Hourly-Scale Modeling of Storm Transitions in Southern Brazil with Markov Chains

Abstract. Southern Brazil faces escalating flood risk, yet intraday storm dynamics remain under-characterized. This work presents a season-resolved, intraday Variable Length Markov Chain analysis (VLMC) of storm state transitions, using hourly precipitation from 15 INMET stations from 2007 to 2024. Storms were identified and quantified by depth, duration, and intensity, then classified into Moderate, Strong, and Very Strong states. Peak mean intensities reach 9 to 11 mm h⁻¹ in mountainous units, compared with maxima near 7 mm h⁻¹ on the coast, where mean intensities are about 3 mm h⁻¹. Storms last roughly 11 to 13 hours in the Southern Plateau and Southern Shield, and 15 to 17 hours in the Central Depression and Coastal Plain, indicating greater lowland persistence. Upward transition probabilities increase in summer, with Moderate to Strong and Strong to Very Strong reaching about 0.20 in orographic areas, while persistence of Very Strong ranges from 0.10 to 0.20 in the Southern Shield. In winter, downward transitions to Moderate exceed 0.90 across most of the domain. Chi-square diagnostics support first-order, season-specific chains with stable transition structure. These intraday, spatially resolved probabilities link geomorphology to storm persistence and provide actionable inputs for early warning, zoning, and climate risk management.