چكيده فارسي
Abstract
This study investigates vegetation cover changes over a 30-year period using satellite remote sensing data and various vegetation indices derived from them. Landsat (5 to 9) and Sentinel-2 satellite imagery served as the primary data sources. The main objectives include analyzing phenological changes in vegetation, evaluating different vegetation indices, and assessing how the choice of indices and sensors influences the final time series. Key vegetation indices employed were NDVI (Normalized Difference Vegetation Index), EVI (Enhanced Vegetation Index), and AFRI (Aerosol-Free Vegetation Index). Data processing methods encompassed preprocessing and corrections, time series smoothing using moving average algorithms, and multi-sensor data harmonization through regression models. Additionally, data simulation techniques were applied to integrate information from different sensors and mitigate discrepancies arising from sensor variations. Time series analysis revealed that variations in vegetation indices depend on climatic conditions, vegetation type, and density. NDVI exhibited higher values in agricultural areas compared to grasslands and forests, demonstrating greater temporal stability than AFRI and EVI. Comparative analysis of multi-sensor data highlighted spectral band differences, particularly in near-infrared (NIR) resolution between Landsat and Sentinel-2, leading to variations in computed vegetation indices. Linear regression models proved more effective for Landsat data, whereas quadratic nonlinear regression was better suited for Sentinel-2 data. Stationarity tests indicated that Sentinel-2 time series exhibited weaker stability and more structural breaks than Landsat, underscoring its higher sensitivity to environmental changes. To detect change points in vegetation index time series, the Dynamic Time Warping (DTW) algorithm was applied, identifying significant shifts often coinciding with sensor calibration changes. Multi-sensor data integration enhanced time series accuracy, while statistical models and time series analysis algorithms improved change-point detection and vegetation stability assessment. This study underscores the importance of harmonizing multi-sensor datasets for robust long-term vegetation monitoring.
Keywords: Time series, Phenology, Vegetation Index, Landsat 5 to 9, Sentinel 2