Objective sampling design in a highly heterogeneous landscape - characterizing environmental determinants of malaria vector distribution in French Guiana, in the Amazonian region

Abstract

BackgroundSampling design is a key issue when establishing species inventories and characterizing habitats within highly heterogeneous landscapes. Sampling efforts in such environments may be constrained and many field studies only rely on subjective and/or qualitative approaches to design collection strategy. The region of Cacao, in French Guiana, provides an excellent study site to understand the presence and abundance of Anopheles mosquitoes, their species dynamics and the transmission risk of malaria across various environments. We propose an objective methodology to define a stratified sampling design. Following thorough environmental characterization, a factorial analysis of mixed groups allows the data to be reduced and non-collinear principal components to be identified while balancing the influences of the different environmental factors. Such components defined new variables which could then be used in a robust k-means clustering procedure. Then, we identified five clusters that corresponded to our sampling strata and selected sampling sites in each stratum.ResultsWe validated our method by comparing the species overlap of entomological collections from selected sites and the environmental similarities of the same sites. The Morisita index was significantly correlated (Pearson linear correlation) with environmental similarity based on i) the balanced environmental variable groups considered jointly (p = 0.001) and ii) land cover/use (p-value << 0.001). The Jaccard index was significantly correlated with land cover/use-based environmental similarity (p-value = 0.001).ConclusionsThe results validate our sampling approach. Land cover/use maps (based on high spatial resolution satellite images) were shown to be particularly useful when studying the presence, density and diversity of Anopheles mosquitoes at local scales and in very heterogeneous landscapes.

Improved spatial ecological sampling using open data and standardization: an example from malaria mosquito surveillance

“…Our literature review shows that the specifics of sampling design are poorly reported and we therefore suggest that even when sampling is based on expert-opinion decisions, a full description of the sampling design should be provided to make the sampling repeatable or comparable or usable for subsequent similar studies. For example, field constraints such as presence of the disease or vector or host, vegetation type and density, elevation, field hostility, logistic feasibility, potential interference, human proximity, breeding sites, and risk of trapping material theft (13,20,21,24), which often are the major influence in the sampling design need to be declared and described. In fact, previous sampling campaigns are often used to inform future sampling design, and therefore standardization of sampling designs and protocols are now a priority (12).…”

“…One method is to describe ecological strata in terms of transformed environmental variation (i.e. factorial analyses) (13), but the results can be difficult to interpret. By contrast, discriminant analysis (DA) requires less computational time and resources because no parameter tuning is required (54).…”

“…Our literature search in Web of Science on spatial sampling of mosquitoes (search terms: mosquito or anopheles AND sampling AND spatial, in title/keywords/abstract) shows that while all studies provide a general description of the sampling design, only a limited number of papers (i.e. (8)(9)(10)(11)) give a detailed description of the rationale, decisions and calculations related to the "Where, When, How and How many" samples to collect (see for example the reviews from (12) and (13)). In other literature, partial justification of the sampling design is provided.…”

“…Quantitative stratification is usually performed by identifying a set of (independent) environmental variables that can be used to define strata within which the property or properties under study (i.e. insecticide resistance) is/are relatively homogeneous (13). Unless the spatial or spatiotemporal process (i.e.…”

“…This availability of open data (largely remote sensing) for almost every global location, combined with appropriate spatiotemporal algorithms (15), make quantitative ecological stratification more accessible as a preliminary step to any sampling programme. Nevertheless "very few studies propose, at an early phase of research work, objective sampling strategies that are consistent with both study goals and constraints" (13).…”