علوم محیطی (Apr 2024)
Integrating Sentinel 1 and 2 Satellite Data with Spectral Indices to Improve Classification Methods (Anzali Wetland)
Abstract
Introduction: Technical limitations in classifying heterogeneous wetland environments, characterized by diverse vegetation cover, land use, and species diversity, often lead to interference in classification results and reduced accuracy in differentiating vegetation classes within wetland ecosystems. There is limited research available to improve classification methods in wetland environments. The main objective of this study is to investigate the combination of multi-spectral and radar data in improving the classification methods of wetland environments and to provide a method for fine separation of different plant covers in these biodiversity environments. In order to better examine the changes of the spectral index during a year, the open-source system of Google Earth Engine is used so that the spectral behavior of the phenomena during the year can be accurately studied. Material and Methods: In this study, a combination of Sentinel-1 and Sentinel-2 data was used as the first data series, and a combination of Sentinel-2 data with spectral indices such as NDVI, SAVI, and mNDWI was used as the second data series. The best image for each season (summer, autumn, winter, and spring) from 2016 to 2022 was selected to create classification maps and examine detailed changes in the wetland. For image classification, training areas were selected based on field sampling, combining satellite imagery and Google Earth images. Classification was performed using three supervised algorithms: Support Vector Machine, Artificial Neural Network, and Maximum Likelihood. Also, the index map was prepared in the Google Earth Engine system and the indices were calculated using the ready-made products available in this system and were reviewed monthly for one year. To ensure the classification and to evaluate the classification accuracy, the most common accuracy estimation parameters, overall accuracy, producer accuracy, user accuracy and Kappa coefficient were used. Results and Discussion: The results indicated that the combination of Sentinel-1 and Sentinel-2 data yielded better results compared to the combination of Sentinel-2 data with spectral indices. The overall accuracy and Kappa coefficient for the four periods were 92.99%, 87.43%, 83.80%, and 97.90% (in 2016, 2017, January 2022, and July 2022, respectively) when using the combination of Sentinel-1 and Sentinel-2 data, which were significantly higher than the results obtained with the combination of Sentinel-2 data and spectral indices. Furthermore, the combination of Sentinel-1 and Sentinel-2 data resulted in better detection of water bodies and lotus habitats within the wetland. NDVI, SAVI and mNDWI have a high correlation in examining the changes, so that an increasing trend was observed in the first six months of the year and a decreasing trend in the second six months, and the trend of vegetation and water changes is the same. Conclusion: Due to the complexity of wetland spatial structures and existing threats, identifying land cover types is challenging. This study demonstrates the use of multi-temporal Sentinel-1 and Sentinel-2 data to comprehensively assess wetland characteristics. The accuracy assessment for the four study periods from 2016 to 2022 using three classification algorithms, Support Vector Machine, Maximum Likelihood, and Artificial Neural Network, showed that the combination of Sentinel-2 and Sentinel-1 data outperformed the combination of Sentinel-2 data with spectral indices in terms of overall accuracy and Kappa coefficient. Among the three algorithms used, the Maximum Likelihood algorithm consistently achieved the highest overall accuracy and Kappa coefficient compared to the other two algorithms.
Keywords
