|
Evaluation of the Stability of the Maximum Entropy Algorithm for Modeling the Potential of Groundwater Springs in the Farahroud watershed, Tehran province.
|
Narges Javidan *    , Ataollah Kavian |
|
|
|
Abstract: (196 Views) |
Abstract
Introduction
A better understanding and assessment of groundwater resources are key principles in water resource management. Identifying areas susceptible to groundwater extraction can provide more water of better quality and reduce the costs associated with drilling and extracting groundwater. There are various methods for determining groundwater potential; however, it is noteworthy that the existing methods, despite their high accuracy, require significant skill, time, and expense. Therefore, utilizing new spatial modeling algorithms that incorporate environmental parameters (such as precipitation and lithology) and geographic information systems to determine groundwater potential represents a significant advancement in the field of groundwater resource management. The objective of this study is to evaluate the stability of a machine learning algorithm used to generate a groundwater spring potential map in the Farahroud watershed, Tehran Province. In this region, the increasing socio-economic problems due to drought and the depletion of water resources have heightened the need for greater attention to underground water resources. The Maximum Entropy Model (MaxEnt) is one of the advanced data mining models that has been employed in various research studies due to its advantages and capabilities. Accordingly, this advanced and well-regarded model was chosen as the benchmark and basis for its ability to predict sensitivity.
Methodology
After conducting spatial modeling, a groundwater potential prediction map was generated. In all three model runs, the accuracy of the model was calculated to be above 87% during the training phase. In the validation phase, accuracy ratios exceeding 79% were achieved. These results indicate very good performance in the training phase and good performance in the validation phase. The results, based on the area under the receiver operating characteristic (ROC) curve, demonstrated that the model performed with greater accuracy (95.0% in the training stage and 92.0% in the validation stage) when modeled with an 80:20 ratio. The superior accuracy and stability of the model with a training-to-validation ratio of 80/20 can be attributed to the increased learning capacity facilitated by a larger training set, better generalization to unused data, and reduced variability in model performance. These factors collectively contribute to a more reliable and effective modeling approach for predicting groundwater spring potential in the Farahroud watershed. Furthermore, based on the results obtained with this ratio, the model demonstrated higher stability. According to the groundwater spring potential map, approximately 16.78% and 6.29% of the area were classified as high and very high potential zones, respectively. Additionally, based on the Jackknife test, the topographic moisture index, drainage density, and distance from faults were identified as the most significant predictive factors for groundwater spring occurrence in the study area.
Results and Discussion
After performing spatial modeling, a groundwater potential prediction map was prepared. In all three model runs, the accuracy of the model was calculated to be above 87% in the training phase. In the validation phase, accuracy ratios above 79% were obtained. These results indicate very good and good performance in the training and validation phases, respectively. The results, based on the area under the receiver operating characteristic (ROC) curve, showed that the model performed with higher accuracy (95.0% in the training stage and 92.0% in the validation stage) when modeled with an 80:20 ratio. The superior accuracy and stability of the model for a training to validation ratio of 80/20 can be attributed to the increased learning capacity facilitated by a larger training set, better generalization to unused data, and reduced variability in model performance. These factors collectively contribute to a more reliable and effective modeling approach for predicting groundwater spring potential in the Farahroud watershed. Furthermore, based on the results obtained with this ratio, the model demonstrated higher stability. According to the groundwater spring potential map, approximately 16.78% and 6.29% of the area were classified as high and very high potential zones, respectively. Additionally, based on the Jackknife test, the topographic moisture index, drainage density, and distance from faults were identified as the most significant predictive factors for groundwater spring occurrence in the study area.
Conclusion
The evaluation of the MaxEnt model's accuracy and stability indicated that the model maintains high predictive power across different scenarios and exhibits robustness to changes in input data. This robustness is critical for practical applications, as it demonstrates that the model can reliably inform water management strategies even under fluctuating environmental conditions. By utilizing a diverse set of input parameters, including geological composition, topographic features, and climatic factors, we demonstrated that the MaxEnt model captures the complex relationships governing the occurrence of groundwater. This research not only identifies the interplay between key environmental factors and spring events but also provides a framework for future studies to further refine these models. We hope that our findings will stimulate additional research on the use of advanced algorithms in hydrological modeling and contribute to more informed decision-making processes regarding water conservation and management practices. Extending this research beyond the Farahroud watershed will enhance the understanding of groundwater resources in different geological and climatic regions and underscore the broad benefits of utilizing the maximum entropy algorithm in hydrology.
|
Article number: 2 |
| Keywords: Area under the curve, Geographic information system, Maximum Entropy model, Sample ratio, Stability |
|
|
Full-Text [PDF 1309 kb]
(35 Downloads)
|
Type of Study: Research |
Subject:
Special
Received: 2024/11/12 | Accepted: 2025/02/10 | Published: 2025/07/29
|
|
|
|
|
|
|
| Send email to the article author |
|
|
|
| Add your comments about this article |
|
|
|
