Detection of Eichhornia crassipes and Pistia stratiotes in the São Francisco River Basin via Remote Sensing

Abstract

Infestations of Eichhornia crassipes (Mart.) Solms and Pistia stratiotes L. compromise the ecological equilibrium and the safety of water management within the São Francisco River Basin. This study evaluated the performance of Random Forest (RF), Support Vector Machine (SVM), and Classification and Regression Trees (CART) algorithms for mapping these invasive macrophytes in the sub-middle region of the basin (Jatobá, Pernambuco, Brazil). Five land cover classes were identified using Landsat 8 imagery and Normalized Difference Vegetation Index (NDVI) datasets processed in QGIS 3.16. RF and CART displayed identical overall accuracy rates of 94%, and all three algorithms achieved 100% accuracy in isolating Pistia stratiotes. Conversely, persistent spectral confusion between Eichhornia crassipes and soil led the RF model to misclassify 33% of soil points due to spatial resolution limitations (30 m). SVM achieved an overall accuracy of 88%, yielding a high recall rate (86%) for water hyacinth but overestimating its area. The results validate orbital remote sensing coupled with machine learning as an effective architecture for continuous bioinvasion monitoring in target zones of the Brazilian semiarid region.