Olsson, Hàkan
The lines are local Loess regressions for pine-dominated and spruce-dominated plots. The purpose is to get an idea of
the general spectral development profile over the long term. The interpolated lines are shown on a scatterplot of band 4
and 5 (fig 3c) to illustrate the changing spectral signature with age. A monotonic decrease is apparent in both bands,
with a plausible exponential decay form. There is large variance around the mean trend, which may reflect differences
in site conditions and growth rate.
=
2 % s = ©
- tt ; 100
F 100 ] 2 -
L 80
hi wn
© © x
e = c
© a ©
s So 3
© © d
z z z
ke i E
L 40 +
20 1 r 20 1
T T T T
0 ; : ; 0 > : ; ; 20 40 60 80 100
0 50 100 150 0 50 100 150 TM94 Band 5
Age Age
Figure 3. a) Scatterplot of TM band 4 (1994) vs age with Loess interpolation for pine- and spruce-dominated
sample plots. b) same with TM band 5. c) Band 4-5 spectral space with the Loess interpolations from a) and b)
plotted against all pixel values in background.
4.2 Longitudinal comparison between plots
Now we focus on comparing the temporal behavior of individual
plots over the 12-year period. From the cross-sectional view
above, we would expect that time trajectories for individual plots
would be generally decreasing over the period, with a rate or slope
that is influenced by the age of the forest on the plot. Over this
short period, we can just examine the linear component of the trend
as a simplification. If
the rate of decrease is a
function of age, then
we could theorize that
for a given age, ee m un n e
3 differences in the slope Dos
would be an indication Figure 4. Subset of pine dominated
2 of the productivity of plot profiles in TM5 (normalized)
1 the site.
3 0 For this investigation, we selected first plots that were pine- dominated and
a between 5 and 50 yrs of age in order to reduce the variance from species
4 differences. There were 251 plots in this subset, and it is rather difficult to
2 display all of the temporal profiles graphically, so a subset of 20 are shown
in figure 4 to give an idea of the variance around individual lines.
-3
In longitudinal analysis it is often useful to first ignore serial correlation
and fit functions using ordinary least squares to all profiles or groups. The
20 40 60 80 100 . ; . . .
bSint model for this written in longitudinal form would be
Figure 5. Plot of fitted linear y; =a, + Bit, T£. (1)
coefficients for Pine-dominated ) e ]
plots in TM5.
1086 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.