Automated 10-m Resolution In-season Crop-type Data Layer Mapping for Contiguous United States
摘要
Nationwide in-season crop planting data is critical for timely agricultural decision-making and application development in the U.S. Currently, the primary source of crop planting data is the Cropland Data Layer (CDL), an annual product from the U.S. Department of Agriculture (USDA) that is available to public in Feburary of following year, mainly supporting post-season applications. To address the need for high-resolution, in-season crop planting information, we developed an automated crop-type mapping workflow to produce a new data product: 10 m resolution In-season Crop-type Data Layer (ICDL) maps for June, July, and August of current, available publicly with a delay of only 5 days. The workflow extracts training labels from historical CDL data and incorporates Sentinel-2 and Landsat 8/9 observations to conduct supervised time-series classifications. The outputs are assembled using a multilevel mosaicking process to produce the Contiguous U.S. ICDL. Validation of the ICDL product demonstrated its high accuracy. Training labels accuracies ranged from 0.825 to 0.937, while classification accuracies improved from 0.807 in June to 0.984 in August, consistently outperforming the annual CDL. Moreover, ICDL-based acreage estimates for major crops showed close agreement with official USDA National Agricultural Statistics Service (NASS) statistics. The ICDL datasets are publicly available on the CropSmart web portal, providing timely, high-resolution crop-type information that can directly support national-scale agricultural monitoring, management, and decision-making.