Modeling Uncertainty With Engression: A Deep Generative Time‐Series Approach

Abstract Deep learning enables precise environmental predictions across spatial and temporal scales. However, reliable uncertainty estimation with generative capabilities remains crucial for actionable forecasting and simulation, yet robust quantification methods remain challenging. Recently, engression , a generative approach for model‐agnostic uncertainty quantification, has been proposed. We evaluate its feasibility for environmental time‐series modeling, specifically applying it to rainfall‐runoff prediction using long short‐term memory (LSTM) networks. As a benchmark, we use quantile regression, a generalization of the mean absolute error. In our analysis, we focus on the quality of the uncertainty estimates and realism of generated sequences, while including a comparison to point‐prediction performance as a sanity check. Our results show that LSTM‐engression is an effective and easy‐to‐use method for generative modeling, outperforming quantile regression in uncertainty quantification. Qualitative analysis of the generated time series indicates high dynamic fidelity. These findings highlight LSTM‐engression as a promising approach for incorporating uncertainty into environmental modeling.