تحليل مقايسه اي مدل هاي آماري و شبكه عصبي مصنوعي جهت برآورد حجم رسوبات نبكا (مطالعه موردي: نبكاهاي درختچه گز در كوير ابراهيم‌آباد سيرجان)

A comparative analysis of statistical models and artificial neural network for estimating the nebkhas sediments volume (case study: tamarix mascatensis nebkhas in Ebrahim abad desert of Sirjan)

روابط موجود در درون سيستم در قالب مدل‌هاي متعددي جمع بندي مي‌شوند كه اين مدل‌ها ابزاري مهم جهت تبيين پديده هايي هستند كه در سيستم‌ها ديده مي شوند. ازاين‌رو مدل‌سازي به‌عنوان ابزاري جهت درك ارتباطات اكوژيومورفولوژيكي پيچيده كه در سير تكامل ناهمواري و پوشش گياهي حاكم مي‌باشد مي‌تواند در مديريت تغييرات محيطي يا انساني در سيستم‌هاي مناطق خشك و نيمه‌خشك موثر واقع شود. چشم اندازهاي نبكايي ازجمله سيستم هاي اكوژيومورفيك پيچيده در مناطق بياباني هستند كه در اثر تجمع رسوبات بادي در اطراف گياهان شكل مي گيرند. هدف اين پژوهش مدل‌سازي حجم رسوبات نبكا با روش‌هاي آماري و شبكه عصبي است. بدين منظور خصوصيات مورفومتري نبكاها و مورفولوژي گياهي شامل، ارتفاع نبكا، قطر قاعده نبكا، حجم نبكا، قطر تاج پوشش و ارتفاع گياه به روش طولي اندازه‌گيري گرديد. سپس از بين روش‌هاي ساده رگرسيوني روش تواني به دليل برخورداري از R² بالاتر انتخاب گرديد. همچنين شبكه مورداستفاده جهت مدل سازي از نوع شبكه‌هاي پيش‌خور با الگوريتم آموزشي پس انتشار خطا مي‌باشد. تابع آموزشي استفاده‌شده در شبكه Trainlm و تابع انتقال از نوع log sig مي‌باشد. جهت آموزش شبكه از 75% داده‌ها و جهت آزمون شبكه از 35% داده‌ها استفاده‌شده است. نتايج نشان مي دهد كه مدل شبكه عصبي مصنوعي با ضريب تبيين 926/0 و ميزان خطاي 16/1 نسبت به مدل رگرسيوني با ضريب تبيين 868/0 و ميزان خطاي 3/3 از برتري بيشتري جهت برآورد حجم رسوبات نبكاهاي مطالعاتي برخوردار است.

Introduction Relations within a system are summarized in several models; these models are important tools to explain the phenomena that occur in the system. Some of the current tools for modeling are those of statistical modeling methods based on regression analysis. The results of these models are based on the most practical processing function among variables, and this feature is the reason for limiting the distribution of time - place variables, and needs subordination of data to a particular function or equation. Today artificial nerve networks (ANN) as a tool to develop models, to predict phenomena, to simulate and to estimate data has been entered into the realm of science. ANN was introduced primarily in 1962 by Frank Rosenblatt and then was developed and improved to the world in 1986 by Rommel Hart and Mac Koland perceptron model. This method by using a neurotic and intelligent structure and appropriate modeling of neurons in brain, tries to simulate behavior of intracellular neurons in the brain via mathematical defined functions, or in other words this method makes a model of synaptic function in natural neurons, by calculating accessible calculations in artificial neurons. This flexible and experimental nature of the method makes it very usable in some areas of problem; like prediction category that has an experimental approach to issues and a nonlinear behavior is found in its structure. Modeling is important in understanding complex ecogeomorphological relationships for useful applicationsinclude: predicting the likely success of remobilizing dormant dune systems, managing and mitigating desertification and degradation of semi-arid land , assessing the impact of climate change , investigating the effects of changes in land use and reconstructing conditions responsible for the formation and stabilization of relict systems. Nebkas are generally appearing in regions that amount of sand is average and sufficient moisture was existed for life of vegetation. In general, Nebka is self-organization reaction of ecosystem against windy erosion stress. In the other word, environmental system is trying to adjust the pressure of windy erosion by creating this feature. Jasem and Al-Awadhi (2013) says Nabkha is a type of Aeolian landforms, which is commonly developed as a result of sand accumulation around coastal and desert shrubs. The morphology of nabk-has is controlled by growth patterns of shrub [2]. The height of nabkha, to some extent, is related to the height of the shrub crown, while its length is related to the overall height of the shrub [3], width of the basal shrub and wind velocity [4]. Other important factors control-ling nabkhas morphology includes type of sediment supply and climate [5] as well as the porosity of the shrub crown. This study investigates the relations between morphometric Nebkha and phormic characteristics of Tamarix Mascatensis species Nebkhas in the study area, and designs mathematical models of neural networks in order to recognize the patterns governing the Nebkha landscape system, and compares the results of statistical methods with techniques of artificial neural networks (ANN). Methodology First use of aerial region, range and then study specific reference to attendance area development and Nebkhas territory were determined. Sampling took place along transects that cover the entire area. for obtaining objective of study we measure canopy cover and height of plant and elevation, diameter for nebkha from 10 transects and 105 samples in Ebrahim abad desert area in Sirjan. Non-linear simple regression: among simple regression methods, power method was chosen due to higher R² factor and the equation is y = bx ?. Multiple linear regression: this method establishes a correlation between the dependent variables (get predicted) and independent variables (predictors) so that the dependent variable is predicted by the independent variables. Artificial neural network: the network used for modeling is that of feedforward network with back propagation algorithm. This type of network is the most useful in experimental methods modeling. Training function used in the network is that of Trainlm this function has high levels of speed and accuracy. Transfer function used in the network is that of log sig. To train the network 35% of the data, and to test the network 75% of the data have been used. Results and discussion In general the compilation of the conceptual model, Quantitative model and Assessment and operate model, are the essential steps in the systemic analysis. Results of regression models and artificial neural networks to estimate of the tamarix mascatensis nebkha sediment volume show artificial neural networks whit 0.926 coefficients, the best correlation to predict the sediment volume. Power regression model show, the following equation to calculate the nebkha sediment volume: Vn=0.002?Vv?^1.263 Vn= nebkha sediment volume Vv= plant volume Conclusion In geomorphological studies, figures of the earth are reflections of surface processes and systems structure in the process. Investigating structure and function of these systems would cause access to their past and also ability to draw their future evolutionary path. Research results show that artificial neural networks whit 0.926 coefficient and 1.16 errors, compared with the power regression model with 0.868 coefficients and 3.3 errors, is better for estimate the nebkha sediment volume in the study area.