Notes
Slide Show
Outline
1
The Mid-Western Cropland Data Layer Project
  • Methodology
2
Project Purpose
  • Remote sensing based cropland acreage indications
    • County and state level “major crops”
    • Internal to NASS
    • Considered for setting official estimates
    • Target fall/winter delivery for major commodities


  • Output categorized crop specific Cropland Data Layer
    • Distribute to public at the cost of reproduction
    • "Freeware" viewer – ESRI’s ArcReader
    • Accuracy statistics
3
Agency Background
  • United States Dept. of Agriculture
    • National Agricultural Statistics Service
    • June Agricultural Survey (JAS) – National in Scope
      • 41,000 farms visited
      • 11,000 one-square mile sample area segments visited
      • Most states contain between 150 – 400 segments
      • Derive planted acreage estimate


    • Cropland Data Layer piggybacks on JAS
      • Unbiased statistical estimator of crop area
        • State and county level estimates




4
Project Players
  • USDA/NASS Research Division
    • Spatial Analysis Research Section
      • Remote sensing analysts
      • Software developers


  • USDA/Foreign Ag Service
    • PECAD (Production Estimates & Crop Assessment Division)
    • Landsat 5 & 7 imagery source


  • State/Federal Cooperators
    • NASS Field Offices
      • Remote sensing analysts
      • Digitizers & ground truth editors



5
19 Cropland Data Layer States
6
Cropland Data Layer Benefits
  • Internal
    • Digitize the entire June Agricultural Survey (JAS)
    • Cleans up JAS problems
    • Produce numerical cropland acreage indications at the county and state level
    • Unbiased statistical estimator of crop area
    • Data sharing among partners
  • External
    • Ortho-rectified image product Commercial image formats
    • Released annually after county estimates are published
    • Detailed breakdown of cropland area via large training sample
    • Distribute to public at the cost of reproduction

7
Program Resources
  • Hardware


  • Computational intensive jobs (i.e. cluster/classify/mosaic)
    • Windows XP
  • Digitizing/ground truth editing
    • Windows XP
  • Software


  • Image processing
    • PEDITOR developed in-house
  • Digitizing/ground truth editing
    • Remote Sensing Project
  • Batch job processing
    • XLNT – COTS
    • www.advsyscon.com


8
Program “In-House” Software
  • PEDITOR - 1970’s
    • Developed in Delphi, Pascal and Fortran
    • Performs
      • Digitizing/clustering/classification/estimation/mosaicking
      • Optimized for dual processors
      • “Expert” rules built into processing routines
    • Capable of importing/exporting to commercial world
  • Remote Sensing Project (RSP) – 1990’s
    • Developed in Microsoft Visual FoxPro
    • Manages ground truth database
    • Performs digitizing and ground truth editing



9
Program Timeline
10
Acreage Estimation Inputs
11
Landsat Platform
  • Landsat 5 launched 1984 (3 yr design life!)
    • Thematic Mapper (TM) Sensor
  • Landsat 7 launched April 15, 1999
    • Enhanced Thematic Mapper (ETM+) Sensor
12
Landsat 5 TM and 7 ETM+ Specs
  • 185 kilometer swath width
  • 16 day repeat coverage (each)
    • Synchronized for 8 day repeat coverage
  • 7 imaging bands* per sensor
    • 6 bands @ 30m resolution
      • 3 Visible & 1 Near Infrared (IR)
      • 2 Short-wave IR
      • 1 thermal IR band
        •  TM: @120m resolution
        •  ETM+: @60m resolution

      • *ETM+ has one 15m Panchromatic Band

13
Landsat Full Scenes
14
Queen Annes
County Maryland
15
Area Sampling Frame
  • Stratify based on percent cultivated land
  • Subdivide strata into primary sampling units or PSU's
    • Selected PSU's divided into secondary sampling units or segments
    • Segments are selected randomly & used in
    •     successive years
    • 20 percent segment rotation/year
      • Entire sample replaced every 5 years
16
Segment Boundaries
  • June Agricultural Survey (JAS) segments
    • Enumerator records field extents, cover types, and acreage
17
Segment Digitization/Editing
  • June Agricultural Survey (JAS) + Agriculture Coverage Evaluation Survey (ACES)
    • Between 100 - 425 one square mile segments per state
    • Use most fields for training classifier
  • Digitize & label all fields in each segment
    • Uses heads up on screen digitizing
  • Compare digitized acres to JAS/ACES reported acres
    • Check fields for accuracy & consistency on current years’ imagery
    • Eliminate fields considered bad for training
18
JAS/ACES Questionnaire
  • Enumerators account for all land usage in segment
    • Draw off field location by direct observation
    • Directly link questionnaire to segment photo
    • Able to ask questions not related to acreage



19
Begin Data/Image Analysis
  • Ground data finalized
  • Clustering, Classification and Regression Estimation
  • Use Batch processing capabilities
    • Built in set of “Expert” rules
    • Share network resources
20
Scene Analysis Districts
  • Can be defined by
    • Contiguous same date coverage
    • Scene edge or county boundary
  • Cloud free or near
    •     cloud free dates
  • Unitemporal
    • One date
  • Multi-temporal
    • Two date(s) overlay
21
Clustering
  • Modified ISODATA clustering
    • Allows for merging and splitting of clusters
  • Cluster pixels within each cover type (corn, soybeans, rice ...)
    • Exclude bad fields
    • Principal component clipping to remove outliers
    • Merge & split clusters within cover types
  • Combine all the cover type clusters back into one signature file
22
Classification Overview
23
Multitemporal
Classification
 6/10 & 9/6 2002
24
Categorized Segment
25
Categorized Segment
26
Regression Template
27
Regression Estimator
  • Regression used to relate categorized pixel counts to the ground reference data
    • Independent variable - satellite data - pixels
    • Dependent variable - JAS acreage estimate
  • Satellite data - lower variance than with only JAS
  • Outlier segment detection - correction or removal from regression analysis
28
State & County Estimates
  • Run large-scale classification across analysis districts/entire state/project area
  • Compare large-scale regression with direct expansion estimates of JAS data
  • Sum satellite and non-satellite areas for state total
  • Produce county and state indications of major crops
    • For consideration in setting official estimates
29
Program Summary
30
Program Mosaicking
  • Initial project focus on acreage indications not mosaic
  • Six – fifteen categorized full base Landsat scenes per state
    • Radiometric & systematic correction only
  • Each unique scene date (Analysis District) produces unique sets of signatures/categories
  • Project repeated each year


  • Need to create mosaicked state-wide image
31
Mosaic Method
  • Run block correlation between all raw images and EarthSat's GeoCover Stock Mosaic
    • Co-register raw images to GeoCover’s band 2
  • Register all categorized scenes to GeoCover base
    • Use calibration coefficients
  • Mosaic all categorized images
    • Establish scene overlap priorities
    • Clip by scene edge or county boundary
    • Mask out clouds via priority schemes
  • Export to ERDAS’s .LAN image format
    • Distribute in GeoTiff format
32
Erdas Enhancement
  • Import categorized mosaic (.LAN) to .img
    • Re-build statistics
    • Add class names
    • Re-build projection metadata
    • Clip to state boundary
    • Colorize categories
      • Maximize contrast between crops
      • Colors maintained within state
      • Category DN’s consistent across all states
    • Export to GeoTiff


33
Color
Contrast
34
State-wide CDL
 Distribution
  • Held until county estimates released
  • Bundle images with ESRI’s ArcReader
    • Ancillary vector layers
    • Area Sampling Frame where available
  • No copyright restrictions
  • Publish accuracy statistics


35
Pocahontas County, IA CDL
36
McLean County, IL CDL
37
Cass County, ND CDL
38
Benton County, IN CDL
39
Rock County, WI CDL
40
Cuming County, NE CDL
41
CDL Conclusion
  • Combine satellite and ground truth data to produce acreage indications
    • County & state level estimates input to official estimates
  • Continue PEDITOR/RSP development
    • Reduce end-user burden
    • Increase functionality
  • Evaluate new sensors (AWiFS) for operational use



  • Remember, in no case is farmer reported data revealed or derivable from the public use Cropland Data Layer CD-ROM's