Three-step neural network approach for predicting monsoon flood preparedness and adaptation: Application in urban communities of Lahore, Pakistan

Abstract

Climate change has caused aberrations in precipitation patterns globally. The increasing variations in heavy monsoon rains in South Asia have been primarily linked to this human-induced phenomenon. Torrential rains in the monsoon season cause regular flash flooding in many urban areas of Pakistan. Poor infrastructure, weak governance, and lack of corrective disaster risk reduction have exacerbated climate risk and vulnerabilities. Thus, it is pertinent to undertake preparedness and adaptation measures to safeguard lives and reduce economic damages. Numerous methodologies have been developed to identify factors affecting such actions. This study proposes a novel methodology by integrating the statistical and artificial intelligence techniques to identify determinants of disaster preparedness and climate change adaptation, i.e., the three-step neural approach. The methodology was tested in two monsoon-affected areas of Lahore metropolitan, Pakistan. Using the Yamane sampling method, 400 samples were collected through household questionnaires. Data regarding monsoon flood risk perception, psychological distance to climate change, preparedness, and adaptation measures were collected. Regression analysis was used to shortlist the influential socioeconomic indicators. A multi-layer perceptron neural network was then used to reconfirm the influence of these indicators. The prediction testing shows the high accuracy of this approach. The proposed methodology was found robust and operational for successfully predicting the socioeconomic determinants. It can be easily modified and streamlined for testing in the context of other natural hazards and regions.