Estimating the cover of Phragmites australis using unmanned aerial vehicles and neural networks in a semi-arid wetland

Unmanned aerial vehicles (UAVs) provide high-spatial-resolution imagery and allow the collection of data in locations or periods of time where field-based data collection is challenging or impossible, such as in wetlands and floodplains. Computational deep learning techniques are transforming the way in which remotely sensed imagery and data can be used and are having an increasing role in remote sensing. Here, we describe a method using UAV and machine learning technique convolutional neural networks (CNNs) to estimate the cover of wetland features Phragmites australis reeds, leaf litter, water, bareground, and other vegetation in a large inland floodplain wetland in Western New South Wales (NSW), Australia. We firstly describe the process we took to train, validate, and test the model. We describe the model's performance by calculating a range of performance indicators and provide density maps and results from individual sites. The model had an overall accuracy of 0.947 and recognized and estimated Phragmites australis reeds to a very high accuracy (>98%). Here, we show an effective, accurate, and reproducible way to estimate the cover of Phragmites australis reeds and other wetland features using UAV and CNNs in a semi-arid wetland.