Forecasting Floods in the River Basins of Odisha Using Machine Learning

Abstract

Frequent floods cause huge disruption and affect lakhs of people in Odisha, India. Every year, due to floods, many lives are lost, and thousands of people are evacuated to safer places. Frequent floods not only affect people but also damage crops and other infrastructure resulting in a huge economic loss. Therefore, to attenuate the losses and impact of floods, mitigation strategies such as flood forecasting are required. However, erratic weather conditions due to global warming make forecasting of floods strenuous. Further, traditional flood forecasting models rely on domain proficiency and are very complex. Alternately, machine learning models have been widely used for forecasting floods. These models take rainfall-runoff data into consideration for forecasting floods with small lead time. Along with rainfall-runoff data, other meteorological parameters such as temperature, evapotranspiration, crop evapotranspiration, cloud cover, wet day frequency, and vapor pressure can be used for forecasting floods with larger lead time. This paper focuses on using these meteorological parameters related to ten flood affected districts of Odisha to design ML-based flood forecasting models that aim to forecast floods with larger lead times. Experimental-based comparisons based on performance metrics such as accuracy, precision, recall, F-measure, and AUC-ROC showed that random forest-based flood forecasting model performed comparatively better than Naïve Bayes, logistic regression, k-nearest neighbor, artificial neural networks, and support vector machine.