Lead-Time Sensitivity of GFS Forecasts for the Deadly 2011 Nova Friburgo Disaster (Brazil): A Hindcast-Based Synoptic and Thermodynamic Assessment

Abstract

Abstract The catastrophic rainfall of 11–12 January 2011 triggered hundreds of landslides and flash floods in the mountainous region of Rio de Janeiro state, Brazil, with Nova Friburgo among the affected municipalities. This paper assesses how the predictability of this event depends on forecast lead time, focusing on the deterministic Global Forecast System (GFS) and the representation of the spatiotemporal structure of extreme precipitation in complex terrain. A hindcast experiment is performed using GFS initializations relative to 18 UTC on 11 and 12 January 2011, with rainfall verified against rain gauges in Nova Friburgo. GOES-12 infrared imagery is used to contextualize convective development, and synoptic dynamic and thermodynamic fields are analyzed to link forecast behavior to environmental evolution. All runs reproduce a broad-scale signal of enhanced precipitation over the coastal-mountain transition, but the local 6-h and 24-h rainfall maxima are sensitive to displacement errors. Skill is not monotonic with lead time: the intermediate-range run better captures the local core than the shorter- and longer-range runs, which displace their cores away from the municipality. The synoptic environment is consistently depicted across forecasts, with a surface low deepening to nearly 998 hPa, 850-hPa specific humidity of roughly 15 g/kg over the moisture-convergence axis toward Southeast Brazil, and an upper-level cyclonic vortex supporting divergence aloft; this configuration is corroborated by Galeão radiosonde soundings showing CAPE close to 660 J/kg ahead of the event, rising to about 2200 J/kg approaching the peak rainfall period. Mesoscale positioning of the moisture-convergence maximum relative to topography controls where extreme rainfall is realized. These results point to the need for verification approaches that separate timing from displacement, and for complementary tools such as ensembles and convection-permitting models to better characterize lead-time-dependent uncertainty in extreme rainfall forecasts over complex terrain.