The study focused on Pithoragarh, the easternmost administrative district of Uttarakhand, bounded by latitudes 29 degrees 21 minutes and 30 degrees 40 minutes north and longitudes 79 degrees 50 minutes and 81 degrees 15 minutes east. Elevation ranges from 416 metres in deeply eroded valleys to 7,378 metres at glaciated alpine peaks, creating an exceptionally broad range of climatic, geomorphic and geological conditions. The authors noted that the district lies in seismic zone V and has more than 1,000 recorded mass-movement events over nearly two decades. The maps are susceptibility assessments rather than forecasts of individual landslides: they estimate where terrain conditions have historically coincided with past failures. The research does not establish that any mapped location will fail at a specified time, nor does it directly test the models against future events.

To construct the models, the researchers assembled a spatial database containing 14 conditioning factors. A 30-metre-resolution digital elevation model derived from Shuttle Radar Topography Mission data was used to map elevation, slope, contours and drainage characteristics. Land use and land cover were classified from Sentinel-2 imagery using a supervised maximum-likelihood method. Geology, geomorphology, lithology, soil, earthquake locations, faults, lineaments and distance from roads were obtained from the Geological Survey of India’s Bhukosh portal, while rainfall information came from the India Meteorological Department. Stream-distance layers were also generated with Euclidean-distance analysis in ArcMap. The two susceptibility approaches were implemented in ArcGIS and evaluated by plotting known landslide locations against their predicted hazard zones.

The fuzzy-logic model represented the study’s knowledge-driven approach. Each thematic layer was assigned a membership or risk rating through a 14-by-14 pairwise comparison matrix based on Saaty’s analytical hierarchy process. A scale from 1 to 9 expressed the relative importance of the factors, after which the matrix was normalised and a principal eigenvector was calculated to derive percentage weights. Rainfall intensity received the largest weight at 23.1%. Earthquakes and faults each received 8.9%, geology 7.2%, lineaments 6.7%, distance from roads and distance from rivers 6.5% each, soil 5.5%, slope 5.3%, land use and land cover and geomorphology 4.9% each, contours 4.1%, lithology 4.0%, and elevation 3.4%. The consistency ratio was 0.064, below the study’s stated maximum threshold of 0.10 for the pairwise matrix. The resulting fuzzy overlay classified 20.5% of the district as high hazard, 25.6% as moderate hazard and 53.9% as low hazard.

The frequency-ratio method instead calculated an empirical association between each class of a conditioning factor and the historical landslide inventory. A ratio of one represents a proportional distribution of landslides relative to the area occupied by that class; values above one indicate more landslides than expected from area alone, while values below one indicate fewer. These class-specific ratios were combined into a Landslide Susceptibility Index and then classified into susceptibility zones using natural breaks. Several reported ratios illustrate the spatial pattern captured by the data. The 0–100 metre band around streams had a frequency ratio of 3.08, observed faults had 2.49, observed thrusts 3.50, and the highest earthquake-magnitude class had 2.43. Lateritic soil had a ratio of 7.71, while unstable anthropogenic or saturated geomorphic terrain reached 6.65. In contrast, several classes containing no inventory points had ratios of zero, illustrating how strongly the empirical model depends on the distribution and completeness of recorded landslides.

Historical events were distributed unevenly across the two classifications. The fuzzy-logic map placed 31 landslides in the very-low zone, 30 in the low zone, 120 in the moderate zone, 73 in the high zone and 89 in the very-high zone. Under the frequency-ratio map, the corresponding counts were 8, 11, 98, 109 and 117. The ROC analysis expresses how well a model separates mapped landslide points from non-landslide locations. Its AUC can also be interpreted as the probability that a randomly selected landslide point receives a higher susceptibility value than a randomly selected non-landslide point. The frequency-ratio result was therefore substantially stronger for this inventory-based comparison, although the study did not present a formal uncertainty interval around either AUC.

Spatial analysis identified a pronounced north–south contrast. The central and southern parts of Pithoragarh contain extensive unconsolidated and weathered soils, lateritic terrain, active river and stream corridors, dense fault and thrust networks, and areas of high rainfall intensity. These characteristics coincided with the largest concentration of high and very-high susceptibility zones. The model’s empirical layer-by-layer associations were particularly strong for lateritic soil, landforms described as unstable or heavily saturated, built-up areas and terrain close to major lineaments. The northern part was generally less susceptible in the model, especially where resistant crystalline rocks, including amphibolite and gneiss, dominate. The authors emphasised that geological age alone does not determine stability: older Proterozoic masses have been intensely fractured and sheared by tectonic deformation, while some younger Tertiary units are poorly lithified and friable.

Field observations supplied qualitative support for the mapped patterns. The researchers reported that debris slides near Munsyari and active rock fall along a national highway were located within high-hazard corridors produced by the models. Photographic documentation cannot establish the predictive accuracy of a full regional model, but the reported locations are consistent with the frequency-ratio map’s correspondence between historical landslides and high-susceptibility terrain. The mapped patterns also fit the broader physical context. Intense rainfall infiltrates soil and structural discontinuities, raises pore-water pressure and reduces effective shear strength, while faults, thrusts and lineaments can create fractured or permeable zones. Road construction and slope modification can further disturb unstable ground, although the study used distance from roads as a spatial proxy for disturbance rather than a direct measurement of causal effects.

The findings are nevertheless constrained by the design of the inventory and the regional setting of the analysis. The authors used the complete historical landslide inventory for validation because the knowledge-driven and semi-quantitative models were not trained on a machine-learning partition, meaning the same historical evidence informed the frequency-ratio construction and its performance assessment. The district’s extreme elevation range also complicates any inference about what caused a particular failure. Satellite detection may miss older scars that have disappeared from imagery, while inventory records can be biased toward failures near roads and settlements and away from remote terrain. The study itself identifies cross-region testing as an important next step: evaluating models developed in Pithoragarh in entirely different geographic areas. Within these limits, the research identifies the frequency-ratio map as the better validated of the two approaches for Pithoragarh and presents it as a regional decision-support tool for identifying where landslide susceptibility is concentrated. It does not claim that road distance, rainfall, soil type or any other mapped factor alone causes failure.

Subject of Research: Comparison of frequency-ratio and fuzzy-logic models for regional landslide susceptibility mapping in Pithoragarh, Uttarakhand

Article Title: Fuzzy logic and frequency ratio based landslide susceptibility mapping using geo environmental and hydro meteorological factors in pithoragarh, uttarakhand

Article References: Pal, R., Akhas, P. K., & Porchelvan, P. (2026). Fuzzy logic and frequency ratio based landslide susceptibility mapping using geo environmental and hydro meteorological factors in pithoragarh, uttarakhand. Frontiers in Remote Sensing, 7, Article 1944229. https://doi.org/10.3389/frsen.2026.1944229

Image Credits: AI Generated

DOI: 10.3389/frsen.2026.1944229

Keywords: landslide susceptibility, frequency ratio, fuzzy logic, GIS mapping, Pithoragarh, Himalayan hazards, monsoon rainfall, ROC-AUC validation, remote sensing, disaster risk, Fuzzy, logic