1- university
Abstract: (55 Views)
Extended Abstract
Background: Land subsidence (LS) is a morphological phenomenon occurring worldwide. In general, land subsidence can lead to a decrease in the Earth's surface in all regions of the globe, caused by either natural or human activities. Land subsidence and human activities have been intertwined throughout history. Recently, more than 150 major countries, including Mexico, Australia, Colombia, China, and the United States, have reported instances of land subsidence. This phenomenon affects many regions of the world and can result from various natural or human factors, particularly the extraction of groundwater, mining activities, and the dissolution of mineral resources.
Numerous reports emphasize that groundwater resources in extensive areas of central, eastern, and southern Iran are utilized as the sole source of water for agricultural, drinking, and industrial purposes. Iran consists of six major watersheds and 609 plains, with approximately 267 of them facing water scarcity. Given the significance of the Darab region in Fars Province as a key area prone to land subsidence, the objectives of this study are: (1) spatial modeling of land subsidence in the Darab plain of Fars Province; (2) evaluating machine learning models individually for spatial modeling of land subsidence; and (3) identifying key factors affecting land subsidence, including the effects of climate change and land use changes in the study area.
Methods: Sentinel-2 images from the years 2015 and 2024 were utilized for image classification to identify land use patterns and create Land Use/Land Cover (LULC) maps for the Darab region. The land uses, such as orchards, pastures, agricultural lands, urban areas, and barren lands, are observable in the region and have been analyzed for their impact on land subsidence.
The data used to extract the land subsidence map in this study included the C band of the ascending Sentinel-1A, the descending Sentinel-1A band, Interferometric Wide (IW) images, and strip mode images collected before and after subsidence, selected as empirical data. Climate data on temperature and precipitation from six existing stations in the area for the years 2015 to 2024 were prepared and interpolated using the IDW (Inverse Distance Weighting) method within ArcGIS, with a spatial resolution of 30 meters for digital modeling. Additionally, groundwater level maps of the plain were interpolated using the IDW method with the assistance of ArcGIS, based on piezometric well statistics from 2002 to 2021. All informational or auxiliary layers were prepared in raster format with a pixel size of 10 meters.
To perform the modeling process, the satellite imagery and DEM (Digital Elevation Model) of the area were first extracted using ArcGIS 10.7, after which the satellite images and DEM were imported into the SAGAGIS software. The obtained data were then input into JMP software and the R environment for random forest modeling.
Results: Seven land subsidence maps from 2015 to 2024 were extracted using Snap software. Using ArcGIS, an average was calculated from the images, revealing an average land subsidence of about 11 cm during this period. Based on the land use change map from 2015 to 2024 and its correlation with the corresponding land subsidence (LS) map for the same period, the greatest subsidence was observed in areas designated as orchards, residential, and agricultural lands.
According to the results, the parameters that had the most significant impact on the modeling were groundwater level, temperature, and precipitation, respectively, with wetness index being noteworthy. Among the machine learning models assessed in this study, the Random Forest (RF) model provided the best results, showing a higher coefficient of determination ((R² = 0.95 (Training), 0.93 (Validation)) and a lower Root Mean Square Error (RMSE = 0.001 (Training), 0.002 (Validation)).
Conclusion: In the present study, an increase in groundwater depth has a direct relationship with the increase in land subsidence in the region. The parameters that had the most significant impact on the modeling, in order, were temperature and precipitation, groundwater level, and the topographic wetness index. Based on the regression maps and validation parameters, among the predictive models used in this study, the Random Forest (RF) model provided the best results, showing a higher coefficient of determination (R² = 0.95 (Training), 0.93 (Validation)) and a lower Root Mean Square Error (RMSE = 0.001 (Training), 0.002 (Validation)).
Creating such conditions and the situation surrounding land subsidence and its hazardous consequences for the country requires a national commitment to acknowledge and address this phenomenon as a significant threat that could lead to disaster. Understanding this issue necessitates an increase in studies aimed at fully identifying prone areas as part of a national plan.
Background: Land subsidence (LS) is a morphological phenomenon occurring worldwide. In general, land subsidence can lead to a decrease in the Earth's surface in all regions of the globe, caused by either natural or human activities. Land subsidence and human activities have been intertwined throughout history. Recently, more than 150 major countries, including Mexico, Australia, Colombia, China, and the United States, have reported instances of land subsidence. This phenomenon affects many regions of the world and can result from various natural or human factors, particularly the extraction of groundwater, mining activities, and the dissolution of mineral resources.
Numerous reports emphasize that groundwater resources in extensive areas of central, eastern, and southern Iran are utilized as the sole source of water for agricultural, drinking, and industrial purposes. Iran consists of six major watersheds and 609 plains, with approximately 267 of them facing water scarcity. Given the significance of the Darab region in Fars Province as a key area prone to land subsidence, the objectives of this study are: (1) spatial modeling of land subsidence in the Darab plain of Fars Province; (2) evaluating machine learning models individually for spatial modeling of land subsidence; and (3) identifying key factors affecting land subsidence, including the effects of climate change and land use changes in the study area.
Methods: Sentinel-2 images from the years 2015 and 2024 were utilized for image classification to identify land use patterns and create Land Use/Land Cover (LULC) maps for the Darab region. The land uses, such as orchards, pastures, agricultural lands, urban areas, and barren lands, are observable in the region and have been analyzed for their impact on land subsidence.
The data used to extract the land subsidence map in this study included the C band of the ascending Sentinel-1A, the descending Sentinel-1A band, Interferometric Wide (IW) images, and strip mode images collected before and after subsidence, selected as empirical data. Climate data on temperature and precipitation from six existing stations in the area for the years 2015 to 2024 were prepared and interpolated using the IDW (Inverse Distance Weighting) method within ArcGIS, with a spatial resolution of 30 meters for digital modeling. Additionally, groundwater level maps of the plain were interpolated using the IDW method with the assistance of ArcGIS, based on piezometric well statistics from 2002 to 2021. All informational or auxiliary layers were prepared in raster format with a pixel size of 10 meters.
To perform the modeling process, the satellite imagery and DEM (Digital Elevation Model) of the area were first extracted using ArcGIS 10.7, after which the satellite images and DEM were imported into the SAGAGIS software. The obtained data were then input into JMP software and the R environment for random forest modeling.
Results: Seven land subsidence maps from 2015 to 2024 were extracted using Snap software. Using ArcGIS, an average was calculated from the images, revealing an average land subsidence of about 11 cm during this period. Based on the land use change map from 2015 to 2024 and its correlation with the corresponding land subsidence (LS) map for the same period, the greatest subsidence was observed in areas designated as orchards, residential, and agricultural lands.
According to the results, the parameters that had the most significant impact on the modeling were groundwater level, temperature, and precipitation, respectively, with wetness index being noteworthy. Among the machine learning models assessed in this study, the Random Forest (RF) model provided the best results, showing a higher coefficient of determination ((R² = 0.95 (Training), 0.93 (Validation)) and a lower Root Mean Square Error (RMSE = 0.001 (Training), 0.002 (Validation)).
Conclusion: In the present study, an increase in groundwater depth has a direct relationship with the increase in land subsidence in the region. The parameters that had the most significant impact on the modeling, in order, were temperature and precipitation, groundwater level, and the topographic wetness index. Based on the regression maps and validation parameters, among the predictive models used in this study, the Random Forest (RF) model provided the best results, showing a higher coefficient of determination (R² = 0.95 (Training), 0.93 (Validation)) and a lower Root Mean Square Error (RMSE = 0.001 (Training), 0.002 (Validation)).
Creating such conditions and the situation surrounding land subsidence and its hazardous consequences for the country requires a national commitment to acknowledge and address this phenomenon as a significant threat that could lead to disaster. Understanding this issue necessitates an increase in studies aimed at fully identifying prone areas as part of a national plan.
Type of Study: Research |
Subject:
سنجش از دور و سامانه های اطلاعات جغرافيايی
Received: 2024/10/1 | Accepted: 2025/02/3
Received: 2024/10/1 | Accepted: 2025/02/3
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |
