مدل سازی فرونشست زمین در دشت سلماس (آذربایجان غربی) با استفاده از ANFIS (مقاله علمی وزارت علوم)
درجه علمی: نشریه علمی (وزارت علوم)
آرشیو
چکیده
دشت سلماس از محدوده های مهم کشاورزی شمال غرب کشور و استان آذربایجان غربی، محسوب می شود. این دشت در دهه های اخیر، به لحاظ برداشت بیش ازحد از آب های زیرزمینی دچار بحران شدید فرونشست زمین شده است. شناسایی مناطق مستعد فرونشست در محدوده دشت سلماس، از دیدگاه شناخت علت وقوع، مدیریت و کنترل پدیده فرونشست و همچنین جلوگیری از فرسایش خاک، از اهمیت زیادی برخوردار است. در این تحقیق، با هدف بررسی علل و ابعاد فرونشست زمین و همچنین پیش بینی مناطق مستعد به فرونشست، از مدل سیستم استنتاج عصبی فازی و برای رسیدن به هدف، از هفت عامل تأثیرگذار در پدیده فرونشست دشت (شامل: شیب زمین، ارتفاع، پوشش گیاهی، آب های زیرزمینی، فاصله از جاده، فاصله از رودخانه و فاصله از چاه های پیزومتری) استفاده شد. در روند تحقیق، اطلاعات جمع آوری شده به محیط ArcGis وارد و برای پیاده سازی مدل به نرم افزار متلب منتقل گردیدند و با استفاده از روشC-flodCV داده ها با دقت آموزش و اعتبار سنجی شدند. داده ها در چند تابع عضویت مختلف، شامل توابع عضویت ذوزنقه ای، مثلثی، گوسی، گوسی دو طرفه و تابع زنگوله ای مدل سازی شدند. نتایج بررسی ها نشان داد تابع عضویت ذوزنقه ای با ضریب همبستگی رگرسیون 0.86 و تابع عضویت گوسی با ضریب همبستگی رگرسیون 0.81 بهترین عملکرد را در شناسایی مناطق مستعد فرونشست دارند. با توجه به نتایج مطالعات در دشت سلماس، می توان گفت در طی سال های موردبررسی، مناطق شرقی و خروجی دشت با افت زیاد آب های زیرزمینی و متعاقب آن با پدیده فرونشست مواجه شده اند. همچنین بررسی های آماری نشان داد در بازه زمانی موردبررسی، آب های زیرزمینی در کل بیش از 18 متر افت سطح ایستابی داشته اند و این افت به طور متوسط در هر سال 1.23 متر گزارش شده است. در دو دهه گذشته، با محاسبه ضریب ذخیره یا تغذیه 0.03 درصدی و شرایط خشکی اقلیمی و افزایش سطح زیر کشت، این بحران در سال های اخیر به شدت تشدید شده است.Modeling of land subsidence of Salmas plain by using adaptive neuro-fuzzy inference system (ANFIS)
Extended Abstarct
IntroductionSalmas Plain is one of the important agricultural regions of West Azerbaijan province, playing a significant role in the region's agriculture and development. In recent decades, the plain has been severely impacted by the phenomenon of subsidence due to the over-extraction of groundwater from aquifers. Therefore, identifying areas prone to subsidence in the Salmas Plain is of particular importance for the management and control of this phenomenon. In this study, a fuzzy neural inference system model was used to predict subsidence-prone areas. To achieve this goal, seven important factors contributing to subsidence in the region were analyzed: land slope, digital elevation model, vegetation, groundwater depth, distance from roads, distance from rivers, and distance from piezometric wells. The information related to these factors was collected in ArcGIS software and transferred to MATLAB software for model implementation. Using the C-fold CV method, the data were randomly divided into three groups: 70% for training, 20% for testing, and 10% for validation. These data were introduced to MATLAB for training, testing, and validation. The data were accurately trained and validated, achieving a precision of 10⁻⁸. Several different membership functions, including trapezoidal, triangular, Gaussian, two-sided Gaussian, and bell curve functions, were tested in the model. The results showed that the trapezoidal membership function, with a regression correlation coefficient of 0.86, and the Gaussian membership function, with a regression correlation coefficient of 0.81, performed best in identifying areas prone to subsidence.Based on the results, the eastern areas and the outlet of the plain have experienced significant drops in groundwater levels and subsidence. Statistical studies also revealed that groundwater levels have dropped by over 18 meters, averaging 1.23 meters per year in recent years. The crisis has been exacerbated over the past two decades by over-extraction of groundwater for agriculture, combined with a storage-to-recharge ratio of 0.03% and unfavorable climatic conditions.The spatial pattern of land subsidence in Iran indicates that this phenomenon is strongly linked to the uncontrolled abstraction of groundwater resources, driven by increasing urban water consumption and low agricultural water efficiency. Drought conditions and excessive groundwater withdrawal have not only affected arid and central regions of the country but have also extended to semi-arid and humid areas in the northwest. Geological experts have reported that 300 out of 600 plains in the country are affected by subsidence. Due to the importance of this issue, many researchers have studied land subsidence. Their results indicate that the Salmas Plain, as part of the Lake Urmia catchment area, is experiencing a major subsidence crisis. This crisis has caused significant problems for infrastructure in many parts of the Salmas Plain, especially in the Qara Gheshlagh region. The main cause of subsidence is the over-extraction of water from wells constructed for agricultural purposes, which has led to a steady decline in the volume of aquifers.
Material and Methods
To evaluate the susceptibility of the Salmas Plain to subsidence, the required data and statistics up to 2019 (1398 in the Iranian calendar) were collected as follows:
Digital topographic maps (1:25,000 scale), a digital geological map (1:100,000 scale) of Salmas city, and thematic maps (e.g., slope, land use, geology, distance from wells, groundwater depth, distance from rivers, and distance from faults).
GPS data, meteorological information, soil maps, and aerial photographs (1:20,000 scale).
Digital Elevation Model (DEM) with a resolution of 30 x 30 meters.In this study, seven criteria were analyzed: vegetation type, groundwater depth, distance from roads, distance from rivers, distance from piezometric wells, digital elevation model, and slope. All criteria were first standardized in ArcGIS software and converted into an 800 x 800-meter grid. The data from each grid were then compiled into a matrix and transferred to MATLAB for analysis using the Adaptive Neuro-Fuzzy Inference System (ANFIS) model.
To ensure accurate modeling, the satellite data were divided into three parts: 70% for training, 20% for testing, and 10% for validation. This process ensured precise measurement of modeling accuracy and error rates using MATLAB software.
Results and Discussion
Studies revealed that over the past 41 years, groundwater levels in the Salmas Plain have dropped by 18 meters, averaging 1.23 meters per year from 2014 to 2015 (93-94 in the Iranian calendar). The plain, characterized by clay and silty layers, has experienced significant aquifer compression and reduced porosity due to over-extraction of groundwater. This has resulted in land subsidence across the study area.The model results demonstrated that trapezoidal and triangular membership functions performed best, achieving error coefficients of 10⁻⁷ and 10⁻⁸, with mean error rates of 0.18 and 0.20, respectively. The regression correlation coefficients were 0.86 and 0.80, confirming the model's accuracy in identifying areas of high subsidence sensitivity. Approximately 8.9 and 6.3 square kilometers of the plain were classified as very high-risk areas, while 1.8 and 3.2 square kilometers were classified as high-risk zones. These findings indicate that the proposed model can effectively predict subsidence with high accuracy and reliability.
Furthermore, the results highlight that over-extraction of groundwater has caused aquifer compaction and subsidence. Linear regression analysis confirmed the model's accuracy, with a correlation coefficient of 0.86.
ConclusionThe Salmas Plain is a vital agricultural region where groundwater serves as the primary water source. Over 70% of extracted groundwater is used for agriculture. Irregular groundwater extraction and reduced rainfall have significantly lowered groundwater levels, exacerbating subsidence in the plain. Since 1993, the annual water table drop has increased to 1.23 meters.The fuzzy neural inference system model demonstrated that trapezoidal and triangular membership functions with regression correlations of 0.86 and 0.81 provided reliable results, predicting subsidence-prone areas effectively. The findings suggest that 8.6% and 6.3% of the study area are at very high risk of subsidence. Immediate measures are required to control groundwater extraction and manage the subsidence crisis in the Salmas Plain.
تبلیغات
