Landslide Intelligence: The Evidence Edition

Section 01

Why Local Government Should Care

Section 02

The Technology Shift: From Static Susceptibility to Dynamic Spatial Intelligence

Early susceptibility work combined terrain, geology and historical incidents in GIS to classify an area as high, medium or low. That work remains foundational. Since roughly 2019, a convergence of technologies has made something considerably more dynamic possible.^[1]

A 2016 comparative study of support vector machines, random forests and logistic regression for shallow landslide prediction in Vietnam set a widely referenced benchmark.^[2] By 2017, multilayer perceptron networks were integrated with GIS at Himalayan scale.^[3] By 2021, deep learning was evaluated across Iran's full national inventory using satellite-derived topographic data, with accuracy comparable to or better than earlier statistical methods.^{[9], [10]}

01

GIS

The organizing environment. Combines terrain, geology, parcels, infrastructure, land use and operational data—increasingly where information is interpreted in context, not just displayed.

02

LiDAR

Detailed elevation revealing subtle slope features and drainage. Repeat acquisition enables landscape-change comparison—detecting movement before failure.

03

InSAR

Interferometric radar measures millimeter-scale surface deformation, identifying terrain that may be creeping long before catastrophic failure. [11]

04

UAVs

Drones supply ultra-high-resolution imagery at targeted sites—most valuable for detailed investigation at priority locations identified by other methods. [12]

05

Satellite Earth Observation

Repeat passes bring time-series capability, tracking whether conditions are changing. Sentinel and Landsat data are freely available.

06

Cloud Computing

Platforms such as Google Earth Engine allow analysis of massive Earth-observation datasets without local infrastructure, enabling regional-to-national processing. [1]

07

Artificial Intelligence

Machine learning associates combinations of environmental conditions with previous failures. Random forest, deep learning and ensemble methods dominate recent high-citation research. [7]

Diagram 1 — From Map to Intelligence

1. Traditional: terrain + geology + historic incidents → GIS analysis → susceptibility map
   ↓
2. Emerging: terrain + Earth observation + rainfall + land use + infrastructure + history → GIS + cloud + AI → changing risk, exposure and consequence
   ↓
3. Applied: inspection · planning · mitigation · emergency decisions

AI does not replace GIS. It expands what GIS can help government understand.

Section 03

From Hazard to Consequence: A Better Map Is Useful, a Better Decision Is More Valuable

A highly susceptible slope in an isolated area may need monitoring and periodic inspection. A similar slope above a major evacuation route, water-treatment facility or dense neighborhood carries entirely different operational significance.

A multi-hazard suitability study integrating susceptibility with flood risk and development pressure demonstrated that consequence mapping changes which locations agencies should prioritize.^[4] More recent quantitative vulnerability modeling extends analysis to individual structures, estimating building damage under extreme rainfall—specificity that supports capital planning, insurance decisions and development review in ways a terrain-based map cannot.^[13]

Diagram 2 — The Local-Government Risk Chain

1. Hazard — where is terrain susceptible? Geology, slope, rainfall, soils, historical incidents.
   ↓
2. Exposure — what people, assets and infrastructure are present? Roads, utilities, facilities, population, development.
   ↓
3. Consequence — what happens if failure occurs? Route loss, service disruption, isolation, damage, cascading effects.
   ↓
4. Priority — where should government focus first? High consequence plus high susceptibility plus changing conditions.
   ↓
5. Action — inspect, engineer, maintain, mitigate, plan, prepare, with clear ownership and decision triggers.

GIS is moving from describing risk to helping agencies prioritize what to investigate, protect and monitor.

GIS is most powerful at the Exposure step. Overlaying susceptibility with what the agency actually manages—roads, bridges, water and sewer lines, utilities, facilities, parcels, structures, evacuation corridors—converts geological analysis into an operational tool. The review of 2,000+ articles confirms this integration as the defining trend of the decade.^[1]

Section 04

What Each Department Asks

Section 05

The Geospatial-AI Maturity Model: You Do Not Need to Start With AI

Sophisticated technology is not the starting point. A jurisdiction with incomplete inventories, fragmented asset information and inconsistent terrain data gains more from improving its GIS foundation than from deploying an advanced predictive model. The path forward is incremental.

An organization cannot become meaningfully predictive until it first becomes connected.

Level 1

Map — know where susceptibility exists

Elevation, slope, geology, soils, historical landslides and basic susceptibility mapping. The starting point for every organization—and still missing or outdated in many jurisdictions.

Where are our potentially unstable areas?

Level 2

Connect — relate the hazard to what government manages

Add roads, bridges, water and sewer infrastructure, utilities, parcels, buildings, critical facilities and population. Even this foundational overlay reveals priorities that were not previously obvious.

What could be affected?

Level 3

Prioritize — move from hazard to consequence

Layer in infrastructure criticality, emergency route designation, service-disruption consequences, maintenance history, replacement cost and development pressure. Research confirms this step changes which locations agencies act on first. [4], [13]

Where should we focus first?

Level 4

Observe — add change through time

Repeat imagery, LiDAR comparison, UAV surveys, InSAR deformation analysis, rainfall monitoring and land-cover change detection. Sentinel-1 InSAR has been used to identify and classify active slow-moving landslides across regional extents. [11] Multi-sensor studies have tracked the transition from slow creep to rapid failure. [12]

What appears to be changing?

Level 5

Predict — introduce analytics and AI

Machine learning, automated change detection, spatio-temporal analysis and predictive risk scoring. Hyperparameter-optimized random forest models perform strongly from remote sensing and topographic inputs. [7] Ensemble and deep learning methods extend prediction to national scale. [9], [10] The operational question is whether models can be validated locally and explained to the staff who must act on them.

Where could risk be increasing?

Level 6

Operationalize — connect risk intelligence to workflows

Inspection scheduling, work orders, engineering investigations, planning review, capital prioritization, preparedness plans, alerts and coordinated response. Geospatial intelligence improves inspection efficiency—helping agencies determine where to look rather than inspecting every slope equally. [8]

What should we do?

The objective is not to reach Level 6 as quickly as possible. It is to move to the level that produces better decisions for your organization.

Section 06

AI, Trust and Governance: Predictive Does Not Automatically Mean Trustworthy

Section 07

A Practical Action Plan: Five Things to Do Now

Section 08

Questions for Local-Government Leaders

Conclusion: From Mapping Risk to Managing Risk

References

Annotated Bibliography

These 13 sources form the primary research basis for this white paper. Each entry includes a full citation and an annotation explaining its relevance to local-government GIS practice.

1. [1] Primary Review

   Sri, M., Daud, Y., & Muhammad, K. (2026). Mapping global trends in landslide susceptibility research using geospatial technologies. Environmental Earth Sciences, 85, 367.

   The foundational source for all statistical claims and trend descriptions here. A bibliometric analysis of 2,002 Scopus-indexed English-language journal articles (2015–2025) using VOSviewer network analysis. It identified five thematic clusters—remote sensing and geospatial automation; machine learning modeling; geotechnics and slope dynamics; multi-hazard and climate adaptation; GIS-based empirical approaches—and documented the post-2019 acceleration in publication volume, with China, India, Italy and the United States dominant. Its core contribution, describing the shift from conventional GIS mapping toward integrated Geospatial-AI infrastructure, is the conceptual frame this paper applies to local government practice.

2. [2] Machine Learning Benchmark

   Bui, D.T., Tuan, T.A., Klempe, H., Pradhan, B., & Revhaug, I. (2016). Spatial prediction models for shallow landslide hazards: a comparative assessment of support vector machines, artificial neural networks, kernel logistic regression, and logistic model tree. Landslides, 13, 361–378.

   The most-cited article in the dataset (1,123 citations). Established the benchmark for applying support vector machines, random forests and logistic regression to shallow landslide prediction in Vietnam, showing that ensemble and ML approaches outperformed traditional statistical susceptibility models. Referenced here as evidence that the move from mapping to prediction rests on a decade of empirical validation—and, in the governance section, that models must still be validated against local conditions.

3. [3] Neural Network + GIS Integration

   Pham, B.T., Bui, D.T., Prakash, I., & Dholakia, M.B. (2017). Hybrid integration of multilayer perceptron neural networks and machine learning ensembles for landslide susceptibility assessment at Himalayan area (India) using GIS. Catena, 149, 52–63.

   Second-most-cited article (549 citations). Integrated multilayer perceptron neural networks with GIS at Himalayan scale, establishing that neural approaches inside GIS workflows achieve strong accuracy across large, complex terrain. Supports the argument that AI and GIS are complementary rather than competing frameworks.

4. [4] Multi-Hazard + Urban Consequence

   Bathrellos, G.D., Skilodimou, H.D., Chousianitis, K., Youssef, A.M., & Pradhan, B. (2017). Suitability estimation for urban development using landslide susceptibility mapping and multi-hazard assessment map. Science of the Total Environment, 575, 119–134.

   Fifth-most-cited article (428 citations). Combined landslide susceptibility with flood risk and urban development pressure for planning decisions in Greece. The clearest peer-reviewed demonstration that consequence-based mapping produces different—and more actionable—priority rankings than terrain-based susceptibility alone.

5. [5] GIS Machine Learning — Ensemble

   Chen, W., Peng, J., Hong, H., Shahabi, H., Pradhan, B., Liu, J., Zhu, A.X., Pei, X., & Duan, Z. (2018). Landslide susceptibility modelling using GIS-based machine learning techniques for Chongren County, Jiangxi Province, China. Science of the Total Environment, 626, 1121–1135.

   Sixth-most-cited article (393 citations). Confirmed that ensemble methods combining multiple algorithms consistently outperform any single model. Useful context when evaluating vendor or research-partner models: require multi-model approaches and documented performance metrics tied to local validation.

6. [6] Training Data & Model Governance

   Kalantar, B., Pradhan, B., Naghibi, S.A., Motevalli, A., & Mansor, S. (2018). Assessment of the effects of training data selection on landslide susceptibility mapping: a comparison between SVM, LR and ANN. Geomatics, Natural Hazards and Risk, 9(1), 49–69.

   Third-most-cited article (454 citations). Showed that how training samples are selected—not just which algorithm is used—materially changes the resulting susceptibility classification. The strongest support for the governance questions on training data and local validation; any AI-based susceptibility product should document sources, sampling methodology and local validation results before operational deployment.

7. [7] Random Forest + Hyperparameter Optimization

   Sun, D., Wen, H., Wang, D., & Xu, J. (2020). A random forest model of landslide susceptibility mapping based on hyperparameter optimization using Bayes algorithm. Geomorphology, 362, 107201.

   Seventh-most-cited article (381 citations). Bayesian-tuned random forest models substantially outperform default settings using remote sensing and topographic inputs—evidence that the predictive step has matured beyond experimentation. Relevant to agencies weighing off-the-shelf against locally calibrated tools.

8. [8] Prediction, Monitoring & Early Warning

   Chae, B.G., Park, H.J., Catani, F., Simoni, A., & Berti, M. (2017). Landslide prediction, monitoring and early warning: a concise review of state-of-the-art. Geosciences Journal, 21(6), 1033–1070.

   Ninth-most-cited article (320 citations). Synthesizes instrumentation, remote sensing, geotechnical monitoring and alert-threshold methods. The operational takeaway: pairing susceptibility analysis with targeted monitoring at priority sites improves inspection efficiency—agencies that know where to look allocate limited field capacity far better than those inspecting uniformly.

9. [9] Deep Learning at National Scale

   Ngo, P.T.T., Panahi, M., Khosravi, K., Ghorbanzadeh, O., Kariminejad, N., Cerda, A., & Lee, S. (2021). Evaluation of deep learning algorithms for national scale landslide susceptibility mapping of Iran. Geoscience Frontiers, 12(2), 505–519.

   Eighth-most-cited article (333 citations). Deep learning applied to a full national landslide inventory with strong accuracy—susceptibility modeling is no longer confined to small study areas. State and federal regional products may increasingly use these methods, so local GIS teams need to know how to validate and apply them.

10. [10] Deep Learning + Satellite Topographic Data

    Azarafza, M., Azarafza, M., Akgun, H., Atkinson, P.M., & Derakhshani, R. (2021). Deep learning-based landslide susceptibility mapping using satellite data and topographic analysis. Scientific Reports, 11, 24112.

    Tenth-most-cited article (312 citations). Open-access demonstration that satellite-plus-AI susceptibility mapping is validated, published methodology rather than experiment—directly usable by agency staff preparing internal briefings or procurement specifications.

11. [11] InSAR Surface Deformation — Regional

    Meng, Q., Confuorto, P., Peng, Y., Raspini, F., Bianchini, S., Han, S., et al. (2020). Regional recognition and classification of active loess landslides using two-dimensional deformation derived from Sentinel-1 interferometric radar data. Remote Sensing, 12(10), 1541.

    Shows that freely available Sentinel-1 InSAR data can detect, map and classify active slow-moving landslides across regional extents—movement invisible to conventional inspection or optical imagery. InSAR monitoring is accessible: agencies can partner with universities or state programs for processed products as part of an Observe-level program.

12. [12] Multi-Sensor — Slow to Rapid Failure

    Handwerger, A.L., Lacroix, P., Bell, A., Booth, A.M., Huang, M., Mudd, S.M., et al. (2025). Multi-sensor remote sensing captures geometry and slow-to-fast sliding transition of the 2017 Mud Creek landslide. Scientific Reports, 15(1).

    Combining satellite imagery, InSAR and ground sensors tracked the transition from slow creep to rapid failure at a California site—precisely where early warning has the greatest potential to protect life and infrastructure. Supports prioritizing sites with documented prior movement for multi-sensor monitoring.

13. [13] Building-Level Vulnerability

    Li, G., Liu, D., Ruan, M., Zhang, Y., He, J., Guo, Z., et al. (2025). Quantitative vulnerability assessment of buildings exposed to landslides under extreme rainfall scenarios. Buildings, 15(11), 1838.

    Extends analysis from terrain susceptibility to quantitative damage estimation at the individual building level, showing that susceptibility alone understates consequence in dense or high-value built environments. Directly applicable to capital prioritization, development review and conversations with owners, insurers and elected officials.