1 Introduction

Plant species diversity is a fundamental goal for ecological restoration particularly in mining areas where functional and often biodiverse ecosystems need to be reinstated (Brancalion and Holl, 2020; McKenna et al., 2020; Han et al., 2021; Chen et al., 2022). Ensuring complete species mixes are reinstated in restoration programs can improve ecosystem function and stability, particularly for alleviating impacts of extreme or unexpected environmental events (Naeem et al., 1994; Gaston, 2000; Isbell et al., 2015; Hauser et al., 2021; Fremout et al., 2022). The UN Decade of Ecosystem Restoration highlights the importance of re-vegetating with complex species mixes (Zhang et al., 2021), while various ecological restoration standards (Gann et al. 2019, Young et al. 2022) emphasise the importance of matching native reference sites with species mixes to ensure restored areas are resilient and functional (Yang et al., 2022; Young et al., 2022). Despite major investments by governments, mining companies, and land managers to restore ecosystems after mining, restoration for many mines particularly those in native ecosystems remains difficult and complex to achieve (Menz et al., 2013). Importantly, once resource extraction has been completed many companies are not willing to invest additional time, resources, and funding to monitor and evaluate the effectiveness of restoration practices, severely hindering the capacity to keep projects on track (Galatowitsch and Bohnen, 2020). An observed phenomenon is that restoration projects tend to prioritize short-term vegetation cover over long-term biodiversity protection (Hoffmann, 2022), ultimately leading to their failure (Chen et al., 2022). Thus, accurate and efficient vegetation diversity monitoring is crucial for tracking the success of restoration projects and guiding adaptive management strategies (Hoffmann, 2022).
Traditional vegetation survey approaches rely primarily on manual ground surveys at plot level to obtain plant species diversity and abundance and/or structural data but these are often costly, time-consuming, and limited in their ability to monitor large areas (Anderson, 2018; Reddy, 2021). Satellite imagery, on the other hand, has proven to be a valuable tool for automated or semi-automated and repeatable vegetation mapping across a range of spatial scales, spanning from regional to community-level, and from low to high spatial resolutions (Li et al., 2021; Villoslada et al., 2020; Randin et al., 2020). However, mining areas are often subject to intense disturbance (Hou et al., 2021; Jiang et al., 2022), resulting in relatively small spatial extent confounded by the often dense regeneration of uniformly aged plants in mine restoration that make it difficult to accurately differentiate taxa when deploying high altitude satellite approaches. In particular, the optical detection of α diversity, which is the foundational biological information in any restoration program (Rocchini, 2007).
To meet the requirements for mine site restoration where high temporal and spatial resolution and site-based data integration are necessary (Lawley et al., 2016), the more nuanced and higher resolution capacity from drone-based analyses may provide useful capacity (Miller et al., 2017; de Almeida et al., 2020). Unmanned Aerial Vehicles (UAVs) are an increasingly common platform for remote sensing data acquisition (Johansen et al., 2019). UAVs enable low-cost, rapid acquisition of high-resolution data at a centimeter scale (Lu and He, 2017; Ren et al., 2019; Alvarez-Vanhard et al., 2020; Belmonte et al., 2020), and can coincide with the timing of site-based assessments (Lawley et al., 2016), making them a convenient tool to aid in vegetation surveys in mining areas.
The applications of UAVs in vegetation assessments in mining areas are wide-ranging, including monitoring vegetation growth conditions such as soil temperature (Ruan et al., 2022); mapping vegetation communities, vegetation structure and height (Banerjee and Raval, 2022; Tang et al., 2022); and assessing biomass (Ren et al., 2022). Multispectral sensors can be used to monitor ecological indicators such as vegetation coverage, aboveground biomass, and tree crown coverage (Villoslada et al., 2020; Fernandez-Guisuraga et al., 2022), and can provide an alternative to high-cost airborne hyperspectral and LiDAR approaches. However, few studies have used spectral information provided by UAV remote sensing integrated with site-based data to identify plant species and provide estimates of diversity.
Studies estimating plant diversity using remote sensing can typically be divided into two categories: direct identification of plant species and their distribution through visual interpretation or image classification algorithms, and indirect methods that establish a relationship between diversity and spectral data, or derive species distribution through habitat mapping (Rocchini, 2007; Madonsela et al., 2017; Wang and Gamon, 2019; Villoslada et al., 2020; Zhu et al., 2022). Many studies use vegetation indices such as NDVI estimate species diversity indirectly, although they do not discriminate well between vegetation communities (Gillespie, 2005; Madonsela et al., 2018; Kacic and Kuenzer, 2022; Tian and Fu, 2022). However, less attention has been given to the sensitivity of vegetation indices to species distributional patterns through visual interpretation or image classification algorithms (Lu and He, 2017; Reddy, 2021).
In this study, we combine on-ground vegetation surveys with UAV multispectral data to map vegetation diversity in mining areas in Northwestern China on a sub-meter scale over tens of hectares and develop classification algorithms to identify plant communities The restoration biome is characterized by the co-existence of trees, shrubs and herbaceous plants with high floral diversity compared to the surrounding area, making fine-scale monitoring necessary to analyze the spatial configuration and diversity distribution of vegetation communities. Our objectives are to: (i) compare the performance of three industry-standard supervised classification methods in identifying plant species and mapping their distributions based on reflectance values, (ii) estimate α-diversity at scales determined by the species-area curve to assess the ability of UAV data, and vegetation indices calculated from this data, to qualitatively and quantitatively map alpha floral diversity. This gives restoration practitioners a robust method to significantly improve the spatial detail and coverage of floral diversity mapping for effective monitoring of restoration projects using UAV-based inventory.