Linking Groundwater Depletion to Land Subsidence; An Integrated Approach Using D- InSAR, Fuzzy Logic, and Machine Learning in the Isfahan-Borkhar Plain, Central Iran
This study integrates D-InSAR observations with fuzzy logic and machine learning models to map land subsidence potential in Iran's Isfahan-Borkhar plain, demonstrating that Support Vector Machines (SVM) achieve the highest accuracy (R² = 0.93) in linking groundwater depletion to surface deformation for improved hazard mitigation.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the arid heart of Iran, the ground is slowly sinking. This phenomenon, known as land subsidence, occurs when the earth's surface gradually drops, often because water is being pumped out from underground aquifers faster than nature can replace it. As the water leaves the spaces between soil grains, the weight of the land above compresses the soft, clay-rich sediments, causing the surface to settle. While this process happens silently, its consequences are loud and damaging: roads crack, buildings tilt, and vital infrastructure like airports and sewage networks face collapse. In regions where water is scarce and demand is high, understanding exactly where and how fast the ground is sinking has become a critical task for engineers and city planners.
For decades, scientists have used satellite technology to watch the earth move. One powerful method involves sending radar signals from space to the ground and measuring the tiny changes in the time it takes for the signal to bounce back. By comparing images taken months apart, researchers can map the vertical movement of the land with incredible precision. However, knowing that the ground is moving is only half the battle; the harder question is predicting where it will happen next and why. Traditional methods for making these predictions often rely on simple calculations that add up different risk factors, such as soil type, how much water is being pumped, and the depth of the underground layers. While useful, these straightforward calculations sometimes miss the complex, messy ways these factors interact in the real world.
A team of researchers from the University of Tehran and Tarbiat Modares University set out to improve these predictions for the Isfahan-Borkhar plain, a vital region in central Iran that hosts major cities, industrial zones, and the Shahid Beheshti Airport. They wanted to move beyond simple addition and see if they could use advanced computer models to better understand the relationship between groundwater depletion and the sinking ground. Their approach was a three-step process that started with a basic map and refined it using increasingly sophisticated techniques. First, they created a standard risk map by weighing seven different factors: the type of rock and soil underground, how the land is used, the rate at which water is pumped out, how much rain recharges the aquifer, the thickness of the sediment layers, the distance from geological faults, and how much the water table has dropped. This initial map gave them a general idea of where the risk was highest.
To make this map more accurate, the researchers brought in a new layer of reality: actual measurements of the ground sinking. Using satellite data collected between April 2023 and April 2024, they mapped the exact amount of subsidence across the plain. They then used these real-world measurements to "condition" their initial risk map, essentially teaching the model to align its predictions with what was actually happening on the ground. With this refined target in place, they tested three different types of fuzzy logic models. These models are designed to handle uncertainty and gradual changes, much like how a human might describe a temperature as "warm" rather than a specific number, allowing for a more nuanced understanding of how the risk factors blend together. They also tested two powerful artificial intelligence tools: one that mimics the way the human brain processes information through connected layers, and another that finds the best mathematical boundary to separate different levels of risk.
The results showed a clear progression in accuracy. The initial, simple map provided a decent starting point, correctly identifying broad areas of risk, but it struggled to match the precise details of the satellite measurements. When the researchers introduced the fuzzy logic models, the predictions became much sharper, capturing the subtle variations in the landscape. The most successful model, however, was the artificial intelligence tool based on support vector machines. This model achieved the highest level of agreement with the satellite data, successfully reproducing the complex patterns of subsidence that the simpler methods missed. It identified specific high-risk zones with remarkable clarity, pinpointing areas in the southwestern part of the plain, northern Isfahan, and the eastern side of the airport as the places where the ground is most likely to continue sinking.
The study confirms that while traditional methods can offer a general overview, they are not enough to protect critical infrastructure in a rapidly changing environment. By combining satellite observations with advanced machine learning, the researchers have created a tool that can see the ground's behavior with much greater fidelity. This approach does not just tell us that the land is sinking; it helps us understand the specific combination of soil, water, and human activity that drives the process in each neighborhood. For the city of Isfahan and its surrounding plains, this level of detail is essential. It allows authorities to focus their monitoring and mitigation efforts on the exact spots where the ground is most vulnerable, helping to safeguard the airport, the power plants, and the homes built on this shifting foundation. The work demonstrates that when we let computers learn from the complex reality of the earth, rather than just applying simple rules, we gain a much clearer picture of the risks we face.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.