Monte Carlo testing identifies stable multi-hazard priorities in Kerala watershed

An uncertainty analysis of watershed prioritization in Kerala’s Pampa River basin found that five sub-watersheds—W10, W14, W7, W4, and W17—consistently remained among the highest priorities when the weights assigned to morphometric, hydrological, and hazard-related indicators were varied. The study combines terrain measurements derived from a 30 m digital elevation model, expert judgments organized through the Analytical Hierarchy Process, and 10,000 Monte Carlo simulations designed to test whether the resulting rankings withstand plausible changes in decision weights. The findings identify W4, W7, W10, W14, and W17 as the most stable candidates for priority attention, although the rankings represent relative physical susceptibility rather than direct estimates of disaster losses.

The researchers divided the approximately 2,093 km² Pampa basin into 31 sub-watersheds and evaluated 15 morphometric, hydrological, and hazard-related criteria. These included drainage-network properties, basin shape, relief, slope, ruggedness, overland-flow length, rainfall variability, and other indicators intended to describe the terrain’s response to water and the occurrence of floods, droughts, and landslides. A 30 m Shuttle Radar Topography Mission digital elevation model was processed in ArcGIS through sink filling, flow-direction and flow-accumulation analysis, stream extraction, stream ordering, and watershed delineation. Daily rainfall records from 2015–2025 were also interpolated in GIS to characterize long-term annual and seasonal variability across the basin.

Relative importance was assigned with AHP, which uses pairwise comparisons to derive criterion weights. Five specialists with more than 10 years of experience in watershed assessment and natural-hazard analysis independently compared the 15 parameters on Saaty’s nine-point scale. Their matrices were combined by geometric averaging, and the resulting consensus matrix was normalized to obtain weights. The consistency ratio was 0.08, below the study’s threshold of 0.10 for acceptable consistency. Each criterion was normalized to a dimensionless scale, then combined through a weighted linear equation to generate a priority score for every sub-watershed. Jenks natural-breaks classification was used to divide the scores into relative priority classes, but the numerical boundaries are specific to this basin and cannot be transferred unchanged to other watersheds.

The terrain analysis described a highly dissected basin with 560 identified stream segments and a Strahler stream-order system extending to sixth order. The drainage network included 405 first-order streams, 103 second-order streams, 37 third-order streams, 13 fourth-order streams, two fifth-order streams, and one sixth-order stream. Mean stream length ranged from 1.5 to 3.8 km, while the basin-wide bifurcation ratio averaged 2.38. Drainage density was below 5.0 km/km² and drainage texture was 8.10, placing the network in an intermediate texture class. Most Rho coefficient values were below 0.2, and more than 60 percent of the basin fell below 0.15, indicating relatively limited storage and a tendency toward rapid overland flow according to the study’s interpretation of the index.

Basin geometry also varied substantially. Pampa’s circularity ratio was 0.36, consistent with an elongated form, while its elongation ratio was 0.65. The highest terrain occurred in sub-watershed W31, and the lowest points appeared in W8 and W9. Sub-watersheds W5 and W18 had comparatively high ruggedness numbers, whereas W31 had a lower value. At the sub-basin scale, infiltration-number values were lowest in W31 and highest in W20, W21, and W24. These contrasts helped differentiate terrain controls on runoff, erosion, slope instability, and water retention, although the researchers cautioned that the AHP model represents these relationships through independent weighted criteria rather than fully coupled physical processes.

The uncertainty experiment began from the baseline AHP weight vector. In each of 10,000 iterations, all 15 weights were randomly perturbed by as much as plus or minus 10 percent using a uniform distribution, then renormalized to sum to one. Priority scores and rankings were recalculated for all 31 sub-watersheds after every perturbation. W10, W14, W7, W4, and W17 repeatedly occupied the highest ranks and were therefore classified as robust. W1, W2, and W22 showed moderate sensitivity. The simulation was designed specifically to examine variability in expert weights; its uniform perturbation range was not intended as a complete probability distribution of uncertainty, and it did not quantify errors in the elevation model, stream-extraction choices, spatial resolution, threshold selection, or model structure.

Household-level records were used as an independent check on the modeled priorities. District Disaster Management Authority data covered flood damage in 2018 and 2019, landslide damage in 2018, and drought impacts reported for the study area. About 75.0 percent of households affected by floods were located in sub-watersheds classified as moderate to very high priority. All 76 households affected by landslides were in moderate- or high-priority areas, while 67.9 percent of the 321 drought-affected households were in moderate- to very-high-priority classes. These results show concentration of reported impacts in relatively high-priority areas, but they do not amount to direct proof of physical susceptibility because household damage also reflects population density, settlement patterns, housing quality, land use, infrastructure, reporting practices, and community resilience.

The quantitative correlation also requires caution. Spearman’s rank correlation between the AHP priority scores and the number of flood-affected households was weak and statistically nonsignificant, with ρ = −0.011 and p = 0.952. This means the composite ranking did not closely track flood damage counts in individual sub-watersheds. The authors attributed the mismatch partly to the framework’s emphasis on intrinsic physical vulnerability, whereas damage records reflect hazard intensity together with exposure, socioeconomic conditions, infrastructure resilience, and post-disaster reporting. The household records were therefore treated as evidence of consistency and practical utility rather than as a direct proxy for susceptibility or a causal validation of the model.

The study’s principal contribution is therefore methodological as much as geographic. Integrating morphometric analysis with AHP creates a transparent way to combine terrain, drainage, rainfall, and hazard indicators when detailed process-model inputs are limited, while Monte Carlo analysis reveals whether the resulting priorities depend heavily on a particular set of expert weights. Even so, the framework remains a first-order regional screening tool. It does not simulate event-scale rainfall intensity, antecedent soil moisture, groundwater behavior, changing land use, cascading hazards, or the full mechanics of floods and slope failures. Transfer to other basins would require recalibrating the natural-breaks classes and examining alternative elevation products, raster resolutions, watershed boundaries, and preprocessing choices. The authors also identify population and infrastructure layers, dynamic land-cover data, climate and land-use scenarios, and process-based flood, drought, and slope-stability models as necessary additions if the approach is to evolve from relative susceptibility mapping into comprehensive risk assessment. The present results provide a reproducible screening tool for comparing 31 tropical monsoon watersheds while explicitly revealing which priorities remain stable and which depend on judgment-based weighting.

Subject of Research: Uncertainty-tested prioritization of 31 Kerala sub-watersheds for flood, drought, and landslide susceptibility

Article Title: Integrating morphometric analysis and AHP–Monte Carlo framework for multi-hazard risk reduction

Article References: Sekar, R., Natarajan, S., & Joseph, N. (2026). Integrating morphometric analysis and AHP–Monte Carlo framework for multi-hazard risk reduction. Natural Hazards, 122(22), Article 689. https://doi.org/10.1007/s11069-026-08468-0

Image Credits: AI Generated

DOI: 10.1007/s11069-026-08468-0

Keywords: multi-hazard susceptibility, watershed prioritization, Monte Carlo simulation, Analytical Hierarchy Process, morphometric analysis, Pampa River basin, Kerala, flood risk, drought risk, landslide risk, GIS, uncertainty analysis