پهنه‌بندي پراكنش مكاني نوعي آويشن (Thymus kotschianus)و بوماران Achilla millefolium) )با استفاده از شبكه عصبي مصنوعي (مطالعه موردي: مراتع دونا استان مازندران)

Spatial distribution mapping of common yarrow (Achilla millefolium) andthyme (Thymus kotschianus)using artificial neural network(Case study: Donna rangelands, Mazandaran province)

ﯿﻨﻪ و ﻫﺪف: ﻫﺪف از اﯾﻦ ﭘﮋوﻫﺶ، اﺳﺘﻔﺎده از ﻣﺪل ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ در ﺗﻬﯿﻪ ﻧﻘﺸﻪ ﭘﺮاﮐﻨﺶ ﻣﮑﺎﻧﯽ ﮔﻮﻧﻪﻫـﺎي ﻧـﻮﻋﯽ آوﯾﺸـﻦ و ﺑـ ﻣﺎدران در ﻣﺮاﺗﻊ دوﻧﺎ اﺳﺘﺎن ﻣﺎزﻧﺪران اﺳﺖ. روش ﺑﺮرﺳﯽ: ﻧﻤﻮﻧﻪﺑﺮداري از ﭘﻮﺷﺶ ﮔﯿﺎﻫﯽ ﺑﻪ روش ﻃﺒﻘﻪﺑﻨﺪي ﺗﺼﺎدﻓﯽ در 29 واﺣﺪ ﻫﻤﮕﻦ اﻧﺠﺎم ﺷـﺪ. در ﻫـﺮ واﺣـﺪ ﻫﻤﮕـﻦ، 3 ﻧﻤﻮﻧـ ﺧﺎك ﻧﯿﺰ از ﻋﻤﻖ 0-30 ﺳﺎﻧﺘﯽﻣﺘﺮي ﺑﺮداﺷﺖ ﺷﺪﻧﺪ. در ﭘﮋوﻫﺶ ﺣﺎﺿﺮ، از 20 ﻋﺎﻣﻞ ﻣﺤﯿﻄﯽ ﺑﻪ ﻋﻨﻮان ﻣﺘﻐﯿﺮ ﻣﺴﺘﻘﻞ و دادهﻫﺎي ﻣﺮﺑـﻮط ﺑـﻪ ﺣﻀﻮر ﮔﻮﻧﻪﻫﺎي ﮔﯿﺎﻫﯽ ﻣﻄﺎﻟﻌﻪ ﺷﺪه ﺑﻪ ﻋﻨﻮان ﻣﺘﻐﯿﺮ واﺑﺴﺘﻪ اﺳﺘﻔﺎده ﮔﺮدﯾﺪ. ﺑﺮاي ﺗﻬﯿﮥ ﻧﻘﺸﻪ ﭘﯿﺶﺑﯿﻨﯽ ﻣﮑﺎﻧﯽ ﮔﻮﻧﻪﻫﺎ، اﻃﻼﻋﺎت ﻣﺤﯿﻄﯽ در GIS ﺗﺒﺪﯾﻞ ﺑﻪ ﻧﻘﺸﻪ ﺷﺪه و ﺑﺎ اﺳﺘﻔﺎده از روش ﻧﺴﺒﺖ ﻓﺮاواﻧﯽ ﻫﺮ ﮐﺪام از آن ﻫﺎ ﮐﻼﺳﻪﺑﻨﺪي ﺷﺪﻧﺪ. در اﯾـﻦ ﭘـﮋوﻫﺶ از ﺷـﺒﮑﻪ ﭘﺮ ﺳـﭙﺘﺮون ﭼﻨﺪ ﻻﯾﻪ، ﻣﺘﺪاولﺗﺮﯾﻦ ﺷﺒﮑﻪﻫﺎي ﻋﺼﺒﯽﻣﺼﻨﻮﻋﯽ ﭘﯿﺶﺧﻮر، اﺳﺘﻔﺎده ﮔﺮدﯾﺪ. ﺳﺎﺧﺘﺎر ﺑﻬﯿﻨﻪ ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ، 20 ،20 ،1 ﺗﻌﯿـﯿﻦ ﺷـﺪ. ﺧﺮوﺟﯽ ﺑﻪ دﺳﺖ آﻣﺪه از ﺷﺒﮑﻪ در ﻧﺮم اﻓﺰار GIS ﺗﺒﺪﯾﻞ ﺑﻪ ﻧﻘﺸﻪﻫﺎي ﭘﻬﻨﻪﺑﻨﺪي ﮔﻮﻧﻪﻫﺎي ﮔﯿﺎﻫﯽ ﺑﺎ 4 ﭘﻬﻨﻪ ﻋﺪم ﺣﻀﻮر، ﺣﻀﻮر ﮐﻢ، ﺣﻀﻮر ﺘﻮﺳﻂ و ﺣﻀﻮر زﯾﺎد ﺷﺪ. ارزﯾﺎﺑﯽ ﻣﺪل ﺑﻪ دو روش ﻣﻨﺤﻨﯽ ROC و ﺿﺮﯾﺐ ﮐﺎﭘﺎ اﻧﺠﺎم ﺷﺪ. ﯾﺎﻓﺘﻪﻫﺎ:ﺑﺎ اﺳﺘﻔﺎده از روش ﻣﻨﺤﻨﯽROC، ﻣﻘﺪار AUC ﺑﺮاي ﮔﻮﻧﻪ ﺑﻮﻣﺎدران ﺑﺮاﺑﺮ 96/8، و ﺑﺮاي ﮔﻮﻧﻪ ﻧﻮﻋﯽ آوﯾﺸﻦ ﺑﺮاﺑـﺮ 84/7 ﺷـﺪ ﮐـ ﻧﺸﺎندﻫﻨﺪة ارزﯾﺎﺑﯽ ﻋﺎﻟﯽ و ﺧﯿﻠﯽ ﺧﻮب ﻣﺪل در ﭘﯿﺶﺑﯿﻨﯽ اﺳﺖ. ﺑﺤﺚ وﻧﺘﯿﺠﻪﮔﯿﺮي: ارزﯾﺎﺑﯽ ﺑﻪ روش ﺿﺮﯾﺐ ﮐﺎﭘﺎ ﻧﺸﺎن داد ﮐﻪ اﯾﻦ ﺿﺮﯾﺐ ﺑﺮاي ﮔﻮﻧﻪ ﺑﻮﻣﺎدران، و ﮔﻮﻧﻪ ﻧﻮﻋﯽ آوﯾﺸﻦ ، ﺑﻪ ﺗﺮﺗﯿﺐ ﺑﺮاﺑﺮ 0/89 0/76 ﺑﻮد ﮐﻪ ﻧﺸﺎندﻫﻨﺪة ارزﯾﺎﺑﯽ ﺑﺴﯿﺎر ﺧﻮب و ﺧﻮب ﻣﺪل اﺳﺖ.

Background and Objective:The purpose of this study was to map the spatial distribution of common yarrow (Achilla millefolium) and thyme (Thymus kotschianus) using artificial neural network model in rangelands Donna, Mazandaran Province. Method: Sampling was carried out with equal random classification in 29 homogenous units. In each unit, 3 soil samples were harvested from depth of 0-30 cm. In this study, 20 environmental factors were the independent variables and the presence of plant species were the dependent variable. For the preparation spatial distribution map of the species, environmental data were converted to maps in GIS. Then each of these factors was classified using the frequency. In this research, network Multilayer Perceptron that is the most common feed forward neural network was used. Optimal structure for the network was determined 1, 20, and 20. Then distribution maps of studied species were prepared with 4 class absence and low presence, medium presence and high presence in the GIS software. Models were evaluated using ROC curves and Kappa coefficient. Findings: AUC were 96.8 and 84.7 for the species Achilla millefolium and Thymus kotschianus was, respectively that indicates models are excellent or very good for the prediction. Discussion and Conclusion: Also kappa coefficient were calculated as 89.0 and 76.0 for Achilla millefolium and Thymus kotschyanus, respectively which indicate very good and good prediction.