Flood hazard mapping in Jamaica using principal component analysis and logistic regression

TLDR: In this article, a GIS-based multi-criteria statistical methodology was developed to quantify hazard potential and to map flood characteristics on the island of Jamaica, where 14 factors potentially responsible for flooding were identified and used as initial input in a hybrid model that combined principal component analysis with logistic regression and frequency distribution analysis.

Abstract: Jamaica, the third largest island in the Caribbean, has been affected significantly by flooding and flood-related damage. Hence assessing the probability of flooding and susceptibility of a place to flood hazard has become a vital part of planning and development. In addition to heavy rainfall from tropical storms and Atlantic hurricanes, several terrestrial factors play significant roles in flooding, including local geology, geomorphology, hydrology and land-use. In this study, a GIS-based multi-criteria statistical methodology was developed to quantify hazard potential and to map flood characteristics. Fourteen factors potentially responsible for flooding were identified and used as initial input in a hybrid model that combined principal component analysis with logistic regression and frequency distribution analysis. Of these factors, seven explained 65 % of the variation in the data: elevation, slope angle, slope aspect, flow accumulation, a topographic wetness index, proximity to a stream network, and hydro-stratigraphic units. These were used to prepare the island’s first map of flood hazard potential. Hazard potential was classified from very low to very high, nearly one-fifth (19.4 %) of the island was included within high or very high flood hazard zones. Further analysis revealed that areas prone to flooding are often low-lying and flat, or have shallow north- or northwest-facing slopes, are in close proximity to the stream network, and are situated on underlying impermeable lithology. The multi-criteria hybrid approach developed could classify 86.8 % of flood events correctly and produced a satisfactory validation result based on the receiver operating characteristic curve. The statistical method can be easily repeated and refined upon the availability of additional or higher quality data such as a high resolution digital elevation model. Additionally, the approach used in this study can be adopted to evaluate flood hazard in countries with similar characteristics, landscapes and climatic conditions, such as other Caribbean or Pacific Small Island Developing States.