Application of several data-driven techniques to predict a standardized precipitation index

Bahram Choubin, Arash Malekian, Mohammad Gloshan


Climate modeling and prediction is important in water resources management, especially in arid and semi-arid regions that frequently suffer further from water shortages. The Maharlu-Bakhtegan basin, with an area of 31 000 km2 is a semi-arid and arid region located in southwestern Iran. Therefore, precipitation and water shortage in this area have many problems. This study presents a drought index modeling approach based on large-scale climate indices by using the adaptive neuro-fuzzy inference system (ANFIS), the M5P model tree and the multilayer perceptron (MLP). First, most of the climate signals were determined from 25 climate signals using factor analysis, and subsequently, the standardized precipitation index (SPI) was predicted one to 12 months in advance with ANFIS, the M5P model tree and MLP. The evaluation of the models performance by error parameters and Taylor diagrams demonstrated that performance of the MLP is better than the other models. The results also revealed that the accuracy of prediction increased considerably by using climate indices of the previous month (t – 1) (RMSE = 0.802, ME = –0.002 and PBIAS = –0.47).


Standardized precipitation index (SPI); climate signals; multi-layer perceptron (MLP); adaptive neuro-fuzzy inference system (ANFIS); M5P model tree; Taylor diagrams

Full Text: