Application of Sentinel-1 Data for Classifying Croplands Using Google Earth Engine

Abstract

The agricultural sector is an important source of national income in many tropical countries, including Indonesia. Monitoring agricultural is essential to support the agricultural activities. However, these countries have tropical monsoon climate with heavy amounts of rainfall. Therefore, temporal monitoring using microwave remote sensing is beneficial to overcome the heavy cloud coverage, which is often an obstacle for applying optical remote sensing. Recently, the growing trend of cloud-based geospatial platforms, such as Google Earth Engine (GEE), provides processing tools and cloud storage for remote sensing data without high specification hardware. In this study, Sentinel-1 synthetic aperture radar (SAR) sensor data from 2017 and supplementary Sentinel-2 optical sensor data was obtained and processed in GEE to identify two types of cropping patterns in paddy fields and to classify agricultural croplands. Four types of polarization combination datasets and a random forest classifier set with number of trees to be 25 and 50 were used for the classification process. The classification results show that the VH, VV, and the subtraction of VH and VV polarization with a random forest of 50 trees was achieved with 76.88% of overall accuracy and the kappa value equaled 0.728. The random forest with 50 trees significantly increased the classification accuracy of the dataset with fewer band compositions. The Sentinel-1 images are believed to be satisfactorily accurate enough for agricultural cropland classification and are sufficient for identifying the two cropping patterns in paddy fields.