Print Masthead

content | navigation | back


United States Department of Agriculture
National Agricultural Statistics Service

images of U.S. agricuture
Search NASS Advanced SearchSearch TipsBrowse NASS by Subject Crops and Plants Demographics Economics Environmental Livestock and Animals Charts and Maps Research and Science Education and Outreach Statistics by State #
You are here:  Home /

  History of Remote Sensing for Crop Acreage


The National Agricultural Statistics Service (NASS) is occasionally asked why it does not use satellite imagery to replace or enhance its present program of crop acreage estimates. The short answer is that NASS does use satellite imagery to enhance, but not replace, its program of crop acreage estimates and has for the last two decades. There are now three major applications of remote sensing with respect to crop acreage estimates. First is the operational construction of the nation's area sampling frame for agricultural statistics, which has used satellite imagery as a major input since 1978. The area sampling frame is the statistical foundation for providing agricultural estimates with complete coverage of American agriculture. Crop acreage estimation is only one part of this system. The second application, which is now done for seven-ten states per year, has been the use of satellite imagery to improve the statistical precision of crop acreage estimate indicators, especially at the county level in those states. This was the first NASS application of Landsat data and it began in 1972.

The third application, and most popular with Geographic Information System (GIS) data users, is the formation of a public use GIS data file called the Cropland Data Layer. The Cropland Data Layer is the crop specific categorization of the "best available" set of Landsat (30 meter resolution) digital imagery for the crop(s) season of interest. Data users have recently used the Cropland Data Layer to aid in watershed monitoring, soils utilization analysis, agribusiness planning, crop rotation practices analysis, animal habitat monitoring, prairie water pothole monitoring, and in the remote sensing/GIS value added industry.

There are several factors which comprise the limits of remote sensing in the context of crop acreage and contribute to decisions which have been made in NASS. This paper will discuss those factors as well and provide an overview of the NASS current applications of satellite imagery, for area frame sampling, crop area estimation and the more recent public use Cropland Data Layer.


Since 1978, satellite imagery has been the major input into stratification of land based on broad land cover definitions. Previously, aerial photography mosaics were used. Current satellite imagery provided a much better base for strata boundaries than the aerial photo mosaics obtained from flights several, sometimes more, years prior to the year of the new frame. This has led to improved statistical precision of numerous area frame-based estimates, including coverage estimates for major probability surveys and the 1997 Census of Agriculture. In addition, beginning in 1978 and continuing today, area sampling frames have been converted from paper-based products, subject to fire and loss, to digital versions which are more accurate and better protected from loss. New area sampling frames for Puerto Rico and all but several states have been built since 1978 using these improved methods due primarily to satellite imagery. Recently procedures have been developed to use satellite imagery that has been categorized into major crop types for more efficient and "deep stratification".


The second crop related application mentioned is the use of Landsat satellite imagery, along with an area frame based ground data sample, for direct crop acreage estimation in seven to ten states. The reasoning for the limitations, such as timing of the estimates, cloud cover, image delivery schedules, uniqueness (or lack thereof) of spectral signatures for crops will be presented next in more detail, as well as the benefits for the States included.


Perhaps the single greatest reason that NASS has not replaced any conventional crop acreage estimation surveys (farmer reported data) is that satellite collected data cannot meet the time requirements of the crop estimates system that American agriculture has come to expect. For Spring planted crops, the first crop acreage estimates are published at the end of March in the Prospective Plantings Report. Actual plantings are released in the Acreage Report at the end of June. Yield forecasts are published about July 10 for oats, barley, and spring wheat and August 10 for corn, soybeans, sorghum, and cotton, along with forecasts of acres for harvest. Early in the growing season, until crops reach nearly their full canopies, there would be considerable misclassification among different crops if acreages were estimated by remote sensing. Essentially, the satellite would be predicting acreages based on soil types. Some studies have shown that the optimum time to accurately separate corn from soybeans in the upper Midwest is about mid-August. Thus, NASS has published three major crop acreage indications (prospective plantings, actual plantings, and forecasted acreage for harvest ) before satellite imagery could be accurately used for the season final acreage for harvest. However, since county crop acreages are not released until early in the calendar year following the crop season, remote sensing based acreage estimates have been used in seven to ten States as an input to small area statistics (county level). Along with farmer reported data from an end of year survey and administrative data from the Farm Service Agency, the remote sensing based estimates are considered when setting an official county estimate.


NASS has been the lead organization in the development of probability based procedures for remote sensing based categorization, or classification, of crop types and estimating crop areas. Its remote sensing research began in 1968, well before the launch of the first Landsat satellite. NASS was the first organization to do full frame satellite classifications. (Most other early research efforts classified the pixels in a sampling of fixed size target areas within a scene.) NASS also found that its June Area Frame Survey, which identified the field boundaries and crop types in randomly selected sample segments, was ideal for training classifiers to identify crops. Segments in the Midwest typically are one square mile in size in heavily cultivated areas and there are often 30 or more segments available in a single Landsat scene.

NASS found that it was initially necessary to perform additional rectification of satellite scenes in order to ensure one-for-one matches of satellite pixels with corresponding fields on the ground. Staff members programmed the automatic rectification processes, as well as routines for automatically checking reGIStration of sample segments with the satellite data. NASS also coded routines which allowed unsupervised classification of all pixels in all wavelength bands from Landsat scenes once training had been done with the pixels within sample segments. Crop classification experimentation began as soon as the Landsat 1 data became available. By 1975, the first attempt was made to classify all scenes for an entire state (Illinois). That effort was successful in classifying all but two counties for which cloud free imagery was never available during the growing season, by the end of the calendar year. It should be pointed out that, since the estimator used by NASS is a regression estimator based on the June Area Survey data, NASS is able to make estimates for an entire state by substituting the Area Survey results for counties with no satellite coverage. In 1978, Iowa was the first entire state processed in time for end-of-year acreage estimates and the new figures were a significant input.

Many State level applications have shown relative efficiencies of 3.0 or more. That means that the combination of ground data and satellite data produces a reduction in sampling errors for acreage estimates comparable to collecting three times as much ground data. Rarely in statistics does one get such precision gains. However, because of the timing issues previously mentioned and the already high quality of area and list-based sample data available, the number of States was limited. This has fueled the incentive to develop additional outputs or products from remote sensing projects. First, the development of county level acreage estimates with measurable statistical precision and, more recently, the categorized satellite scenes as a data layer for direct GIS input. These efforts greatly add to the value of the projects and the internal and external uses now justify an eventual, funding permitting, 10-20 major crop state program.

The costs associated with extracting the crop type information from the raw satellite imagery data, which were once a major concern, are no longer an issue due to continual emphasis on this by the NASS research team assigned to these projects. The NASS regression estimator and the associated software, now called PEDITOR, was originally designed to run on ILLIAC and CRAY supercomputers and DEC mini-computers as the computations required were quite demanding. NASS researchers have continually and dramatically improved the cost efficiency of the IT aspect of State level acreage projects and PEDITOR now runs large volume classification and clustering on Windows 2000 on dual CPU Pentium processors with adequate data storage devices.

NASS has also studied many other sensors, as they became available, but the sensor of preference for crop acreage estimation remains the Landsat Thematic Mapper, with 8 day coverage.


There is a common misunderstanding that crop type signatures are so unique that they could be determined once and for all later classifications would be a matter of running a new satellite data file against known parameters; this is called signature extension. Satellite-based crop classification is based on the measurement of energy emitted or reflected by plants. Those readings do differ somewhat from one crop type to others in different wavelengths. However, that pattern differs throughout the growing season of a particular crop. There can also be considerable differences between healthy plants and plants of the same crop but under serious stress. The density of crop planting and the presence or absence of weeds and recent precipitation also can affect crop response. On top of all other factors, the atmosphere through which the crop response is being measured is not the same from one day to the next.


An important partnership was developed between NASS and USDA's Foreign Agricultural Service and Farm Service Agency in 1996. Joint use of Landsat imagery for the continental U.S. and sharing of outputs has benefited the programs of all three Agencies.

Since NASS has state of the art processing and analytical capabilities but not the total funding and staffing to create remote sensing based estimates and GIS data layers for many states each year, it has recently begun formal partnerships with other USDA agencies and State governments and universities to jointly increase the number of states involved. The partnership with the Foreign Agricultural Service and the Farm Service Agency in sharing Landsat data has been instrumental in expanding the number of states NASS can address. In addition, all NASS field offices operate through Federal-State cooperative agreements with State Departments of Agriculture and land grant universities. NASS conducted a search of these cooperators to identify those interested in joint ventures to provide products useful to both organizations through cost and resource sharing arrangements. This triangular arrangement involving NASS, FAS/FSA, and State governments or land-grant universities is a win-win-win situation and is enabling more States to benefit from these applications. In addition, NASS continues to partner with the Agricultural Research Service to evaluate new sensors for crop monitoring.

These new cooperative agreements were started in 1999 and analysts for several States have been trained. States currently involved in the Cropland Data Layer Program are New Mexico, Arkansas, North Dakota, Mississippi, Illinois, Indiana, and Iowa. Pilot studies are being conducted in Missouri, Nebraska, Wisconsin and Maryland. Details of the arrangements vary in each case, but the cooperating organization is essentially providing extra staffing and or hardware and peripherals and/or ground data for non-cropland areas. The State governments of Illinois, Mississippi, and North Dakota are currently formal partners for those States.


The Cropland Data Layer Program, which began in 1996 along with the NASS/FAS partnership, has proven to be quite popular with a broad spectrum of GIS proficient data users. Among the users are other Federal agencies, State and Local governments, Crop Farm Growers Associations, Crop Insurance Companies, Seed and Fertilizer Companies, Farm Chemical Companies, Universities and Value Added Remote Sensing/GIS Companies. The Cropland Data Layer is provided with substantial metadata on the dates of Landsat imagery used and the accuracy of crop specific categorization. The accuracy, for major crops of interest, generally varies from 80 percent to the high 90's for kappa coefficients and the statistical correlations with ground data ranging from 0.6 to nearly 1. Of course, this application performs significantly better with 8-day Landsat coverage versus 16 day coverage. See the NASS Web Site at:
for categorized image examples, methodology descriptions, and metadata for the Cropland Data Layer Program of NASS.



Last modified: 05/27/09