چكيده لاتين
Wetlands as the kidneys of the earth play a vital role in environmental balance, biodiversity maintenance and water resource management. Thus, land cover mapping and detection of their changes through modern methods, especially remote sensing, are of particular importance. In Iran, many international wetlands have been registered. Among them, Mighan wetland in Markazi province and Gavkhoni wetland in Isfahan province are very important. Therefore, this study aimed to study the land cover zoning of these two wetlands in six classes of water, vegetation, bare soil, salinized, mountainous areas and urban areas and then their changes during 2017 to 2023. The data and images were obtained from satellite images of radar Sentinel-1 satellite and images of optical Sentinel-2 satellite. Moreover, digital elevation model (DEM) obtained from SRTM mission was used as an auxiliary data. Extracted features from radar satellite images were include VV and VH polarization data and extracted features from optical satellite images were include normalized difference vegetation index (NDVI), normalized difference water index (NDWI), modified NDWI (MNDWI), soil-adjusted vegetation index (SAVI), bare soil index (BSI) and water ratio index (WRI). Also, the elevation data was used from the DEM. For mapping, machine learning algorithms such as decision tree (DT), random forest (RF) and support vector machine (SVM) were used. For classification of Gavkhooni wetland, 11 strategies and for classification of Mighan wetland, 4 strategies were considered from combination of optical, radar and altitudinal features. The samples were collected through high quality images from Google Earth software and field operations which 70 % of them were randomly selected as training samples and 30 % as the test samples. The results indicated that RF algorithm had the highest accuracy and efficiency in land cover mapping of both wetlands. In the classification of Gavkhooni wetland, the use of all optical and radar features and altitudinal gradient obtained highest overall accuracy (between 93% and 97%). By contrast, the use of only the main bands of optical images and radar and altitudinal obtained highest overall accuracy (between 92% to 98%) in Mighan wetland. The most important note in classification of both wetlands was the advantage of using combination of optical and radar features along with DEM which increased accuracy (sometimes up to 40%) than single use of each feature. The results of changes in the period between 2017 and 2023 showed that a decrease of 93% and 7% in Gavkhooni and Mighan wetlands, respectively at water level. It was due to severe drought in this period. By contrast, 23% and 5% increase of salinized areas in these two wetlands showed the development of desertification process and reducing the land utilization in these two regions. The findings of this research can be used as a basis for optimal management and protection of threatened wetlands.
