Mapping the spatial distribution of Lippia javanica (Burm. f.) Spreng using Sentinel-2 and SRTM-derived topographic data in malaria endemic environment

Abstract

Lippia javanica (L. javanica) is one of the commonly used ethnobotanical plant species for controlling malaria globally. Accurate mapping of L. javanica species is important for malaria control interventions that require geospatial information for the assessment of malaria distribution and monitoring especially in communities that have limited access to western malaria medicine. Currently, high spatial resolution information pertaining the distribution and habitat suitability of L. javanica species is very rare. The high resolution mapping could assist in identifying potential niche areas of ethnobotanically important species and to facilitate community health and wellness against malaria. In this study, we tested the ability of high spatial resolution Sentinel-2 (S-2) derived variables and Shuttle Radar Topography Mission (SRTM)-derived topographic variables to predict the distribution of L. javanica in the Vhembe District Municipality (South Africa). The relationship between remote sensing variables and the occurrence data of L. javanica was assessed using coefficient of determination (R2). Here, for the first time we compared three commonly used species distribution models (logistic regression, Maxent and ensemble models) to derive the best possible subsets of environmental predictors, and to produce the species distribution map that could aid in identifying areas were L. javanica occurs for use against malaria vectors. Various validation matrices such as the overall accuracy (OA), model sensitivity (Sn) and specificity (Sp), and the true skill statistics (TSS) were employed to test the robustness of the resultant models. The results showed a superior performance of weighted ensemble model, which yielded higher overall accuracy (91.3%, TSS = 0.66) than both logistic regression (OA = 84.4%, TSS = 0.42) and Maxent (95.6%, TSS = 0.73). The indices derived from the Sentinel’s red edge bands were the most contributory variables in both logistic regression and Maxent. The normalized difference averaged red edge vegetation index (NDARVI) and the normalized difference red edge1 vegetation index (NDVIre) contributed 39.25% (p = 0.0002) and 32.50% in LR and Maxent models respectively. Slope was the most significant SRTM-derived variable correlated to L. javanica occurrence in all models. The results of this study show that high resolution Sentinel-2 data can be used to map hardy shrub species at higher accuracies using ensemble model. The derived occurrence map of L. javanica could assist in updating the currently coarse resolution distribution map of species required for its aromatic ecological service against malaria vectors.