This PhD provides a comprehensive knowledge on the impacts of anthropogenic disturbances on freshwater and mammal biodiversity, but also the impacts of a modified biodiversity on human populations in French Guiana. To answer this question, biodiversity inventories are realised using eDNA metabarcoding, an innovative method to assess biodiversity. The PhD is divided into three chapters: The first chapter is articulated in two parts. In the first part, I investigated the robustness of our eDNA metabarcoding approach by comparing fish community richness and composition of several streams and rivers, sampled with two distinct protocols. Testing the robustness of datasets to protocol variations is crucial to valorize old data, time series, or data collected in difficult to access locations. Here, I show that ecological results are poorly influenced by the differences between protocols, therefore underlining the robustness of our method. In the second part of this chapter, I tested a new application of the sampling methods. We used eDNA collected from running water to inventory non-aquatic mammals along three rivers of the Amazonian biome. To estimate the reliability of this application, I used a comparative analysis between aquatic eDNA metabarcoding and standardized visual faunistic inventories classically used for mammal monitoring in French Guiana. Here, I demonstrate that aquatic eDNA metabarcoding provide relevant information on mammal distribution, complementary to that obtained with traditional method. The second chapter gathers two studies centred around the assessment of anthropogenic impacts on fish and mammal communities. In the first study, I measured the spatial extent and the strength of anthropisation on biodiversity by analysing the relationships between the diversity of local fauna and disturbance intensity measured at multiple scales. Indeed, given the hydrologic connectivity of river networks, the fauna inhabiting river catchments may be affected by remote disturbances via the water-mediated downstream transfer of matter, energy, and chemicals. We show that low levels of anthropisation are responsible for drastic declines of fish and mammals species. In addition, we demonstrate that these declines are better explained by the integration of deforested surfaces over an area of 30 km upstream the sites than the deforested surfaces in the vicinity of the sampling sites, therefore illustrating the spread of anthropogenic disturbances along the river. In the second part of this chapter, I shifted to a regional investigation of biodiversity by quantifying the relative contribution of geographical, environmental and anthropic factors to the spatial distribution of freshwater fish. While the basin isolation mainly explained the taxonomic fish distribution between basins, I show that fish are also structured by limited dispersal capacities within basins. Contrastingly, I demonstrated that environmental sorting and anthropogenic activities mainly explain the functional distribution. Importantly, modifications of functional assemblages generated by human activities are directed towards fish species that possess unique functional attributes in the assemblages. The last chapter of the PhD is exploratory. We used a multidisciplinary approach combining sociological and ecological sciences to understand how human population are affected by the modification of biodiversity. To answers this question, we sampled sites along a river facing different level of anthropogenic pressures. In parallel, at the same sites, we realised social surveys of biodiversity perception with local Amerindians from the Wayãpi and Teko ethnic groups. Here we demonstrated that despite possessing accurate knowledge of biodiversity distributions, the biodiversity perceptions of Amerindian populations is influenced by the presence of gold mining activities, by subsistence hunting and fishing, and by environmental factors.