Spatiotemporal variations of aquatic vegetation in Maracaibo Lake: Remote sensing and machine learning approach with Google Earth Engine

Abstract

Aquatic plant invasions endanger lake biodiversity and ecosystem services, resulting in significant economic losses for local communities. Therefore, it is crucial to accurately delineate the extent and frequency of development, but traditional methods are costly and in remote areas. Cost-effective methods, such as satellite monitoring are required. This study uses a Random Forest classification model in the Google Earth Engine (GEE) with Landsat 7 and 8 images to monitoring aquatic vegetation invasion in freshwater ecosystems. The methodology automates the selection of training samples through a dynamic adjustment that incorporates the Otsu-Canny Edge algorithms applied to a vegetation index, allowing for monthly updates while minimizing human bias. Applying this methodology to Lake Maracaibo, Venezuela, between 2013 and 2021, there was a significant increase in floating aquatic vegetation cover, from ≤10 % in 2013 to 25.63 % in 2021, particularly along the northwest coast and the Strait of Maracaibo. This increase could be attributed to a combination of natural processes like precipitation patterns and increased anthropogenic inputs from human activities. The model achieved high accuracy (>0.80), as evidenced by the confusion matrix and cross-sensor comparison. This approach provides a tool for continuous long-term monitoring that can be applied to other eutrophic lakes, improving our understanding of the effects of invasive vegetation, and assisting resource managers and policymakers in developing sustainable management strategies.