Drought prediction using co-active neuro-fuzzy inference system, validation, and uncertainty analysis (case study: Birjand, Iran)

详细信息    查看全文

• 作者：Hadi Memarian ; Mohsen Pourreza Bilondi ; Majid Rezaei
• 刊名：Theoretical and Applied Climatology
• 出版年：2016
• 出版时间：August 2016
• 年：2016
• 卷：125
• 期：3-4
• 页码：541-554
• 全文大小：5,392 KB
• 刊物类别：Earth and Environmental Science
• 刊物主题：Earth sciences
  Meteorology and Climatology
  Atmospheric Protection, Air Quality Control and Air Pollution
  Climate Change
  Waste Water Technology, Water Pollution Control, Water Management and Aquatic Pollution
  
• 出版者：Springer Wien
• ISSN：1434-4483
• 卷排序：125

文摘

This work aims to assess the capability of co-active neuro-fuzzy inference system (CANFIS) for drought forecasting of Birjand, Iran through the combination of global climatic signals with rainfall and lagged values of Standardized Precipitation Index (SPI) index. Using stepwise regression and correlation analyses, the signals NINO 1 + 2, NINO 3, Multivariate Enso Index, Tropical Southern Atlantic index, Atlantic Multi-decadal Oscillation index, and NINO 3.4 were recognized as the effective signals on the drought event in Birjand. Based on the results from stepwise regression analysis and regarding the processor limitations, eight models were extracted for further processing by CANFIS. The metrics P-factor and D-factor were utilized for uncertainty analysis, based on the sequential uncertainty fitting algorithm. Sensitivity analysis showed that for all models, NINO indices and rainfall variable had the largest impact on network performance. In model 4 (as the model with the lowest error during training and testing processes), NINO 1 + 2(t-5) with an average sensitivity of 0.7 showed the highest impact on network performance. Next, the variables rainfall, NINO 1 + 2(t), and NINO 3(t-6) with the average sensitivity of 0.59, 0.28, and 0.28, respectively, could have the highest effect on network performance. The findings based on network performance metrics indicated that the global indices with a time lag represented a better correlation with El Niño Southern Oscillation (ENSO). Uncertainty analysis of the model 4 demonstrated that 68 % of the observed data were bracketed by the 95PPU and D-Factor value (0.79) was also within a reasonable range. Therefore, the fourth model with a combination of the input variables NINO 1 + 2 (with 5 months of lag and without any lag), monthly rainfall, and NINO 3 (with 6 months of lag) and correlation coefficient of 0.903 (between observed and simulated SPI) was selected as the most accurate model for drought forecasting using CANFIS in the climatic region of Birjand.