Phenology-based decision tree classification of rice-crayfish fields from Sentinel-2 imagery in Qianjiang, China
Tian Xia, Wenwen Ji, Weidong Li, Chuanrong Zhang, Wenbin Wu
Abstract
Rice-crayfish farming systems combining rice cultivation and crayfish breeding have been a widely adopted unique cropping system in Central and Southern China. Despite the importance of understanding and monitoring the rice-crayfish fields, mapping rice-crayfish fields and identifying it from pure rice fields are challenging due to their spectral similarity and complexity. To solve this problem, we propose a new rice-crayfish mapping method based on Sentinel-2 data, a decision tree model and phenological characteristics. In doing so, segmentation of Sentinel-2 image data was first performed using an object-oriented approach, and spectral features of the images were extracted as segmented image objects. Crops were then classified using a decision tree classifier by selecting spectral features and texture features as classification indices. Summer and winter datasets for two special periods were finally used to construct a decision tree classifier to distinguish rice-crayfish fields from midseason rice fields. We tested this method in the birthplace of rice-crayfish farming system, Qianjiang county, China. A comparison with surveyed sample points showed a rice-crayfish field classification producer accuracy of 0.84 and user accuracy of 0.92. This indicates that the proposed approach is suitable for mapping regional-scale rice-crayfish fields and could be valuable for the monitoring and management of rice-crayfish fields.