Global connections between El Niño and landslide impacts

The study investigates how inter-annual climate variability associated with the El Niño–Southern Oscillation (ENSO) modulates exposure to rainfall-triggered landslides globally. Landslides cause thousands of fatalities annually and are commonly triggered by intense rainfall. While seasonal patterns (e.g., monsoons and tropical cyclones) control much of the annual variability in landsliding, the extent to which multi-year climate oscillations such as ENSO alter landslide impacts has remained unclear. ENSO drives significant spatial and temporal changes in rainfall across many landslide-prone regions, including India, Indonesia, Colombia, and the Philippines, potentially affecting both total and extreme precipitation. The purpose of this study is to quantify and map the global connections between ENSO strength and landslide exposure to people and infrastructure, thereby informing disaster risk reduction on seasonal timescales.

Prior work shows: (1) Seasonal peaks in landslide fatalities, notably during the Northern Hemisphere summer aligned with monsoon and tropical cyclone seasons. (2) ENSO substantially alters rainfall patterns and extremes globally, with teleconnections affecting Latin America, the Caribbean, Southeast Asia, and India. (3) Studies identify conventional and possible non-standard El Niño types (Cold-Tongue vs Central Pacific/Modoki), which may produce differing precipitation responses, though their distinction remains debated and standardized indices are limited. (4) Other decadal teleconnections (PDO, NAO) affect precipitation but operate on longer timescales, complicating attribution within short observational windows. (5) Global landslide inventories are sparse and uneven, limiting empirical assessment; satellite-based rainfall with model-based hazard proxies offer a means to bridge data gaps.

Data and model framework: The study combines an 18-year (2001–2018/2019) global satellite rainfall record (NASA GPM IMERG v06B, merging GPM and TRMM) with an updated NASA LHASA (Landslide Hazard Assessment for Situational Awareness) model and global exposure datasets (population, roads, critical infrastructure). LHASA computes daily hazard nowcasts at 30 arc-second resolution when 7-day exponentially weighted accumulated rainfall (exponent=2) exceeds the historical 95th percentile (2001–2019 baseline) and a minimum 6.6 mm threshold, and only where susceptibility is at least ‘moderate’ (heuristic susceptibility using slope, land cover, roads proximity, lithology; validated against the NASA Global Landslide Catalog). Hazard is treated as a binary daily nowcast, aggregated to monthly counts per grid cell. Exposure estimation: Hazard outputs are combined with static exposure rasters: population (GPWv4, 2015), roads (GRIP, 2015; converted to km per cell), and critical infrastructure (OpenStreetMap, snapshot July 2018; point density per cell). Exposure units represent days exposed to hazard multiplied by the exposed element per cell, aggregated to admin-2 districts. Outputs are daily, summed to monthly time series (216 months) for population, roads, and infrastructure exposure. Spatial aggregation: Exposure is aggregated to admin-2 districts (GADM) for interpretability and to reduce noise from single-pixel anomalies. District-level totals can be normalized by population or long-term monthly exposure for intercomparison. Teleconnection indices and smoothing: ENSO strength is represented by NOAA’s Multivariate ENSO Index version 2 (MVEI v2). For each month, 12-month moving averages are computed for exposure and MVEI (window includes the month and prior 11 months) to remove annual seasonality. Similar comparisons are made for total monthly rainfall, number of extreme-rainfall days (>95th percentile) per month, and LHASA hazard nowcasts per month. Statistical analysis: For each admin-2 district, linear least-squares regression between smoothed MVEI and smoothed exposure is fitted to obtain slope (magnitude of exposure change per unit MVEI) and p-value (significance). P-values are also computed for relationships between MVEI and total rainfall, extreme rainfall days, and hazard nowcasts to diagnose which components (total vs extreme rainfall) are most linked to ENSO. Maps display districts with significant relationships (p<0.05) and the slope sign/magnitude. Relative change maps express slope normalized by long-term monthly average exposure. Comparisons with PDO and NAO are performed but interpreted cautiously due to timescale limitations. Validation with impacts: Country-level comparison against the Global Fatal Landslide Dataset (GFLD, 2004–2016; >30 events per country) examines correspondence of fatality frequency across MVEI intervals with modeled exposure distributions. Performance is summarized qualitatively and via variability (standard deviation) of the ratio between observed fatalities and modeled exposure by MVEI bin. Assumptions: ENSO impacts are assumed to affect landslides primarily via rainfall changes; susceptibility and exposure layers are static over the analysis period; linear relationships are applied consistently across districts.

• ENSO–exposure linkage and significance: Many admin-2 regions exhibit statistically significant relationships (p<0.05, often far lower) between MVEI and both total rainfall and modeled landslide exposure. Strongest associations occur in northern South America, southern Brazil/Uruguay, Central Asia, northern Australia, and Southeast Asia.
• Magnitude and direction of effects: Linear slope maps show spatially coherent patterns. La Niña conditions are associated with greater landslide exposure across northern South America, Central America and the Caribbean, Indonesia, Papua New Guinea, and the Philippines. El Niño conditions correspond to increased exposure in parts of Central Asia (e.g., Tajikistan, Kyrgyzstan), eastern/southeastern China, and Mexico.
• Example magnitude: In Planadas, Colombia, ENSO-driven changes in population exposure are on the order of ~60,000 person-days per month, comparable to but somewhat smaller than seasonal variability (~100,000 person-days per month). Slopes are presented as change in average days each person is exposed per month per unit MVEI; large-magnitude, significant slopes cluster in parts of SE Asia and northern Latin America.
• Relative changes: Normalized relative-change maps highlight that regions with large absolute slopes also experience large percentage changes per unit MVEI; additional hotspots of large relative change appear in Mexico, southern Brazil, and Central Asia during El Niño.
• Total vs extreme rainfall pathways: In many key regions (e.g., Central America, northern South America, Indonesia, southern Philippines), ENSO’s effect on total rainfall better explains exposure variability (lower p-values for total rainfall than extreme rainfall). In contrast, in eastern China, Luzon, and parts of Thailand/Cambodia/Vietnam, extreme rainfall shows a stronger ENSO linkage than total rainfall.
• Model–impact comparison: In 11 of 20 assessed countries with sufficient fatality records, the MVEI intervals of highest modeled exposure match those with the highest observed fatal landslide frequency. Modeled exposure tends to be less variable (more conservative) than fatalities, potentially missing peaks linked to increases in extreme rainfall intensity. Lowest deviation between modeled exposure and fatalities occurs in Indonesia, China, Mexico, and Uganda; larger deviations in Thailand, Myanmar, and India.
• Other teleconnections: Given the 18-year record, robust conclusions about PDO and NAO impacts on landslide exposure are not possible, despite some localized associations.

The findings demonstrate that ENSO exerts a meaningful and regionally variable control on landslide exposure by modulating rainfall. The analysis clarifies when total rainfall anomalies versus changes in the frequency of extreme rainfall are the dominant pathways linking ENSO to landslide hazard. Regions where La Niña increases exposure often coincide with stronger total rainfall–ENSO correlations, whereas regions with El Niño–related increases (e.g., Central Asia, eastern China) align more with extreme-rainfall sensitivity. Incorporation of susceptibility and static exposure layers only modestly alters the strength of the ENSO relationships, suggesting the core ENSO–rainfall link drives exposure variability. Comparison with fatality datasets indicates that the modeled exposure captures broad ENSO-related patterns of landslide impacts, albeit with a tendency to under-represent peak intensities, implying conservative estimates. Overall, these results provide a foundation for seasonal risk assessment informed by ENSO forecasts and highlight priority regions for preparedness and mitigation.

This study provides the first globally consistent quantification of how ENSO modulates exposure to rainfall-triggered landslides, integrating satellite rainfall, a global landslide hazard model, and exposure datasets. Key contributions include (i) mapping significant, directionally consistent ENSO–exposure relationships, (ii) distinguishing where total versus extreme rainfall changes drive exposure shifts, and (iii) demonstrating qualitative agreement with observed fatality patterns in multiple countries. The results support the feasibility of seasonal-to-interannual landslide risk outlooks informed by ENSO forecasts and offer actionable insights for disaster risk reduction. Future work should: extend analyses with longer rainfall records (capturing major events such as 1997–1998), explore lead–lag relationships and potential nonlinearity, differentiate El Niño flavors where data permit, integrate time-varying exposure/vulnerability, and validate with expanded, consistent landslide impact inventories. Interactions between climate change, evolving ENSO behavior, and landslide hazard should be jointly assessed in forward-looking risk studies.

• Temporal coverage: The ~18-year satellite rainfall record (2001–2018/2019) is short for robust assessment of multi-decadal PDO/NAO impacts and excludes the major 1997–1998 El Niño event.
• Rainfall measurement limits: Satellite precipitation uncertainties are larger for orographic enhancement, very short-duration high-intensity storms, and mixed rain–snow events (e.g., U.S. Pacific Northwest), potentially under-resolving relevant extremes.
• Hazard model scope: LHASA is optimized for shallow, short-duration rainfall-triggered landslides; it is less suited for slow-moving or deep-seated failures driven by long-term moisture accumulation.
• Extreme intensity insensitivity: The hazard trigger depends on exceedance frequency over the 95th percentile, not the intensity above that threshold; increases in peak intensity may not raise hazard/ exposure estimates.
• Static exposure/susceptibility: Population, roads, and infrastructure layers are treated as static over the analysis period; susceptibility parameters are heuristic and calibrated to catalog performance.
• Lead–lag not modeled: Correlations use contemporaneous 12-month moving averages, potentially missing region-specific lags between ENSO peaks and local rainfall/landslide responses.
• ENSO event types: The study does not differentiate canonical versus Central Pacific (Modoki) El Niño types due to limited cycles and index standardization; potential differences are unexamined.
• Spatial aggregation: Admin-2 aggregation may mask heterogeneous signals within large or topographically diverse districts.
• Comparability to impacts: Fatalities reflect vulnerability in addition to exposure; country-level comparisons are limited by relatively small event counts and differing time spans.