Assessing Rainfall Variability and Atmospheric Dynamics Using a Multisite Hidden Markov Model for Hydrometeorological Hazard Assessment in Yogyakarta, Indonesia

Rainfall variability is a major driver of hydrometeorological disasters, including droughts, floods, and landslides. This study analyzes rainfall and atmospheric dynamics in Yogyakarta using a Multisite Hidden Markov Model (MHMM) estimated via Expectation-Maximization (EM), identifying hidden states (dry, normal, and wet) from decadal (10-day) rainfall data during the September-October-November (SON) season across 18 stations from 2004 to 2023. The results show high persistence of dry and wet states, with estimated transition probabilities of 76% and 69.8%, respectively. Rainfall occurrence probabilities and expected rainfall vary spatially across stations, ranging from 6-25% (2-9 mm/decade) in dry states and 86-94% (85-197 mm/decade) in wet states. These ranges reflect spatial variability among stations rather than formal statistical uncertainty, such as confidence intervals. MHMM-generated maps identify dry and wet states, which can be associated with hydrometeorological hazards such as droughts, floods, and landslides. The temporal dynamics of these states show qualitative correspondence with ENSO and IOD phases. This study focuses on meteorological and climatological hazards as physical phenomena, without modeling exposure or vulnerability, and therefore does not quantify risk. The findings demonstrate MHMM's potential to support hydrometeorological hazard assessment and improve climate-based early warning systems for disaster preparedness in Yogyakarta.