Extended Abstract
Introduction
The coastline is defined as the contact line between the land and the lake. Any fluctuation in the lake level creates significant changes in the beach. Extreme changes in the status of different wetlands in the world have occurred mainly through the expansion of urbanization and the growth of agriculture. Changes in the coastline are supposedly essential in environmental management. This is important in coastal erosion monitoring, flood forecasting, and water resources assessment. Remote sensing images are the most reliable type of information data in the world. Various satellite sensors have been placed in Earth’s orbit since 1960 to monitor the Earth's surface. Regular monitoring of lakes can provide a basis for understanding human impacts and managing the lake more efficiently. Nowadays, remote sensing and geographic information systems have presented new methods for ecosystem management. These systems allow the development of an automated system for extracting specific elements such as coastlines. GIS system is a useful tool in spatial analysis, accurate description of spatial correlations, and providing effective outputs. Remote sensing systems have been widely used to analyze and evaluate images in different intervals and have made it possible to examine beaches by comparing old and new images. Researchers so far have used no learning algorithm to check the changes in the coastline and the level of Zarivar Lake. This research uses the maximum probability model (MLC) and support vector machine (SVM) to evaluate the change in the coastlin e and water level of Zarivar Lake.
 
Methodology
At first, 3 Landsat 5 and 8 satellite images were selected for Zarivar Lake in the periods of May 1993, 2010, and 2023. The amount of rainfall reaches its peak in May. Therefore, the processing is done in the best conditions so that the maximum rate of expansion and retreat of the lake water level can be obtained in 30 years. The specifications of the images should be free of cloud cover to assess shoreline changes over time. Satellite images depict the state of Zarivar Lake from 1993 to 2023. The 1993 TM sensor map of the Landsat 5 satellite was supposedly a base map to obtain and compare the amount of changes in the years under study. This research uses a multivariate spatial analysis tool. At first, the lake water level was obtained for all three images through the maximum probability method. The support vector machine technique was also used to classify the area under study and gain the amount of surface reduction. The extracted vector layers determined the regression of the lake since 1993. This research used the Kappa coefficient to evaluate the accuracy and reliability of each model and calculated its values ​​ for each of the desired years. It estimated, as the last step, the correlation between the NWI and WRI spectral indices and the bands used in their preparation and drew a scatter diagram to determine the relationship between the spectral indices.
Results and Discussion
The water level of the lake in 1993 is supposedly a basis for evaluating the amount of regression. The total area of ​​the lake according to the MLC model in 1993 was equal to 5.4 square kilometers. The surface area of ​​the lake in 2010 was approximately 8.28 square kilometers, which has increased by 52.7% compared to 1993. The area of ​​surface water reached 21.6 square kilometers in 2023, which has decreased by 25% compared to 2010. The maximum retreat occurred in the southern part of the lake, and a significant deformation was observed in the western part of the lake. Both methods have shown similar results. NWI and WRI spectral indices besides MLC and SVM algorithms were also used to investigate changes in the water’s broad stretch of Zarivar Lake. The above indices correctly separate the blue broad stretches from other ones, but it is necessary to reclassify the above maps to gain the area of ​​the blue broad stretches. As the classified map shows, it is possible to estimate the area of ​​Zarivar Lake in spectral indices. The area of ​​the lake has increased from 8.28 square kilometers to 9.26
 
square kilometers and has increased by 11.8 percent according to the NWI spectral index. The area of ​​the lake in the WRI index has increased from 2.8 square kilometers to 9.06 square kilometers, showing an increase of 10.48%. This shows that the lake level has increased in both indices and the resulting values ​​are very close to each other and only differ by 1.32%. A correlation diagram was established between indicators and the spectral bands used in their preparation to investigate their correlation. The results show that the Pearson correlation coefficient between the two spectral indices NWI and WRI is equal to 0.97 percent and the R2 coefficient is equal to 0.95.
Conclusion
New techniques based on learning algorithms and spectral indices make it possible to investigate the changes of water broad stretches in intervals of two years. Free satellite images such as Landsat 5 and 8 enable long-term observations. Different spectral bands can show the reflective behavior of water and help to identify them. Likewise, the changes in water broad stretches in semi-arid land like Iran are of great importance. Thus, changes in the water area of Zarivar Lake were investigated in the period 1993-2023. The area under study was divided into three water broad stretches, agricultural lands, and barren lands, through the SVM and MLC algorithms. It showed that the water broad stretch in the SVM model for the years 1993, 2010, and 2023 is equivalent to 8.62, and 10.13, respectively and it was 12.75 square kilometers. The above values in the MLC model are equivalent to 5.42, 8.28, and 6.21. Both models showed an increase in the lake level, but the obtained values had significant differences. For example, the changes in 2023 showed different values for both models. The results of the Kappa coefficient for the SVM model were equal to 0.94 in 2023.