PREDICTING DROUGHT USING ARTIFICIAL INTELLIGENCE TECHNIQUES

The research dealt with climate drought, its types, the causes that lead to drought, and how to reduce drought. Drought in the city of Mosul, northern Iraq, was also analyzed and studied by obtaining monthly rainfall data for the Mosul climate station located within the study area for the period from 1981-2018.

Previous studies and research conducted to study drought in different regions of the world were taken into consideration. Most of these studies and researches use drought indices, which are among the most widely used indices in estimating the amounts of deficit, their use, severity, and their impact on the water balance. The standard rainfall index (SPI) is one of the most widely used indices in estimating climate drought. (SPI) is characterized by many characteristics that distinguish it from other indicators. The standard rainfall index (SPI) technique was used in analyzing rain records. The analysis principle using the standard rainfall index is statistically based on the principle of converting the gamma distribution of the data series to the normal distribution. Positive SPI values mean that there is an increase in rainfall above the average rainfall, i.e. wet years, while negative values mean that there is a decrease in rainfall below the average rainfall, i.e. dry years. SPI values for a period of 12 months were adopted in the analysis because they cover the annual rainfall amount falling on the station during a year. Using the MATLAB program, several networks were created and tested, and the network with the best performance was selected from among the networks. 30 annual rainfall values were used against the SIP values calculated using equations and the Excel program to train the neural network on the data. While the rainfall data for the remaining 8 years were used to verify the results of the neural network by comparing the results of the neural network with the actual values recorded at the measuring station. This network was able to obtain the index value by simply entering the annual rainfall value. By comparing the index value with the drought classification table, the drought class can be determined without resorting to the calculation method. The network with the 1-7-1 structure (input layer, hidden layer containing seven neurons, and output layer) with the TRAINLM training function and the LEARNGDM learning function gave the best performance, as the correlation coefficient between its results and the actual results (which were not included in the training) was equal to 0.99 and the square error rate was 0.014, meaning that the results of this network can be adopted for the purpose of calculating the standard rain index with high confidence in the outputs