Volume 28, Issue 2 (2024)
MJSP 2024, 28(2): 36-67 |
Back to browse issues page
Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:
Karami J, Babaee F, Mahmud Niya P, Sharifi kia M. Analysis and modeling of ground water using spatio-Temporal Data mining and deep learning in order to explanation of it with subsidence hazard. MJSP 2024; 28 (2) :36-67
URL: http://hsmsp.modares.ac.ir/article-21-76871-en.html
URL: http://hsmsp.modares.ac.ir/article-21-76871-en.html
1- Assistant Professor in Remote Sensing and GIS, Tarbiat Modares University, Tehran, Iran. , jl.karami@modares.ac.ir
2- Msc in Remote Sensing and GIS, Tarbiat Modares University, Tehran, Iran.
3- Tarbiat Modares & Msc in Remote Sensing and GIS, Tarbiat Modares University, Tehran, Iran
4- Associate Professor in Remote Sensing and GIS, Tarbiat Modares University, Tehran, Iran.
2- Msc in Remote Sensing and GIS, Tarbiat Modares University, Tehran, Iran.
3- Tarbiat Modares & Msc in Remote Sensing and GIS, Tarbiat Modares University, Tehran, Iran
4- Associate Professor in Remote Sensing and GIS, Tarbiat Modares University, Tehran, Iran.
Abstract: (421 Views)
The deficiency of surface water in arid and semi-arid territories has exacerbated the dependence on groundwater resources, resulting in considerable reductions in groundwater levels. This phenomenon has been particularly pronounced in numerous plains throughout Iran, where the diminution has exacerbated issues related to land subsidence. A comprehensive understanding of groundwater level variations is imperative for enhancing water management strategies and alleviating the associated hazards. A range of statistical, mathematical, and machine-learning methodologies have been utilized to model the dynamics of groundwater aquifers. Recently, deep neural network algorithms have gained prominence in the investigation of surface and groundwater resources, particularly in light of the spatiotemporal characteristics inherent to groundwater.
In the present investigation, a hybrid spatiotemporal data mining framework, denoted as Wavelet-PCA, was employed to analyze data acquired from 44 piezometric wells situated in the Qahavand plain over a span of three decades (1988-2018) for the purpose of elucidating temporal and spatial patterns associated with fluctuations in groundwater levels. Subsequently, a sophisticated deep recurrent neural network architecture incorporating Long Short-Term Memory (LSTM) was implemented to model the time series data resulting from the data mining procedure. Various degrees of wavelet transformation were applied to effectively capture the intricate trends in groundwater levels. The LSTM model exhibited a coefficient of determination (R²) of 0.85 for the training dataset while achieving an R² of 0.62 for the testing dataset.
The research additionally examined regional patterns of land subsidence utilizing radar interferometry data obtained from the Sentinel-1 satellite during the period from 2014 to 2019. The results revealed an average maximum subsidence measurement of 9 centimeters, with the most pronounced subsidence noted in regions that are undergoing the most substantial declines in groundwater levels. This observed relationship between groundwater depletion and land subsidence underscores the necessity for judicious land use planning and the implementation of effective water resource management strategies in analogous regions.
In the present investigation, a hybrid spatiotemporal data mining framework, denoted as Wavelet-PCA, was employed to analyze data acquired from 44 piezometric wells situated in the Qahavand plain over a span of three decades (1988-2018) for the purpose of elucidating temporal and spatial patterns associated with fluctuations in groundwater levels. Subsequently, a sophisticated deep recurrent neural network architecture incorporating Long Short-Term Memory (LSTM) was implemented to model the time series data resulting from the data mining procedure. Various degrees of wavelet transformation were applied to effectively capture the intricate trends in groundwater levels. The LSTM model exhibited a coefficient of determination (R²) of 0.85 for the training dataset while achieving an R² of 0.62 for the testing dataset.
The research additionally examined regional patterns of land subsidence utilizing radar interferometry data obtained from the Sentinel-1 satellite during the period from 2014 to 2019. The results revealed an average maximum subsidence measurement of 9 centimeters, with the most pronounced subsidence noted in regions that are undergoing the most substantial declines in groundwater levels. This observed relationship between groundwater depletion and land subsidence underscores the necessity for judicious land use planning and the implementation of effective water resource management strategies in analogous regions.
Keywords: Spatiotemporal Data Mining, Groundwater Modeling, Land Subsidence, Deep Learning, Ghahavand Plain
Article Type: Original Research |
Subject:
ecological planning and sastaining devlopment
Received: 2024/09/4 | Accepted: 2024/10/5 | Published: 2024/06/30
Received: 2024/09/4 | Accepted: 2024/10/5 | Published: 2024/06/30
Send email to the article author
| Rights and permissions | |
 |
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. |