Many practical time series forecasting (TSF) tasks are plagued by data limitations. To alleviate this challenge, we design a data-level augmentation framework.