Hydrologically Informed Machine Learning for Rainfall‐Runoff Modeling: A Genetic Programming‐Based Toolkit for Automatic Model Induction

Abstract

Models of water resources systems are conceived to capture the underlying environmental dynamics occurring within watersheds. All such models can be regarded as working hypotheses, differing in the aspects of process representation and conceptualization. Most of the associated efforts in the water resources research community is dedicated to development of new models that perform well under specific atmospheric conditions and catchment properties. In this context, flexible modeling frameworks are gaining importance as they facilitate the model building process by providing the model building blocks, whereby the hydrologist is free to assemble the model for task at hand. Such flexible models have high degree of transferability, which in turn aid in progressing toward a unified hydrological theory at catchment scale. However, in cases without sufficient insights regarding a catchment characteristics and/or lack of expert's knowledge, one may have to try a large number of model configurations based on available model building blocks to construct an appropriate model for the catchment of interest. Undoubtedly, this may be time consuming and computationally intensive. This paper proposes a novel model building algorithm, which uses the full potential of flexible modeling frameworks by searching the model space and inferring suitable model configurations relying on machine learning. Proposed machine learning algorithm is based on evolutionary computation approach using genetic programming (GP). State-of-art GP applications in rainfall-runoff modeling so far used the algorithm as a short-term forecasting tool that generates an expected future time series very similar to neural networks application. In this case, the proposed algorithm develops a physically meaningful rainfall-runoff model. Although at the moment we learn models using two flexible modeling frameworks (SUPERFLEX and FUSE), the model induction toolkit can be armed with any internal coherence building blocks. The model induction capabilities of the proposed framework have been evaluated on the Blackwater River basin, Alabama, United States. The model configurations evolved through the model induction toolkit are consistent with the fieldwork investigations and previously reported research findings. Fixed Models Versus Flexible ModelsDesign of conceptual models traditionally begins with a perceptual model derived from insights gained on the basis of fieldwork and experience, proceeding through a mathematical formulation of the hypothesized structure to the numerically robust implementation in a computer code . Conceptual hydrological modeling can be broadly classified into single-and multiple-hypothesis (often referred to as flexible) modeling approaches.Development of models with fixed structure is based on the identification of a general model structure that is physically realistic and applicable to a reasonably wide range of catchments and climatic conditions. Several

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

“…Additionally, studies have argued for an upper limit on the description length of a model (Vanneschi et al., 2010) as done in Chadalawada et al. (2020), though this limit is difficult to identify a priori . Hybrid methods, such as evolutionary strategies to approximate a gradient, are promising for tractable search in combined model‐parameter spaces (Conti et al., 2018; Miikkulainen et al., 2019), as well as approaches that asynchronously tune parameters and structure (Frankle & Carbin, 2019).…”

“…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.…”

“…Symbolic regression uses linear and nonlinear operators as base functions, and can, for example, learn to compose nested functions and automate the process of feature engineering. Symbolic trees can also incorporate noise (M. D. Schmidt & Lipson, 2007), can be seeded with relations of interest during optimization (e.g., M. D. Schmidt & Lipson, 2009; Chadalawada et al., 2020), and can be strongly typed to incorporate and handle heterogeneous data types or function outputs (Montana, 1995). Model evaluations of symbolic regression trees are generally faster than traditional feed‐forward neural networks because each model evolves a sparse input representation based only on the inputs that improve performance.…”

Application of gradient tree boosting regressor for the prediction of scour depth around bridge piers

“…In recent years, machine learning models (Azmathullah et al 2005;Babovic 2009;Karpatne et al 2017;Chadalawada et al 2020;Nearing et al 2020) which allow computers to learn patterns from existing data sets without being explicitly programmed are being very widely used for this purpose (Azamathulla 2012). Artificial neural networks (ANN), which act as a universal approximator have been widely used for estimating bridge scour (Muzzammil 2008;Kaya 2010;Toth & Brandimarte 2011;Choi et al 2015;Kızılöz et al 2015).…”

Post processing the U.S. National Water Model with a Long Short-Term Memory network

“…One potential problem with ML, however, is that it lacks a physical basis. While there are emerging efforts in hydrology to merge physical understanding with machine learning (Karpatne et al, 2017a;Daw et al, 2020;Pelissier et al, 2019;Chadalawada et al, 2020;Tartakovsky et al, 2020), theory informed machine learning (Karpatne et al, 2017b) is still relatively immature in hydrology.…”