A Comprehensive Assessment Framework for Drought Risk in Taiwan Using a Combined ANP-ANN Approach

Drought is a natural hazard that has serious impacts on the environment and human society including agricultural, industrial, and domestic sectors, especially in the era of climate change. For Taiwan, drought poses a challenge particularly to the water-intensive semiconductor manufacturing industry. Comprehensive assessment is therefore necessary to identify key regions and sectors with high risk. This study utilized a combination of Analytic Network Process (ANP) and Artificial Neural Network (ANN) in an ensemble learning method to evaluate and map drought risk in Taiwan. ANP constructs a network and assigns weights to indicators while the ANN model uses these indicators to predict drought risk classes. Twenty indicators were selected representing socio-economic and environmental factors which are categorized into hazard, exposure, and vulnerability components for risk assessment. The environmental condition during the 2021 spring drought was selected to represent the drought hazard in Taiwan. The trained ANN model showed effective prediction of drought risk as indicated by performance metrics of accuracy, precision, recall, F1 score, and Kappa Index with values 0.940, 0.946, 0.938, 0.942, and 0.923, respectively. The final drought risk map was validated through fieldwork and independent statistical data. Overall accuracy values ranging 0.717-0.851 by comparing drought risk classes with indicators related to damaged crops, converted damage areas, and estimated product losses. The prediction and validation results highlight the reliability of the framework for rapid and accurate risk assessment. The framework can be applied to different natural and socioeconomic backgrounds for effective drought management to inform future long-term adaptation strategies.