Automatic Model Structure Identification for Conceptual Hydrologic Models

Abstract

Choosing (an) adequate model structure(s) for a given purpose, catchment, and data situation is a critical task in the modeling chain. However, despite model intercomparison studies, hypothesis testing approaches with modular modeling frameworks, and continuous efforts in model development and improvement, there are still no clear guidelines for identifying a preferred model structure. By introducing a framework for Automatic Model Structure Identification (AMSI), we support the process of identifying (a) suitable model structure(s) for a given task. The proposed AMSI framework employs a combination of the modular hydrological model RAVEN and the heuristic global optimization algorithm dynamically dimensioned search (DDS). It is the first demonstration of a mixed-integer optimization algorithm applied to simultaneously optimize model structure choices (integer decision variables) and parameter values (continuous decision variables) in hydrological modeling. The AMSI framework is thus able to sift through a vast number of model structure and parameter choices for identifying the most adequate model structure(s) for representing the rainfall-runoff behavior of a catchment. We demonstrate the feasibility of the approach by reidentifying given model structures that produced a specific hydrograph and show the limits of the current setup via a real-world application of AMSI on 12 MOPEX catchments. Results show that the AMSI framework is capable of inferring feasible model structure(s) reproducing the rainfall-runoff behavior of a given catchment. However, it is a complex optimization problem to identify model structure and parameters simultaneously. The variance in the identified structures is high due to near equivalent diagnostic measures for multiple model structures, reflecting substantial model equifinality. Future work with AMSI should consider the use of hydrologic signatures, case studies with multiple types of observation data, and the use of mixed-integer multiobjective optimization algorithms.

Toward Data‐Driven Generation and Evaluation of Model Structure for Integrated Representations of Human Behavior in Water Resources Systems

“…The relative performance demonstrated here thus forms a basis for the analysis of model structural uncertainty (Walker et al, 2003) by considering model structures as competing hypotheses (Beven, 2019), which could be compared alongside theory-based models. feature data and primitives (Bongard & Lipson, 2007;, though informing and bounding search through process understanding and structural priors (Knüsel et al, 2019), constrained problem framings (e.g., Dobson et al, 2019;Müller & Levy, 2019), and structured generation schemes (e.g., Chadalawada et al, 2020;Spieler et al, 2020), and using advanced interpretation tools postsearch (e.g., Quinn et al, 2019;Worland et al, 2019) could uncover more specific emergent phenomena in the data and resulting models. However, framing model structural experimentation according to this generic framework enables a baseline contextualization of the complex integrated systems problem.…”

“…Methods have been demonstrated for systems in which the target relationships are well‐known, such as the double pendulum (M. D. Schmidt & Lipson, 2009) and the Navier‐Stokes equations (Rudy et al., 2017). In hydrology, data‐driven system identification methods have been used to infer rainfall‐runoff transfer functions (Klotz et al., 2017) and to automate the identification of rainfall‐runoff model structures using global optimization (Spieler et al., 2020).…”

Flood spatial coherence, triggers, and performance in hydrological simulations: large-sample evaluation of four streamflow-calibrated models

“…Our climate sensitivity analysis shows that the simulation of floods becomes even more challenging under climate con- ditions different from the current ones as the hydrological models employed in this study have limited capability in reproducing observed hydrologic sensitivities during flooding. These limitations may be related to input uncertainties (Te Linde et al, 2007), insufficient model calibration (Fowler et al, 2016), or equifinality in process contributions for simulations with (very) similar efficiency scores, leading to an inability to unambiguously identify the appropriate relative process contributions (Khatami et al, 2019).…”

“…To improve uncertainty estimates, such uncertainty should be accounted for by explicitly considering streamflow measurement uncertainty in model calibration (McMillan et al, 2010). In addition, the uncertainty of the precipitation product used to drive a hydrological model can lead to differences in observed and simulated flows (Te Linde et al, 2007;Renard et al, 2011). Precipitation products may show observation uncertainties (Mcmillan et al, 2012) and underestimate extreme rainfall or the spatial dependence of extreme precipitation at different locations because spatial smoothing or averaging during the gridding process reduces variability (Haylock et al, 2008;Risser et al, 2019).…”

Construction of a method for representing an approximation model of an object as a set of linear differential models

“…Although the cited work considers the structure of a model but, from the point of view of choosing among several, the task of its construction is not addressed. In addition to expanding the scope of application, the issues of automation of the modeling process itself, the automation of the process of finding approximating models of objects and processes, are discussed [4]. However, even in this case, the authors consider, although automated, but the choice of model structure from the list of those available.…”