An analysis of the spatial variations and dynamics of vegetation recovery using low-altitude remote sensing in actively restored forests of the Amathole region

Dataset Title:
UAV-based RGB and point cloud dataset for assessing vegetation recovery in actively restored Mistbelt forests of the Amathole region, South Africa

Author(s):
Andisiwe Manase ^a, Alen Manyevere ^a,*, Mohamed A.M Abd Elbasit ^b, Jessica Leaver ^c

Corresponding author affiliation:
Department of Agronomy, University of Fort Hare, Private Bag X1314, Alice, 5700, South Africa

Journal submission:
GIScience & Remote Sensing

Description:

This dataset supports the research article entitled "An analysis of the spatial variations and dynamics of vegetation recovery using low-altitude remote sensing in actively restored forests of the Amathole region," submitted to GIScience & Remote Sensing. It comprises high-resolution unmanned aerial vehicle (UAV) imagery and derived point cloud metrics, alongside ground-based leaf spectral data, collected to evaluate the effectiveness of active forest restoration in three Mistbelt forests (Izingcuka, Swallowtail, and Madonna and Child) in the Amathole region, South Africa.

Data were acquired during the spring season of 2024 using a DJI Mavic 2 Enterprise Dual drone. The dataset includes:

• RGB orthomosaics and 3D point clouds processed from UAV flights over paired reference (intact) and restoration (actively revegetated) sites, following a 2 × 3 factorial design (two site types × three blocks).
• Derived vegetation indices computed from RGB imagery: VEG (Vegetative Index), TGI (Triangular Greenness Index), GRVI (Green-Red Vegetation Index), NGRDI (Normalized Green-Red Difference Index), ExG (Excess Green), ExR (Excess Red), and GLI (Green Leaf Index).
• Point cloud structural metrics including above-ground biomass (AGB), canopy height model (CHM), canopy cover (%), and gap fraction.
• Ground reference data consisting of spectral and biophysical measurements from 270 Podocarpus latifolius leaf samples.
• Binary vegetation classification outputs (Random Trees and Support Vector Machine) produced in ArcGIS Pro 3.5.0, with associated accuracy metrics (user’s accuracy: 95–99%; Kappa: 0.76–0.96).
• RStudio analysis scripts (R version 4.1.2, 2024) for two-way ANOVA, Principal Component Analysis (PCA), and stepwise regression used to assess spatial variation and vegetation dynamics.

Key variables:

• Site type (reference vs. restoration)
• Vegetation indices (spectral health proxies)
• Point cloud metrics (structural complexity)
• Leaf-level ground truth data

Spatial and temporal coverage:

• Amathole Mistbelt forests, Eastern Cape, South Africa
• Single-season spring survey (2024); dynamics interpreted as within- and between-site variability in vegetation indices.

Data usage notes:
These data are suitable for comparative analyses of active forest restoration outcomes, methodological assessments of UAV-based monitoring techniques, and studies investigating the relationship between spectral vegetation indices and structural forest attributes in temperate or Mistbelt ecosystems. The inclusion of both RGB-derived indices and point cloud metrics enables users to evaluate multifactorial drivers of vegetation recovery.

Methodological context:
Full experimental design, preprocessing steps, and analytical procedures are detailed in the associated manuscript. All coordinate reference systems and file formats (CSV, R script) are documented in the accompanying README file.

Licence:
[CC BY 4.0]

Keywords:
Unmanned aerial vehicle, Photogrammetry, Forest recovery, Spectral signatures, Machine learning