Climate data has been vastly accumulated over the past several years, making climate science one of the most data-rich domains. Despite the abundance of data to process, data science has not had a lot of impact on climate research so far, partly due to the fact that ample expert knowledge is rarely exploited. The main goal of this project is bridging the gap between deep learning and causal graphical models while using domain knowledge which could prove to be of significant importance for facilitating an understanding of the Earth system. We aim to develop a sequential version of the Causal Effect Variational Auto-Encoder (CEVAE) and apply it to time series of ecological or climate variables having suitable underlying causal graph structure. When this is accomplished, we intend to apply our method to time series anomaly detection, as well as to variables having more general causal structures.