An Insight into Theory-Guided Climate Data Science—A Literature Review

Abstract

Data science models, though successful in a large number of commercial domains, have found limited applications in scientific problems that involve complex physical phenomena. Most of these problems comprise of multi-spectral data composites. Climate science and hydrology is one such scientific domain that faces several big data challenges. Climate data poses many challenges in research because of its spatiotemporal characteristics, high degree of variance, and predominantly its physical nature. One such challenging data in climate science and hydrology is precipitation data. Precipitation data is vast, and generated at a fast pace from several sources, but due to the lack of underlying principles, the models in data science to address climatic issues such as precipitation are dysfunctional. These challenges call for a novel approach that integrates domain knowledge and data science models. To do so, the paper surveys an evolving paradigm of theory-guided data science (TGDS). It is a new paradigm in data science and analytics that aims to improve the generalization of data science models and improve their effectiveness in scientific discovery. The authors, through the survey, present the challenges imposed by climate data, which is representative of the precipitation data, and limitations of traditional data science methods. The paper suggests a shift in data science practices to adapt theory-guided data science for climate and hydrology domain of precipitation data, by providing insights on TGDS, its models and approaches.