Prediction of Water Hyacinth Coverage on the Hartbeespoort Dam

Abstract

Water hyacinth is an invasive weed that contributes to theHartbeespoort Dam’s poor water quality. Although biological control isthe most effective and sustainable method of controlling waterhyacinth, a prediction model to plan the biological controls is essentialfor successful intervention. The literature shows that mathematicalmodels and remote sensing have been used successfully in the past toestimate plant growth rates in similar applications. This study presentsvarious machine-learning models that were investigated to predict waterhyacinth coverage.The complex relationships of water hyacinth growth were simplified tofocus on the most influential factors: temperature and nutrients. Missingdata were imputed using the multiple k-nearest neighbours imputation.The nutrient datasets were extrapolated to the correct timeline usingMonte Carlo simulations and seasonal patterns. Ensemble learning,decision trees, artificial neural networks, and support vector machinemodels were developed, with ensemble learning (bag algorithm)resulting in the best predictions.