Using Sentinel-2 imagery to differentiate between spring wheat and alfalfa in Qingtongxia Irrigation District

Abstract

【Objective】 Spring wheat and alfalfa and two crops widely grown by farmers in Northwestern China. A knowledge of their planting areas is important for agricultural management but challenging at regional scale. This paper investigates the feasibility of using air-born technologies to identify their areas at different growing stages. 【Method】 The studies were based on Sentinel-2 imagery acquired from the Qingtongxia Irrigation District, with which we developed a decision tree classification algorithm to identify spring wheat and alfalfa. Accuracy of the method was tested against ground-truth data. 【Result】 The accuracy of the model for identifying spring wheat and alfalfa in early April was 69% and 75%, respectively, due to the similarities of the canopies of the two plants. With the growth of the plants and increase in available data, the accuracy of the model improved gradually, with its accuracy for identifying the spring wheat and alfalfa exceeding 90% on 14th May. Using all five satellite imageries available by 13th May, the accuracy of the model for identifying spring wheat and alfalfa reached 94% and 97%, with their associated Kappa coefficient being 0.75 and 0.86, respectively. The estimated planting areas of the spring wheat and alfalfa in Qingtongxia Irrigation District was 24,000 hm2 and 2,000 hm2, respectively. The spatial distribution of spring wheat was complex, characterized by a large number of fragmented planting zones. 【Conclusion】 The decision tree classification method combined with the Sentinel-2 images can preliminarily identify spring wheat and alfalfa in early April. Its accuracy improves steadily as more data become available, with the accuracy exceeding 90% after the middle of May.

Keywords

• remote sensing
• early identification
• decision tree
• time series
• spring wheat
• alfalfa