Modeling water hyacinth (Eichhornia crassipes) distribution in Lake Tana, Ethiopia, using machine learning

Aquatic invasive plant, water hyacinth poses serious environmental and socioeconomic challenges. Understanding and predicting the spatiotemporal distribution of this species is important for reducing its environmental impact. Therefore, the present study aimed to model the distribution of water hyacinths in an important ecological region (Lake Tana) of Ethiopia using four machine learning models. We used 11 variables obtained from Sentinel-1 SAR bands, Sentinel-2A bands and indices, and bioclimate data sources. The models use 458 presence and 458 randomly generated pseudoabsence data as response variables and employ a tenfold bootstrap sampling method. The area under the curve (AUC), receiver operator curve (ROC), true skill statistics (TSS), coefficient of rank correlation (COR), sensitivity, specificity, and kappa coefficient were used to evaluate the models. The findings demonstrate that the random forest model outperforms the other models, with AUC values of 0.93 and 0.95, TSS values of 0.77 and 0.82, and kappa values of 0.76 and 0.82 in the wet and dry seasons, respectively. B12 (16% and 20%), NDWI (15% and 12%), mean annual temperature (13% and 14%), and B5 (11% and 12%) were found to be the most relevant variables during the wet and dry seasons, respectively. Water hyacinths have greater spatial coverage during the wet season than during the dry season because of high rainfall, high water levels and nutrient runoff. We can conclude that to detect and predict the spatiotemporal conditions of water hyacinth accurately, integrating Sentinel image indices and bands with bioclimatic variables and using machine learning models are crucial.