EXPLORING LAND COVER/CROP CLASSIFICATION IN EASTERN AND NORTHEASTERN CHINA USING SENTINEL-1 AND SENTINEL-2 DATA

摘要

Timely and accurate identification of crop type at field level is crucial to state and local governments, as well as agricultural, food, and insurance industries. High temporal frequency observations are often required to distinguish between various crop types. This type of data is often available from coarse spatial resolution instruments such as the Moderate Resolution Imaging Spectroradiometer (MODIS) (250-500m) and medium spatial resolution instruments such as the Landsat series (30m), but with many applications finer spatial details are desired. With the launch of European Space Agency (ESA)'s Sentinel series, higher spatial resolution data (10m) at high temporal frequency provide new opportunities for agricultural applications. To improve available regional information and methods, we explored high temporal resolution radar and optical observations from the Sentinel-1 C-band Synthetic Aperture Radar (SAR) instrument and Sentinel-2 MultiSpectral Instrument (MSI) for mapping major crop types (corn, soybean, and rice) in Eastern and Northeastern China. To accomplish this, we construct Sentinel-1 and Sentinel-2 growing season time series data sets, and used unsupervised clustering for our exploratory land cover/crop type classification. Moving forward, we plan to build supervised models for Eastern and Northeastern China to account for difference in crop phenology. Different set of inputs including SAR only, optical only, and SAR and optical combined will be assessed to determine the most suitable for each region. Additionally, the crop type information generated using the Sentinel fusion approach will be migrated to field/patch level through object-based approach.