Comparing Pixel-based and Object-based Classification for Mapping Smallholders Rice Field Types Using Planetscope Imagery in Sekampung District, East Lampung, Indonesia

Abstract It is imperative to obtain precise and up-to-date information on the area of rice fields, as the yield of rice is a fundamental indicator of national food security. The mapping of rice fields based on water supply is still limited, particularly in Indonesia, which is characterised by ecological and management diversity. The launch of Planetscope imagery, which offers high temporal and spatial resolution, provides the opportunity to map the types of rice fields occupied by people. The objective of this study is to compare the accuracy of land use mapping using pixel-based (maximum likelihood) and object-based (SNIC segmentation) classification using random forest classification in order to generate a rice field map. The results of the object-based classification showed a total accuracy of 92.8% and a kappa coefficient of 0.82. In contrast, the pixel-based maximum likelihood classification showed an accuracy of 74.34% and a low kappa coefficient of 0.43. The mapping of rice field types relies on remote sensing data and field interview data, with an accuracy of 88.05% and 91.09% in planting season 1 and planting season 2, respectively.