Efficient riparian planting monitoring using remote sensing trained convolutional neural networks

This study examines the effectiveness of combining multiband remote sensing (RS) and machine learning (ML) methods for monitoring riparian planting success. A 0.5m, 8-band Maxar satellite imagery-based RS model classifies land cover across five pilot sites. Using RS model outputs, a modified DeepLabV3+ ML model is trained with 4-band imagery. This novel approach allows the ML model to replicate the 8-band RS model's capabilities at a lower cost, achieving a 72-80% classification accuracy and F1 scores of 0.8–0.9 for different land cover classes, exceeding those of 3-band ML methods alone. This method demonstrates potential for national-level riparian survival monitoring using cost-effective 4-band imagery and an efficient ML algorithm.