Forecasting the Reconnaissance Drought Index (RDI) Using Artificial Neural Networks (ANNs)

Abstract

The study of drought and its forecasting plays an important role in planning and managing water resource systems, especially in extreme climatic periods. This study aims to analyze and forecast drought characteristics, through the use of the Reconnaissance Drought Index (RDI) in order to analyze temporal and spatiotemporal climatic drought in nine climate stations in the Kurdistan Region of Iraq for the period (1973-2020) to detect the beginning and end of the drought period, as well as forecasting future droughts using two types of artificial neural networks: Recursive Multi-Step Neural Networks (RMSNN) and Direct Multi-Step Neural Network (DMSNN). The results revealed that the driest years were in the years (1998-99) for Amadiyah, Erbil and Sulaymaniyah stations, and the years (2007-08) for the rest of the stations in the study area. Moreover, the results of the two models depending on the simulation methods adopted have shown the ability of these models with regard to the forecasting for the last six years, and the ability of both models to forecast with an increase in the amount of error as we go forward. However, the (DMSNN) model was more accurate, as shown by the results of the statistical tests.