Crop Biomass Estimation with NRT MODIS Data

1

• Crop Biomass Estimation with NRT MODIS Data
• Integrated remote sensing technologies for
  improved farm management
• Dr Andrew Rodger
• arodger_at_agric.wa.gov.au

2
Project Goal and Requirements
GOAL To improve the estimation of crop biomass
(mainly wheat at this point) at the farm level
using Near-Real-Time MODIS data. HOW Using a
combination of MODIS-derived reflectance data,
local spectral libraries, and a stress index crop
growth model. Specifics 1). Gather in-situ
surface spectra using the ASD Fieldspec 2).
Acquire MODIS data (supplied by DLI) 3). STIN
(crop growth model) 4). Spectral Mixture Analysis
and Libraries (replace the ASTER spectral library
with a local one)
3
Reflectance Spectra Normalisation
The image to the left is a reflectance spectrum
that has had a 20 offset added to each channel
based on the same channel for the spectra before
it..
After normalising the spectra using Equation 1
(1)
We get the image to the left. This tells us that
the spectra in the upper image are in fact the
same and are only subject to a wavelength
independent offset
4
Merredin
Department of Agriculture Western Australia
(DAWA) Research Station
Old Stubble
Pasture
Wheat
Lupins
Wheat
Stubble
Approximately 5.6km long
Approximately 2.7 km across
Field Peas
Latitude -31.494 N, Longitude 118.222 E
5
Merredin
6
Automatic Soil End-Member (AUTOSEM) Selection
Using MODIS Imagery

• Uses the first 5 MODIS bands
• Uses a time series of MODIS reflectance imagery
• Each MODIS reflectance spectrum is compared to
  the a known soils spectral library (ASTER)
• An error of fit is applied to the data that is
  selected as being a possible soil.
• All spectra having an error of fit greater than a
  predefined tolerance level are removed an the
  resulting pixels are deemed as non-soil.

7
AUTOSEM _at_ Merredin
8
AUTOSEM _at_ Merredin
Veg
Post histogram rejection
Prior to histogram rejection
Possible Soil Texture (Unconfirmed)
9
Wavelength versus Zeiss AUTOSEM Reflectance

• Good agreement between MODIS-derived soil spectra
  and those measured at Merredin
• Comparable variation in the first 4 MODIS bands.

10
Field Sites _at_ Wongan Hills

• 7 properties in total were chosen.
• property areas range from 2255-11391 ha.
• All owners/managers have been contacted and all
  have indicated their consent to let us perform
  spectral measurements over the next 1-1.5 years
  on a fortnightly basis.
• Close proximity to Perth will reduce travel time
  and expenses required to sample the sites.
• Close proximity to each other will reduce the
  time needed to move from site-to-site.

11
Selected Farms _at_ Wongan Hills after Application
of the AUTOSEM
2255 ha
2769 ha 2923 ha
3399 ha 5270 ha 5685 ha 11391 ha
12
Preliminary Results of a Fractional Analysis
Model Using a Zeiss Vegetation End-Member and the
AUTOSEM Sharpees Farm Using the AUTOSEM in
combination with a Zeiss-derived vegetation
end-member yields the following time series of
fractional vegetation cover. These results do not
distinguish between crop species, remanent
vegetation or natural vegetation.
13
While the overall trend is what we expect the
confidence in the vegetation end-member is not
high.
14
Potential STIN yield prediction for Wongan Hills
as of the 22/7/05 is 2.305 (T/ha). The potential
mass of the crop Mc can be estimated at any point
in time as McVF6.252.305, where VFvegetation
fraction and the factor 6.25 is the size of a
MODIS 250m pixel in hectares. The assumption is
that the peak tonnage corresponds to the end of
year yield. This is no different than the
assumption of mid-season NDVI being the best
indicator of yield.
15
QUESTIONS?