International Transaction Journal of Engineering,
Management, & Applied Sciences & Technologies

FEATURE PEER-REVIEWED ARTICLE

• AN INVESTIGATION OF CAPABILITY OF ARTIFICIAL INTELLIGENCE APPROACH IN DROUGHT FORECAST

  

  Uruya Weesakul, Shekhar Mahat (Department of Civil Engineering, Thammasat School of Engineering, Thammasat University, THAILAND).
  

  Disciplinary: Water Resources Engineering.

  Full text

  DOI: 10.14456/ITJEMAST.2020.179

  Keywords: Deep neural network; Rainfall forecasting; Large-scale atmospheric variables; Drought forecasting modelling; Klong Yai river basin; Thailand.

  Abstract
  In Thailand, droughts frequently occur during the last 5 decades with increases in frequency occurrence and their impact on the agricultural sector as well as human economic activities, in recent decades due to climate change. It is necessary to develop drought forecasting models for efficient water resources management. Rare drought forecast models, with high accuracy, are developed in Thailand. The study aims to investigate the capability of an Artificial Intelligence Approach in drought forecasting by using Deep Neural Network for monthly rainfall forecast. Eastern river basin in Thailand was selected as a case study, due to its frequent occurrence of drought. Monthly rainfall from Plauk Daeng station during 1991-2016 was collected for analysis of drought situations. Drought in the study is defined as low rainfall than normal conditions leading to insufficient water to meet normal needs. The percentile range method was adopted in the study for drought identification when monthly rainfall is lower than the 40th percentile then drought condition is classified. Monthly rainfall during 1991-2010 was used for the training process of the DNN model while monthly rainfall during 2011-2016 was used for the validation process. Drought forecast during the validation process reveals that Artificial Intelligence Approach can predict drought situations with acceptable accuracy for only three months ahead with 60% accuracy. When the leading time of the forecast increases to be 6 months, the accuracy of the forecast decreases to be only 55%. Further study to improve model performance will be conducted using the Input Selection Technique so that the model is applicable in engineering practice.
  
  Paper ID: 11A9Q