 |
Change Detection Exercise:
Image Deviation
|
Introduction
While techniques appear to be quite rich for the pairwise comparison of
images, far fewer methods exist for cases where much larger time series
data are being examined. Here we are concerned with a whole sequence of
images and the ability to examine trends in environmental change or the
abstraction of significant anomalies from the general trend (Eastman et
al, 1995).
One of the techniques for analyzing long time series data; image deviation
will be addressed in this exercise. With image deviation, it is assumed
that change areas are identified by contrast to a long-term average or
characteristic condition. Given such a characteristic image, the deviation
of any particular time from this long-term average can then be assessed
by simple differencing. One possibility would be to produce a mean image
(i.e., a simple average) over the whole series. This would then allow change
for any specific image to be evaluated by subtracting it from the mean
(Eastman et al, 1995).
The Data Set
NDVI Images for the month of March from 1982 to 1991 will be used along
with the NDVI image for March of 1992. This data set is the same as the
one used in the image differencing exercise. The data set was produced
by USAID/FEWS from compositing dekadal (10 day) images into a monthly maximum
value composite, where the maximum value over the three dekads in each
month for each pixel is used to represent that pixels monthly vegetation
index.
Procedure
The objective of this exercise is to be able to understand change in one
time slice as a deviation from its own long term mean. We will difference
the March 1992 NDVI image from the long term average NDVI image for March.
We would first need to add the ten March images (1982 to 1991) and then
divide them by ten to get the average March NDVI image.
P1 Use the Map Algebra procedures of your software to add the ten March
images (1982 to 1991) and then divide them by ten. Call the resulting image
SMARLAV (march long term average). Display it with a continuous 256 color
palette. Display SMARLAV along with the country boundaries vector file
and the country names file.
Q1 Given this long term average March image, what can you say about
the spatial location of the desert areas and rainforests.
P2 If your system permits it, simultaneously display SMAR92 and see
if you can visually detect anomalies from the long term average image.
Use the inquiry option in your software to check the NDVI values across
the images. Now that we have the average March and the March 1992 images,
we could do a difference operation between them to detect the status of
the current growing season in relation to the long term mean.
P3 Use the Map Algebra procedures of your software to subtract SMARLAV
from SMAR92. Call the output image SDMAR92. Display it with a continuous
256 color palette and overlay the country boundary vector file.
Q2 Can you detect areas of negative anomaly in March of 92? Hint: Use
the inquiry option to check the values along the Zimbabwe-Mozambique border,
and Central Botswana.
We could reclassify the SDMAR92 image into four categories to better identify
areas of change.
P4 Use the reclassification procedure in your software to assign the
following values:
| Assign a new value of |
to all old values |
| ranging from |
to |
| 1 |
-202 |
-21 |
| 2 |
-20 |
0 |
| 3 |
1 |
19 |
| 4 |
20 |
120 |
| 0 |
0 |
1 |
The threshold of 20 was based on expert opinion to identify areas of significant
change. Call the output SDMAR92R. Display SDMAR92R with a any color palette.
Categories 2 and 3 in this image represent areas of non-significant change,
while 1 and 4 correspond to areas of significant departure from the long
term mean.
You should now be able to clearly see the areas of negative anomalies
in March of 1992 as compared to the March long term mean. The entire north
and eastern parts of Zimbabwe, Northern Mozambique, Central Botswana, and
Central to Northern Namibia seems to be the worst affected areas. In the
next exercise, we will analyze a relatively more advanced differencing
technique and explore the combination of NDVI difference with socio-economic
factors.
Observations
Image deviation is a commonly used technique for expressing current month
anomalies in its long term setting. It is very helpful as a monitoring
tool during a growing season to understand the spatial extent and progress
of vegetation. In practice, agencies such as USAID/Famine Early Warning
System, FAO/Global Information and Early Warning System, etc. publish dekadal
(10 days) and monthly images for most of the African countries as deviations
from their long term mean. It is an important source of information for
seeking donor funding. For more information on these, the following web
sites might be useful:
USAID/Famine Early
Warning System
Food and Agriculture Organization
Back
to Module 8 - Digital Change Detection
Credits
This exercise was written by Mahadevan Ramachandran, Research Fellow at
the IDRISI Project/Clark University. A major portion of this exercise was
culled from the NDVI workbook and the Change and Time series Analysis workbook
(Ramachandran et al, 1996 and Eastman et al, 1995).
References
Eastman et al. (1995) "Change and Time Series Analysis" United Nations
Institute for Training and Research/Clark Labs for Cartographic Technology
and Geographic Analysis, Clark University, Worcester, MA 01610, USA.
Ramachandran et al. (1996) "An African NDVI Archive, 1982-1996: Techniques
for Environmental Monitoring and Analysis" United Nations Institute for
Training and Research/Clark Labs for Cartographic Technology and Geographic
Analysis, Clark University, Worcester, MA 01610, USA (forthcoming).