AiNet- and GIS-based regional prediction system for the spatial and temporal probability of rainfall-triggered landslides

Abstract

We developed a real-time forecasting system, aiNet-GISPSRIL, for evaluating the spatiotemporal probability of occurrence of rainfall-triggered landslides. In this system, the aiNet (a kind of artificial neutral network based on a self-organizing system) and GIS are merged for integrating the rainfall conditions into various environmental factors that influence the landslide occurrence and for simulating the complex non-linear relationships between landslide occurrence and its related conditions. Zhejiang Province (101,800 km2 in area), located in the southeast coastal region of China, is highly prone to the occurrence of landslides during intensive rainfall. Since 2003, the aiNet-GISPSRIL has been used to predict landslides during the rainy seasons in the region. The aiNet-GISPSRIL uses the regional 24-h forecast rainfall information and the real-time rainfall monitoring data from the rain-gauge network as its inputs, and then provides 24-h forecast of the landslide probability for every 1 × 1-km grid cell within the region. Verification studies on the performance of the aiNet-GISPSRIL show that the system has successfully predicted the dates and localities of 304 landslides (accounting for 66.2% of reported landslides during the period). During the period from 2003 to 2007, because the system provided the probability levels of landslide occurrences up to 24-h in advance, gave locations of potential landslides, and timely warned those individuals at high-risk areas, more than 1700 persons living in the risk sites had been evacuated to safe ground before the landslides occurred and thus casualty was avoided. This highly computerized, easy-operating system can be used as a prototype for developing forecasting systems in other regions that are prone to rainfall-triggered landslides.