مدلسازي تبخير ماهانه با استفاده از روش هاي منفرد و هيبريدي-موجك داده كاوي در حوضه هاي آبريز ايران با تنوع اقليمي

Modeling Of Monthly Evaporation Using Single and Hybrid-Wavelet Data-Driven Methods in Basins of Iran with Climate Variety

ﭼﮑﯿﺪه ﺗﺒﺨﯿﺮ ﺑﻪ ﻋﻨﻮان ﯾﮑﯽ از ﭘﺎراﻣﺘﺮﻫﺎي ﻃﺒﯿﻌﯽ، ﻫﻤﻮاره ﻣﻮرد ﺗﻮﺟﻪ ﻣﺤﻘﻘﯿﻦ ﺑﻮده اﺳﺖ. در اﯾﻦ ﭘﮋوﻫﺶ، ﻣﺘﻐﯿﺮ ﺗﺒﺨﯿﺮ ﻣﺎﻫﺎﻧﻪ ﺑﺎ اﺳﺘﻔﺎده از روشﻫﺎي ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ، ﺳﺎﻣﺎﻧﻪ اﺳﺘﻨﺘﺎﺟﯽ ﻓﺎزي-ﻋﺼﺒﯽ ﺗﻄﺒﯿﻘﯽ و ﺑﺮﻧﺎﻣﻪرﯾﺰي ﺑﯿﺎن ژن و ﺗﺮﮐﯿﺐ روشﻫﺎي ﻣﺬﮐﻮر ﺑﺎ ﺗﺌﻮري ﻣﻮﺟﮏ، در دو اﻗﻠﯿﻢ ﻣﺘﻔﺎوت اﯾﺮان ﻣﺪﻟﺴﺎزي ﺷﺪ. ﺑﺪﯾﻦ ﻣﻨﻈﻮر، دادهﻫﺎي ﻫﻮاﺷﻨﺎﺳﯽ ﺑﺎرش رﻃﻮﺑﺖ ﻧﺴﺒﯽ، دﻣﺎي ﻣﯿﺎﻧﮕﯿﻦ، دﻣﺎي ﺑﯿﺸﯿﻨﻪ، دﻣﺎي ﮐﻤﯿﻨﻪ و ﺳﺮﻋﺖ ﺑﺎد، در ﻃﻮل دوره آﻣﺎري 1384-1397 ﻣﺮﺑﻮط ﺑﻪ دو ﺣﻮﺿﻪ آﺑﺮﯾﺰ درﯾﺎﭼﻪ اروﻣﯿﻪ و ﮔﺎوﺧﻮﻧﯽ ﺑﻪ ﮐﺎر ﮔﺮﻓﺘﻪ ﺷﺪ. در اﯾﻦ ﻣﻄﺎﻟﻌﻪ، اﺛﺮ ﻓﺼﻠﯽ و ﻧﻮﯾﺰزداﯾﯽ دادهﻫﺎ اﻋﻤﺎل ﺷﺪ. دﻗﺖ روشﻫﺎي ﻣﻮرد ﻣﻄﺎﻟﻌﻪ ﺑﺮ اﺳﺎس ﺷﺎﺧﺺﻫﺎي آﻣﺎري ﺿﺮﯾﺐ ﻫﻤﺒﺴﺘﮕﯽ )R(، رﯾﺸﻪ ﻣﯿﺎﻧﮕﯿﻦ ﻣﺮﺑﻌﺎت ﺧﻄﺎ )RMSE(، ﻣﯿﺎﻧﮕﯿﻦ ﺧﻄﺎي ﻣﻄﻠﻖ )MAE( و ﺿﺮﯾﺐ ﮐﺎراﯾﯽ ﻧﺶ-ﺳﺎﺗﮑﻠﯿﻒ )NSE( ﻣﻮرد ﺑﺮرﺳﯽ ﻗﺮار ﮔﺮﻓﺖ. ﻧﺘﺎﯾﺞ ﺣﺎﮐﯽ از اﯾﻦ اﺳﺖ ﮐﻪ در دو اﻗﻠﯿﻢ ﻣﺨﺘﻠﻒ روشﻫﺎي ﻫﯿﺒﺮﯾﺪ ﻣﻮﺟﮏ ﺑﺮﻧﺎﻣﻪرﯾﺰي ﺑﯿﺎن ژن و ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ ﻣﻨﻔﺮد ﺑﻪ ﺗﺮﺗﯿﺐ داراي ﺑﺎﻻﺗﺮﯾﻦ و ﺿﻌﯿﻒﺗﺮﯾﻦ ﻋﻤﻠﮑﺮد در ﻣﯿﺎن ﺳﺎﯾﺮ ﻣﺪلﻫﺎي داده ﮐﺎوي ﺑﻪ ﮐﺎر رﻓﺘﻪ در اﯾﻦ ﺗﺤﻘﯿﻖ ﻫﺴﺘﻨﺪ. ﻣﺪل ﻫﯿﺒﺮﯾﺪي ﻣﻮﺟﮏ-ﺑﺮﻧﺎﻣﻪرﯾﺰي ﺑﯿﺎن ژن ﺑﺎ ﻣﻘﺪار RMSE ﺑﺮاﺑﺮ ﺑﺎ 20/870و156/884 ﻣﯿﻠﯽﻣﺘﺮ ﺑﻪ ﺗﺮﺗﯿﺐ ﺑﺮاي اﯾﺴﺘﮕﺎهﻫﺎي ﺗﺎزهﮐﻨﺪ در ﺣﻮﺿﻪ آﺑﺮﯾﺰ درﯾﺎﭼﻪ اروﻣﯿﻪ و ﮐﻮﻫﭙﺎﯾﻪ در ﺣﻮﺿﻪ آﺑﺮﯾﺰ ﮔﺎوﺧﻮﻧﯽ ﻋﻤﻠﮑﺮد ﺑﺎﻻﺗﺮي را داﺷﺘﻪ اﺳﺖ. ﻫﻤﭽﻨﯿﻦ، ﻧﺘﺎﯾﺞ ﻧﺸﺎن داد ﮐﻪ ﺗﺎًﺛﯿﺮ ﺑﻪ ﮐﺎرﮔﯿﺮي ﺿﺮﯾﺐ ﻓﺼﻠﯽ و ﻧﻮﯾﺰزداﯾﯽ دادهﻫﺎ در ارﺗﻘﺎء ﻋﻤﻠﮑﺮد ﻣﺪلﻫﺎ ﻗﺎﺑﻞ ﺗﻮﺟﻪ اﺳﺖ. ﺑﺮ اﺳﺎس ﻧﺘﺎﯾﺞ، ﻋﻤﻠﮑﺮد ﻣﺪلﻫﺎ در ﺣﻮﺿﻪ آﺑﺮﯾﺰ درﯾﺎﭼﻪ اروﻣﯿﻪ ﺑﺎ اﻗﻠﯿﻢ Dsaﺑﻬﺘﺮ ﺑﻮده اﺳﺖ. ﻫﻤﭽﻨﯿﻦ، روش ﻫﺎي دادهﮐﺎوي ﻫﯿﺒﺮﯾﺪي را ﻣﯽﺗﻮان ﺑﻪ ﻋﻨﻮان ﺟﺎﯾﮕﺰﯾﻦ ﻣﻨﺎﺳﺒﯽ ﺑﺮاي روشﻫﺎي ﻗﺪﯾﻤﯽ ﻣﻌﺮﻓﯽ ﻧﻤﻮد.

Evaporation as one of the natural parameters has always been considered by researchers. In this study, the monthly evaporation variable was modeled in two different climates of Iran using artificial neural network, adaptive fuzzy-neural inference system and gene expression programming methods and combining these methods with wavelet theory. For this purpose, meteorological data of precipitation, relative humidity, average temperature, maximum temperature, minimum temperature and wind speed were used during the statistical period of 1384-1397 related to the two catchments of Urmia Lake and Gavkhouni. In this study, the seasonal effect and data noise reduction were applied. The accuracy of the studied methods was evaluated based on statistical correlation coefficient (R), root mean square error (RMSE), mean absolute error (MAE) and nash-sutcliffe efficiency (NSE). The results show that in two different climates, the wavelet-hybrid gene expression programming and the single artificial neural network have the highest and weakest performance, respectively, among other data mining models used in this study. The hybrid wavelet-gene expression programming model with RMSE value of 20.870 and 156.884 had higher performance for Tazehkand station in Urmia Lake catchment area and Kuhpayeh catchment in Gavkhouni catchment area, respectively. Also, the results showed that the effect of seasonal factor utilization and data noise reduction in model performance improvement is significant. Based on the results of the models performance Urmia Lake catchment area with Dsa climate has been better. However, hybrid data mining methods can be introduced as a good alternative to the old methods.