Eichhornia crassipes management in a eutrophic reservoir: Cover reduction, rapid recolonization, and cyanobacterial dynamics

Abstract Mechanical harvesting is widely used to control Eichhornia crassipes in reservoirs, yet its performance is judged by biomass removed rather than the reduction and stabilization of plant cover. We integrated a 129 month Sentinel-1 time series derived using U-Net segmentation with harvesting records and water quality monitoring to evaluate harvest campaigns, recurrent expansion and phytoplankton responses in a 177 ha eutrophic tropical urban reservoir. Harvesting effort was not proportional to cover change: campaigns with comparable biomass removal produced contrasting outcomes. Cover reductions were followed by recolonization, with the two fastest episodes doubling reservoir scale cover in 19 and 20 days. Expansion was fastest after low cover phases and declined as open water surface became occupied, whereas harvesting imposed a weaker constraint on cover gain. At the reservoir outlet, higher cover was associated with lower cyanobacterial density and dominance during the dry season, while chlorophyll-a remained unrelated to cover. Station specific Random Forest models interpreted with explainable AI (SHAP) retained true color and macrophyte cover as main predictors of cyanobacterial density. These findings distinguish three dimensions often conflated in macrophyte management: operational effort, spatial outcome and ecological state. Harvested biomass should therefore be treated as an effort metric, whereas net cover change, residual cover and rebound provide direct measures of management performance. Combining deep learning satellite mapping, operational records, explainable machine learning and in situ monitoring can support adaptive harvesting, but durable control will also require reductions in nutrient and suspended solid loads.