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

Abstract

Simulations of human behavior in water resources systems are challenged by uncertainty in model structure and parameters. The increasing availability of observations describing these systems provides the opportunity to infer a set of plausible model structures using data‐driven approaches. This study develops a three‐phase approach to the inference of model structures and parameterizations from data: problem definition, model generation, and model evaluation, illustrated on a case study of land use decisions in the Tulare Basin, California. We encode the generalized decision problem as an arbitrary mapping from a high‐dimensional data space to the action of interest and use multiobjective genetic programming to search over a family of functions that perform this mapping for both regression and classification tasks. To facilitate the discovery of models that are both realistic and interpretable, the algorithm selects model structures based on multiobjective optimization of (1) their performance on a training set and (2) complexity, measured by the number of variables, constants, and operations composing the model. After training, optimal model structures are further evaluated according to their ability to generalize to held‐out test data and clustered based on their performance, complexity, and generalization properties. Finally, we diagnose the causes of good and bad generalization by performing sensitivity analysis across model inputs and within model clusters. This study serves as a template to inform and automate the problem‐dependent task of constructing robust data‐driven model structures to describe human behavior in water resources systems.

Dynamic Adaptation of Water Resources Systems Under Uncertainty by Learning Policy Structure and Indicators

“…However, this would also pose a challenge in generating transient forcing scenarios and indicators that are physically consistent. One approach would be to make land use an endogenous response to climate and other changes, rather than exogenous scenarios (e.g., Ekblad & Herman, 2020; Giuliani et al., 2016).…”

A Typology for Characterizing Human Action in MultiSector Dynamics Models

“…For example, AI/ML techniques can be used in both descriptive and prescriptive forms (axis 1a), either to mimic human actors as they behave in the real world based on observed data or to simulate actors as they might ideally behave given a specific goal as they respond to their environment and adapt to change over time. In the former descriptive mode, AI/ ML methods could be deployed alongside big social data (Lazer et al, 2009), for the realistic representation of human actors in multisector systems such as mimicking mobility patterns through a city (Moro et al, 2021) or to infer real-world management practices (Ekblad & Herman, 2021). In prescriptive form, modeled actors could be simulated using AI/ML techniques as state-aware agents that selectively and dynamically react to system states via reinforcement learning (e.g., see model free policy approximation methods in Powell, 2019;Bertsekas, 2019;and food-energy-water examples in Giuliani et al, 2021;Zaniolo et al, 2021;Cohen & Herman, 2021).…”

Uncertainty Analysis in Multi‐Sector Systems: Considerations for Risk Analysis, Projection, and Planning for Complex Systems

“…Additionally, The critical transitions literature provides tools for modeling and empirically detecting shifts in endogenous dynamics (Lade et al, 2013;Scheffer et al, 2009). Models incorporating human and institutional decisions may also be able to incorporate the data-driven generation of model structure (Ekblad & Herman, 2021) coupled with dimension reduction to support feature engineering for dynamic multi-sector data sets (Cominola et al, 2019;Giuliani & Herman, 2018) to generate structural and parametric variants which are consistent with past observations.…”