Crop classification based on phenology information by using time series of optical and synthetic-aperture radar images

Abstract

Crop classification provides essential information for ensuring global food security, allowing early crop monitoring practices and water irrigation management. Despite of high importance of satellite imagery in identifying agricultural products, a significant number of optical images could become inaccessible due to cloud cover, which makes mapping agricultural products difficult. Therefore, the provision of up-to-date data integration methods and the availability of time series of optical and radar images such as Sentinel-1 data have provided new opportunities for mapping agricultural products with high spatial and temporal resolution. In this study, the results of classification accuracy were investigated based on the phenology of agricultural products and using time series of Landsat 8, Sentinel-1 and the digital elevation model (DEM). To do so, the key phenological parameters of the products such as start of season, length of season and end of season were first extracted using time series of the NDVI vegetation index calculated from the optical images and the TIMESAT model. Then, based on vegetation phenology of crops using optical (Landsat-8) and radar (Sentinel-1) time series data along with Digital Elevation Model (DEM) to increase the accuracy of classification, we applied the support vector machine method based on machine learning. The performance of this model was investigated on a part of Miandoab catchment basin located in the catchment of Lake Urmia in the northwest of Iran. Also, five major agricultural products of the study area, such as alfalfa, wheat, sugar beet, apple and grape were identified with overall accuracy of 89% using the support vector machine algorithm. Classification results demonstrated that using a combination of different data (overall accuracy (OA) = 89%, Kappa = 0.78) were more accurate than the only single-sensor inputs (OA = 77%, Kappa = 0.64), and when differences between crop types were largest, classification performance increased throughout the season until July.