Mapping and Classification of Invasive Tree Species Chinese Celtis in Riparian Ecosystems

Mapping and Classification of Invasive Tree Species Chinese Celtis in Riparian Ecosystems

Chinese Celtis are tree species that have, over time, become a threat to the flora and faunal habitat of native Australian species in New South Wales and Queensland. Early detection has become essential for monitoring its spread and implementing targeted management strategies for the conservation of native ecosystems. Observing the distribution of this weed is made easier by data collected and processed using remote sensing methods. Furthermore, high-resolution images are favoured since they are more likely to differentiate Chinese Celtis from other native grasses and trees. In this research project, we investigate the use of UAV-acquired very high-resolution multispectral imagery to map the spatial distribution of invasive trees in a relatively small region along the Brisbane River. The spectral details of the vegetation were emphasised by calculating vegetation indices like NDVI and NDWI, while spatial patterns were quantified by extracting texture-based features. Several machine-learning models were then trained for classification using these features as input. AutoML tool in ArcGIS Pro was employed to automate model selection and tuning, helping streamline the workflow. Metrices such as confusion matrix, precision, recall and f1-score were used for model evaluation. Among the tested models, LightGBM (Light Gradient Boosting Machine) achieved the best classification performance of 96.64%. While UAVs are limited in large-scale coverage, their use in localised or high-priority areas provides an efficient and cost-effective option for detailed ecological monitoring. Future research could combine UAV imagery with images from other sensors, like LiDAR or hyperspectral, to improve classification accuracy in case of larger regions and address similar issues occurring in the environment.

The accessible file is the accepted version of the paper. Please refer to the URL for the published version.