ارزيابي قابليت روش شبكه هاي عصبي مصنوعي در تهيه نقشه پيش بيني پراكنش رويشگاه گونه هاي گياهي (مطالعه موردي: مراتع پشتكوه استان يزد)

Evaluation of artificial neural network capability in preparation of the predictive map of plant habitat distribution (Case study: Poshtkouh rangelands of Yazd province)

ﭘﮋوﻫﺶ ﺣﺎﺿﺮ ﺑﺎ ﻫﺪف ﺑﺮرﺳﯽ ﻗﺎﺑﻠﯿﺖ روش ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﻣﺼﻨﻮﻋﯽ در ﺗﻬﯿﻪ ﻧﻘﺸﻪ ﭘﯿﺶ ﺑﯿﻨﯽ ﭘﺮاﮐﻨﺶ روﯾﺸﮕﺎه ﮔﻮﻧﻪ ﻫﺎي ﮔﯿﺎﻫﯽ در ﻣﺮاﺗﻊ ﭘﺸﺘﮑﻮه اﺳﺘﺎن ﯾﺰد اﻧﺠﺎم ﺷﺪ. ﺑﺪﯾﻦ ﻣﻨﻈﻮر ﺑﻌﺪ از ﺗﻌﯿﯿﻦ واﺣﺪﻫﺎي ﻫﻤﮕﻦ ﺑﻮم ﺷﻨﺎﺧﺘﯽ، ﺑﺎ اﺳﺘﻔﺎده از ﻣﺪل رﻗﻮﻣﯽ ارﺗﻔﺎع و ﻧﻘﺸﻪ زﻣﯿﻦ ﺷﻨﺎﺳﯽ ﺑﺎ ﻣﻘﯿﺎس 1:25000، ﻧﻤﻮﻧﻪ ﺑﺮداري از ﭘﻮﺷﺶ ﮔﯿﺎﻫﯽ و ﻋﻮاﻣﻞ ﻣﺤﯿﻄﯽ درون واﺣﺪﻫﺎي ﻫﻤﮕﻦ اﻧﺠﺎم ﮔﺮﻓﺖ و ﻧﻘﺸﻪ ﻣﺮﺑﻮط ﺑﻪ ﻣﺘﻐﯿﺮﻫﺎي ﻣﺤﯿﻄﯽ ﺑﺎ اﺳﺘﻔﺎده از ﺳﯿﺴﺘﻢ اﻃﻼﻋﺎت ﺟﻐﺮاﻓﯿﺎﯾﯽ و زﻣﯿﻦ آﻣﺎر ﺗﻬﯿﻪ ﺷﺪ. ﻣﺪﻟﺴﺎزي ﭘﯿﺶ ﺑﯿﻨﯽ ﭘﺮاﮐﻨﺶ ﺑﺎ اﺳﺘﻔﺎده از ﺷﺒﮑﻪ ﭘﺮﺳﭙﺘﺮون ﭼﻨﺪﻻﯾﻪ اﻧﺠﺎم ﺷﺪ. ﺑﻌﺪ از ﺗﻌﯿﯿﻦ ﺑﻬﺘﺮﯾﻦ ﺳﺎﺧﺘﺎر ﺷﺒﮑﻪ ﻋﺼﺒﯽ ﺑﺎ اﺳﺘﻔﺎده از ﻣﯿﺎﻧﮕﯿﻦ ﻣﺮﺑﻌﺎت ﺧﻄﺎ، ﺷﺒﯿﻪ ﺳﺎزي اﺣﺘﻤﺎل ﺣﻀﻮر و ﻏﯿﺎب ﮔﻮﻧﻪ ﻫﺎ ﺑﺎ ﺷﺒﮑﻪ ﺑﻬﯿﻨﻪ اﻧﺠﺎم ﺷﺪ. در ﻫﺮ روﯾﺸﮕﺎه ﻧﻘﺸﻪ ﭘﯿﻮﺳﺘﻪ اﺣﺘﻤﺎل ﺣﻀﻮر و ﻏﯿﺎب ﮔﻮﻧﻪ ﻫﺎ ﺑﺎ اﺳﺘﻔﺎده از ﻧﺮم اﻓﺰار Arc GIS ﺗﻬﯿﻪ و آﺳﺘﺎﻧﻪ ﺑﻬﯿﻨﻪ ﺣﻀﻮر ﺗﻌﯿﯿﻦ ﺷﺪ. ﺑﺮرﺳﯽ ﻣﯿﺰان ﺗﻄﺎﺑﻖ ﻧﻘﺸﻪ ﻫﺎي ﺑﻪ دﺳﺖ آﻣﺪه ﺑﺎ ﻧﻘﺸﻪ ﻫﺎي واﻗﻌﯽ ﺑﺮ اﺳﺎس ﺷﺎﺧﺺ ﮐﺎﭘﺎ ﻧﺸﺎن داد ﮐﻪ نقشه ­هاي پيش­ بيني رويشگاه­ Seidlitzia rosmarinus داراي تطابق عالي؛ رويشگاه­ هاي Rheum ribes- Artemisia sieberi و Cornulaca monacantha داراي تطابق خيلي ­خوب؛ رويشگاه­ Ephedra. strobilacea- Zygophyllumeurypterum و Artemisia aucheri داراي تطابق خوب؛ رويشگاه S. orientalis - Artemisia sieberi ،Scariola orientalis-Astragalus albispinus، A. sieberi1، A. sieberi2 و Tamarix ramosissima داراي تطابق متوسط و رويشگاه ­هاي Artemisia sieberi- Scariola orientalis و Artemisia sieberi- Zygophyllm eurypterum داراي تطابق ضعيف با نقشه ­هاي واقعيت زميني است.

The present study aimed to investigate the possibility of using artificial neural network to estimate the spatial distribution, preparation of predictive distribution of plant species and recognizing the strengths and weaknesses of this method. For this purpose, sampling of vegetation and environmental factors was carried out after determining homogeneous units using digital elevation model and of Geological map scale of 1: 25,000. Then, environmental variables maps were prepared using GIS and geostatistics. The input variables of the neural network were selected based on the results of logistic regression. Predictive distribution modeling was done by Multilayer Perceptron network. Simulation of the presence and absence probability was done after determining the optimal structure of the neural network using mean square error, then, continuous map of the presence or absence probability of species in each habitat was developed using Arc GIS software and optimal threshold was determined. Assessment of the predictive and actual map agreement by calculation of kappa coefficient showed that predictive map of Seidlitzia rosmarinus has excellent and Cornulaca monacantha has very good correspondence, predictive maps of the S.orientalis - Artemisia sieberi، S. orientalis-Astragalus albispinus، A. sieberi1، A. sieberi2, R.ribes-A. seiberi predictive maps of habitat has moderate agreement and predictive maps of the A.sieberi-S. orientalis and A. sieberi- Z. eurypterum has the poor agreement with actual maps of these species.