Alignment of Growth Seasons from Satellite Data

1
Alignment of Growth Seasons from Satellite Data
Ragnar Bang Huseby, Lars Aurdal, Dagrun Vikhamar,
Line Eikvil, Anne Solberg, Rune Solberg
2
What?

• Determine the correspondence between the
  chronological time of the image acquisition and
  the time at which the phenological state of the
  vegetation cover shown in the image would
  typically occur.

?
?
Landsat image
3
Why?

• The work is motivated by a high mountain
  vegetation classification problem in Norway.
• Vegetation classes are characterized by their
  temporal evolution through a growth season.
• Data of high spatial resolution are often
  temporally sparse.
• In order to get a longer sequence of images,
  data from different years should be combined into
  one single synthetic sequence.

4
Combining image data
May 19th May 23rd June 5th June 25th July
24th July 29th August 17th August 19th October
18th
Naive approach
5
Problems with naive approach...
May 23rd 2004
June 5th 1997
6
How?
NDVI The normalized difference vegetation index
Extract the maximum NDVI from the region of
interest.
7
NDVI in various growth seasons
alignment of growth seasons alignment of NDVI
curves
8
Model of observations

is the expected NDVI as a function of
phenological time.
9
Double logistic function
The parameters can be estimated from data from a
given year.
10
Time warping functions
Early spring
Normal spring
Late spring
11
Minimization problem
We wish to determine the warping function .
The discrepancy between model and observations
should be small.
Minimizing this object function produces weird
results.
12
Object function
We wish to determine the warping function .
The warping function minimizes the sum
13
Dynamic programming

• Discretize time
• Finite but huge number of feasible solutions
• Find optimal solution by dynamic programming
• Similar approaches have been used in speech
  recognition.

14
Experiments

• Dataset for time warping
• Global Inventory Modeling and Mapping Studies
  (GIMMS)
• A sequence of NDVI composites extracted from
• NOAA AVHRR data from the period 1982 2003.
• 24 images from each year.
• In addition, MODIS data from 2003 and 2004.
• Test site Region in Norway including mountain
  areas.

15
Results
16
Images in correct order!
Day 156, 1997
Day 144, 2004
17
Application

• The results from this study have been used in
  the work titled
• Classification of Multitemporal Satellite Images
  Using Phenological Models
• also a talk at MultiTemp 2005.

18
Conclusions

• We have proposed a methodology for alignment of
  growth seasons from satellite data.
• The proposed methodology has been tested on data
  from several years covering a region in Norway
  including mountain areas.
• The results show that it is possible to combine
  data from several years into a sequence of
  observations from one season.
• We expect that the methodology will serve as a
  useful tool in multi-temporal classification.