Crop Surveillance Demonstration Using a Near-Daily MODIS Vegetation Index Time Series

1
Crop Surveillance Demonstration Using a
Near-Daily MODIS Vegetation Index Time Series

• Robert E. Ryan
• Science Systems and Applications Inc.
• NASA Applied Sciences Directorate
• Stennis Space Center
• May 16, 2005
• Multi-temp 2005

2
Contributors

• Rodney McKellip NASA ASD
• Donald Prados CSC
• Slawek Blonski SSAI

3
Socioeconomic Human Tollof Crop Diseases

• In wealthy nations where monocultures are grown,
  large economic damage can result from outbreaks
• In 1970, naturally occurring leaf blight
  destroyed 1 billion of corn crop in the southern
  United States
• Present U.S. soybean rust outbreak could cost
  billions
• Crop bio-terrorism is another threat
• In poor countries, damage to a crop such as rice
  and/or food staple could cause famine
• Bengal famine in India 1942-1943, 2 million
  starved
• Irish potato famine (Late Blight) 1848, 1 million
  starved

Problem We do not have a well-developed approach
for monitoring crops for wide-scale disease
outbreaks.
4
Crop Surveillance System Requirements
System must detect subtle changes in plant health
and environmental conditions in their earliest
stages, before the effects of a disease outbreak
or environmental conditions can become widespread
and devastating.

• Spatial resolution
• Sufficient to detect significant changes to crop
  health
• Coarse enough to be practical.
• Temporal resolution
• Sufficient to detect unexpected changes in crop
  health within days.
• Affordable

5
Crop Surveillance Methods
In-situ methods
Remote Sensing

• Small number of data points
• Limited coverage
• Labor intensive
• Expensive
• Vulnerable to politics
• Accurate
• Not limited to surface
• Day/night all-weather

• Large number of data points
• Large coverage
• Can be automated
• Relatively inexpensive
• Ignores politics
• Can be inaccurate
• Limited to surface
• Affected by atmosphere

Remote sensing is the only method that can be
used today to monitor large areas.
6
Architecture Speculations

• Total system will require all source solution.
• Remote sensing will be dominant component.
• Cloud statistics and cost of launching future
  systems will require use of many systems
  including international ones.
• Some places in the world could require active
  airborne solutions
• Problem can be cast as a detection/estimation
  problem.

7
Architecture Issues

• Interoperability of a variety systems
• Spatial Scales
• 1-1000 m range
• Time Scales
• Products
• What should they be?
• Radiometric calibrations
• Spectral bandpasses
• Solar and viewing geometry
• Detection
• How do we set thresholds or detect anomalies?
• What are the natural variations?

8
Potential Crop Surveillance System Architecture
Near-Daily Wide Area Surveillance MODIS/AVHRR
Inexpensive Anomaly Surveillance
Large Swath Systems
Coarse Resolution Spectral Time Series Analysis
Crop Models Geospatial Detection (Time Series
Coarse Spatial/Spectra)
Agriculture Database
Meteorology In Situ Data
Moderate High Spatial Resolution
Surveillance Commercial Satellite Systems
Targeted Moderate/High Spatial Resolution Systems
(Expensive Acquisitions)
Moderate Small Swath Systems
High Spatial Resolution Analysis
Scouts/Remediation
9
Relevant Coarse Resolution Systems
MODIS is the highest resolution (250 m) large
swath system
10
Relevant Moderate Resolution Systems
11
Relevant High Spatial Resolution Systems
Geolocated and radiometrically corrected products
available.
12
Multi-resolution Crop Surveillance

• Coarse-resolution systems (250-1000 m GSD), such
  as the Moderate Resolution Imaging
  Spectroradiometer (MODIS), provide regional and
  continental views with frequent revisits.
• Medium-resolution systems (10-100 m GSD) provide
  field-level information but typically have
  infrequent revisits.
• High-spatial-resolution systems (lt10 m GSD) can
  provide spatial analysis at the row level.

An integrated, multi-resolution, remote-sensing
monitoring system will probably include a
Normalized Difference Vegetation Index (NDVI)
Product. (Red and NIR bands are available for all
systems)
13
Normalized Difference Vegetation Index (NDVI)

• NDVI based on reflectance maps (atmospheric
  correction)
• Reduced influence of atmosphere and solar
  illumination variations
• Improved intercomparisons between different
  instruments and acquisitions

Most commonly used vegetation index
14
Objectives

• Examine utility of the Moderate Resolution
  Imaging Spectroradiometer (MODIS) for
  understanding signatures and specifications for
  wide area crop surveillance
• Can we measure ideal crop signatures?

15
Approach

• Produce coarse spatial resolution, near-daily
  NDVI time series using the atmospherically-correct
  ed MODIS surface-reflectance product.
• Study areas having high number of cloud-free days
  and known ground truth
• Use standard products for validation
• Generate new products where necessary
• Examine known crop models (cotton fields with
  defoliants applied as rapidly propagating disease
  analogs).

16
MODIS(Moderate Resolution Imaging
Spectroradiometer)
MODIS provides long-term observations from which
an enhanced knowledge of global dynamics and
processes occurring on the surface of the Earth
and in the lower atmosphere can be derived.

• MISSIONS
• Terra Dec 1999
• Aqua May 2002

• PRODUCT SUMMARY
• Congruent observations of high-priority
  atmospheric, oceanic, and land-surface features

• OWNER
• U.S., NASA

• HERITAGE
• AVHRR
• High Resolution Infrared Radiation Sounder (HIRS)
• Landsat TM
• Coastal Zone Color Scanner

• VITAL FACTS
• Instrument Whiskbroom imaging radiometer
• Bands 36 from 0.4 and 14.5 µm
• Spatial Resolution 250 m (2), 500 m (5), 1000 m
  (29)
• Swath 2,300 km (55) from 705 km orbit
• Repeat Time Global coverage in 1 to 2 days
• Design Life 6 years

• LINKS
• Sensor Sitehttp//modis.gsfc.nasa.gov/
• Data Siteshttp//daac.gsfc.nasa.gov/ (ocean and
  atmospheric)http//edcdaac.usgs.gov/main.html
  (land)

17
MODIS Spectral Bands

• Day Bands
• Bands 1 2, 250 m GSD, broadband red and NIR
• Bands 37, 500 m GSD, blue short-wave infrared
• Bands 819, 1000 m GSD, blue NIR
• 1 short-wave H2O absorption band for cirrus cloud
  detection
• Night Bands
• Bands 2036, 1000 m GSD, long-wave infrared

18
MOD13 Vegetation Indices
The MODIS Vegetation Indices (VI) are robust
spectral measures of the amount of vegetation
present on the ground. They involve
transformations of the red, near-infrared, and
blue bands designed to enhance the "vegetation
signal" and allow for precise inter-comparisons
of spatial and temporal variations in terrestrial
photosynthetic activity. The NDVI output
represents a "continuity index" for existing
AVHRR-derived NDVI. The EVI is MODIS-specific and
offers improved sensitivity in high biomass
regions and improved vegetation monitoring
through a de-coupling of the canopy background
signal and a reduction in atmospheric influences.

• Principal Investigator Alfredo Huete
• Science Quality Status Provisional as of Nov 1,
  2000

• Intrinsic Spatial Resolution 250 m, 500 m,
• Applications monitor photosynthetic vegetative
  activity for phenologic and biophysical
  interpretations and change detection.
• Parameters
• Normalized Difference Vegetation Index (NDVI)
• Enhanced Vegetation Index (EVI).
• Web Link http//edcdaac.usgs.gov/modis/myd13q1v4
  .asp

Vegetation Indices
19
MOD09 Surface Reflectance
The MOD09 building block of MODIS products is an
estimate of the surface spectral reflectance for
each band as it would be measured at ground level
if there were no atmospheric scattering or
absorption. A correction scheme reducing the
effects of atmospheric gases, aerosols, and thin
cirrus clouds is applied to all pixels passing
Level 1B quality control.

• Principal Investigator Eric Vermonte
• Science Quality Status Provisional as of
  November 1, 2002
• Distribution Land Processes Distributed Active
  Archive Center (LPDAAC)

• Intrinsic Spatial Resolution 250 m, 500 m
• Applications Input for vegetation indices,
  thermal anomaly products, leaf area indices, and
  the bi-directional reflectance distribution
  function
• Parameter Surface reflectance
• Web Link
• http//edcdaac.usgs.gov/modis/myd09gqkv4.asp

20
Near-daily MODIS NDVI and Other Crop Monitoring
Products

• Near-daily NDVI product
• Based on MODIS MOD09 daily surface reflectance
  product
• Pixel quality assurance, temporal interpolation
  and smoothing usedto reduce noise
• Remove all non-ideal pixels
• Median and Savitzky-Golay filtering
• Single pixel time series (MOD13 and MOD09 NDVI)
• AVI Image Movies (MOD13 and MOD09 NDVI)
• Complements MOD13 16-day composite NDVI
• Standard Science Team validated product
• Used for validation of MOD09 NDVI generated
  product

21
Study Area
San Joaquin Valley, California Sheely
Farm (Ground truth available for cotton fields
with defoliants applied for rapidly propagating
disease analogs)
QuickBird color-infrared imagery acquired over
San Joaquin Valley, CA, on July 26, 2003
Application VV data Crop shape file and growing
season data by field
22
Why Sheely Farm?

• Located in San Joaquin, CA
• High value agricultural region
• Large number of clear days with excellent cloud
  statistics.
• Existing ground truthing data exists
• AG20/20 project through NASAs Stennis Space
  Center.
• Shape files of fields
• Spray schedules associated with the application
  of defoliants to the cotton crop.
• QuickBird imagery available over most of 2003
  growing season

23
Wheat Field Raw Data
24
Wheat Field Processed Data
25
Daily MOD09 NDVI vs.16-Day MOD13 NDVI
26
Noise Analysis Examples
Raw Data
NDVI values drop during cloudy days
27
Noise Analysis Examples
Pixel metadata can be used to ignore bad pixels
and pixels affected by clouds
28
Noise Analysis Examples
Savitky-Golay or Median filter can further reduce
the noise.
29
Cotton Field 5-3 Crop Management Dates
Date Action
8/3/2003 Insecticide Aerial
8/15/2003 Weeding
8/18/2003 Irrigation Furrow
8/21/2003 Insecticide Aerial
8/30/2003 Irrigation Furrow
9/25/2003 Close Ditches
10/02/2003 Chemical Prep (Defoliant)
10/12/2003 Chemical Sodium Chlorate
10/12/2003 Chemical Shark
10/20/2003 Harvesting
10/21/2003 Shred Cotton Stalks
10/22/2003 Root Knife
Validation Data
30
Near-Daily NDVI Time Series Example
Raw MOD09 NDVI Time Series Sheely Farm Cotton
Field 5-3
Digitally Filtered MOD09 NDVI Time Series Sheely
Farm Cotton Field 5-3
Digital filtering significantly improves
Signal-to-Noise ratio while preserving shape
31
Terra/Aqua Comparison
32
Terra/Aqua Comparison
33
MOD09 Quality Assurance Flag
34
Sample QA Analysis
MOD09GQK.A2003218.h08v05.004.2003231133953.hdf At
mospheric Correction Performed 84.2033 band
1 missing input 15.7967 correction out of
bounds, pixel constrained to extreme allowable
value 0.484436 band 2 missing input
15.7967 correction out of bounds, pixel
constrained to extreme allowable value
0.286649 L1B data faulty 0.0790625 cloud
state cloudy 37.6309 mixed 2.04824 QA
bits ideal quality, all bands 83.5977 less
than ideal quality, some or all bands 0.526467
other reasons some or all bands may be fill
value 15.8758
35
MOD09 NDVI AVI Movie
Fields are slightly under 1 km in size
36
Noise Analysis Examples
37
Noise Analysis Examples
38
Noise Analysis Examples
39
High Spatial Resolution Products
40
High Spatial ResolutionCommercial Products

• Standard true color and color-infrared products
• High resolution NDVI based on atmospherically
  corrected commercial products (IKONOS and
  QuickBird)
• Standard commercial products are radiometrically
  and geopositionally corrected only.

41
San Joaquin QuickBird 05/20/03

42
San Joaquin QuickBird 06/02/03

43
San Joaquin QuickBird 06/20/03

44
San Joaquin QuickBird 07/08/03

45
San Joaquin QuickBird 07/26/03

46
San Joaquin QuickBird 08/18/03

47
San Joaquin QuickBird 09/05/03

48
San Joaquin QuickBird 09/23/03

49
San Joaquin QuickBird 10/06/03

50
Atmospheric Correction of High Resolution Imagery

• Leverage SSC commercial imagery radiometric
  characterizations
• IKONOS, QuickBird, OrbView-3 (future)
• Use daily coverage from MODIS to provide input
  data for atmospheric correction
• MOD04 Aerosol Optical Thickness
• MOD05 Total Precipitable Water (Water Vapor)
• Generate MODIS-like products
• Surface Reflectance (MOD09)
• Gridded Vegetation Indices NDVI (MOD13)

51
Atmospheric Correction Approach
52
Atmospherically Corrected IKONOS NDVI
High spatial resolution (4 m) commercial-system-de
rived vegetation index using NASA satellite
provided atmospheric data and radiometric
characterization
Interoperable product with MODIS and other
sensors Comparable NDVIs generated from MODIS
53
Summary

• Coarse resolution NDVI products, based on
  near-daily MODIS surface reflectance, crop
  signature demonstrated in Southwest United
  States.
• Requires high percentage of cloud free days
• Good location to develop crop signatures
• Standard 16 day product misses key changes
• Weekly (daily desirable)
• Quality assurance data extremely valuable in
  developing products
• High spatial resolution data (virtual ground
  truthing)
• Simple digital filtering can reduce noise
  (unwanted variations) significantly
• Mixed pixels at 250 m significant at field
  boundaries
• General rule of thumb 3-5 pixels from field
  edges required for pure pixels

54
Plant Spectra Signature

• Plant spectra signature can be used to detect
  stress
• Healthy green plants are highly reflective in the
  Near-infrared (NIR) and sick plants are not
• Combination of Red and NIR bands can be used to
  detect plant illness. Thermal bands can also be
  used to distinguish water stress (drought) from
  other forms of stress

IKONOS 4 m GSD color-infrared imagery acquired
over San Joaquin Valley, CA, on July 19, 2001 Red
areas are healthy crops

• Color infrared color map
• Red NIR band
• Green Red band
• Blue Green band

Remote sensing plays a significant role in crop
health monitoring