پيش‌بيني خشكسالي و سطح بارندگي در ايران جهت مديريت منابع آب مبتني بر مدل‌هاي ماركوفي تركيبي

Drought and Precipitation level forecast in Iran for water source management based on Combinational Markov Models

ﺧﺸﮏ ﺳﺎﻟﯽ ﺑﻪ ﻋﻨﻮان ﯾﮑﯽ از ﻣﻬﻢ ﺗﺮﯾﻦ ﺑﻼﯾﺎي ﻃﺒﯿﻌﯽ اﺳﺖ ﮐﻪ ﻣﻤﮑﻦ اﺳﺖ در ﻫﺮ اﻗﻠﯿﻢ آب وﻫﻮاﯾﯽ اﺗﻔﺎق ﺑﯿﻔﺘﺪ. ازآﻧﺠﺎﮐﻪ وﻗﻮع ﺧﺸﮏ ﺳﺎﻟﯽ اﺟﺘﻨﺎب ﻧﺎﭘﺬﯾﺮ اﺳﺖ، ﺑﻨﺎﺑﺮاﯾﻦ ﺷﻨﺎﺧﺖ آن ﺑﻪ ﻣﻨﻈﻮر ﻣﺪﯾﺮﯾﺖ ﺑﻬﯿﻨﻪ ﻣﻨﺎﺑﻊ آب، از اﻫﻤﯿﺖ ﺑﺴﺰاﯾﯽ ﺑﺮﺧﻮردار اﺳﺖ. در اﯾﻦ ﺗﺤﻘﯿﻖ ﺑﻪ ﭘﯿﺶ ﺑﯿﻨﯽ ﺧﺸﮏ ﺳﺎﻟﯽ در ﺷﻬﺮ ﺗﻬﺮان و ﻣﺸﻬﺪ ﭘﺮداﺧﺘﻪ ﺷﺪه اﺳﺖ؛ ﺑﻪ اﯾﻦ ﻣﻨﻈﻮر از ﻣﺪل ﻫﺎي ﻣﺎرﮐﻮﻓﯽ در دو ﻣﺮﺣﻠﻪ اﺳﺘﻔﺎده ﺷﺪ. در ﻣﺮﺣﻠﻪ اول ﺑﻌﺪ از ﭘﯿﺶ ﭘﺮدازش داده ﻫﺎ، زﻧﺠﯿﺮه ﻣﺎرﮐﻮف ﺑﺮ اﺳﺎس وﯾﮋﮔﯽ ﻫﺎي ﺑﺎرﻧﺪﮔﯽ، ﺣﺪاﻗﻞ دﻣﺎ، ﺣﺪاﮐﺜﺮ دﻣﺎ و ﺑﺎد اﯾﺠﺎد ﺷﺪ و اﺣﺘﻤﺎل رﺧﺪاد ﭘﯿﺶ ﺑﯿﻨﯽ در ﺳﺎل آﯾﻨﺪه ﻣﺤﺎﺳﺒﻪ ﺷﺪ، ﺳﭙﺲ در ﻣﺮﺣﻠﻪ دوم ﻣﯿﺰان ﺑﺎرﻧﺪﮔﯽ، ﺣﺪاﻗﻞ و ﺣﺪاﮐﺜﺮ دﻣﺎ ﺗﻮﺳﻂ ﻣﺪل ﻣﺨﻔﯽ ﻣﺎرﮐﻮف ﭘﯿﺶ ﺑﯿﻨﯽ ﺷﺪ ﺗﺎ ﺑﺮ اﺳﺎس آن ﻫﺎ ﺗﺼﻤﯿﻤﺎت ﻻزم ﺟﻬﺖ ﻣﺪﯾﺮﯾﺖ ﻣﻨﺎﺑﻊ آب اﺗﺨﺎذ ﺷﻮد. روش ﭘﯿﺸﻨﻬﺎدي ﺑﺎ روش ﻫﺎي ﻣﺪل ﻣﺨﻔﯽ ﻣﺎرﮐﻮف اﺳﺘﺎﻧﺪارد و ﺷﺒﮑﻪ ﺑﯿﺰﯾﻦ ﺑﺮ اﺳﺎس دﻗﺖ و ﻧﺮخ ﺧﻄﺎ ﻣﻘﺎﯾﺴﻪ ﺷﺪ. ﻧﺘﺎﯾﺞ ﻧﺸﺎن داد روش ﭘﯿﺸﻨﻬﺎدي ﺑﺮ روي ﻣﺠﻤﻮﻋﻪ داده ﺷﻬﺮ ﻣﺸﻬﺪ در ﻣﻘﺎﯾﺴﻪ ﺑﺎ HMM 14% و در ﻣﻘﺎﯾﺴﻪ ﺑﺎ ﺷﺒﮑﻪ ﺑﯿﺰﯾﻦ 31% اﻓﺰاﯾﺶ دﻗﺖ و ﺑﺮ روي ﻣﺠﻤﻮﻋﻪ داده ﺷﻬﺮ ﺗﻬﺮان در ﻣﻘﺎﯾﺴﻪ ﺑﺎ HMM 10% و در ﻣﻘﺎﯾﺴﻪ ﺑﺎ ﺷﺒﮑﻪ ﺑﯿﺰﯾﻦ 15% اﻓﺰاﯾﺶ دﻗﺖ داﺷﺘﻪ اﺳﺖ

Drought is considered as one of the vital natural disasters that can be occurred in any climate. Given that drought incidence is inevitable, so its recognition is of important in order to efficient management of water. In this article drought forecasts in Tehran and Mashhad are considered. To this Markov models have been applied. At first stage after preprocessing of data, according to precipitation features, lowest temperature, highest temperature and wind Markov chains were created and the possibility of its incidence for next year was calculated, then at second stage precipitation level, lowest and highest temperature were forecasted by Markov hidden models so that based on them necessary decisions for water source management can be taken. Suggested method was compared based on error rate and accuracy by standard hidden Markov models and Bayesian network. The results show that suggested method on Mashhad collection data had 14% and 31% and on Tehran collection data 10% and 15% better accuracy compared to HMM and Bayesian network, respectively.