Deep learning methods have frequently outperformed conceptual hydrologic models in rainfall-runoff modelling. Attempts of investigating such deep learning models internally are being made, but the traceability of model states and processes and their interrelations to model input and output is not yet fully understood. Direct interpretability of mechanistic processes has always been considered an asset of conceptual models that helps to gain system understanding aside of predictability. We introduce hydrologic neural ordinary differential equation (ODE) models that perform as well as state-of-the-art deep learning methods in stream flow prediction while maintaining the ease of interpretability of conceptual hydrologic models. In neural ODEs, internal processes that are represented in differential equations, are substituted by neural networks. Therefore, neural ODE models enable the fusion of deep learning with mechanistic modelling. We demonstrate the basin-specific predictive performance for 569 catchments of the continental United States. For exemplary basins, we analyse the dynamics of states and processes learned by the model-internal neural networks. Finally, we discuss the potential of neural ODE models in hydrology.
This study investigates Bayesian signature‐domain inference of hydrological models using Approximate Bayesian Computation (ABC) algorithms, and compares it to “traditional” time‐domain inference. Our focus is on the quantification of predictive uncertainty in the streamflow time series and on understanding the information content of particular combinations of signatures. A combination of synthetic and real data experiments using conceptual rainfall‐runoff models is employed. Synthetic experiments demonstrate: (i) the general consistency of signature and time‐domain inferences, (ii) the ability to estimate streamflow error model parameters (reliably quantify streamflow uncertainty) even when calibrating in the signature domain, and (iii) the potential robustness of signature‐domain inference when the (probabilistic) hydrological model is misspecified (e.g., by unaccounted timing errors). The experiments also suggest limitations of the signature‐domain approach in terms of information loss when general (nonsufficient) statistics are used, and increased computational costs incurred by the ABC implementation. Real data experiments confirm the viability of Bayesian signature‐domain inference and its general consistency with time‐domain inference in terms of predictive uncertainty quantification. In addition, we demonstrate the utility of the flashiness index for the estimation of streamflow error parameters, and show that signatures based on the Flow Duration Curve alone are insufficient to calibrate parameters controlling streamflow dynamics. Overall, the study further establishes signature‐domain inference (implemented using ABC) as a promising method for comparing the information content of hydrological signatures, for prediction under data‐scarce conditions, and, under certain circumstances, for mitigating the impact of deficiencies in the formulation of the predictive model.
28 October 2013 in Hydrology and Earth System Sciences
Hydrodynamic models are useful tools for urban water management. Unfortunately, it is still challenging to obtain accurate results and plausible uncertainty estimates when using these models. In particular, with the currently applied statistical techniques, flow predictions are usually overconfident and biased. In this study, we present a flexible and relatively efficient methodology (i) to obtain more reliable hydrological simulations in terms of coverage of validation data by the uncertainty bands and (ii) to separate prediction uncertainty into its components. Our approach acknowledges that urban drainage predictions are biased. This is mostly due to input errors and structural deficits of the model. We address this issue by describing model bias in a Bayesian framework. The bias becomes an autoregressive term additional to white measurement noise, the only error type accounted for in traditional uncertainty analysis. To allow for bigger discrepancies during wet weather, we make the variance of bias dependent on the input (rainfall) or/and output (runoff) of the system. Specifically, we present a structured approach to select, among five variants, the optimal bias description for a given urban or natural case study. We tested the methodology in a small monitored stormwater system described with a parsimonious model. Our results clearly show that flow simulations are much more reliable when bias is accounted for than when it is neglected. Furthermore, our probabilistic predictions can discriminate between three uncertainty contributions: parametric uncertainty, bias, and measurement errors. In our case study, the best performing bias description is the output-dependent bias using a log-sinh transformation of data and model results. The limitations of the framework presented are some ambiguity due to the subjective choice of priors for bias parameters and its inability to address the causes of model discrepancies. Further research should focus on quantifying and reducing the causes of bias by improving the model structure and propagating input uncertainty.
This study considers Bayesian calibration of hydrological models using streamflow signatures and its implementation using Approximate Bayesian Computation (ABC). If the modeling objective is to predict streamflow time series and associated uncertainty, a probabilistic model of streamflow must be specified but the inference equations must be developed in the signature domain. However, even starting from simple probabilistic models of streamflow time series, working in the signature domain makes the likelihood function difficult or impractical to evaluate (in particular, as it is unavailable in closed form). This challenge can be tackled using ABC, a general class of numerical algorithms for sampling from conditional distributions, such as (but not limited to) Bayesian posteriors given any calibration data. Using ABC does not avoid the requirement of Bayesian inference to specify a probability model of the data, but rather exchanges the requirement to evaluate the pdf of this model (needed to evaluate the likelihood function) by the requirement to sample model output realizations. For this reason ABC is attractive for inference in the signature domain. We clarify poorly understood aspects of ABC in the hydrological literature, including similarities and differences between ABC and GLUE, and comment on previous applications of ABC in hydrology. An error analysis of ABC approximation errors and their dependence on the tolerance is presented. An empirical case study is used to illustrate the impact of omitting the specification of a probabilistic model (and instead using a deterministic model within the ABC algorithm), and the impact of a coarse ABC tolerance.
The Batalin–Vilkovisky (BV) method is the most powerful method to analyze functional integrals with (infinite-dimensional) gauge symmetries presently known. It has been invented to fix gauges associated with symmetries that do not close off-shell. Homological perturbation theory is introduced and used to develop the integration theory behind BV and to describe the BV quantization of a Lagrangian system with symmetries. Localization (illustrated in terms of Duistermaat–Heckman localization) as well as anomalous symmetries are discussed in the framework of BV.