Evaluating hyperspectral and machine learning approaches to classify biocontrol-induced damage on water hyacinth ( <i>Pontederia crassipes</i> )

Abstract Water hyacinth ( Pontederia crassipes Mart.) is a free-floating aquatic plant native to South America that has spread to nearly 50 countries, becoming one of the world’s most invasive aquatic weeds. In Florida, the biocontrol agents Neochetina eichhorniae and Neochetina bruchi were released in 1970s, while Megamelus scutellaris was released in 2010. Assessing the impact of these biocontrol agents is crucial in evaluating efficacy, distribution, and overall progress in management efforts. The traditional survey and monitoring methods used to evaluate the impact of biocontrol present numerous challenges in data acquisition, especially in remote areas and aquatic habitats. This study aimed to detect damage caused by Neochetina spp. and M. scutellaris on P. crassipes using hyperspectral remote sensing. Plants were exposed to varying levels of Neochetina spp. and M. scutellaris herbivory for 2 and 4 wk under laboratory conditions. After the exposure period, the plants were scanned using a visible and near-infrared hyperspectral imaging system. Two classification algorithms, partial least-squares discriminant analysis (PLS-DA) and support vector machine (SVM) were employed for classification. SVM achieved high classification accuracy at both low and high damage levels, with overall training and validation accuracies of 84.9% and 78.79%, respectively, while PLS-DA only achieved high classification accuracy at high damage levels, with overall training and validation accuracies of 56.3% and 60.38%. Based on the observed performance metrics, both algorithms demonstrated improved classification accuracy as damage increased over time. The results indicated that hyperspectral remote sensing can be used to monitor and assess biocontrol agents damage on P. crassipes .