Spatio-temporal data refers to observations indexed by both spatial locations and time points, emphasizing the importance of accounting for dependencies across these dimensions. Such data finds applications in various fields, including environmental monitoring, transportation, agriculture, public health, and urban planning. However, spatio-temporal data often suffers from incomplete observations due to various factors such as sensor failures and environmental disruptions, making accurate imputation particularly challenging. Traditional imputation methods frequently approach the problem from either a spatial or a temporal perspective. However, recent advancements have introduced numerous methods that leverage both spatial and temporal dependencies to reconstruct missing data, leading to significantly improved results. This paper provides a comprehensive review of spatio-temporal missing data imputation techniques, encompassing an extensive array of methods, from traditional statistical approaches to modern machine learning-based techniques. While other domain-specific approaches (e.g., physics-based or simulation-driven methods) exist, the majority of research in this area falls into these two categories, which are therefore the primary focus of this review. Our contributions also include a discussion of key challenges and research gaps, as well as the identification of promising directions for future research in spatio-temporal data imputation.
Zahmatkesh et al. (Sat,) studied this question.