperature of 29.6˚C. Rainfall is relatively low, ranging from 600 mm to 900 mm. Previous work had indicated the presence of all the potential intermediate host species in this region [19] . Notwithstanding this, the results were encouraging in epidemiological terms [20] [21] . The Hauts-Bassins region is largely located in the Sudanian zone. The extreme north of the region is located in the Sudano-Sahelian zone. The region has an average annual temperature of 27.2˚C. Rainfall ranges from 800 mm to 1200 mm. The relief is made up of plateaux, plains, a few hills, and valleys. The region is home to the two oldest hydro-developments, the Banzon plain and the Kou Valley [22] . Prevalences are still high in this region at 25% according to [21] . According to [23] , the Hauts-Bassins region is known to be an endemic area for Schistosoma mansoni, which has replaced S. haematobium. These observations were consolidated by the work of [24] , who found a prevalence of 8.7% out of 480 pupils examined.

Five physicochemical parameters of the water were measured in situ using a HI 9829 multi-parameter. These were water temperature, electrical conductivity, salinity, pH, and dissolved oxygen. The nature of the bottom and the vegetation cover were also assessed visually. The habitat type, level of human activity, bottom composition, vegetation cover, and water regime at each measurement point were assessed subjectively based on standardized criteria. Habitat type was classified as follows: 1 for river, 2 for stream, and 3 for irrigated plains. Human activity was categorized by intensity: 1 for low, 2 for moderate, and 3 for high activity levels. The nature of the bottom was assessed using the following scale: 1 for mud, 2 for gravel, 3 for a mixture of gravel and mud, 4 for a mixture of gravel and rocks, and 5 for irrigation canals. Vegetation cover was recorded as 1 for low, 2 for medium, and 3 for high. The water regime was classified as 1 for non-permanent, 2 for intermediate, and 3 for permanent flow conditions.

At each site, water samples were taken as described in [25] and analyzed in the laboratory using an ORION AQUAMATE 8000 spectrophotometer. The parameters measured were nitrate, ammonium, total iron, and alkalinity.

At each sampling point on the site, two surveyors collected snails by direct examination of all the micro-habitats available in the water using flexible forceps for 15 minutes. For finer sediments, snails were collected by dredging with an Eckman grab. The snails collected were placed in plastic pillboxes and labeled by site, microhabitat, and collection period. The snails were taken back to the laboratory for identification and enumeration. Identification was carried out in the laboratory by examining the shell and confirmed using the identification guides of [26] [27] , and [28] .

Redundancy analysis (RDA) was used to determine the influence of local environmental variables on the distribution of snail intermediate host using the vegan rda function. Not all environmental variables follow a normal distribution, so a logarithmic transformation log1p() was performed to reduce skewness and improve the power of the statistical tests. As the data were all in different units, we standardized them using vegan’s decostand() function. A Hellinger transformation was applied to the snail abundance data to preserve the Euclidean distance between sites, reduce the influence of very abundant species, and preserve information about rare species [29] . A progressive selection of explanatory variables was carried out using the ordiR2step() function in the vegan package in order to identify significant independent variables and obtain a better RDA model. The statistical significance of the overall model was tested using anova.cca().

The relationship between local biotic and abiotic parameters and the abundance of schistosome intermediate host snails was assessed using the Boosted Regression Trees or BRT method [30] [31] . BRT models are machine learning algorithms built from the combination of the advantages of two algorithms [30] - [32] : models fitted from decision trees [33] and the bagging algorithm for calculating the global prediction from the combination of decision tree results [34] . BRTs have several advantages because: 1) BRT models can handle large databases with predictors having different characteristics, 2) they adapt to complex non-linear response curves, 3) they efficiently handle data with missing values, 4) they are insensitive to tooling, 5) BRTs also have better predictive performance, and 6) predictor independence is not a requirement as interactions are automatically modelled by the hierarchical structure of the tree [35] . Furthermore, BRTs are robust against the presence of uninformative predictors, as these predictors are discarded when selecting the best split [32] . All these advantages make the BRT model useful for ecologists to explore relationships between predictor variables and species abundance dynamics [35] - [37] .

In addition, BRTs are robust against the presence of uninformative predictor variables because these predictors are discarded during the selection of the best split [32] . BRT models were built using the gbm.step() function of the dismo v1.3.3 package [38] . To find the optimal parameters based on the recommended rules for ecological modelling, we regulated the BRT model by jointly optimising the learning rate (lr), which determines the contribution of each tree to the growing model, the tree complexity (tc), which controls the actual level of interaction in the BRT, and the bag fraction (bf) to specify the proportion of data to be selected at each step (stochasticity control) [30] . Cross-validation was used to fit and evaluate the model [30] . It specifies the number of times to randomly divide the data in order to fit and validate the model [39] . A lower value for deviance and standard error for cross-validation [37] . We fitted 20 models based on the combination of these different parameters: tc: 1, 2, 3, 4, 5; lr: 0.01, 0.05, 0.005, 0.001; and bf: 0.5 or 0.75. Among the fitted models, we found that the best model had an lr of 0.001, a tc of 4, and a bf of 0.75, with small cross-validation gaps as well. For each species, the optimal number of trees (nt) was chosen by internal cross-validation to avoid overfitting due to overdispersion. We then fitted all other models using these same optimal parameters. All BRT models were fitted using the Poisson error distribution [39] , which is appropriate for count data [40] . The selected variables are those that explain at least 40% of the variability in species abundance [31] [36] [41] . We assessed the fit of the BRT model to the data by calculating the percentage of deviance explained: 1-(residual deviance/total deviance) [42] .

A total of 14,587 freshwater snails belonging to 7 families and 21 species were collected. Of the specimens collected, 8535 specimens were intermediate hosts of human schistosomes (i.e., 58.51% of the snails). These were B. truncatus, B. forskalii, B. globosus, B. senegalensis and Bi. pfeifferi. B. truncatus (3618 specimens) and B. forskalii (2123 specimens) were the most abundant species, while B. globosus and B. senegalensis were the least abundant ( Figure 2 ).

Species such as B. forskalii, B. senegalensis, and B. truncatus are more abundant in reservoirs and irrigated plains but rare in rivers ( Table 1 ). The Kruskal-Wallis test shows that the difference in abundance is significant between habitat types (p = 0.00072). For B. senegalensis, the difference in abundance was also significant between habitat types (p = 0.019). For B. forskalii, the difference was significant between rivers and irrigated plains (p = 0.038), rivers and reservoirs (p = 0.023), and reservoirs and irrigated plains (p = 0.038). Although Bi. pfeifferi was more abundant in reservoirs and irrigated plains, the Kruskal-Wallis test showed that the difference in abundance was not significant between habitat types (p = 0.069). As for B. globosus, it is more abundant in reservoirs but rare in irrigated plains and rivers (p = 0.0075).

The RDA results show that environmental variables explain 62.32% of the variation in the presence-absence of snail species. The ANOVA permutation test used to test the overall significance of the RDA shows that the overall model is significant (p = 0.003) and that the first axis is also significant ( Table 2 ).

** = Very significant.

The RDA results show that the presence of B. truncatus is positively correlated with conductivity, pH, and latitude, but negatively correlated with altitude. The presence of B. forskalii is more positively correlated with vegetation. As for Bi. pfeifferi and B. senegalensis, their presence is positively correlated with temperature and altitude but negatively correlated with ammonium and total iron ( Figure 3 ).

The optimized BRT model explained 43.359%, 77.047%, 73.906%, 48.169%, and 23.23% of the spatiotemporal variation in abundance of B. senegalensis, B. truncatus, B. globosus, B. forskali, and Bi. pfeifferi, respectively ( Table 3 ).

With the exception of B. globosus, the main factors influencing the abundance of snail intermediate hosts of schistosomes were the physicochemical parameters of the water. Furthermore, the fitted functions reveal a non-linear and sometimes complex relationship between the explanatory variables and snail abundance. The abundance of B. senegalensis and B. truncatus was more influenced by conductivity and/or salinity and alkalinity. These associated factors contributed respectively to 58.4% and 78.2% of the predictive power of the optimized model for B. truncatus and B. senegalensis ( Figure 4 and Figure 5 ). The abundance of Bi. pfeifferi was mainly influenced by electrical conductivity (27.7%), pH (16.6%), water temperature (12.1%), and habitat type (9.2%). These variables contributed 65.6% of the model’s predictive power ( Figure 6 ). For B. globosus, the main factors influencing abundance are permanence (24.9%), vegetation cover (18.6%), alkalinity (12.6%), and salinity (9.3%). These cumulative explanatory variables contribute 65.4% of the predictive power of the optimized model ( Figure 7 ). The species was abundant in water bodies with intermediate regimes. The abundance of B. forskalii was mainly influenced by total iron (26.8%), temperature (20.9%), and pH (8.7%). These variables contributed 56.4% of the model’s predictive power ( Figure 8 ).

This study revealed the presence of five species of snail intermediate hosts of human schistosomes in the Centre and Hauts-Bassins regions. With the exception of B. truncatus, which was absent from the Hauts-Bassins region, the presence of all the other species in these regions had already been reported since 1994 [19] . The dispersal methods of B. truncatus require further study. Nevertheless, their geographical distribution pattern can be deduced by analogy with other snail species and from the limited observations that have been made. The sudden appearance of these snails in isolated basins in the Gezira region of Sudan, up to thirty kilometres from the nearest river or irrigation canal [43] , has led to the deduction that they must have been carried on the feet of aquatic birds or buried under the skin of aquatic animals. B. truncatus could also have been transported to this region by the action of run-off or floodwaters on vegetation or floating debris. It remains to be seen to what extent the country’s other rivers play a role in the spread of B. truncatus. Bulinus forskalii, which is ubiquitous in Africa, is the species most frequently encountered at the various sites surveyed. The high frequency of occurrence of this species could be explained by the natural behavioural adaptation mode, which is different for each species. B. pfeifferi is a calm-water snail that feeds on the surface and can easily be carried to the dam [44] , finding a resting place when the water current is greatly reduced, whereas B. globosus can cling to or settle on the bottom of the water and then rise to the surface [45] . In general, habitat type, vegetation cover, and human disturbance were the most important variables determining the occurrence and abundance of snail intermediate hosts. Our results are in agreement with those of Utzinger and Tanner (2000) [44] , who found that the abundance and distribution of freshwater snails were governed by both biotic and environmental factors. Our results indicate that the preferred sites for snails in terms of occurrence and abundance were habitats disturbed by waste dumps, agriculture, and other human activities (in this case, urban marshes and dam lakes characteristic of Burkina’s hydrographic network). This will lead to an increase in the concentration of detritus and eventually the proliferation of algae and aquatic plants, the basis of the diet and attachment support of Planorbid and Prosobranch snails [46] - [48] . This high occurrence could also be attributed to the low abundance and diversity of other invertebrates and fish in highly disturbed habitats, which can suppress snail populations through predation and competition. Indeed, predators reduce snail abundance directly through predation and avoidance of oviposition or indirectly through competition for food resources [49] .

The main local factors influencing the occurrence of snail populations were salinity, nitrates, pH, temperature, latitude, and vegetation. This can be explained by the fact that factors such as temperature are very important for the survival and reproduction of snail intermediate hosts of schistosomes [50] . Recent studies on the effects of temperature change on the growth, fecundity, and survival of intermediate hosts of schistosomiasis indicate that increasing temperature can modify the distribution, reproduction, growth, and survival of snail intermediate hosts of schistosomes [51] [52] . Snail intermediate hosts of schistosomes are poikilothermic. Their body temperature changes according to that of the surrounding environment [52] [53] . Snail intermediate host appears to be tolerant of salinity, conductivity, and pH. According to Camara et al. (2012), the distribution of snail species was associated with high conductivity and pH values and low dissolved oxygen values. This is the case for B. truncatus and Bi. pfeifferi, which were abundantly encountered in habitats with high electrical conductivity during this study. High conductivity values are generally associated with organic pollution and decreasing water quantity [54] [55] .

In addition to water physicochemistry, the optimised BRT model showed that snail abundance could be influenced by water body regime, vegetation cover, and habitat type. This was the case for B. globosus, whose high abundance was positively correlated with water bodies with intermittent regimes and high vegetation cover. B. globosus has already been reported in previous studies as a species that prefers biotopes with high vegetation cover [10] . In the Bi. pfeifferi species, high abundances of the species were positively correlated with developed plains. Irrigated plains provide a suitable environment for the development of snail intermediate hosts of schistosomes. The development of irrigated plains in Burkina Faso has contributed significantly to the development of snail intermediate hosts of schistosomes and to the explosion of the two forms of schistosomiasis (urinary and intestinal schistosomiasis) [10] [11] [56] . In this study, season was not included as a covariate, which is a limitation. These observations show that the occurrence and abundance of snail intermediate hosts of schistosomes cannot be associated with a single factor but rather are the result of more complex interactions of several habitat factors [57] . Integrated schistosomiasis control combines efforts to fight against morbidity and control intermediate host snails. Identifying these predictors is an important step that can guide the factors to target in implementing strategies to control the abundance of intermediate host snails of human schistosomes in aquatic ecosystems of Burkina Faso.