The long-term persistence of biodiversity is the key to mine ecological restoration. Unmanned Aerial Vehicles (UAVs) offer new opportunities for accurate vegetation assessments, which are needed to adaptively manage the restored mine ecosystem. However, studies on the relationship between UAVs data and biodiversity are limited. We selected the vegetation community in the Manlailiang mining area of Ordos City, Inner Mongolia, in northwestern China, for our study. Here, we used UAVs multispectral images to identify plant types among maximum likelihood, artificial neural network, and support vector machines and compared the data accuracy. And then, we quantified species-level biodiversity by calculating α-diversity indices at the 10m*10m and 30m*30m scales, which consisted of three dimensions: richness, diversity evenness. The results showed that the support vector machine has the highest species classification accuracy among the three classification methods, with overall accuracy and kappa coefficient of 89.38 and 0.87, respectively. The α-diversity indexes were basically consistent with those by the field survey. And 10m*10m-scale data was more accurate than 30m*30m-scale for predicting plant community diversity. Moreover, shrub-dominated planted areas have a low species diversity, and we recommend considering vegetation configuration in restoration projects. Our results demonstrate that a combined approach of UAVs multispectral sensors and ground surveys is an effective method to study plant species and predict the alpha-diversity index in semi-arid mining areas on a fine scale, which can provide guidance for monitoring ecological restoration of mines.
 

Unmanned aerial vehicle; Multispectral imagery; Vegetation identification; Biodiversity; Manlailiang mine