‹header›
‹date/time›
Click to edit Master text styles
Second level
Third level
Fourth level
Fifth level
‹footer›
‹#›
•This is a brief overview of the remote sensing activities and methodology involving acreage estimation and the Cropland Data Layer Program.
•The Cropland Data Layer Program provides the Agency with internal proprietary county and state level indications of major commodities, and secondarily provides the public with "statewide" (where available) mosaicked categorized output products after the public release of county estimates.
•Satellite based estimates will not completely replace surveys of farmers for several reasons:  acreage planted estimates are done in June, before the crop canopy fully develops; cloud problems can cause loss of large areas; and the enumerators ability to ask farmers other questions, not related to acreage during interviews.
•This project builds upon the USDA's National Agricultural Statistics Service (NASS) traditional crop acreage estimation program, and integrates the enumerator collected ground survey data with satellite imagery to create an unbiased statistical estimator of crop area at the state and county level for internal use.
•No farmer reported data is revealed, nor can it be derived in the publicly releasable Cropland Data Layer product.
•Up to two years worth of ortho-rectified categorized images where applicable are published on the DVD, along with associated accuracy statistics.
•A freeware browser “ArcReader” (Environmental Systems Research Institute, Redlands, CA) is bundled onto the DVD, allowing for users without a GIS or image processing software package to be able to view the CDL products.
•An “ArcReader” project is built on the Cropland DVD to assist the user in browsing the CDL and any ancillary vector data layers.
•Every June approximately 41,000 farms are visited by enumerators or receive questionnaires by mail as part of the USDA/NASS June Agricultural Survey (JAS).  These farmers are asked to report the acreage, by crop, that was planted or that they intend to plant, and the acreage they expect to harvest.  Approximately 11,000 area segments are selected nationwide for the JAS.  This represents approximately 2.5 percent of the total land area in the entire United States.
•A segment can range in size from four to eight square miles in open range areas; to about 1 square mile in cultivated areas; to 0.1 of a square mile in urban areas. This division allows intensively cultivated land segments to be selected with a greater frequency than those in less intensively cultivated areas.   Sample segments representing cultivated areas are selected at a rate of about 1 out of 125, whereas sample segments in land use classifications with decreasing amounts of cultivated land are selected at rates ranging from 1 out of 250 to 1 out of 500.
•Every field/land use within each segment is accounted for on the survey.
•The Spatial Analysis Research Section (SARS) in Fairfax, VA has performed remote sensing research & development activities since the early 1970's. The program has undergone various phase changes during its existence, going from research to production mode depending on funding levels and sensor availability.  NASS is actively looking for cooperative partnerships to expand the program into other states. •The Foreign Ag Service has provided imagery under a cooperative agreement for the program since 1997. Joint use of Landsat imagery for the continental U.S. and sharing of analysis/results benefited the programs of all Agencies and cooperating partners.  In 2004, SpaceImaging provided August 2004 imagery over all nine of the Cropland Data Layer states to begin the assessment of the AWiFS platform to supplement our program. •A variety of cooperative partnerships have been established to perform the analytical duties associated with this program.  The 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 in the late 1990’s to identify those interested in joint ventures to provide products useful to both organizations through cost and resource sharing arrangements.  Details of the arrangements vary in each case, but the cooperating organization essentially provides extra staffing and/or hardware and peripherals and/or ground data for non-cropland areas.  The Spatial Analysis Research Section trains and provides technical support for all cooperator sites.
These are the cooperating states and their respective partners.  The year the state began production is listed on the graphic.  The logos of the partners are links to their websites. Towson University sponsored the 2002 project and is a one-time study over 10 states in the mid-Atlantic region.
•This Program is repeated every year in participating CDL states, using the current June Agriculture Survey as training, and the best available Landsat imagery for categorization.  This is a one of a kind program, where the entire agricultural inventory in a state is categorized annually. •A spatial data archive of training data is created each year for participating CDL states.  Ground truth problems such as large acreage variance or misidentified crops in fields can be corrected during the JAS, providing a better official acreage estimate, when the digitizing is performed in concert with the JAS. •It is possible for cooperative partners to now import their own training data sets along with the NASS training set to create a better categorized image of a state.  The partner’s training data is imported via ESRI’s shapefile format.
•
•
•The hardware requirements for this system are as follows:  for digitizing/ground truth editing, any of the 32 bit Microsoft OS’s will work.  For computationally intensive jobs including; scene processing, clustering, classification, estimation and mosaicking a batch type system is utilized where jobs can be queued on different devices, and the minimum requirements are Windows XP.  A dual processor PC is recommended. •Image processing is performed by PEDITOR, where PEDITOR utilizes the Windows console along with environmental variables.  PEDITOR as it is now constituted, will only run under the Microsoft Windows operating systems.  •A Microsoft Visual FoxPro application called the Remote Sensing Project or RSP is used to manage the ground truth collection process, and track each segment to it’s completion. •Commercial off the shelf software XLNT from Advanced Systems Concepts, allows for batch job processing on the XP operating systems.  SARS utilizes XLNT to run computationally intensive jobs that are shared across network resources to expedite processing.
•PEDITOR was originally written in the 1970's, and has been updated and maintained since by NASS. It was developed during the early 70’s using Purdue University’s LARSYS system as a basis for further development.  NASS and the University of Illinois Center for Advanced Computing developed a customized program called EDITOR.  It was ported to other computer platforms by NASS and the name modified to PEDITOR.  •NASS has supported PEDITOR throughout the LACIE and AgRISTARS programs and continues to today, as PEDITOR was updated and modified to run on the latest desktop platforms utilizing some of the original algorithms from the LARSYS project.  •PEDITOR is written mainly in DELPHI (Visual PASCAL) with a few subroutines remaining in FORTRAN.  PEDITOR was written to run on a variety of computers, including; Cray supercomputer, IBM mainframe, Vax mini, UNIX workstations, and now PC microcomputers.  PEDITOR was ported to the desktop once PC's were capable of handling the processing requirements. •The clustering/classification/mosaicking/estimation programs only run on the XP platform using "Batch" type processing features.  “Expert" components or "rules of thumb" have been built into PEDITOR to expedite processing of repetitious tasks.  PEDITOR is now capable of importing/exporting published commercial vendor formats, such as ESRI’s shapefile and Erdas Imagine .lan and .gis formats.  •RSP was developed in the early 1990’s to manage the ground truth data collection effort, tracking segment production, minimize digitizing errors, and provide an expert like interface for the end user.  RSP was originally developed under FoxPro and migrated to Visual as the industry moved in that direction.
The program methodology is a continuous process throughout the year. 
•The first step “Seg Prep” establishes the training segments, digitizes the perimeters, and distributes software and data to the field offices, this goes from February to late May
•Seg digitizing begins during the JAS and continues until all fields and all segments are completely digitized, this may run thru July or even until mid-October in some states depending on human resource availability.
•Scene selection begins in early summer, and could run into the late fall depending on image availability.
•Seg cleanup analyzes the newly digitized segments with the newly acquired imagery.
•Fields that are bad either by digitizing or cover type are corrected or removed from training.
•Scene processing fits each segment onto a scene by shifting and cloud influenced segments are removed.
•The cluster/classification process runs in concert with the scene processing step, as segments are shifted they can be clustered.
•Estimation can be performed as early as August, once an entire state is finished classification, and the outputs are satisfactory.
•The mosaic process runs once estimation is completed. The mosaic for a particular state is released once the county estimates are released for that state.
These are the four major inputs for the Cropland Data Layer Program.
Each input will be discussed briefly in the following slides...
•The Cropland Data Layer program primarily uses the Landsat platform for acreage estimation. However, other platforms such as the Indian IRS Resource-Sat 1 platforms can be used to fill "data acquisition" holes within a state. •Landsat 5 launched in 1984, and is still producing images, it is projected to have enough fuel to last until 2009! •A data sharing partnership was created with the Foreign Ag Service and they have continuously provided NASS satellite images for the Cropland Data Layer Program since 1997. •On May 31, 2003 an instrument anomaly was discovered with Landsat 7.  To obtain up to date status on the anomaly visit:  http://landsat7.usgs.gov/index.php
•
•Landsat observes the earth in a near polar orbit, and covers the same spot on earth every 16 days. The combination of Landsat 5 & 7 synchronized in a 8 day repeat cycle greatly increases the probability of full state coverage during the growing season. •NASS uses seven Landsat TM/ETM+ bands, systematic correction, 30 meter resolution, path/row orientation, and nearest neighbor resampling. •In many cases, it is very difficult to obtain full state cloud free imagery during the growing season, even with two Landsat satellites, therefore, losing one would be very detrimental to the program.  However, adding a third or more Landsat like sensors would greatly increase the odds of obtaining cloud free imagery during the growing season. •The Landsat 7 covers the next adjacent path that the Landsat 5 covered the previous day, so that path 30 is observed on a Friday by Landsat 5, then on Saturday Landsat 7 observes adjacent path 31. •The medium resolution Landsat platform provides excellent results not only because of the temporal coverage but because of spectral properties of the available bands, and how effective they are in separating land cover.
This is a raw TM scene taken on 8/10/2004, bands 4,5,3.
Two acquisition dates are preferred for accurately separating and classifying ag land use.  One spring and one mid-summer are most preferred for optimal crop separation.
This is the secondary TM scene of the raw image on 8/10/04
•The Area Sampling Frame (ASF) is a stratification of each state into broad land use categories according to the percentage of cropland present.  Since 1978, satellite imagery was the major input into stratification of land based on broad land cover definitions. Previously, aerial photography mosaics were used.  Each year NASS replaces some of the state area frames because of land use changes or the need to improve acreage estimates. •The ASF is stratified using visual interpretation of satellite imagery. This 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. •The sampling frames are constructed by defining blocks of land whose boundaries are physical features on the ground (roads, railroads, rivers, etc.). These blocks of land cover the entire state, do not overlap, and are placed in strata based on the percent of land in the block that is cultivated. The strata allow for efficient sampling of the land, as an agriculturally intensive area will be more heavily sampled than a non ag intensive area.
•Every June, approximately 41,000 farms are visited or called by enumerators as part of the June Agricultural Survey.  States that participate in the Cropland Data Layer Program will have every field digitized by the NASS field office staff or cooperating partners. •The unit of observation is the tract, which may contain one or more fields or land uses and represents a particular land operator’s acreage within a segment.  The enumerators draws off field boundaries onto the National Aerial Photography Program’s (NAPP) 1:8,000 scale black and white aerial photos where the segment is located, according to their observations and the farmer reported information.  The fields are labeled and the cover type is recorded using a grease pencil on the aerial photo. •Enumerators account for every field/land use type within a segment.  They assign each field a cover type based upon a fixed set of land use classes for each state.  Every field within a segment must fit into one of the pre-defined classes.
•
•The segments are digitized in the field office using heads-up digitizing techniques, interpolating the enumerators drawing from the NAPP 1:8,000 photo of the segment, to the on-screen satellite image of the segment.  The digitizers are using satellite imagery from the previous growing season, as the crops from the current growing season have not fully emerged, and are not visibly separable during the JAS.  The fields are digitized using PEDITOR, and all of the segments in a state are managed by RSP to track segment production.  •Once digitizing is complete, all fields are labeled and a comparison is made to check for acreage validity between the digitized fields and the reported acreage.  Fields with variances greater than 10 percent when comparing the reported acreage (JAS) to digitized acreage, and fields smaller than 10 acres are both marked as bad for training.  If a field appears to be mislabeled or the reported cover did not emerge, an enumerator can revisit the site during the growing season and update the field level data, or the field can be marked off as bad for training. •A spring and summer date of observation is preferred for maximum crop cover separation for multi-temporal analysis.  If only one date of observation is available (unitemporal), a mid summer date is preferred. •If only an early spring date March-May or a fall date September-October is available (unitemporal) during the growing season, then it is best to not use that scene or analysis district for estimation, as bare soil in the spring and fully senesced crops in the fall will provide erroneous results.
•This is a sample of the questionnaire from which the enumerator asks the farmer for information. •Enumerators record the grower’s responses on cover type and acreage for each field in a segment on the JAS questionnaire.  The questionnaire is directly linked to the NAPP 1:8,000 segment photo by referencing the field number between the questionnaire and the photo. The JAS attribute data is imported into RSP and used to compare the digitized acreages to the reported acreages.  The enumerator observations are compared to the digitized acreage to filter out acreage discrepancies. •Performing segment digitizing during the JAS allows the NASS State Statistical Offices to improve acreage estimates by cleaning up acreage errors before the final official estimate is made. •The farmer reported data is only used internal to NASS and cannot be derived from the public output Cropland Data Layer.  Farmer reported data is held strictly confidential by NASS to calculate aggregated statistics.
•This marks the conclusion of the "ground truthing" process, all fields from here on have been cleaned and ready for clustering/classification/estimation.  Fields that were marked as bad for training were eliminated, but will be retained later for acreage estimation.
The analysis phase now begins ...
•Analysis Districts (AD) can be a individual or multiple scenes footprints that have to be observed on the same date, and analyzed as one.  An AD can be comprised of one or more scenes. An AD can be defined by either a scene edge or a county boundary.  Multi-temporal AD's are possible as long as both dates in all scenes are the same.
•A single or multi-scene AD will use all potential training fields for clustering/classification/estimation. Even though a segment can lie in more than one scene, each segment can be identified as belonging to only one AD.  Therefore, a segment can be used many times during the clustering/classification stage, depending on the amount of scene overlap, but each segment can be used only once for estimation.
•Preliminary definition of AD’s usually only considers the dates of imagery and the existence of cloud cover problems. A slight re-definition might later be based on amount of training data available or parameters required in the regression estimation process.
•Usually all parts of an AD are contiguous, but “holes” might exist due to cloud cover problems.
•Having two dates per scene location is preferred, one early in the growing season of the most important crops and one later. For example, in the Mississippi River Delta Region, one scene from April through May, and another from June through August would be selected; these cover the spring planted crops such as rice, cotton, soybeans, corn and sorghum.
•Two scene dates from the same sensor and same scene location can be overlaid; in addition, Landsat 5 TM and Landsat 7 ETM+ can be overlaid with each other.
•If only one date can be found for a given scene location, it would be selected based on the mid-growing season of the most important crops. Anytime the vegetation of interest is active and not in senescence, may work.
•In a two date image situation, some information is usually gained from scenes that are too early, such as the distinction between woods/pasture and bare soil.
•In a one date situation, scenes that are too late tend to create confusion due to bare soil, harvested crops, or non-vigorous vegetation. Obviously, scenes that are too early have bare soil for crops.
•
•Known pixels are separated by cover type and clustered, within cover type, using a modified ISODATA clustering algorithm, which allows for both merging and splitting of clusters.  
•Clustering is done separately for each cover type (or specified combination of cover types, such as all small grains). The clustered cover types are then assembled together into one signature file.
•Clustering is based on the LARSYS (Purdue University) ISODATA algorithm.  It performs an iterative process to divide pixels into groups based on minimum variance.  The pairs of clusters in close proximity (based on Swain-Fu distance) are merged under a NASS modification.  High variance clusters can be split into two clusters (variance of first principal component is used as a measure).
•The output of any clustering program is a statistics file: stores mean vectors and covariance matrices of final set of clusters for all cover types. 
•Three criteria for program termination (i.e. end of cluster merging and splitting):
•No further merges or splits possible.
•A minimum number of clusters (user specified) is reached.
•An infinite loop (self-detecting) is entered.
•An Analysis District can contain one or more scenes, the clustering is performed on all candidate segments found in the scene.  Clustering is an iterative process as it may take a few runs before a clean data set is successfully run through the clustering routine.
•This is a graphical representation of the inputs to the classifier (1) the clustered signature file, (2) the raw satellite imagery, (3) the area sampling frame. •The maximum likelihood classifier performs a pixel-by-pixel classification based on the final, combined statistics file.  It calculates the probability of each pixel being from each signature; then classifies a pixel to the category with highest probability.  The processing time depends on size of file to be classified (i.e. number of pixels), number of categories in the statistics file and number of input dimensions (number of bands/pixel). •The outputs are a categorized or classified image in PEDITOR format and the associated tabulation of categorized pixels for each cover type.
•This is a full classified scene based on the clustering of the training segments.  This classification was derived from a multi-temporal scene observed 6/10 and 9/6 in 2002.
•For estimation purposes, clouds are minimized by defining Analysis Districts along adjacent scene edges, by cutting the Analysis Districts by county boundary, or cutting out the clouds out by primary sampling units.
•Several factors can lead to problems in a classification, some get corrected in early edits and some do not:
•Poor imagery dates, with respect to the major crops of interest; this may not be correctable.
•Complete training fields that are incorrectly identified in the ground truth.
•Parts of training fields that are not the same as the major crop or cover type.
•Irrigation ditches, wooded areas, low spots filled with water.
•Bare soil areas in an otherwise vegetated field.
•Other ‘waste’ areas, grassy waste fields and idle cropland.
•Crops that look alike to the algorithm(s) due to planting/growing cycle:
•Small grains in general.
•Crops in senescence.
•Cover types that are essentially the same but used differently:
•Wooded pasture versus woods or waste fields (only difference may be the presence of livestock).
•Corn for grain versus corn silage.
•Cover crops such as rye and oats.
•Cover types that change signatures back and forth during the growing season.
•Alfalfa and other hays before and after cutting, with multiple cuttings per year.
•
•This is an enlargement of a 1 square mile training segment. 
•The clustering/classification is an iterative process, as fields get misclassified, they can be fixed or marked as bad for training and reprocessed.
•There are a number of tools that can be used to evaluate classification accuracy: 1) Visual inspection of all categorized training fields within an Analysis District. 2) Review of individual signatures/categories that do not perform as expected. 3) Evaluate the percent correct and confusion matrix, which shows counts of pixels by ground truth labels versus the classification’s output category number. 4) A linear regression may be run, by cover type of interest, to determine classification accuracy with respect to calculation of crop acreage.
•Once the analyst is satisfied with the classification, the next step is acreage estimation.
This is another enlargement of a training segment.
•Regression parameters are derived from four sources.
•The actual area frame survey provides the variable 'y', which is farmer reported acres of some given crop/cover type, within each of 'n' sampled segment areas.
•Sampled segment areas are then located in the categorized data; the variable 'x' comes from counting pixels categorized to that crop type within the segment boundaries.
•The parameter 'b' (or slope) is calculated as a relationship between the x's and y's for each stratum.
•The area frame itself gives the number (known as 'N') of all possible segments in the analysis area.
•Finally classification of full scenes allows counting of all pixels categorized to the given crop within the entire analysis area.
•Where available, regression is chosen as the preferred type of estimation. This approach essentially corrects the area sample (ground only) estimate based on the relationship found between reported data and classified pixels in each stratum where it is used.
•A regression relationship should be based on 10 or more segments for any stratum used, however, as few as five segments have been used in the past.  Where there are not enough segments in each stratum, a pixel based ratio estimator may be used which essentially combines data across stratum to get the relationship.
•Finally, the direct expansion (total number of possible segments times the average for sampled segment) may be used in the absence of pixel based methods.
•Regression adjusts the direct expansion estimate based on pixel information. It usually leads to an estimate with a much lower variance than direct expansion alone.
•Segments, called outliers, which do not fit the linear relationship estimated by the regression are reviewed; if errors are found, they are corrected or that segment may be removed from consideration in the regression analysis.
•This graph shows the approximately linear relationship between acres reported during the ground survey and pixels categorized to rice by the classification process.
•Several possible outliers are visible.
•The R-squared term measures how close the relationship is to a straight line; the closer it gets to 1.00, the better it is.
•Full scene classifications (large scale) are run wherever the regression or pixel ratio estimates are usable. •Estimates derived from the classification are compared to using the ground data only for the same area to make one final check. •State estimates are made by summing pixel based estimators where available and ground data only estimators everywhere else.  County estimates are then derived from the state estimates using a similar approach. •Final numbers are delivered to state field offices and the NASS Agricultural Statistics Board for their use in setting the official final estimates. •The states also have administrative data, such as FSA certified acres at the county level, and other NASS survey data.  Every 5th year, NASS also performs the Census of Agriculture at the county level.
•This is an overall schematic of the inputs and outputs of the acreage estimation and mosaic process. •The mosaic can be performed before the regression estimate, but it is not recommended.  The regression procedure allows for cloud influenced areas on a particular scene to be identified, and removed from consideration.  Adjacent scenes are then considered for inclusion over cloud influenced areas of another scene.
•NASS started producing mosaicked images of the CDL Program in 1997 using Erdas Imagine.  The process is human resource intensive as each scene needs to be geo-registered, and each separate analysis district has to be recoded to a master crop list, before they are stitched together.  Since the CDL program grew by about two states a year since 1997, and is re-created each year, an automated solution was investigated.
•EarthSat Inc. GeoCover Program provides ortho-rectified base scenes that the CDL uses as a basis for registration.  Each scene is co-registered to EarthSat's GeoCover LC imagery (50 meters RMS), and then stitched together using the priorities previously assigned from the scene observation dates/analysis districts map.
•A priority scheme is built into the system to allow for scenes/analysis districts with better quality observation dates to be assigned a higher priority when stitching the images together.
•Scenes can be cut/stitched by county or scene edge.
•Clouds are assigned a null value on all scenes, and scenes of lower priority that are cloud free, take precedence over clouded higher priority images.
•Once cloud cover is established throughout the mosaic, the clouds are assigned a digital value.
•Each GeoCover mosaic image covers an area approximately six degrees of latitude by six degrees of longitude.
•To completely cover a state and the Landsat scene footprints around a state, multiple GeoCover mosaics are used.
•The GeoCover handling steps are 1)  Imported into Erdas Imagine 2) Reduced to band two only 3) Resampled from 28.5 to 30 meters 4) Reprojected to one common UTM zone per state 5) Stitched together with other images to cover the state footprint 6) Exported to .LAN format.
•All CDL products are currently processed entirely by PEDITOR
•Finalized mosaics are exported to Geotiff format upon completion.
•PEDITOR’s Batch program routines process the scenes in as little as a few hours for the simple mosaics, and can run up to 2/3 a day on a large/complex state with many seams/stitch lines/cloud problems.
•The Landsat TM/ETM+ scenes used are radiometrically and systematically corrected.  There is a need to tie down registration points on a continuing basis for every scene within each state in the project.  Without some image/image registration, the scene registration tends to float 2-3 pixels in any given direction, for any given scene.
•An automated registration method was developed to co-register each raw scene to an ortho-base image, and then to apply each raw scene’s correlation coefficients to the categorized scenes. 
•Image recoding is necessary between different analysis districts, to rectify to a common signature set for a state.
•Once an Analysis District is categorized, it possess a unique set of signatures.  These signatures are recoded to a master signature set.  For instance, a corn signature or category may have between 1 and 50 classes for a given Analysis District, and they are condensed into one class or digital number for the final mosaic.  This is done for every cover type in the project.
•Clouds pose a big problem when trying to make acreage estimates, and there are mechanisms within PEDITOR to minimize their extent, as there are ways to minimize cloud coverage in the mosaic process by prioritizing scene overlap.
•Each categorized scene needs to be geo-registered to an ortho-base image. A block correlation is run between band two from each raw scene, and band two of the ortho-base image. The registration of the GeoCover mosaicked scene and the individual raw input scenes are used to get an approximate correspondence.  A correlation procedure is used on the raw Landsat scenes and the mosaicked scene to get an exact mapping of each pixel from the input Landsat scenes to the mosaicked scene.  The results of the correlation are used to remap the pixels from the individual input scenes into the coordinate system of the mosaicked scene.
•The mosaic process now performs:  1)  Automates image registration/rectification, 2)  Converts each categorized image and associated statistics file to a set standard automatically (recode),  3)  Specifies overlap priority by scene or county,  4)  Filters out clouds when possible
•Each scene is co-registered to EarthSat's GeoCover LC imagery (50 meters RMS), and then stitched together using the priorities previously assigned from the scene observation dates/analysis districts map.
•Scenes/analysis districts with better quality observation dates are assigned a higher priority when stitching the images together.
•Clouds are assigned a null value on all scenes, and scenes of lower priority that are cloud free, take precedence over cloudy higher priority images.
•Once cloud cover is established throughout the mosaic the clouds are assigned a digital value.
•Once the mosaic is successfully run in PEDITOR, it is exported into Erdas Imagine .LAN format, to prepare for public distribution.
•It is then necessary to rebuild the image statistics, add color/class name metadata, build pyramid layers, rebuild the projection information, and clip to state boundary.
•The Cropland Data Layer DVD products contain two years (if available) of imagery in two different image formats.
•GeoTiff format and .IMG format (winzip compression).
•The category colors are held consistent across each state each year
•The category digital numbers DN are consistent across each state, so that corn will always be DN 1, and soybeans will always be DN value 5, in all states for all years.
•The CDL images are available in GeoTIFF and ERDAS Imagine .img formats.  The latest two years worth of images are available on the DVD, size limitations prohibit adding additional years.  Each CDL DVD was ortho-rectified to 50 meters RMS, unless noted, this allows for precise comparisons of field and crop changes/rotations over time.
•The latest CDL products are available once the official NASS county estimates are released.  So for any given crop year, the CDL will be available the following March or June depending when the major commodities within a particular state are released.
•
•In order to maximize the visual contrast between different crops in various states, colors that provided the best contrast for the crop mix in a particular state are chosen.  Some of the Midwestern States are dominated by two main crops, while the Great Plains States may have up to 15 different crop types and the Delta States may have four or more main crops.  It is necessary to chose the best color mix given the amount of crops that are grown in a state, and what kind of contrast that you which to achieve. •For instance, permanently assigning the color yellow to corn for all states would limit the ability to contrast that color with other crops within any given state.  Whereas a state that grew little or no corn, would be forced to use up the yellow color, therefore, reducing the amount of contrasting colors. •The color schemes chosen for the Towson University sponsored Mid-Atlantic project are the same as the Midwestern states because of the similarities in the crops that are grown. •However, the digital numbers for each category within every state are the same.  So corn in North Dakota will have the same digital number as corn in Arkansas.  See mastercat.htm in the statinfo directory for a full listing by cover type.
•All CDL distribution for the previous crop year is held until the release of the official NASS county estimates for the major commodities grown within a given state.
•Corn and Soybeans are released in March for the previous crop year – Midwestern States
•Rice and Cotton are released in June for the previous crop year – Delta States
•A freeware browser “ArcReader” from ESRI is bundled onto the DVD, allowing users without a GIS or image processing software package to be able to view the CDL products.
•A demo ArcReader project is on the root directory of each CDL CD called “PublishedMapDocuments”.  The file has a .pmf file extension.  Once loaded, the categorized image appears in the main window, and contains themes from the National Atlas program, also included are NASS Agricultural Statistical Districts boundaries, and Area Sampling Frame.
•For users who have ESRI’s ArcGIS installed on their Desktop a \MapDocuments folder contains a .mxd project.
•The shapefiles and images on this DVD are under no copyright restrictions, as they are considered in the public domain.
•If you choose to reuse and publish any of the data this DVD, NASS would appreciate acknowledgment or credit for the usage of the categorized imagery.
•NASS publishes all available accuracy statistics for end-user viewing.
•The Percent Correct is calculated for each cover type in the ground truth information, it shows how many of the total pixels were correctly classified (i.e. across all cover types).
•“Commission Error” is the calculated percentage of all pixels categorized to a specific cover type that were not of that cover type in the ground truth (i.e. incorrectly categorized).
•CAUTION: a quoted Percent Correct for a specific cover type is worthless unless accompanied by its respective Commission Error.
•Example: if you classify every pixel in a scene to “wheat”, then you have a 100% correct wheat classifier (its Commission Error is also almost 100%).
•The “Kappa Statistic” is an attempt to adjust the Percent Correct using information gained from the confusion matrix for that cover type.
•Many remote sensing groups use the Percent Correct and/or Kappa statistics as their final measure of classification accuracy.
•
•
•The Cropland Data Layer Program is a one of a kind agricultural inventory program, where every state participating in the program is re-surveyed (i.e., ground truth) every June, and thus re-categorized. •Acreage estimation using satellite data combined with the standard NASS June Agricultural Survey (JAS), produces a substantially lower variance than when only using the JAS. •NASS plans to continue to evaluate additional states for inclusion into the Acreage Estimation and Cropland Data Layer Program.  States that are able to obtain cooperative  partnerships to support a remote sensing analyst position, would merit inclusion into the program. •The data on the DVD is in the public domain, and you are free to do with it as you choose.  NASS would appreciate acknowledgment or credit regarding the source of the categorized images in any uses that you may have. •Peditor is a one of a kind light weight GIS, industry standard pixel based image processor and a one of a kind acreage estimator based on an Area Sampling Frame. •The following list are some of the CDL customers:  Farmers, farm org, seed companies, fertilizer & pesticide companies, farm equipment dealers, grain transit/storage companies, farm real estate, global change, water quality, soils, & environmental assessment, crop insurance, universities, federal, state, & county government, value added RS/GIS resellers and agribusinesses. •NASS is going to continue research and product development using the AWiFS sensor and research other sensor opportunities as they become available.