Typhoon Rainstorms Trigger Landslides Differently in Northern and Southern China

When the remnants of Typhoon Doksuri swept across northern China in late July 2023, they delivered a deluge that few in the semi-humid Northeastern Taihang Mountains had ever witnessed. More than 700 millimeters of rain fell at several monitoring stations within days, and the slopes responded catastrophically: over 18,000 shallow landslides scarred the mountains west of Beijing and its surroundings. For a region where such clustered slope failures had long been considered rare, the event was a stark warning. Now, a comparative study published in the International Journal of Disaster Risk Science has dissected why these landslides happened where they did, and how their triggering conditions differ from those in the humid, landslide-prone hills of southern China.

A team of researchers led by Yue Lu and Gang Mei of the China University of Geosciences in Beijing, together with colleagues from Chang’an University, compared two dramatically different landscapes struck by typhoon rainstorms: the semi-humid Northeastern Taihang Mountains in the north and the humid Zixing Hills in Hunan Province, where Typhoon Gaemi unleashed an extreme rainstorm in August 2024. The team mapped 18,494 shallow landslides in the Taihang study area and 14,785 in the Zixing Hills, using manual visual interpretation of high-resolution satellite imagery acquired before and after each rainfall event. These inventories, combined with fifteen geospatial conditioning factors spanning meteorology, geology, hydrology, geomorphology, and ecology, formed the foundation for one of the most detailed regional comparisons of typhoon-induced landslide triggering conducted to date.

The analytical approach combined classical spatial statistics with explainable machine learning. Kernel density estimation and Ripley’s K function revealed that landslides in both regions cluster far more strongly than random chance would allow, with observed clustering curves lying consistently above the 95 percent confidence envelope of a completely spatial random model. At short to medium distances of up to three kilometers, clustering intensity was remarkably similar between the two regions. Beyond that scale, however, the differences emerged: the Zixing Hills showed an extended clustering reach of 6,800 meters compared with 5,337 meters in the Taihang Mountains, while mean clustering intensity was slightly higher in the north at 7.11 versus 5.74 in the south.

To move beyond description and into mechanism, the researchers trained XGBoost classification models for each region and interpreted them using Shapley additive explanations, or SHAP, a technique that quantifies how each conditioning factor pushes landslide probability up or down at individual locations. The models achieved satisfactory validation performance, and their outputs revealed a striking commonality: precipitation was the dominant factor in both regions, underscoring the fundamental role of extreme rainfall in destabilizing slopes regardless of climate zone. Yet the secondary controls diverged sharply. In the Taihang Mountains, elevation and lithology ranked next in importance, whereas in the Zixing Hills, slope aspect took second place, with lithology, elevation, topographic position index, and vegetation cover exerting moderate influence.

The single-factor analysis uncovered hydrological thresholds that differ profoundly between the regions. In the semi-humid Taihang Mountains, SHAP values shifted from negative to positive at a cumulative precipitation of roughly 420 millimeters and climbed steeply toward 700 millimeters. In the humid Zixing Hills, the transition occurred at a much higher level, approximately 590 millimeters, and remained elevated within the 590 to 700 millimeter range. The researchers attribute this contrast to the deep, highly weathered granitic saprolite that mantles the southern hills. This thick regolith acts as an enormous water reservoir, and slope failure there appears to require rainfall sufficient to exceed its substantial storage capacity, a saturation-excess regime rather than a rapid surface response.

Interaction analysis sharpened the picture further. In both regions, the coupling between precipitation and elevation was the most significant factor pair, with interaction values of 0.109 in the north and 0.135 in the south, and landslide probability peaked in the mid-elevation band of 300 to 1,000 meters in both cases. But the accompanying mechanisms diverged. In the Taihang Mountains, where thin colluvial mantles of just 0.5 to 2 meters overlie relatively impermeable bedrock, slopes respond rapidly to intense rainfall, with transient pore-pressure buildup concentrated on gentle concave-to-convex slope positions characterized by topographic position index values between −7 and 9, rather than exclusively in convergent hollows. When rainfall exceeded 450 millimeters, landslides spread across these transitional slope positions, suggesting that topographic focusing of surface runoff drives failure in a system with limited water storage.

Perhaps the most counterintuitive finding concerns vegetation. In the semi-humid north, landslides concentrated in sparsely vegetated areas with normalized difference vegetation index values below 0.4, and rainstorms amplified instability most severely on slopes with NDVI below 0.2. Dense natural forests with deep root systems appear to protect these thin soils through mechanical root reinforcement, rainfall interception, and enhanced evapotranspiration. In the humid Zixing Hills, the relationship inverted: the highest landslide densities occurred in densely vegetated zones with NDVI values between 0.79 and 0.84. The explanation lies in land use and hydrology. The lush slopes of Zixing typically support tea plantations, secondary shrublands, and planted forests with shallow root systems frequently disturbed by human activity. Dense canopies reduce evaporation and keep the thick residual soils near saturation for extended periods, so extreme rainfall generates excessive pore pressure rather than being buffered by the vegetation.

Geomorphology tells a parallel story of regional contrast. The Taihang landslides favored gentle transitional slope positions where shallow soils can quickly saturate, consistent with a rapid hydrological response regime. The Zixing landslides, by contrast, clustered in terrain of moderate local relief between 20 and 35 meters. High-relief zones in the south tend to have thin soils eroded away and drain too efficiently to permit deep saturation, while low-relief areas lack the gravitational driving force for failure. Moderate relief offers the optimal combination of sufficient slope energy and topographic conditions that allow deep groundwater to converge and accumulate within the weathered mantle.

The practical implications are considerable. Rainfall thresholds calibrated for humid southern China, where roughly 590 millimeters of cumulative rain may be needed to trigger failure in deep regolith, would be dangerously misleading if applied to the semi-humid north, where slopes can fail after approximately 420 millimeters or less. Conversely, treating vegetation cover as a universal stabilizing indicator would overestimate safety in the tea-covered hills of Hunan while underestimating it in the sparsely vegetated gullies of the Taihang range. The authors argue that early warning systems must be regionally adaptive, incorporating mechanism-based, context-specific parameters rather than one-size-fits-all thresholds. They also acknowledge limitations: the proposed rapid-response and storage-controlled regimes remain hydrologically plausible interpretations rather than directly verified mechanisms, since no in situ groundwater or pore-pressure monitoring was available. The team plans to establish monitoring networks in sensitive areas and integrate them with numerical modeling to validate the hydraulic coupling processes behind landslide occurrence, a step they argue is essential as climate change intensifies extreme rainfall events across both climatic zones.

Subject of Research: Comparative triggering conditions of typhoon rainstorm-induced clustered shallow landslides in semi-humid northern and humid southern China.

Article Title: Similarities and Differences in Triggering Conditions of Typhoon Rainstorm-Induced Clustered Shallow Landslides in Semi-humid Northern and Humid Southern China

Article References: Lu, Y., Mei, G., Ma, Z., Zhang, Y., & Peng, J. (2026). Similarities and Differences in Triggering Conditions of Typhoon Rainstorm-Induced Clustered Shallow Landslides in Semi-humid Northern and Humid Southern China. International Journal of Disaster Risk Science. https://doi.org/10.1007/s13753-026-00770-6

Image Credits: AI Generated

DOI: 10.1007/s13753-026-00770-6

Keywords: clustered shallow landslides, typhoon rainstorms, triggering conditions, precipitation-elevation coupling, vegetation cover, NDVI, topographic position index, XGBoost, SHAP, landslide early warning, Taihang Mountains, Zixing Hills