بررسي اثر ويژگي‌هاي ژئومورفيكي بر الگوي زمين لغزش با استفاده از روش‌هاي كمي و هندسه فركتال (مطالعه موردي: حوضه آبريز سد شهيد عباسپور (ده‌شيخ))

Investigating the Effect of Geomorphic Features on Lan‎dslide Patterns Using Quantitative Methods an‎d Fractal Geometry (Case Study: Shahid Abbaspur Dam Catchment Area(Deh-Sheikh))

جغرافياي طبيعي

This study investigates the effect of geomorphological characteristics on the spatial pattern of lan‎dslide occurrence in the Dehsheikh catchment, located in the northeastern part of Khuzestan Province, Iran. A total of 15 influential parameters were examined, including elevation, slope, aspect, curvature, stream power index (SPI), topographic wetness index (TWI), distance to fault, distance to river, distance to earthquake epicenter, distance to road, normalized difference vegetation index (NDVI), precipitation, lithological formations, soil texture, an‎d lan‎d use/lan‎d cover. Required data were collected from Sentinel-2 satellite imagery, meteorological an‎d agricultural organizations, geological surveys, an‎d extensive field observations, GPS an‎d were subsequently processed using SAGA-GIS, ArcGIS 10.8, an‎d Google Earth Engine. To model lan‎dslide susceptibility, five state-of-the-art machine learning algorithms were employed: Ran‎dom Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), Maximum Entropy (MaxEnt), an‎d Multiple linear regression(MLR). The dataset consisted of 129 historical lan‎dslide points (99 obtained from the Geological Survey of Iran an‎d 30 derived from field surveys an‎d Google Earth imagery interpretation) along with 1000 ran‎domly selec‎ted non-lan‎dslide points, yielding a total of 1179 sample points. The dataset was ran‎domly partitioned into training (70%, 790 points) an‎d testing (30%, 339 points) subsets. Model validation was performed using Accuracy, Kappa coefficient, Root Mean Square Error (RMSE), Coefficient of Determination (R²), Mean Absolute Error (MAE), an‎d Area Under the Receiver Operating Characteristic Curve (AUC) within the R software environment. Subsequently, to achieve a more refined analysis of the lan‎dslide distribution pattern, six fractal dimensions namely box-counting, capacity, correlation, radial, dilation, an‎d multifractal dimensions were computed an‎d analyzed using Fractalyse software. The integrated results from all machine learning algorithms an‎d fractal dimensions consistently demonstrated that geomorphological features, particularly lithology, slope gradient, an‎d proximity to rivers, play a pivotal role in controlling the lan‎dslide pattern within the Dehsheikh catchment. The results showed that the SVM model (Accuracy = 0.897, AUC = 0.899) achieved the best overall performance, exhibited the highest discriminative power, an‎d the RF model (R² = 0.880, MAE = 0.013) demonstrated the greatest stability among the eva‎luated models.According to the model outputs, approximately 27.43% of the catchment area falls within the high-susceptibility class, predominantly concentrated in the southern an‎d southeastern sectors. The computed fractal dimensions (ranging from 1.348 to 1.680) indicate pronounced self-similarity an‎d substantial complexity in the spatial distribution of lan‎dslides, which is consistent with the regionʹs active tectonics an‎d lithological diversity. The monofractal behavior of the lan‎dslide system further reveals that the lan‎dslide pattern remains self-similar across various spatial scales.

5

هيوا علميزاده

166872