Multi-source and heterogeneous marine hydrometeorology spatio-temporal data analysis with machine learning: a survey

Song Wu, Xiaoyong Li, Wei Dong, Senzhang Wang, Xiaojiang Zhang, Zichen Xu

Abstract: There is a new trend in marine hydrometeorology (MHM) that calls for novel solutions on massive multi-source and heterogeneous spatiotemporal data sets. Traditionally, the research and analysis of MHM objects are generally based on statistical analysis of observation data, numerical simulation, laboratory observation and experiments. However, these methods fail to adopt onto the massive multi-source and multi-modality data analysis problem. To better understand and analyze the new requirements on MHM data, researchers started a data-oriented approach that mines features and patterns from the massive datasets, or so called the machine learning approach. In this paper, we provide a systematic review on the applicability of machine learning approaches in understanding and mining MHM objects. We start with these techniques from the perspective of different data sources and focusing on learning objects like oceanic internal wave, tide, sea ice, typhoon, and red tide from recognition to prediction. First, this paper systematically summarizes the current research methodologies, unique data characteristics, and the challenges of machine learning in MHM. Next, we classify the mainstream data and models, and overview the machine learning models from the perspective of different MHM scenarios and multiple data sources. Then, we summarize the key techniques, with pros and cons of machine learning applications in processing such scenarios. Last, we conclude with the future research trend of machine learning in MHM, especially in model interpretability.