New Bayesian Framework Accounts for Uncertainty in Landslide Forecasts

Published in Natural Hazards and Earth System Sciences on 17 September 2026, the open-access article Bayesian forecasting of triggered landslides introduces a method that expresses landslide occurrence as a probability distribution rather than a single value or a yes-or-no threshold. Authors Flavia Ferriero, IHRR member Professor Fausto Guzzetti and Warner Marzocchi designed the approach to represent uncertainty from incomplete landslide inventories, rainfall measurements and limited knowledge of triggering processes. They tested it with a catalogue of 169 landslides recorded between 2000 and 2022 and up to 22 years of daily observations from 23 rain gauges in Campania, southern Italy.

The results show that landslide probability rises gradually as daily rainfall increases, with no sharp physical threshold. At the highest tested daily rainfall threshold of 70 mm, the posterior probability of landslide occurrence ranged from 0.2 to 0.8 across rain-gauge areas when uncertainty was included. Rainfall information consistently improved forecast skill compared with landslide records alone. The findings indicate that warning thresholds should be treated as operational choices, selected by weighing the costs of false alarms against missed events, rather than as fixed natural boundaries.

The framework offers a transparent way to support early warning and risk management, and its structure could be adapted to other triggering processes, including earthquakes, where sufficient data exist. The current application remains data-intensive and relies on historical daily rainfall, so it cannot provide a next-day operational forecast without meteorological rainfall predictions. Future work will test antecedent rainfall, higher-resolution records, wider regions and the integration of quantitative weather forecasts.

Photo credit: Ferriero, Guzzetti and Marzocchi (2026), CC BY 4.0

Photo caption: Probability gain increases as daily rainfall thresholds rise across the 23 rain-gauge areas. Blue points include uncertainty; green points do not.