نوع مقاله : مقاله پژوهشی
موضوعات
عنوان مقاله English
نویسندگان English
Aim: This study aims to identify flood-prone areas in the semi-arid region of Tabas using LT, K-SVM, and KNN machine learning algorithms and to examine the most critical factors affecting flooding in the region.
Materials & Methods: First, 288 flood points were collected from the Natural Resources and Watershed Management Organization of South Khorasan Province. In the second step, the data were divided into two parts: 70% for training and model development, and 30% for validation. Then, 12 key factors in assessing flood potential were examined, and the accuracy of the algorithms was evaluated using various statistical indicators.
Findings: The flood zoning maps indicate that, according to the LT algorithm, the northern and western regions face a high risk of flooding, estimated at 43%. The western and southern areas are at very high risk (36%). In contrast, the K-SVM algorithm indicates that the central, western, and southern parts of the region have a high risk of flooding (59%), whereas the eastern, northern, and central areas face a very high risk of flooding (41%). According to the KNN algorithm, the northern, central, and eastern regions show a high flood risk, estimated at 38%. The northwestern and western regions are classified as very high-risk zones, accounting for 27% of the total area.
Conclusion: Flood zoning maps indicate that high-risk areas are located near waterways, where subsurface water flow generates moisture fronts. Hydrological factors, such as flood-related indices, indicate that the western, central, northern, northwestern, and southeastern regions are the main flood-prone areas.
Innovation: The present study uses machine learning algorithms, which bring significant innovations. This approach not only increases the accuracy of predictions but also enables the use of multi-purpose data and the identification of complex patterns. The outcomes of this method will play a significant role in flood management and in preserving the health of aquatic ecosystems.
کلیدواژهها English
Extended Abstract
1. Introduction
In recent decades, flooding has become one of the most widespread natural disasters globally. In Iran, due to specific climatic and geographical conditions, ignoring the flood phenomenon can have irreparable consequences. Accurate estimation, assessment, and zoning of flood-prone areas are crucial for minimizing potential risks and damages. Arid regions, characterized by heavy rainfall, poor watershed management, limited vegetation cover, and impermeable soils, experience destructive and hazardous floods. Surface runoff mechanisms in these regions, due to reduced water penetration in the soil, lead to rapid accumulation of water on the ground surface. This causes sudden and severe floods that can have devastating effects on the environment and communities. These unique conditions in arid regions present significant management and environmental challenges that demand a comprehensive understanding and effective solutions for flood management. However, it is possible to reduce the damage and loss caused by floods by predicting the areas at risk. These methods involve locating potential flood areas and evaluating their impacts to mitigate destructive effects. New techniques in the field of GIS (Geographic Information Systems) and remote sensing have led to significant developments in flood prediction modeling.
2. Materials and Methods
In this study, 288 flood-prone locations were first collected from the Natural Resources and Watershed Management Organization of South Khorasan Province. Then, these locations were divided into two groups: 70% of the data for training and model implementation, and 30% for validation. Flood-prone areas, as well as locations that had no flood history, were randomly selected and divided into two groups. In the next step, the factors affecting the occurrence of flooding were examined. In the present study, several variables involved in the flood phenomenon were considered independent variables, and their relationship with flood zones was analyzed. Then, 12 key factors in flood potential assessment were examined. The IGR index was also used to evaluate the predictability of quantitatively effective factors. Then, the multilinear factor test was used to assess the linearity of the variables. Also, five different statistical methods have been used to validate and evaluate the algorithms, including sensitivity, specificity, positive and negative indices, and ROC-AUC analysis.
3. Results and Discussion
The results obtained from the IGR index show that land use, precipitation, slope, lithology, distance from the river, and the SPI index are the most critical factors influencing flooding in the study area. In the multiple collinearity test, the results show that all flood factors have VIF values less than 10 and a Tolerance greater than 0.2. This indicates the lack of interference and overlap of factors in this field. The performance evaluation results of the algorithms indicate that the LT algorithm achieves the highest SST, SPF, PPV, and NPV values in both datasets. The K-SVM algorithm is ranked second, with SST and PPV values of 0.77 and 0.71, respectively. The KNN algorithm demonstrates strong performance, achieving the highest SST (0.68) and PPV (0.65) values among the algorithms studied. These results indicate that all three algorithms show good accuracy and performance in analyzing the dataset. The results of the flood zoning maps suggest that areas with high and very high flood risk are located on slopes exceeding 20 degrees, which has a significant impact on the occurrence of flooding in the region. In addition to the slope of the main river, the slope of the land is also a key factor in the intensification of flooding in the region. In the study area, rivers and their tributaries follow their paths along the slopes. Flood-prone areas are located in low-lying areas. As a result, the rate of erosion and transport of alluvial and sedimentary materials in these areas increases sharply. Areas near the river network are at high risk due to persistent high water flow and spring floods. Studies show that parts of the region with geologically and lithologically resistant surface formations, low permeability, and sparse vegetation cover are prone to flooding.
4. Conclusions
According to the results of flood zoning maps, areas with high flood risk are located near the confluence of watercourses. This phenomenon is caused by the flow of subsurface water from rivers to adjacent slopes, resulting in the formation of moisture in the soil and weakening its resistance. This trend, together with the high density of drainage, exacerbates the risk of flooding in these areas. The western, central, northern, and southeastern regions are known to be prone to flooding due to topographic features such as altitude, steep slopes, high rainfall, and dense watercourse networks. Human activities, such as excessive grazing of livestock and the destruction of vegetation, have reduced soil permeability and increased runoff and flood intensity in these areas. In contrast, low-risk flood areas usually have dense vegetation, are far from significant watercourses, and have a less dense drainage network. These areas have different conditions in terms of soil, altitude, and slope, and, for this reason, are less exposed to flooding. The findings of this study contribute to the preservation of natural resources, soil, and vegetation, as well as to effective crisis management.
5. Acknowledgment & Funding
· The authors are thankful to all interview participants for supporting this research.
· The manuscript did not receive a grant from any organization.
6. Conflict of Interest
