Can causal discovery lead to a more robust prediction model for runoff signatures?

Runoff signatures characterize a catchment's response and provide insight into the hydrological processes. These signatures are governed by the co-evolution of catchment properties and climate processes, making them useful for understanding and explaining hydrological responses. However, catchment behaviors can vary significantly across different spatial scales, which complicates the identification of key drivers of hydrologic response. This study represents catchments as networks of variables linked by cause-and-effect relationships. We examine whether the direct causes of runoff signatures, representing independent causal mechanisms, can explain these catchment responses across different environments. To achieve this goal, we train the models using the causal parents of the runoff signatures and investigate whether it results in more robust, parsimonious, and physically interpretable predictions compared to models that do not use causal information. We compare predictive models that incorporate causal information derived from the relationships between the catchment, climate, and runoff characteristics. The Peter and Clark (PC) causal discovery algorithm is applied separately for 11 runoff signatures to derive causal relationships between catchment attributes, climate indices, and corresponding runoff signatures. Three prediction models – the Bayesian network (BN), generalized additive model (GAM), and random forest (RF) – are used for predictions. The results indicate that among models, BN, a linear model with a structure based on the causal network, exhibits the smallest decline in accuracy between training and test simulations compared to the other models. Across nearly all environments and runoff signatures, using causal parents enhances robustness and parsimony while maintaining the accuracy of GAMs. While RF achieves the highest overall performance, it also demonstrates the most significant drop in accuracy between the training and test phases. When the sample size for training is small, the accuracy of the causal RF model, which uses causal parents as predictors, is comparable to that of the non-causal RF model, which uses all selected variables as predictors, particularly for low-flow duration, high-flow duration, low flows, and high flows. This study demonstrates the potential of causal inference techniques for interpreting and enhancing the prediction of catchment responses by effectively representing the interconnected processes in hydrological systems in a cause-and-effect manner.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.