S. BHALAI – Landslide susceptibility of Portland, Jamaica
subdivided into various classes or ranges to understand the correlation with landslide activity and further detail the analysis. Elevation was split into ranges of 100 m (Figure 9). Slope gradient was grouped into 10° ranges (Figure 10) and slope aspect according to the eight major cardinal directions (Figure 11). Lithology was grouped into thirteen geotechnical classes (Figure 12; Table 1). Distances of 50 m-ranges were outlined from major geological faults. The highest resolution permitted by the datasets was utilized, ensuring no compromise to the true natural characteristics of the features. This was dependent on the scale of data collection.
Figure 8. Monochromatic aerial photograph of the Unity Valley Landslide in the Rio Grande Valley, south-east of Port Antonio. Note the broad flat area in the bottom right hand corner of the image where a landslide dam developed. The landslide is approximately 2 km long (east-west). Long arrow shows flow direction. [1953–54 Aerial Photography by Hunting Aerosurveys Ltd.]
predisposing factors are also necessary, that is, topographical and geological features. Elevation and slope characteristics were derived from 1:12,500 topographic survey maps published in the 1960s. Geological information was extracted from the 1:50,000 geological maps published in the 1990s (Mines and Geology Division, 1997a-d).
coverages of elevation, slope gradient and slope aspect were generated from the topographic maps. In the case of slope coverage the topographic maps were limited in the areas disturbed by older landslides. For landslide areas older than the maps, it is the landslide morphology that defines contour alignment. However, it is necessary to investigate which slope gradients were originally causative. In the areas of older landslides the topographic maps show the final disturbed slope, not the original pre-failure gradient. To overcome this limitation the computer algorithm for generating the slope gradient map was modified in order to carefully regenerate the natural slope before failure. Analysis was conducted using this regenerated topography.
Coverages of rock types, grouped according to similar geotechnical behaviour, and distances from geological faults were also generated. Geotechnical behaviour was determined based on the classification of O’hara and Bryce (1983) confirmed by observations of the surveyor. Geological faults were obtained from the published geological maps. Distance from faults was easily calculated using computer algorithms.
The coverages of the causative factors were each
From the landslide inventory the landslide source areas were extracted for use in the analytical step. The source areas were then unevenly split where 75% was used for analysis and the remaining 25% was reserved for testing the final model. Bivariate statistical methods were used in the first stage of the susceptibility analysis to identify correlations between the predisposing factors and the landslide sources, and to generate a numerical value (numerator of equation 1) indicative of the influence of landslides on the different classes of the predisposing factors. This bivariate statistical approach was introduced by Brabb et al. (1972) where landslide distribution was combined with a slope map and a lithology map to derive the landslide susceptibility for part of California. In the past the technique of overlaying maps was cumbersome but today Geographic Information System software conveniently and efficiently facilitates this step.
The numerical values derived for each causative factor were used in equation 1 to compute the respective weighting value or Susceptibility index.
Where: A1 = area of landslide sources within class of
predisposing factor A2 = area of class of predisposing factor A3 = total area of landslide sources in parish A4 = area of parish
The calculated Susceptibility indices were further applied to the respective coverage of each causative factor as weightings. The weighted maps were all summed to generate the final susceptibility model.
The susceptibility model generated from the summation of maps of all the predisposing factors comprised a wide range of values. These were grouped into five zones of susceptibility, where low