Mapping understory invasive plant species with field and remotely sensed data in Chitwan, Nepal

Abstract

Monitoring invasive species distribution and prevalence is important, but direct field-based assessment is often impractical. In this paper, we introduce and validate a cost-effective method for mapping understory invasive plant species. We utilized Landsat imagery, spectral mixture analysis (SMA) and a maximum entropy (Maxent) modeling framework to map the spatial extent of Mikania micrantha in Chitwan National Park, Nepal and community forests within its buffer zone. We developed a spectral library from reference and image sources and applied multiple endmember SMA (MESMA) to selected Landsat imagery. Incorporating the resultant green vegetation and shade fractions into Maxent, we mapped the distribution of understory M. micrantha in the study area, with training and testing Area under Curve (AUC) values around 0.80, and kappa around 0.55. In vegetated places, especially mature forests, an increase in green vegetation fraction and decrease in shade fraction was associated with higher likelihood of M. micrantha presence. In addition, the inclusion of elevation as a model input further improved map accuracy (AUC around 0.95; kappa around 0.80). Elevation, a surrogate for distance to water in this case, proved to be the determining factor of M. micrantha's distribution in the study area. The combination of MESMA and Maxent can provide significant opportunities for understanding understory vegetation distribution, and contribute to ecological restoration, biodiversity conservation, and provision of sustainable ecosystem services in protected areas.