Tropical cyclones have a significant impact on estuaries, resulting in deteriorating water quality, changes in phytoplankton productivity, as well as scouring and physical harm to mangroves and other vegetation. Furthermore, they have the potential to impede, pause, or even reverse ecosystem restoration and management efforts. As such, it is crucial to devise methods for detecting and measuring the severity of estuarine disturbances caused by tropical cyclones. Several statistical methods have been proposed in the past to detect and quantify disturbances, with varying levels of complexity and accuracy. Recent advancements in data collection and the quality of high-frequency data have opened up the opportunity to employ machine learning models for assessing the impact of tropical cyclones on ecosystems. This study develops a new machine learning-based model for detecting disturbances within estuaries, assessing their severity, and determining the recovery time, primarily focusing on disruptions to estuarine water quality. A Long Short-Term Memory (LSTM)-based deep learning model is developed to detect disturbances and quantify their severity while a Gaussian filter-based algorithm is developed to assess recovery time. The research utilizes data from NOAA's National Estuarine Research Reserve System for training and validating the model. The model demonstrates an ability to distinguish between disturbances in water quality caused by tropical cyclones and those resulting from natural variability. It also assesses the extent of the disruption and the time for the estuary to revert to its pre-event state. Detecting and quantifying disturbances, as well as estimating recovery time in estuaries due to tropical cyclones, represent the initial steps in the development of predictive disturbance models. Detecting and quantifying disturbance can also aid stakeholders in comprehending the severity of tropical cyclones' impacts on aquatic systems, allowing for the development of suitable interventions. The developed model can be applied to detect and quantify anomalies in any time series data, making it useful in various other fields.