FUGRO N.V.

1
USDA-FSA, Fugro EarthData GeoSAR Acquisition
Project ASPRS Portland, May 1st, 2008 Nathan
Pugh Cartographer/GIS Specialist USDA-FSA Aerial
Photography Field Office, Salt Lake City, Utah
Steven Shaffer GeoSAR Project Manager Fugro
EarthData Incorporated, Frederick Maryland
2
USDAFarm Service AgencyAdministers and
manages farm commodity, credit, conservation,
disaster and loan programs through a network of
federal, state and county offices.
3
Aerial Photography Field Office
4
Why is the USDA-FSA interested in IFSAR?

• Can IFSAR counteract the effects of poor weather
  conditions in time sensitive image collection?
• Assess the potential to determine land use and
  crop types.
• Can be collected on large scales regardless of
  time of day.
• How can it enhance our current products?

5
(No Transcript)
6
NAIP enhancements
7
Common Land Units (CLU)
8
CLU combined with compliance data
9
Weather related problems in the Southeastern US
Source NOAA
10
(No Transcript)
11
NAIP flying seasons 2007
12
All-weather collection
Mississippi

• GOES-12 IR Images of Eastern US from Aug 29-31
  2007, about 0900 AM to 0300 PM (Typical Imaging
  window for optical images).
• This demonstrates GeoSARs ability to collect
  during cloudy weather.

13
GeoSAR USDA FSA test program

• Determine the applicability of IFSAR data to the
  NAIP Program
• Collect a county in Mississippi SAR data.
• during NAIP collection window
• Produce image data
• Demonstrate geodetic accuracy compared to NAIP
  imagery and Common Land Unit Data.
• Demonstrate ability to separate between crop
  types in the SAR data.

14
Other federal agencies
15
Now, I turn it over to Fugro EarthData to
discuss GeoSAR and the preliminary results.
16
The GeoSAR aircraft
GeoSAR is an advanced mapping system employing
Interferometric Synthetic Aperture Radar. The
technology was developed at NASA Jet Propulsion
Laboratory. GeoSAR is owned by Fugro EarthData
Incorporated in Frederick, Maryland USA
17
Radar basics

• Radar uses signal echo returns to determine range
• The radar sensor transmits a microwave signal
  (A) in a focused beam (B) towards the target and
  detects the backscattered portion of the signal
  (C).
• The strength of the backscattered signal is
  measured to discriminate between different
  targets and the time delay between the
  transmitted and reflected signals determines the
  distance (or range) to the target

A
18
What is GeoSAR?

• Single-pass, dual frequency, interferometric
  radar mapping system
• X-band (? 3 cm) shows first surface
• P-band (? 85 cm) reveals detail beneath canopy
• Rapid, large area mapping through cloud cover,
  day and night
• Products include
• Ortho-Rectified Intensity (ORI) maps (1.5 m
  resolution)
• Digital Surface Model (DSM) (5 m resolution)
• Large scale maps

19
Major GeoSAR components
Antenna Positioning Measurement Unit
Two P-band Antennas Are Mounted on Each Wingtip
and Use Frequency From 270430 MHz
Two X-band Antenna Are Mounted Under Each Wing
Close to the Fuselage and Use Frequency From
96309790 MHz
20
GeoSAR data collection configuration
Collection Height 9.5 km to 12 km
13km
11.5 km
13km
21
The GeoSAR process
User Request
22
Radar wave properties

• Radar waves will propagate until they encounter
  an object with a radar cross-section as large as
  the wavelength.
• For this reason, X-Band SAR (? 3 cm), will tend
  to scatter off vegetation tops, or have
  relatively shallow penetration.
• The longer wavelength (? 85 cm) of the P-Band
  SAR gives it foliage penetration or FOPEN
  capabilities.
• P-Band will penetrate much deeper into the
  canopy, often all the way to the ground.

23
GeoSAR images
Colombia, 2006
24
Mountain Range
25
Yazoo County overview
26
Flight planning and ground control
27
Aerial survey of Yazoo agriculture

• Images shows typical agriculture of region, with
  catfish ponds, cotton, soybeans, peanuts, and
  corn fields.

28
Integrated GIS analysis
29
Ground data collection

• 25 fields surveyed for crop type, soil roughness,
  row spacing, plant height, width, spacing, and
  compass direction.
• 220 photograph assess local field and crop
  conditions.
• 5 soil types tested for bulk density, soil
  moisture and electrical conductivity.

30
Ground data collection and analysis
31
Comparison of GeoSAR and NAIP imagery
NAIP imagery from 2006 (left, courtesy of the
USDA Farm Service Agency Aerial Photography Field
Office) with the same area of GeoSAR composite
(right, courtesy Fugro EarthData Inc.) from Yazoo
Mississippi. The GeoSAR imagery combines the
backscatter and elevation information to produce
a colorized image.
32
Sweet potato crop
NAIP 2006 Image - Collected after the sweet
potato harvest. - The image shows bare soil and
furrow structure.

• Crop Height .5 m
• Row Spacing 1 m
• Stem Spacing 7.5 cm

33
Sweet potato crop
34
Center pivot irrigation
35
Center pivot irrigation
36
Summary of classification methods
X,Y raster cell
GeoSAR data layers
X-band ORI X-band DSM P-band ORI P-band
cross-pol ORI Volumetric correlation Volumetric
de-correlation
f(x) ? x statistical function
Crop distributions
Where x CART and/or ISODATA
37
ISODATA analysis

• Iterative Self-Organizing Data Analysis Technique
  (yAy!)
• ISODATA is a method of unsupervised
  classification
• Dont need to know the number of clusters
• Algorithm splits and merges clusters
• User defines threshold values for parameters
• Computer runs algorithm through many iterations
  until threshold is reached

38
ISODATA classification result
ISODATA Maximum Likelihood Classification based
on X Band Backscatter Images.
39
CART analysis

• The Classification and Regression Tree (CART)
  analysis is a non-parametric statistical analysis
  that predicts a variable (crop class) from
  multiple independent variables.
• For each class, it determines which independent
  variables are most important for predicting that
  class
• Builds a hierarchical tree diagram (set of
  if-then statements) to predict the class a
  dichotomous tree
• CART is powerful
• Can accept both raster and vector data inputs
• Results are easy to interpret
• No assumptions about independent data
  distributions
• Missing values OK
• Can find complex relationships between variables
  that other techniques might not uncover

40
Image Classification by CART analysis
CART Input Sample Polygons
CART builds classification rules (decision tree)
based on user inputs
Samples input into CART
All portions of image are assigned a class label
based on rules
Classification rules are applied to entire image
Classified Map
41
CART classification result
Reference Reference Reference Reference Reference Reference Reference Reference Reference Reference Reference
Corn Soybeans Cotton Forest Other Other Catfish Ponds Totals User's Accuracies User's Accuracies
Map Corn 8 5     3   16 0.50
Map Soybeans 1 4   5 0.80
Map Cotton   1 10 1   12 0.83
Map Forest Other 1 10   11 0.91
Map Other   3   3 1.00
Map Catfish Ponds           2 2 1.00
Totals 10 10 10 10 7 2 49
Producer's Accuracies 0.80 0.40 1.00 1.00 0.43 1.00 76 Overall Accuracy
42
Conclusions

• GeoSAR has demonstrated to the USDA-FSA the
  ability to collect and process wide-area SAR
  imagery and digital elevation models for
  agricultural mapping.
• Fugro EarthData is conducting research to extend
  the capabilities into land use land cover
  classifications for agriculture, forestry and
  other uses.
• Initial segmentation experiments have yielded a
  reasonable classification result.
• The refinement of data is on-going, the final
  report will be delivered in August 08.