Harmful Algal Blooms (HAB) pose significant challenges to fisheries management and food and water security. The onset of a HAB is notoriously difficult to predict. Traditional methods of algal species identification under a microscope are also laborious and time consuming. A real-time system for identification and concentration measurement of algal bloom species has been developed at a marine fish culture zone (FCZ) in the subtropical coastal waters of Hong Kong. The system is based on analysis of high frequency algal cell images obtained from an underwater Imaging FlowCytobot (IFCB) deployed on the fish farm. An explainable supervised machine learning technique has been successfuly developed. The algal species classifier is trained by presenting a wide range of extracted image features to a random forest algorithm. An optimized set of 25 features is identified by a recursive feature elimination technique. The random forest (RF) classifier can identify 15 target HAB classes with an overall out-of-bag accuracy of 94.2%, with individual $F1$ score ranging from 0.8 to 1.0. The classifier performs equally well as a Convolution Neural Network (CNN) developed using transfer learning techniques. Based on the classifier, an automated real-time species identification and cell counting protocol has been developed, with a response time of 10 min after data collection. This work represents the first successful attempt of continuous algal species monitoring by IFCB and artificial intelligence (AI) – based detection of HAB in subtropical coastal waters.