The Application of a Genetic Algorithm to the Optimization of a Mesoscale Model for Emergency Response

Abstract Besides solving the equations of momentum, heat, and moisture transport on model grid, mesoscale weather models must account for subgrid-scale processes that affect resolved variables. These are simulated with parameterizations, which often rely values preset by user. Such “free” parameters, along others set to initialize model, poorly constrained, requiring a user select each from range plausible values. Finding optimize any forecasting tool can be accomplished search algorithm, one such process—the genetic algorithm (GA)—has become especially popular. As applied modeling, GAs represent Darwinian process: an ensemble simulations is run different parameter member, members subsequently judged most accurate selected as “parents” who pass their parameters onto new generation. At U.S. Department Energy’s Savannah River Site in South Carolina, we applying GA Regional Atmospheric Modeling System (RAMS) supplies input simulate dispersion airborne contaminant part site’s emergency response preparations. An forecasts day, data used “score” individual ensemble, best next day’s forecasts. meteorological conditions change, change well, maintaining configuration adapted atmospheric conditions. Significance Statement We wanted develop system over day repeatedly adjusted according how well it performed previous day. To series (parameters) control will calculate winds, temperatures, other desired Each was several times using combinations these later compared observed Parameters produced were preferentially reused create The process tested summer 2020 exhibited lower errors than default parameters.