The influence of rainfall patterns on shallow landslides in New Zealand

Understanding how rainfall events influence the pattern and magnitude of landslide response is an important research focus from geomorphological and hazard planning perspectives. Few studies quantitatively relate spatial patterns in rainfall and landslides, largely due to difficulties in acquiring landslide inventories and data on rainfall patterns for individual storm events. Here, we aim to a) identify which factors most influence susceptibility to rapid shallow landslides at the event scale and b) assess how the spatial density of landslides varies in relation to rainfall. While we do not know precisely when individual landslides were triggered during an event, we can examine how the overall pattern of landslides varies spatially in relation to rainfall and geo-environmental factors.

Rapid landslides triggered by intense rainfall occur extensively in New Zealand’s hill country (land <1000 m in elevation with slopes generally between 20-30°). These landslides are typically shallow (approximately 1 m deep) and small (median source areas 50-100 m²). Past deforestation for pastoral farming accelerated landslide erosion. As a result, large rainfall events, such as Cyclone Gabrielle in February 2023, may trigger tens to hundreds of thousands of landslides, causing significant damage to land, infrastructure, and sites of cultural significance to Māori, as well as agricultural production losses and degradation of receiving environments from excess sediment.

In the present study, we focus on four large storm events that generated over 26,000 landslides across mostly hill country terrain on the North Island of New Zealand in 2017-18. High-resolution (0.5 m), before/after satellite imagery was used to map landslides within each study area. Ground-based weather radar data was processed to generate high-spatiotemporal-resolution gauge-calibrated rainfall grids and compute a) maximum intra-event intensities (30 min – 24-h), b) total event rainfall, and c) pre-event accumulations (10 – 90 days) that influence antecedent soil moisture. Rainfall variables were included alongside geo-environmental factors in a binary logistic regression model applied with automated variable selection using the least absolute shrinkage selection operator (LASSO) to assess the influence of different explanatory variables.

Land cover and slope most influenced landslide susceptibility ahead of intra-event rainfall intensities and pre-event rainfall accumulations. Of the rainfall variables, maximum 12-h rainfall normalised by the 10-y recurrence interval intensity and the 10-d pre-event accumulation normalised by mean annual rainfall had the most influence. Forest cover reduced the sensitivity of landslide spatial density to variations in slope, rainfall, and rock type, in contrast to pasture. Mean landslide density increased 3.5-fold once the maximum 12-h intensity exceeded the 10-y recurrence interval intensity by ≥25% for pastoral land on weak sedimentary rocks. This threshold is consistent with the increase in 12-h rainfall by late century under the highest levels of projected warming in New Zealand, which suggests the landslide response to storm rainfall could be significantly amplified by climate change.