The following is a Process_Description for the Cropland Data Layer upon with the CDL Confidence Layer is based.
OVERVIEW: NEW 10-METER CDL: The crop classification utilized remote sensing data from harmonized Sentinel-2 MSI Level-2A, Landsat 8, and Landsat 9 Level-2 Collection 2 Tier-1 products, providing surface reflectance (SR) data across multiple spectral bands, including GREEN, RED, NIR, SWIR1, SWIR2, and RedEdge bands 1-4. To mitigate cloud cover, 10-day median composites of surface reflectance and NDVI were created from the cloud-masked Landsat-Sentinel multi-sensor data for the growing season of 2024. An impervious layer from USGS NLCD 2021 and a digital elevation model from USGS 3DEP were also included ancillary input variables. In addition, mixed sampling strategies and localized training and were applied to the 2024 10m CDL production. Additional information: Z. Li, R. Mueller, Z. Yang, D. Johnson and P. Willis, "Cloud-Powered Agricultural Mapping: A Revolution Toward 10m Resolution Cropland Data Layers," IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium, Athens, Greece, 2024, pp. 4081-4084, doi: 10.1109/IGARSS53475.2024.10641079.
FOR MORE TECHNICAL DETAILS AND PROGRAM HISTORY: <https://www.nass.usda.gov/Research_and_Science/Cropland/sarsfaqs2.php> The United States Department of Agriculture (USDA) National Agricultural Statistics Service (NASS) Cropland Data Layer (CDL) Program is a unique agricultural-specific land cover geospatial product that is produced annually in participating states. The CDL Program builds upon NASS' traditional crop acreage estimation program and integrates Farm Service Agency (FSA) grower-reported field data with satellite imagery to create an unbiased statistical estimator of crop area at the state and county level for internal use. It is important to note that the internal CDL acreage estimates, which most closely aligned with planted acres, are not simple pixel counting but regression estimates using NASS survey data. It is more of an 'Adjusted Census by Satellite.'
SOFTWARE: New for the 2024 CDL a random forest classifier in Google Earth Engine was used to create the classification. ERDAS Imagine is used in the pre- and post- processing of all raster-based data. ESRI ArcGIS is used to prepare the vector-based training and validation data.
RANDOM FOREST CLASSIFIER: The 2024 Cropland Data Layer uses a random forest classifier approach. This is a departure from previous CDLs (2008-2023) that used a decision tree classifier using See5 software. Older CDLs (pre-2007) had limited ground reference training and less satellite imagery inputs and used a maximum likelihood classifier approach.
GROUND TRUTH: As with the maximum likelihood method and decision tree classifiers, random forest is a supervised classification technique. Thus, it relies on having a sample of known ground reference areas in which to train the classifier. Older versions of the CDL (prior to 2006) utilized ground reference from the annual June Agricultural Survey (JAS). Beginning in 2006, the CDL utilizes the very comprehensive ground reference provided from the FSA Common Land Unit (CLU) Program as a replacement for the JAS data. The FSA CLU data have the advantage of natively being in a GIS and containing magnitudes more of field level information. Disadvantages include that it is not truly a probability sample of land cover and has bias toward subsidized program crops. Additional information about the FSA data can be found at <https://www.fsa.usda.gov/>. The most current version of the NLCD is used as non-agricultural training and validation data.
INPUTS: The 2024 CDL has a spatial resolution of 10 meters and was produced using satellite imagery from Landsat 8 and 9 OLI/TIRS and ESA SENTINEL-2A and -2B collected throughout the growing season. Additional ancillary inputs were used to supplement and improve the land cover classification including the United States Geological Survey (USGS) 3D Elevation Program (3DEP) data and the USGS National Land Cover Database imperviousness data. Agricultural training and validation data are derived from the Farm Service Agency (FSA) Common Land Unit (CLU) Program. The USGS NLCD is used as non-agricultural training and validation data. Please visit the CDL FAQs and metadata webpages at <https://www.nass.usda.gov/Research_and_Science/Cropland/SARS1a.php> to view complete lists of imagery, ancillary inputs and training and validation used for a specific state and year.
ACCURACY: The accuracy of the land cover classifications are evaluated using independent validations data sets generated from the FSA CLU data (agricultural categories) and the NLCD (non-agricultural categories). The Producer's Accuracy is generally 85% to 95% correct for the major crop-specific land cover categories. Please visit the CDL FAQs and metadata webpages at <https://www.nass.usda.gov/Research_and_Science/Cropland/SARS1a.php> to view or download full accuracy reports by state and year.
PUBLIC RELEASE: The USDA NASS Cropland Data Layer is considered public domain and free to redistribute. The official website is <https://www.nass.usda.gov/Research_and_Science/Cropland/SARS1a.php>. The data is available free for download through CroplandCROS <https://croplandcros.scinet.usda.gov/> and the Geospatial Data Gateway <https://datagateway.nrcs.usda.gov/>. Please note that in no case are farmer reported data revealed or derivable from the public use Cropland Data Layer.