Exploring Calibration of Nextgen Formulations Using Various Objective Functions and Optimization Algorithms

NOAA’s prototype Next Generation Water Resources Modeling Framework (NextGen) provides a model-agnostic modular system to support large-scale hydrologic modeling with mosaics of model formulations, driven by the dominant processes occurring in a given basin or region. NextGen also includes a comprehensive calibration capability (NGEN-CAL) with multiple options of optimization algorithms and cost functions to facilitate formulation selection and parameter optimization for basins with high quality streamflow observations via calibration. Currently available optimization algorithms include Dynamic Dimensioned Search (DDS), Particle Swarm Optimization (PSO), and Grey Wolf Optimization (GWO). NGEN-CAL also supports choosing from over a dozen metrics (e.g., Kling-Gupta Efficiency, Nash-Sutcliffe Efficiency, Pearson Correlation, Critical Success Index, among many others) to be used as the objective function for calibration.

We hypothesize that, for a given basin with a chosen formulation and fixed meteorological forcing and streamflow data, the final optimal parameters and subsequently the performance of model simulations during validation could be critically dependent on the choices of optimization algorithms and objective functions for calibration. To that end, we design and carry out a series of multi-year calibration/validation experiments, each with a different objective function and optimization algorithm. The experiments are carried out for a few carefully chosen study basins from different hydroclimatic regions to help draw more meaningful conclusions. In this presentation, we discuss the relevant details of these experiments, analyze the performance in streamflow simulations, and summarize the findings from the study that could potentially benefit future calibration efforts toward building the next version of the National Water Model (NWM).