Monitoring water hyacinth in Lake Victoria using Sentinel-2, Sentinel-1 and cloud computing: a case study of Winam Gulf, Kenya

Abstract

Water hyacinth is considered one of the world’s worst invasive aquatic weeds. Originally from the Amazon, it was introduced in Lake Victoria in 1989, causing periodic waves of infestation bearing socioeconomic and environmental issues.

Studies monitored water hyacinth invasions in Lake Victoria to inform management practices until 2017. This study proposes a Google Earth Engine framework to monitor the weed in Winam Gulf (Kenya) over the period 2018-2021. Sentinel-2 data was used to test two different water hyacinth classification methods: Red Edge Normalised Difference Vegetation Index (reNDVI) thresholding and Random Forest classification. Both yielded mean Overall Accuracies of ~92% for the 22 images classified. Additionally, Sentinel-1 data was used to classify the weed by thresholding 56 bi-weekly VV mean composites and to create yearly invasion heatmaps. The Sentinel-2 and Sentinel-1-derived water hyacinth timeseries indicated a Nov 2018 water hyacinth peak (~155 km2), which could have been the worst invasion in the Gulf since 2013. In accordance with previous studies, water hyacinth invasions exhibited Autumnal peaks, albeit with a decreasing trend (~95% decrease Dec 2018-Dec 2021). Weather data from Kisumu (Kenya) suggested that increased rainfall and cloudiness in 2019-2021 might be the culprit. The Sentinel-1 yearly composites also revealed that Osodo and Kusa Bay were the areas most affected by water hyacinth invasions in 2018-2021. Limitations of this study include Sentinel-2 L2A data unavailability in GEE and Sentinel-1 biweekly averaging. Finally, the proposed cloud-based water hyacinth monitoring solution could ensure continued observation of the Gulf and/or be applied to other geographical locations.