Improving Sample Applicability for Early-Season Mapping of Winter Wheat Using Geoclimatic Zoning

Accurate and timely information on the early spatial distribution of winter wheat is essential for crop growth monitoring and crop yield prediction. However, the primary challenge in large-scale remote sensing-based early winter wheat mapping lies in the irregular phenology caused by diverse growing environments. Our study aims to explore the suitability of samples in early winter wheat mapping across different geoclimatic zones, with the goal of deriving more accurate early winter wheat distribution maps. Specifically, we chose topography and climate as alternative geoclimatic factors. Using Sentinel-2 imagery as an example, we examined the differences in classification accuracy among samples influenced by geoclimatic zoning. The results show that (1) Training models based on geoclimatic zoning can reduce the omission error due to asynchronous phenology and improve the accuracy of large-scale winter wheat early mapping. (2) Among the different geoclimatic factors, the sample optimization based on both topographical and climate factor zoning had higher accuracy (2~8% higher in OA) and optimization of early identifiable time (about 2~10 days earlier). The training strategy of adjusting the applicable area of the samples according to the geoclimatic factors is of great significance for the accurate early mapping of winter wheat in large-scale.