Impact of red-edge waveband of RapidEye satellite on estimation accuracy of crop planting area

Adding red-edge waveband (690-730 nm) based on traditional visible light and infrared band is a notable trend of the development of current high resolution satellite sensors. RapidEye satellite carries red-edge waveband sensors. Based on the RapidEye satellite remote sensing data of Dongsheng town, Bei′an city, Helongjiang province on July 27, 2014, and by employing the monitoring classification method, this paper comparatively analyzed the impact of the red-edge waveband on the estimation accuracy of crop planting area under 2 kinds of waveband combinations. The indices including the separable measure, the classification accuracy and the degree of landscape fragmentation of the 3 kinds of ground objects i.e. corn, soybean and the others were computed under the conditions of 2 types of band combinations, which were with or without the involvement of red-edge waveband. Training samples of monitoring classification took the 2 km × 2 km grids covering study area as their basic units. Under the principle of equal probability of corn and soybean area proportion, 10 grids were chosen as training samples. Identification of crop within ground samples employed visual interpretation method. Accuracy verification adopted the background survey results of the crop area covering the study area. Based on the preliminary classification of RapidEye data with spatial resolution of 5 m, the background survey was conducted by using visual observation correction method combined with ground survey according to the seasonal changing rule of multi-temporal Landsat-8/OLI (Operational Land Imager) data. The result showed that, the overall identification accuracy of 3 types of ground objects (corn, soybean and the others) with red-edge was 88.4%, the Kappa coefficient was 0.81, and the mapping accuracies of 3 types of ground objects were 93.1%, 86.0% and 87.3% respectively; the overall identification accuracy of 3 types of ground objects without red-edge was 81.7%, the Kappa coefficient was 0.71, and the mapping accuracies of 3 types of ground objects were 83.9%, 73.4% and 84.6% respectively. By introducing red-edge band, the overall identification accuracy of 3 ground objects was improved by 6.7%, and the identification accuracies of 3 types of ground objects were improved by 9.20%, 12.6% and 2.7% respectively. By employing Jeffries-Matusita method, we calculated the degree of separation of 3 types of ground objects. The degrees of separation between corns and soybeans, corns and the others, as well as soybeans and the others were improved from 0.84 to 1.73, from 1.37 to 1.81, and from 1.27 to 1.29 respectively; by employing fragmentation index, we calculated the landscape fragmentation. The number of land parcels was reduced by 69.2%, with the average parcel area increased by 2.2 times, the average parcel perimeter increased by 60.5%, and the ratio between parcel area and perimeter increased by 1.0 time. So, by introducing red-edge band, the study has improved the separable measure of different ground objects, reduced the increase of landscape fragment caused by "Pepper salt" effect, and improved the overall identification accuracy of crop planting area. Currently, more and more satellites carry red-edge devices, and the domestically-produced satellites to be launched also plan to add red-edge band so as to improve crop identification capacity. The result of this paper will provide a reference for the application of domestically-produced red-edge satellite data in agriculture.