Global connections between El Niño and landslide impacts

Landslides, often triggered by intense rainfall, cause widespread devastation globally, resulting in significant loss of life and infrastructure damage. While seasonal rainfall variations are major contributors to landslide occurrences, the influence of multi-year rainfall patterns, particularly the El Niño Southern Oscillation (ENSO), remains largely unexplored. ENSO's significant influence on rainfall intensity in landslide-prone regions like India, Indonesia, Colombia, and the Philippines makes understanding its impact crucial. This study aims to quantify the global effects of ENSO on landslide exposure by combining updated satellite rainfall data with a global landslide exposure model. This approach allows for a more comprehensive understanding of hazard variability, improving disaster mitigation strategies on seasonal timescales. The study acknowledges the complexity of ocean-atmosphere interactions and simplifies the analysis by focusing solely on the impact of rainfall changes associated with ENSO. Figure 1 illustrates the contrasting rainfall patterns during peak El Niño and La Niña conditions based on NASA's IMERG data, highlighting the significant rainfall shifts across the Pacific and other regions vulnerable to landslides. The study acknowledges the existence of different El Niño types ('Cold-Tongue' and 'Warm-Pool') but does not differentiate between them due to limited data and the ongoing debate surrounding their classification.

Existing research highlights increased landslide fatalities during the Northern Hemisphere summer, linked to the Indian Summer Monsoon and tropical cyclone seasons. However, the extent to which rainfall seasonality drives landslide impacts in many countries remains unclear. Previous studies have demonstrated ENSO's significant impacts on global rainfall patterns and extreme rainfall events. Other climate oscillations like the Pacific Decadal Oscillation (PDO) and the North Atlantic Oscillation (NAO) also influence rainfall patterns, but their effects on landslide-prone regions and timescales are less pronounced compared to ENSO's relatively shorter (2-5 years) cycles. The challenge in connecting global climate patterns with localized landslide impacts stems partly from limited landslide event inventories, particularly in developing countries. Model-based estimates of rainfall-triggered landsliding, such as those provided by the NASA Landslide Hazard Assessment for Situational Awareness (LHASA) model, are valuable tools for addressing this data gap. The availability of nearly two decades of consistent global satellite rainfall data now allows for a detailed examination of ENSO's impact on landslide exposure.

This study combines an updated global landslide exposure model with empirical observations of ENSO, NAO, and PDO. The landslide exposure model utilizes the LHASA model, leveraging nearly two decades of high-resolution, consistent satellite rainfall data from the IMERG dataset. The LHASA model now incorporates a revised global susceptibility map reflecting recent deforestation changes, extending its coverage to higher latitudes. The model identifies landslide hazard 'nowcasts' when 7-day accumulated rainfall exceeds the historical 95th percentile and surpasses a minimum threshold. These nowcasts, representing periods of elevated landslide hazard, are then combined with datasets for population (GPWv4), roads (GRIP), and critical infrastructure (OpenStreetMap) to estimate exposure. For each of the 38,257 administrative districts (admin-2 level), the study calculates the seasonality of exposure for population, roads, and infrastructure. To isolate the effect of ENSO (and PDO/NAO), annual variability is removed by calculating 12-month moving averages of both exposure and ENSO index (MVEI v2). The correlation between smoothed ENSO index and smoothed exposure data is then analyzed, along with the slope of the fit to assess the significance and magnitude of the ENSO influence. P-values are calculated to determine the statistical significance of the relationships between MVEI and various model outputs (total rainfall, extreme rainfall, hazard nowcasts, and exposure). The magnitude of change in exposure due to a unit shift in MVEI is calculated using linear least-squares regression, focusing on areas with statistically significant and large-magnitude effects. The study also analyzes the relative change in exposure due to a unit change in MVEI, normalized by the monthly average exposure, to identify areas with substantial relative changes. Finally, the study compares model outputs with fatality data from the Global Fatal Landslide Dataset (GFLD) to assess the model's predictive performance. The comparison uses average monthly fatal landslide events and average modelled landslide exposure for each MVEI interval to reveal patterns of agreement and any discrepancies between the model and observed landslide impacts.

The study finds that ENSO significantly modulates landslide exposure in numerous regions. In locations such as Colombia, Philippines, and Indonesia, the impact of ENSO on landslide exposure is comparable to or even exceeds seasonal variations. Figure 2 showcases examples of how ENSO can significantly influence landslide exposure in specific administrative districts, illustrating differences in seasonal variability and the relationship between MVEI and population exposure. Figure 3 maps the p-values for the relationship between MVEI and total rainfall/modelled population exposure, indicating significant relationships in various regions. Figure 4 maps the magnitude of exposure change driven by a unit shift in MVEI, highlighting Southeast Asia and northern Latin America as key regions where ENSO has a substantial impact. Figure 5 displays the relative change in exposure due to a unit change in MVEI, revealing areas with substantial relative changes during El Niño periods. The analysis of the different model components shows that the impact of ENSO on total rainfall is the main driver of exposure changes in many regions, while extreme rainfall plays a more significant role in others, such as parts of China, Mexico, and Central Asia. In several locations like Central Asia, Mexico, Iran, Colombia, Indonesia, and the Southern Philippines, the relationship between ENSO and landslide exposure is stronger for total rainfall than for extreme rainfall, suggesting that ENSO's impact on total rainfall may be the primary driver of exposure changes. The study also compared the impact of considering extreme rainfall versus total rainfall on the relationship with MVEI. In some geographical areas, including parts of Central Asia, Mexico, Iran, Colombia, Indonesia, and the Southern Philippines, total rainfall showed a stronger relationship than extreme rainfall. In contrast, the relationship between MVEI and extreme rainfall was stronger in Southern Brazil, Mexico, Eastern China, and parts of Central Asia. A comparison of the model's predictions with fatality data from the GFLD reveals that in most countries, ENSO intervals with the highest modelled exposure align with periods of high fatal landslides, although the model's exposure estimates show less variability than fatality data. This suggests that the model may be overly conservative in estimating impacts, potentially underestimating the effects of increased peak rainfall intensity.

The significant relationship between ENSO and landslide exposure in specific regions highlights the need to understand the underlying mechanisms. The differing roles of total and extreme rainfall changes in mediating ENSO's influence suggests that the two ENSO modes may lead to distinct landslide impacts through different mechanisms. The model's sensitivity to total rainfall rather than extreme rainfall in certain areas implies that even relatively small changes in total rainfall can strongly influence landslide exposure when not masked by other climatological trends. The study's findings offer a valuable first-order estimate of ENSO's impact on landslide exposure, providing valuable insights for disaster risk reduction. The consistency of these findings with landslide fatality data demonstrates the model's utility in forecasting landslide risks. However, limitations exist, such as the model's insensitivity to increased intensity of extreme rainfall events exceeding the 95th percentile and the lack of data from the 1997–1998 El Niño event.

This study presents the first globally consistent model connecting ENSO variability with rainfall-triggered landslide impacts, revealing substantial effects in various regions. Medium-term ENSO forecasting capabilities allow for preliminary estimations of landslide impacts. The model's findings, along with seasonal exposure data, enable nuanced seasonal landslide forecasts globally. Future research should validate the model with specific landslide data and develop a 'baseline' exposure level for future change contextualization. Considering lead-lag relationships between ENSO and rainfall patterns and incorporating a longer rainfall time-series are crucial for improved model accuracy and comprehensiveness. Furthermore, future research should investigate the potential influence of ENSO-driven land cover changes on landslide exposure.

The model's reliance on rainfall exceeding the historical 95th percentile limits its sensitivity to changes in the intensity of extreme events. The relatively short satellite rainfall record prevents inclusion of the significant 1997-1998 El Niño event, which might affect the accuracy of the findings. The model's suitability for specific landslide types (shallow, rainfall-triggered) may also limit the scope of the results. The accuracy of satellite precipitation estimates for extreme rainfall in complex scenarios might also introduce uncertainties. Furthermore, the model does not account for potential lead-lag effects between ENSO and rainfall changes, and it uses static parameters for population, roads, and infrastructure which may affect the generalizability of the findings. Finally, while this study focuses on the role of rainfall changes, other ENSO-related factors (land cover) might also influence landslide exposure.