Document Type : Research article
Authors
1
Assistant Prof., Research Institute of Forests and Rangelands, Agricultural Research, Education and Extension Organization (AREEO), Tehran, Iran
2
Assistant Prof., West Azarbaijan Agricultural and Natural Resources Research and Education Center, Agricultural Research, Education and Extension Organization (AREEO), Uremia, Iran
3
Associate Prof., Research Institute of Forests and Rangelands, Agricultural Research, Education and Extension Organization (AREEO), Tehran, Iran
4
Researcher, Research Institute of Forests and Rangelands, Agricultural Research, Education and Extension Organization (AREEO), Tehran, Iran
10.22092/ijfpr.2024.363758.2123
Abstract
Background and Objectives: Given the significance of investigating and monitoring riparian ecosystems, this project was designed to identify and map the land cover, including tree and shrub species classes, around the Zarineh Rood River in West Azerbaijan province, Iran. Recognizing that the separation of lands with high spectral similarity using single-time images is not precise, this study utilized a time series of satellite images, capitalizing on the phenological differences of plant species.
Methodology: The research separated the land cover classes into two stages. In the first stage, the time series data from Sentinel 1 and 2 were used to map different classes of tree cover (natural, wood farming, orchard), shrub cover (natural, orchard), grass or pasture, agriculture, residential lands, soil, and water bodies. Given that seasonal changes in the images can provide valuable information about land cover classes, a one-year (2021) time series of Sentinel 2 optical images and Sentinel 1 radar polarizations for 2021, in the form of median in each season, were processed on the Google Earth Engine platform. The data were classified using four composites of input features and four classifiers. In the second stage, to separate the vegetation classes into Tamarix, willows, orchard, and poplar plantation, the trend of one-year changes of normalized difference vegetation index (NDVI), normalized green red difference index (NGRD), normalized difference red edge index (NDREI), and green normalized difference vegetation index (GNDVI) combined with HV polarization of Sentinel 1 radar in the form of median in seasons, was used as an input feature. The land cover map produced contained Tamarix, willows, orchard, poplar plantation, grass or pasture, agriculture, residential lands, soil, and water bodies.
Results: In the first stage of classification, the input feature of NDVI (Monthly)_ Radar (Seasonal)_ Sentinel 2 (Seasonal) and the random forest classifier were the best feature and the most accurate classification algorithm, separating the classes from each other with an overall accuracy and Kappa coefficient of 88% and 0.85, respectively. In the second stage of classification, the NDVI index between the months of April and November enabled the separation of all four tree and shrub covers. GNDVI between December and April was the best indicator for separating willows. Also, between May to November, it effectively separated Tamarix. NGRDI was suitable between May and November for separating Tamarix and also separated the poplar plantations between April and November. The GNDVI index between April and September effectively separated the two categories of orchards and poplar plantations from Tamarix and willows. The map was generated using the mentioned input feature and random forest algorithm. The overall accuracy and Kappa coefficient obtained from the validation relying on ground samples and Google Earth images were 80% and 0.77, respectively. The main diagonal of the error matrix shows the highest separation between water, soil, and urban land classes. Among the vegetation classes, willows and agricultural lands exhibited the best distinction.
Conclusion: The variation in a plant’s phenology, encompassing leafing, blossoming, fruiting, fall, and sleep cycle, leads to changes in the values of vegetation indicators during the seasons, which can be utilized in mapping vegetation to enhance separability. Consequently, if tree and shrub stands are pure and exhibit a different phenological behavior from their neighbors, they can be distinguished with higher accuracy using time series of satellite images.
Keywords
Main Subjects
)