Dynamic Calibration of a Spatial SEIQR Model via Bayesian Optimization: Analyzing the Behavioral Gap in Argentina's COVID-19 Pandemic

The COVID-19 pandemic highlighted the limitations of static epidemiological models in capturing shifting human behaviors and non-pharmaceutical interventions. This paper presents a spatial stochastic SEIQR model calibrated dynamically using Bayesian Optimization via the Optuna framework. By dividing a 600-day simulation into 30 temporal windows of 20 days, we optimized key behavioral parameters: mobility ($\nu$) and social friction/care ($KT$). This dynamic calibration achieved a 20% to 30% reduction in the Mean Squared Error (MSE) compared to static approaches. Furthermore, our analysis introduces the "Vulnerability Gap" ($\Delta = \nu_t - KT_t$), demonstrating that the pandemic's severity was largely driven by behavioral fatigue and late social reactivity, transforming abstract mathematical variables into direct proxies for historical epidemiological milestones.