Extended Abstract:
Introduction:
  Vegetation is an indicator of ecosystem health and plays an important role in maintaining ecological balance, regulating the water cycle, and facilitating the flow of matter and energy. The dynamics of vegetation caused by environmental stresses and human activities are increasing and require continuous long-term monitoring and surveillance. With the development of remote sensing technology, remote sensing images have been widely used in vegetation survey work.
Materials & Methods:
  In this study, multispectral images from the Landsat satellite were used, including time series from 1994 (Landsat 5), 1999 (Landsat 5), 2004 (Landsat 5), 2009 (Landsat 7), 2014 (Landsat 8), 2019 (Landsat 8), and 2024 (Landsat 8). After obtaining the desired time series images from the Google Earth Engine system, the desired images were called in the ENVI 5.6 software environment and cropped using the shapefile of the study area. Then, the desired images were saved in TIFF format. Pre-processing operations, including radiometric and atmospheric corrections, were performed on all images. eCognition software was used for segmentation and also classification based on object-oriented fuzzy operators. In order to perform object-oriented processing of images, image segmentation was performed at this stage. In the segmentation stage, satellite images were segmented based on different scales, shape factors, and compression factors to achieve the optimal scale, shape factors, and compression factors, and after achieving the optimal segmentation conditions, two classes were created under the title of vegetation cover and non-vegetation cover. Subsequently, the training points required for each class were determined to train the algorithm used. Then, using fuzzy operators, including AND, OR, MGE, MAR, MGWE, and ALP, the classification process was performed based on object-oriented processing of satellite images.
Results & Discussion
  With the aim of producing image objects, the segmentation process was performed at different scales, shape factors and compression factors to achieve optimal segmentation. At this stage, the multi-resolution segmentation method was used for image segmentation. The most optimal segmentation included a scale of 40, a shape factor of 0.3 and a compression factor of 0.7. Because at scales higher than 40, the creation of image objects was not performed well. After the segmentation and generation of image objects were completed, two classes of vegetation cover and non-vegetation cover were created, and a number of image objects produced in the segmentation stage were selected as training points and a number were selected as ground reality points. At this stage, classification was performed based on the nearest neighbor algorithm and by selecting parameters such as mean, standard deviation, skewness, and texture for each band.
Conclusion:
  The results showed that the fuzzy AND operator, with an overall accuracy of 97.38% and a kappa coefficient of 0.9818, performed best in monitoring vegetation cover in the Karkheh watershed. A 30-year Landsat time series analysis revealed an approximately 18% decline in vegetation cover. This trend, driven by both human and natural factors, poses a serious threat to ecological sustainability and local communities, highlighting the urgent need for comprehensive policies on vegetation protection and restoration.