Using Sentinel-2 Sensing Imagery to Estimate Planting Structure in Fragmented Irrigated Lands

Abstract

【Objective】 Understanding planting structure and crop growth in a region is important to assess its food supply and security. The objective of this paper is to investigate the feasibility of a decision-tree model derived from the Sentinel-2 remote sensing imagery to map cropping structure in fragmented irrigation regions. 【Method】 The study site was Alagou irrigation area in Xinjiang. The planting areas of major crops in 2021 were estimated using the Sentinel-2 remote sensing imagery. We then compared these with both field investigation and visual interpretation from the Google HD images. Critical growth stage for identifying each crop was determined based on the phenological information and the NDVI time series, from which we derived a decision tree classification model. Accuracy of the model was verified against ground-truth data. 【Result】 The planting structure mapped from the Sentinel-2 remote sensing imagery had sharp textures, meeting the requirements for agricultural water management. The decision tree classification model can accurately classify crops at the scale required for irrigation management. The model is simple and feasible. Compared with ground-truth data, its average accuracy is 81.56% and the Kappa coefficient is 0.716 6. 【Conclusion】 The Sentinel-2 remote sensing imagery and the decision tree classification method derived form it can be used to accurately identify planting structure in fragmented lands. They can provide support information for decision-making in water management, and improve agricultural water usage in irrigation districts.

Keywords

• sentinel-2
• crops classification in irrigated areas
• ndvi time series
• decision tree
• fragmented land parcel