The shale gas temporal production prediction (SGTPP) aims to estimate the capacity of gas resources. However, existing SGTPP methods suffer from data sparsity and neglect the correlations across different wells, resulting in performance degradation. To overcome these limitations, in this paper, we propose a novel Granger Causality-based Temporal Production Prediction model, named as GCTPP. In GCTPP, we firstly propose a Granger causality detection module, which utilizes the historical production data of shale gas wells to identify the Granger causal well pairs. Secondly, we propose a temporal attention mechanism to selectively extract the most important temporal features from the Granger causal well pairs across different timestamps. Finally, we leverage the extracted temporal features of the Granger reason well to help predict the future production of the Granger result well. The results of extensive experiments on real-world datasets demonstrate that our GCTPP model significantly outperforms the state-of-the-art methods in terms of prediction accuracy.