پهنه بندی خطر زمین لغزش با استفاده از فرایند تحلیل شبکه ANP و شبکه مصنوعی ANN در محیط GIS ( مطالعه موردی : حوضه آبخیز رودخانه رودبال
محورهای موضوعی : کاربرد GIS&RS در برنامه ریزیمحمدابراهیم عفیفی 1 , مهین صدری زاده 2
1 - استادیار داشگاه آزاد اسلامی واحد لارستان
2 - کارشناس ارشد جغرافیای طبیعی ودبیردبیرستانهای لار،لارستان،ایران
کلید واژه: پهنهبندی زمین لغزش, شبکه عصبی, منطق فازی, مدلAHP, حوضه رودخانه رود بال.,
چکیده مقاله :
زمین لغزش از معمولترین مخاطرات زمین شناختی در جهان است که همواره باعث تلفات جانی و آسیبهای مالی زیاد می گردد. شناسایی مناطق مستعد وقوع زمین لغزش از طریق پهنهبندی توان خطر با مدلهای تجربی مناسب، یکی از اقدامات اولیه در کاهش خسارات احتمالی و مدیریت خطر است. حوضه رودبال که در شهرستان داراب در استان فارس واقع شده به دلیل موقعیت جغرافیایی و ویژگیهای طبیعی آن یکی از مناطق تقریبا مستعد کشور نسبت به وقوع زمین لغزش و سایر پدیده های مرتبط با لغزش دامنه ها میباشد. ما در این پژوهش با استفاده از روش شبکه عصبی، مدلهای AHP ،ANP و منطق فازی با عملگرهای OR، And،Sum، Product و گاما به ارزیابی خطر زمین لغزش در حوضه رودبال داراب پرداختیم. مجموعاً 9 معیار اصلی مرتبط با وقوع پدیده زمین لغزش از جمله لیتولوژی، فاصله ازگسل، شیب، جهت شیب، باش، فاصله ازجاده، ارتفاع، فاصله از آبراهه و کاربری اراضی مورد تحلیل قرارگرفت. این معیارها به عنوان نقشه های عامل، هر کدام جداگانه کلاس بندی شده و از روش های آماری ارزش گذاری شدند. نقشه نهایی تولید شده برای پهنهبندی خطر زمین لغزش در حوضه نشان داد که مجموعاً حدود 70 درصد از حوضه در معرض خطر خیلی زیاد و زیاد زمین لغزش قرار دارد. همچنین زمین لغزه های موجود از طریق مشاهده زمینی، به عنوان شاهد زمینی(مرجع خطر) نقشهبندی شد که انطباق پذیری بالایی با روش مورد استفاده شبکه عصبی و تلفیق منطق فازی با عملگر 3/0 و ANP نشان داد.
Introduction
Landslides are among the most destructive natural hazards in mountainous and semi-mountainous regions, frequently causing environmental degradation, infrastructure damage, economic losses, and threats to human life. Effective landslide hazard zonation plays a crucial role in land-use planning, infrastructure development, watershed management, and disaster risk reduction. Identifying landslide-prone areas using reliable and scientifically robust modeling approaches is, therefore a fundamental step toward minimizing potential damages.
The Roudbal River watershed, located in Darab County, Fars Province, Iran, is particularly susceptible to landslides due to its geological structure, tectonic activity, topographic conditions, and climatic characteristics. Given its vulnerability, comprehensive hazard assessment is essential for sustainable regional development.
This study aims to produce a landslide hazard zonation map for the Roudbal watershed by integrating the Analytic Network Process (ANP) and Artificial Neural Network (ANN) models within a Geographic Information System (GIS) environment. Additionally, comparative analyses incorporating AHP and fuzzy logic operators (OR, AND, SUM, PRODUCT, and GAMMA) were performed to enhance modeling reliability and evaluate methodological performance.
. Materials and Methods
This research adopts a spatial multi-criteria decision-making and machine learning approach. A comprehensive landslide inventory map was first prepared using field surveys, historical landslide records, and high-resolution satellite imagery. The inventory map served as the reference dataset for model training and validation.
Based on literature review and expert consultation, nine primary landslide conditioning factors were identified and prepared as thematic layers in the GIS environment:
-
Lithology
-
Distance to faults
-
Slope gradient
-
Slope aspect
-
Elevation
-
Distance to roads
-
Distance to drainage networks
-
Rainfall
-
Land use/land cover
All factor layers were standardized and converted into raster format with a unified spatial resolution to ensure compatibility in spatial analysis.
The Analytic Network Process (ANP) was applied to determine the relative importance of the conditioning factors while accounting for interdependencies and feedback relationships among criteria. Unlike hierarchical approaches such as AHP, ANP considers complex network structures, making it particularly suitable for natural hazard modeling. Pairwise comparison matrices were constructed based on expert judgment, and consistency ratios were calculated to ensure the reliability of the weighting process.
The ANP-derived weights were then used as input parameters for the Artificial Neural Network (ANN) model. The ANN model was trained using landslide occurrence data and conditioning factors to identify nonlinear relationships between variables and landslide probability. Furthermore, fuzzy logic models using OR, AND, SUM, PRODUCT, and GAMMA operators (with gamma value 0.3) were implemented for comparative hazard assessment.
Model validation was conducted by overlaying the predicted hazard zones with the landslide inventory map to evaluate spatial agreement and predictive performance.
. Discussion
The integration of ANP and ANN provided a comprehensive framework combining expert-based weighting and data-driven machine learning. ANP effectively captured the interrelationships among conditioning factors, while ANN modeled complex nonlinear interactions influencing landslide occurrence.
Results indicate that lithology, slope gradient, and distance to faults were among the most influential factors in landslide occurrence within the Roudbal watershed. Areas characterized by steep slopes, weak lithological formations, and proximity to tectonic faults exhibited significantly higher landslide susceptibility.
The ANN model demonstrated strong predictive capability, particularly when combined with ANP-derived weights. The comparison with fuzzy logic operators revealed that the GAMMA operator (γ = 0.3) produced high spatial consistency with observed landslides. Field-verified landslide locations showed strong agreement with hazard zones predicted by the ANN and ANP-integrated models.
Overall, the hybrid modeling approach reduced uncertainty and improved the reliability of hazard zonation compared to single-method applications.
. Results and Conclusion
The final landslide hazard zonation map classified the Roudbal watershed into several risk categories, including very low, low, moderate, high, and very high hazard zones. The results revealed that approximately 70% of the basin area falls within high and very high landslide risk classes, indicating substantial vulnerability across the watershed.
Validation results demonstrated strong spatial conformity between mapped hazard zones and observed landslide locations, confirming the robustness of the integrated ANN–ANP approach. The fuzzy GAMMA (0.3) and neural network models showed particularly high adaptability to ground-truth data.
In conclusion, the integration of ANP and ANN within a GIS framework provides a powerful and reliable tool for landslide hazard assessment. The proposed methodology enhances predictive accuracy by combining expert knowledge with machine learning techniques and accounting for interdependent conditioning factors.
The findings offer valuable guidance for regional planners, engineers, and disaster management authorities in implementing preventive measures, optimizing land-use decisions, and improving risk mitigation strategies in landslide-prone regions. The methodological framework can be extended to other mountainous watersheds in Iran and similar geological settings to support sustainable development and hazard resilience
