Missing data imputation of climate time series: A review

Area of Science:

• Climatology
• Data Science
• Environmental Science

Background:

• Missing data in climate time series poses significant challenges for monitoring and prediction.
• Accurate climate data is crucial for understanding and mitigating climate change impacts.

Purpose of the Study:

• To systematically review and describe the most relevant imputation methods for missing climate data over the past decade.
• To identify geographical concentrations and key climate variables addressed in climate data imputation research.

Main Methods:

• Literature review focusing on imputation techniques for climate time series data.
• Analysis of research distribution by geographical region and climate variables.
• Evaluation of conventional statistical methods (e.g., mean, regression, PCA) and advanced machine learning approaches (e.g., ANNs, GANs).

Main Results:

• Research on climate data imputation is concentrated in Asia and Europe, with significant contributions from countries like Malaysia, China, Italy, Brazil, and Australia.
• Temperature and precipitation are the most frequently studied climate variables.
• Monitoring networks are the primary data source, with conventional statistical methods and artificial neural networks being widely used, while Generative Adversarial Networks show superior performance.

Conclusions:

• Generative Adversarial Networks (GANs) demonstrate superior capability in imputing missing climate data compared to other deep learning methods.
• Continued research is needed to refine imputation techniques, especially for complex climate patterns and diverse geographical regions.