Rethinking geohazard monitoring under extreme weather: a call for a coupled weather–terrain framework

Introduction

China has developed one of the world’s most extensive geohazard monitoring networks, with more than 70,000 automated monitoring stations deployed at high-risk sites and nearly 300,000 potential geohazard sites identified through field investigation and remote sensing (Zhou et al., 2026; Xu, 2020). This site-centered strategy has substantially strengthened the monitoring work of known unstable slopes or landslides. However, its dependence on historically identified hazard locations creates an inherent spatial blind spot, as approximately 80% of newly occurring geohazards have historically occurred outside previously identified hazard points (Department of Natural Resources of Sichuan Province, 2021; Yin 2024). This limitation becomes increasingly important as short-duration rainfall extremes intensify under a warming climate (Fowler et al., 2021). In mountainous regions, rainfall forcing is far from spatially uniform. Terrain dynamically interacts with moisture-bearing boundary-layer flows through orographic lifting, flow blocking and channeling, and moisture convergence, redistributing and locally amplifying precipitation (Lin et al., 2001; Houze, 2012). Importantly, the orographic influence on precipitation extremes is strongly dependent on rainfall duration and atmospheric conditions, with particularly pronounced effects at sub-daily to hourly timescales (Marra et al., 2021; Formetta et al., 2022). Consequently, rainfall hotspots can shift between slopes and catchments as low-level wind and moisture transport evolve during an extreme event, and may not coincide with historically identified geohazard sites.

This perspective suggests that terrain should not be regarded only as a static control on landslide susceptibility; it also actively modulates the meteorological forcing that triggers slope failure. Where terrain-amplified rainfall intersects steep and hydrologically susceptible slopes, rapid infiltration, pore-pressure increases, and runoff concentration can initiate clusters of landslides and subsequently mobilize sediment into channels, promoting debris flows and broader cascading hazards (Guzzetti et al., 2008; Gariano and Guzzetti, 2016). Geohazard hotspots may therefore emerge dynamically from the coupling between atmospheric forcing and terrain, rather than simply reproducing the spatial distribution of past failures. Monitoring networks optimized for known unstable slopes may thus have limited capacity to anticipate newly emerging hazard hotspots beyond their existing observational footprint.

Addressing this challenge requires a transition from site-based monitoring toward a dynamic weather–terrain–geohazard framework that integrates synoptic moisture supply, boundary-layer airflow, orographic rainfall amplification, hydrological response, hazard-prone geological structure, and hillslope–channel connectivity. Terrain-amplified rainfall can increase hillslope saturation and runoff, while failures connected to channels may be entrained into debris flows and downstream sediment transport. The objective should shift from monitoring where slopes are already known to be unstable toward anticipating where slopes may become unstable under an evolving extreme-weather event. Such a transition will require closer collaboration among meteorologists, hydrogeologists, geomorphologists, data scientists, engineers, decision-makers, grid liaison officers and emergency responders, together with the integration of ground sensors, weather radar, satellite observations, numerical weather prediction, and terrain-geohazard aware hazard models. It also requires greater public awareness that safety in mountainous regions cannot rely solely on installed sensors, because changing meteorological forcing and its dynamic interaction with local topography can generate new areas of instability beyond historical records and established monitoring networks.

Blind spots in history-based hazard system

Climate warming is reshaping the spatial distribution of geological hazards. During China’s 2024 flood season, GeoStudio simulations of representative fully weathered granite slopes (∼6 m thick) showed that the factor of safety (FOS) could decrease from >1.8 to ∼0.92 under extreme rainfall. A complementary infinite-slope illustration for the Southern Appalachians showed a decrease from 1.22 under dry conditions to ∼1.00 as the normalized water-table height reached 0.44 (Tao and Barros, 2014). The above simplified FOS calculation represents hydrological sensitivity rather than a direct propagation of the terrain-induced rainfall anomaly; in a fully coupled implementation, terrain-resolved rainfall would instead provide the surface-flux boundary for transient infiltration and pore-pressure modeling before FOS calculation (Baum et al., 2010). These examples illustrate how rainfall-driven hydrological changes can destabilize previously stable slopes beyond existing monitoring footprints.

This pattern exposes a fundamental blind spot in history-based hazard monitoring: monitoring resources are preferentially allocated to locations where hazards have already been recognized, whereas extreme weather can generate new instability in terrain with little or no documented failure history. Such a reactive paradigm resembles the traditional Chinese expression, “treat the head when the head aches and the foot when the foot hurts,” addressing hazards only after they become visible rather than anticipating where they may emerge. Current guidelines for monitoring-site selection remain strongly informed by historical geohazard inventories and previous survey results. Under increasingly non-stationary climatic conditions, however, past hazard occurrence may no longer provide a sufficient proxy for future risk. Extreme rainfall, particularly when amplified by local weather–terrain interactions, can dynamically transform previously stable slopes into new geohazards.

Limitations of geohazard monitoring in high-altitude terrain

In complex high-altitude terrain, geohazard risk is not determined solely by the historical stability of a valley or slope. Orographic interactions between moisture-bearing airflow and local topography can dynamically reorganize rainfall patterns, particularly when slope orientation becomes aligned with the prevailing low-level wind. Under such conditions, terrain can enhance uplift and moisture convergence, creating intense precipitation hotspots over upper catchments even when rainfall measured at valley entrances remains moderate. This dynamic windwardness is rarely represented in conventional regional warning systems, which typically rely on sparse rainfall observations, historical hazard inventories, and static terrain susceptibility.

The catastrophic debris flow that struck Guza Town, Sichuan, on 3 August 2024 illustrates this monitoring blind spot. The event originated in high-altitude catchments approaching 5,000 m and caused severe damage to settlements and infrastructure, resulting in 27 fatalities. Rainfall recorded near the valley entrance reached only 25.5 mm, apparently remaining below conventional debris-flow warning thresholds (Yin 2024; Chang et al., 2024). Post-event investigations, however, indicated substantial rainfall amplification in the upper mountainous catchment, where terrain-enhanced precipitation was not adequately captured by the existing observation network. Consequently, a threshold derived from low-elevation or valley-based measurements failed to represent the actual meteorological forcing acting on the debris-flow source area.

This case highlights a fundamental limitation of history- and site-based geohazard monitoring: the location of the sensor may not coincide with the location where meteorological forcing is most strongly amplified. In high mountain environments, monitoring strategies need to move beyond fixed observations at known hazard sites toward a weather–terrain coupled framework that identifies dynamically evolving windward slopes, orographic rainfall hotspots, and geomorphologically sensitive source areas. Strategically deployed high-altitude observations, together with radar, satellite precipitation, atmospheric modelling, and terrain-aware forecasting, could help close this spatial gap and anticipate emerging landslide and debris-flow hazards beyond existing monitoring footprints.

Challenges of small and micro-scale geohazards

Extreme rainfall events are transforming formerly stable slopes into unstable or critically stable conditions, with particularly significant effects on small- and micro-scale hazard bodies. Currently, the majority of annually occurring geological disasters are characterized by volumes below 100,000 cubic meters, classifying them as small and micro-scale geohazards.

For instance, during the 2023 flood season, western Beijing experienced unprecedented torrential rainfall, with Fangshan District recording 1,025.5 mm of precipitation over four days—far exceeding the city’s annual average of approximately 600 mm—and triggering more than 1,200 clustered, small-scale flow slides across Fangshan and Mentougou districts in Figure 1 (Nan et al., 2024). Cui et al. (2026) conducted terrain-removal experiments on the “23·7” event in Beijing and found that the actual topography increased cumulative precipitation over the southwestern mountainous area of Beijing by approximately 229.4 mm, corresponding to an increase of about 41.3%. In contrast, precipitation over the plain areas decreased. The study further indicated that the Taihang Mountains not only uplifted the warm and moist airflows but also impeded the movement of the low-pressure system, thereby prolonging the duration of the extreme precipitation process by approximately 12 h.

As summarized in Table 1, Hurricane Helene provides a representative example of terrain–rainfall–landslide coupling. Orographic simulations showed that the Appalachians locally increased accumulated rainfall by more than 300 mm on upslope terrain (Bercos-Hickey et al., 2026). Post-event mapping identified 2,217 landslides, 64% of which initiated on southeast-facing, predominantly windward slopes (Schaefer et al., 2025). This correspondence illustrates how terrain-modified rainfall can contribute to the spatial concentration of slope failures.

TABLE 1. Representative evidence for weather–terrain–geohazard coupling.

Reported associations summarize evidence from the cited studies and should not be interpreted as multivariable causal attribution of landslide initiation to terrain-induced rainfall after controlling for geological and topographic susceptibility.

Small- and micro-scale geohazards are prevalent in the southwestern, southeastern, and northern mountainous regions. Although individually limited in volume, these events are often highly concealed and spatially scattered, increasing the difficulty of early identification and hindering the precise deployment of monitoring instrumentation.

Although the governing weather–terrain–hydrological coupling is transferable across mountain regions, its operational implementation requires region-specific calibration of rainfall characteristics, geological susceptibility, hydrological response, monitoring density, and warning thresholds.

Barriers to proactive monitoring

The challenge is further compounded by the limited mobilization of social capital and monitoring resources. Although private-sector actors, particularly within the insurance and risk-management industries, have begun developing catastrophe models to quantify geohazard exposure and potential losses, the development of advanced early-warning algorithms remains largely concentrated within academia. Enterprises therefore tend to focus on hardware development, sensor manufacturing, and engineering implementation, while sustained mechanisms for large-scale field validation, data sharing, algorithmic benchmarking, and iterative model improvement remain underdeveloped. A more effective geohazard-monitoring paradigm will require moving beyond this fragmented model toward an open and collaborative ecosystem in which academia, industry, government agencies, insurers, local communities, and emergency responders jointly contribute data, field observations, modelling capacity, and operational feedback. Such broad societal participation is essential for translating emerging monitoring technologies from isolated technical solutions into adaptive and scalable systems for proactive geohazard risk reduction.

Discussion

Existing warning systems often incorporate terrain through susceptibility maps, terrain attributes, or spatial rainfall fields. However, terrain is generally treated as a static susceptibility factor rather than as a dynamic modifier of meteorological forcing. The proposed framework therefore extends current practice by explicitly considering wind-dependent orographic uplift, blocking, channeling, and moisture convergence to anticipate shifting rainfall and instability hotspots.

We advocate a paradigm shift from hazard-site-centered monitoring toward monitoring and early-warning systems grounded in the physical mechanisms of localized severe weather–terrain–geohazard chains. This transition requires a multiscale framework linking regional circulation, boundary-layer processes, terrain-modulated precipitation, hydrological response, and slope instability, rather than treating meteorological conditions as external forcing on a static geological system. A practical implementation could couple high-resolution atmospheric modeling with hydrological and slope-stability models: terrain-resolved rainfall fields would drive infiltration and pore-pressure simulations, which would then be propagated into time-dependent factor-of-safety calculations and dynamic susceptibility or risk zoning. Counterfactual terrain experiments could further isolate orographic contributions by comparing precipitation, pore pressure, factor of safety, and unstable area under actual and reduced-terrain scenarios. Such a process-based vulnerability chain would provide a quantitative pathway for tracing risk from meteorological anomalies through orographic rainfall amplification and hydro-mechanical perturbation to landslides, debris flows, and cascading downstream impacts. The added value of this framework should ultimately be tested against conventional site-based warning using the same independent events and quantitative metrics, including detection and missed- and false-alarm rates, warning lead time, and, critically, the detection of failures outside existing hazard inventories or monitoring footprints (Calvello and Piciullo, 2016; Piciullo et al., 2017).

A prerequisite for such advances is the systematic curation, classification, and integration of historical observations and geohazard records. We propose establishing a nationally coordinated database of benchmark extreme events, with particular emphasis on compound and cascading geohazards triggered by climate extremes. Representative cases should encompass diverse climatic and geomorphic settings, including extreme rainfall in northern China, typhoon-related precipitation in southern and southeastern regions, and cryosphere-related changes such as snowline retreat and intensified freeze–thaw processes across western high-altitude plateaus. These benchmark events should integrate meteorological observations, radar and satellite products, geological and geomorphological information, soil-moisture measurements, slope-deformation records, remote sensing, and post-event field surveys. Such a database would provide a common basis for model development, intercomparison, and validation across regions and hazard types.

Building on this data infrastructure, physics-based simulations could sample combinations of terrain-resolved rainfall, slope geometry, initial hydrological state, and hydraulic and geotechnical properties to generate time-dependent pore pressure and factor of safety (FOS). These simulations could train an ML surrogate for rapid prediction of FOS or instability probability from forecast rainfall and current slope conditions, followed by calibration with monitoring observations (Li et al., 2025). Operationally, gauges, radar, satellites, and kilometer-scale numerical weather prediction could provide terrain-resolved rainfall at sub-hourly to hourly intervals, while soil moisture and deformation observations update evolving slope conditions. Validation should withhold independent storm events or regions and compare intermediate hydrological states and final instability predictions against physics-only and data-only models. Supporting observations across meteorological, geological, hydrological, and related sectors should therefore be progressively integrated into an interoperable natural-hazard observation network.

To support this transition, we propose establishing a National Intelligent Geohazard Monitoring and Early Warning Laboratory, complemented by regional field-observation stations in representative high-risk environments. These platforms should bring together meteorologists, hydrogeologists, geomorphologists, numerical-modelling specialists, AI researchers, data scientists, instrumentation engineers, emergency responders, and policymakers. Their role should extend beyond algorithm development to long-term field observation, instrument testing, model benchmarking, cross-regional validation, and the translation of scientific advances into operational warning systems. Particular attention should be given to improving observations in topographically complex mountain regions, where sparse conventional rain gauges may fail to capture strong spatial gradients in precipitation. Advanced weather radar, satellite retrievals, dense high-altitude precipitation observations, and other three-dimensional sensing technologies should therefore be developed and strategically deployed to resolve terrain-induced rainfall heterogeneity in near real time. In parallel with monitoring known hazard sites, greater emphasis should be placed on observing dynamic triggering factors—particularly rainfall intensity, antecedent precipitation, soil moisture, groundwater response, and terrain-conditioned hydrological processes—in areas where extreme weather may generate previously unrecognized instability. Major deployments should nevertheless be preceded by systematic assessments of cost-effectiveness, social benefit, operational feasibility, and residual risk.

Government investment alone is unlikely to sustain the scale of technological development and operational expansion required. Carefully structured participation by private capital could help create a market-based ecosystem for advanced sensors, forecasting services, data products, and risk-management technologies, supported where appropriate by public procurement of monitoring and data services. The insurance and financial sectors could further accelerate the adoption of quantitative geohazard-risk assessment by developing products specifically designed for losses associated with extreme weather-induced landslides, debris flows, and cascading hazards. At the same time, open benchmark challenges and field-validation competitions could provide mechanisms through which universities, technology companies, and independent research teams test competing models against common datasets and real-world events. Carefully designed incentive programs for successful prospective prediction could further encourage rapid methodological innovation, provided that performance is evaluated using transparent and scientifically rigorous criteria.

Finally, technological innovation must be accompanied by an appropriate governance and accountability framework. A scientifically informed responsibility-assessment system should clearly distinguish the roles and obligations associated with general regional surveillance, specialized site monitoring, forecasting, and emergency decision-making. Importantly, regulatory mechanisms should differentiate unavoidable missed detections arising from recognized technical limitations from failures attributable to inadequate implementation, misuse of available information, or professional negligence. Without such distinctions, excessive liability may discourage experimentation and the deployment of emerging technologies in precisely the high-uncertainty environments where innovation is most urgently needed. A balanced framework—one that preserves accountability while recognizing scientific uncertainty—will be essential for transforming geohazard monitoring from a reactive, site-based practice into an adaptive and anticipatory system capable of responding to an era of increasingly extreme and spatially unpredictable hazards.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. We acknowledge the use of AI-based tools for assistance with grammar and language editing.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Hongzhi Cui, Shaoxing University, China

Reviewed by: Shi-lin Luo, Changsha University, China

Tong Zhang, International University of Catalonia, Spain

Thawatchai Chalermpornchai, National Central University, Taiwan