| چکیده انگلیسی مقاله |
Extended Abstract
1- Introduction
Water bodies are crucial in Earth's ecosystems, human life, agriculture, and various industries. However, in recent years, global challenges like urbanization, climate change, and over-extraction of groundwater have significantly affected these vital resources. Iran, in particular, is dealing with a severe water crisis, made worse by reduced precipitation and changing climate patterns, leading to visible declines in its lakes, rivers, and wetlands. Consequently, effective monitoring and management of water bodies are essential to battle this crisis. Remote sensing (RS) technology provides a cost-effective, long-term solution for large-scale environmental monitoring. The Google Earth Engine (GEE) cloud platform enables rapid and accurate analysis of satellite images, facilitating effective monitoring of temporal changes in water bodies. GEE's ability to process freely accessible satellite data, such as Sentinel-2 imagery, makes it particularly useful and efficient for monitoring surface water area through spectral indices specifically designed for water detection and extraction. Therefore, the main objective of this study is to monitor the 6-year time series of changes in Gorgan Gulf and Miankaleh Wetland using Sentinel-2 imagery in the GEE platform. The study utilizes GEE’s capabilities to provide accurate, up-to-date information, which is important for water resource management in the region.
2- Materials & Methods
2-1- Study Area
The present study focuses on Gorgan Gulf and Miankaleh Wetland, both located below the Caspian Sea in northern Iran.
2-2- Data
To extract water bodies in the current study, Sentinel-2 imagery was used along with 440 validation samples (220 water and 220 non-water) to assess the accuracy of the spectral index classification. The validation samples were extracted from the 6-year mean (2018–2024) RGB Sentinel-2 image within the study area.
2-3- Methodology
A spectral index-based approach was employed to extract water bodies using Sentinel-2 satellite images. The spectral indices used to identify water bodies in the study area include the Modified Normalized Difference Water Index (MNDWI), Water Ratio Index (WRI), New Water Index (NWI), and Enhanced Water Index (EWI). The study was implemented and executed in GEE. After preprocessing Sentinel-2 images using the SCL and QA60 bands, a mean reducer function was applied to generate seasonal composite images for each year. Based on these seasonal composites, the spectral indices (MNDWI, EWI, NWI, and WRI) were calculated. Optimal threshold values for each index were determined using the Edge Otsu thresholding algorithm. The steps of the Edge Otsu thresholding method include binary thresholding with initial thresholds, Canny edge detection, edge length filter, edge buffering, sampling within the buffer, histogram creation, and finally applying Otsu's method to calculate optimal thresholds. After calculating and applying the optimal thresholds to the spectral index images, binary maps (water and non-water classes) were generated. Finally, the accuracy of the extracted water bodies was assessed both quantitatively and qualitatively.
3- Results & Discussion
According to the quantitative evaluation results, the WRI spectral index achieved the highest accuracy with an overall accuracy (OA) of 99% and a Kappa coefficient (KC) of 0.99, while the MNDWI index had the lowest accuracy, with an OA of 98% and a KC of 0.96. When applying the default threshold values, the WRI and NWI indices had the highest and lowest accuracy metrics, with overall accuracies of 94% and 86%, and Kappa coefficients of 0.88 and 0.65, respectively. The results also suggest that the NWI and EWI indices can be used interchangeably due to their high accuracy similarity (98% for optimal threshold and 83% and 85% for default threshold). Qualitative and visual accuracy assessments confirmed the quantitative accuracy values of the different spectral indices in extracting water bodies. Furthermore, the study shows a significant reduction in the area of the studied water bodies over the past six years, especially in between the years 2021 and 2022. The annual mean surface water area from 2018 to 2024 steadily declined, with areas of 399.73, 381.52, 374.18, 357.99, 311.63, and 293.60 square kilometers, respectively. In addition, the annual rate of change for the study period was estimated at -4.55%, -1.92%, -4.32%, -12.94%, and -5.78%. Based on the analyses, the most significant reduction in water area occurred in the Miankaleh Wetland. Visual analysis of the results indicated that the fall of 2023 had the smallest surface water area, while the summer of 2018 had the largest.
4- Conclusion
Gorgan Gulf and Miankaleh Wetland, two of the most important water resources of the Caspian Sea and Iran, have faced serious challenges from drought and reduced surface water in recent years, highlighting the need for continuous and accurate monitoring. Therefore, this study utilized Sentinel-2 satellite imagery from 2018 to 2024 to extract water bodies using spectral indices and the Edge Otsu thresholding algorithm. The proposed method was implemented in GEE, which enables rapid cloud-based calculations. The findings of this study demonstrate that the Edge Otsu thresholding method achieved optimal accuracy in water body extraction compared to default thresholds (such as 0 and 1). In addition, the spatiotemporal changes in the study area were analyzed, revealing a significant decrease in the water area, particularly in the Miankaleh Wetland. Thus, this study illustrates the effectiveness of the Edge Otsu algorithm in improving accuracy and suggests that combining spectral indices with machine learning models in future research could further enhance water body extraction. |