Winter wheat is one of the major cereal crops. Monitoring and mapping its spatial distribution has significant implications for agriculture management, water resources utilization, and food security. Generally, winter wheat has the representative phenological stage during the growing season, which forms the unique EVI (Enhanced Vegetation Index) time series curve and differs considerably from other natural vegetation and crop types. Since early 2000, the MODIS product has become the primary dataset for satellite-based crop monitoring at large scales due to its high temporal resolution, huge observation scope, and timely availability. However, the intra-class variability of winter wheat caused by the field conditions and management might lower the mapping accuracy, which has received little attention in existing studies. In this study, we proposed a winter wheat mapping approach that integrates the characteristics of angle and distance of MODIS EVI time series and combines with the intra-class variability. We applied this approach in two winter wheat concentration regions, i.e., Kansas State in the U.S. and the North China Plain (NCP) region. We also analyzed the effects of landscapes structure on the accuracy of winter wheat mapping using landscape indices.
Results/Conclusions
The results were evaluated against crop-specific maps or survey data at the state/regional level, county level, and site level. The accuracies in Kansas and NCP region were 95.09% and 92.88% at the state/regional level, 90.33% and 85.00% at the site level, and 0.96 and 0.71 (R2) at the county level respectively. The comparison with other studies demonstrates that considering the intra-class variability can increase the winter wheat classification accuracy. The analysis of landscape structure showed that the approach performed better in areas with lower landscape fragmentation, which may partly explain the relatively higher accuracy of winter wheat mapping in Kansas. Our study provides a new perspective for generating multiple sub-classes as training inputs to decrease the intra-class differences for winter wheat detection based on MODIS EVI time series. The proposed approach could be applied for winter wheat mapping in other areas that ground datasets are available. A flexible framework with simple variables and fewer training samples facilitates this approach to expand to multi-crop-type mapping at large scales.