Detection and monitoring of water hyacinth in large lakes of five Indian urban agglomerates using Sentinel-2 and machine learning models.

Rapid urbanisation has intensified ecological pressures on urban lakes, promoting the spread of invasive aquatic macrophytes such as water hyacinth (Eichhornia crassipes). This study presents a multi-year, satellite-based framework for detecting and monitoring water hyacinth in urban lakes using Sentinel-2 multispectral imagery and supervised machine learning. Four classifiers (Support Vector Machine, XGBoost, Random Forest, and K-Nearest Neighbours) were evaluated using classification-based accuracy metrics, including accuracy, precision, recall, F1-score, Cohen's kappa, and cross-validation F1-score. Although all models achieved high quantitative accuracy, the Support Vector Machine (SVM) classifier demonstrated the most consistent performance, combining high classification accuracy with spatially coherent and cartographically consistent mapping outputs. Independent validation using two lakes excluded from model training and hyperparameter optimisation demonstrated promising transferability of the SVM classifier, with differences between model-derived and manually digitised water hyacinth extents remaining below 2%, primarily due to pixel-level boundary effects. The validated SVM model was subsequently applied to all eligible urban lakes across five major Indian agglomerates (Mumbai, Kolkata, Bengaluru, Chennai, and Hyderabad) for the post-monsoon period from 2021 to 2025. The results reveal pronounced spatial heterogeneity in infestation persistence, with Kolkata and Bengaluru exhibiting the highest number of persistently infested lakes, Mumbai characterised by a single consistently infested lake, and Chennai and Hyderabad showing intermediate levels of persistence. Importantly, infestation was not confined to urban cores but frequently extended into peri-urban and suburban zones. This spatial pattern is consistent with mechanisms reported in previous studies, including nutrient enrichment, wastewater inflows, altered catchment characteristics, and fragmented governance; however, these potential drivers were not evaluated directly in the present study. Overall, this study demonstrates the value of combining classification-based accuracy assessment with spatially explicit, multi-year satellite mapping to support evidence-based prioritisation of urban lake restoration and invasive macrophyte management in rapidly developing cities.