Integrating drone and deep learning technology to monitor floodplain changes in response to environmental flows

Woody-perennial vegetation forms a critical component of floodplains, which are commonly managed with environmental flows to mitigate the impacts of flow regulation. Drone-based imagery analysed with deep machine learning provides an accurate and repeatable approach to monitor vegetation responses in environments with limited access. This study provides a method integrating drone imagery and convolutional neural networks (CNNs) to classify and estimate floodplain vegetation and evaluate its response to environmental flows—with a focus on the floodplain shrub Duma florulenta (tangled lignum). Drone RGB imagery, collected from 18 sites in 2023 and 2024 in the Mallee Region, Victoria, and four sites in 2024 in Gaynor Swamp, Victoria were used as the input for the framework. Four sites in the Macquarie Marshes, New South Wales, were used to demonstrate the method’s generalization. The CNN model was constructed from kernel design to classification results, followed by a sensitivity analysis and compared in operational performance with four CNN frameworks. Our CNN model outperformed the other CNN frameworks in Overall Accuracy (91.12%) and running efficiency. The drone-CNN spatial framework offers an efficient, accurate and repeatable method to estimate Duma florulenta extent and condition (dormant or vigorous) and assess its response to flooding and drying.