Advances in Time Series Forecasting

The field of time series forecasting is moving towards more realistic and comprehensive benchmarks, with a focus on addressing the challenges of real-world settings such as tasks with covariates and long-tailed distributions. Researchers are also exploring new models and techniques to improve forecasting accuracy, particularly for complex and practically relevant tasks like rainfall nowcasting and heavy rainfall prediction. Noteworthy papers include Fidel-TS, which introduces a new large-scale benchmark for multimodal time series forecasting, and DPSformer, which proposes a long-tail-aware model for improving heavy rainfall prediction. Additionally, the development of new metrics and post-processing techniques, such as the expected thresholded calibration error, is enhancing the reliability of probabilistic forecasts.

Sources

Fidel-TS: A High-Fidelity Benchmark for Multimodal Time Series Forecasting

DPSformer: A long-tail-aware model for improving heavy rainfall prediction

How Effective Are Time-Series Models for Rainfall Nowcasting? A Comprehensive Benchmark for Rainfall Nowcasting Incorporating PWV Data

fev-bench: A Realistic Benchmark for Time Series Forecasting

Probability calibration for precipitation nowcasting

Built with on top of