Development of a multitemporal image differencing approach for identifying invasive water hyacinth (Eichhornia...

Invasive aquatic plants such as water hyacinth (Eichhornia crassipes [Mart.] Solms) pose significant threats to freshwater ecosystems through obstruction of waterways, alteration of hydrodynamics, and impacts on biodiversity. Effective monitoring is critical for ecosystem protection and sustainable water management. However, learning-based approaches, such as machine learning and deep learning, require extensive annotated training data and substantial computational resources, limiting their suitability for large-scale operational monitoring. To address this, we developed a multitemporal image differencing (MID) approach that leverages the highly dynamic nature of water hyacinth in flowing waters. By analyzing a sequence of multitemporal satellite images, MID directly captures spatial and temporal changes, enabling timely and reliable detection of both established and newly forming patches. MID achieved an overall accuracy of 90.7%, comparable to the random forest classifier at 91.0%, while providing better discrimination of dynamic water hyacinth from stable nontarget vegetation, reflecting its ability to exploit temporal dynamics that single-date classifiers cannot capture. With high computational efficiency, simplicity, and scalability, MID is highly suitable for large-scale operational monitoring of invasive aquatic plants. These results demonstrate that MID is a robust and scalable framework for supporting ecological management and timely decision-making in dynamic freshwater systems.