An Object-Based Crop Classification Using Optimum Remotely Sensed Phenological and Multi-Spectral Data in Pakistan

Abstract Accurate and timely mapping of crops is essential for water resource management and to ensure sustainable food security. Satellite remote sensing has the well-documented ability to provide crop-type maps based on multispectral temporal datasets. However, due to the highly heterogeneous cropping practices, crop type mapping usually involves large phenological datasets, complex procedures, extensive field data, and resources. Therefore, we developed a simple and efficient image object hierarchy to delineate agricultural field boundaries and identify different crops. The combination of multispectral and temporal profiles was assembled using 23 Landsat 8 images over the period of two cropping seasons in Sahiwal district, Pakistan. The crop calendar information was also used to retrieve unique features distinguishing various crops through rule set development in object-based image analysis (OBIA). The approach incorporated the optimum phenological information at the start, senescence, and peak of the growing season to map major crops (wheat, maize, rice, cotton, sugarcane, orchards, fodder, and other land-cover land-use types including built-up areas, bare soil, grasses, and bushes) in the study area. The overall accuracy for crop maps was reported greater than 84% for both cropping seasons and ranged from 80 to 96% when compared with crop areas, as reported by the agriculture department and through independent accuracy assessment. The proposed workflow not only applies the Earth observation data to generate accurate and reproducible crop and land cover maps but also is an auspicious step to reduce the extensive field work and resources.