14 . Kongress der Internationalen Gesellschaft fur Photogramme trie Hamburg 1980 , Kornrnission VII , Arbeitsgruppe 2
A SAMPLING TECHNIQUE TO ASSESS SITE , STAND , AND DAMAGE
CHARACTERISTICS OF PINE FORESTS ON CIR AERIAL PHOTOGRAPHS
H. U. Scherrer , H. Fltihler, F . Mahrer , and o. u. Braker
Federal Institute of Forestry Research , CH - 8903 Birrnensdorf
Switzerland
Abstract
A test area of the severely damaged pine forests in the lower
Swiss Rhone Valley was sampled systematically . Nine site,
stand , and damage characteristics were rated and encoded for
each sample plot using medium scale color infrared (CIR) aerial
photographs . The objective of this study is to quantify the
interdependences between these variables . Using a multiple
linear regression in connection with a principal component
analysis , the damage variable " pine mortality " .is expressed
in terms of site and stand characteristics .
Introduction
This study is part of a forest damage investigation carried
out in the lower Rhone Valley , Switzerland (SCHERRER et al .,
in prep . ) . Some of its pine forests are severely damaged .
This study was undertaken with the following question in mind :
To what extent is it possible to explain " pine mortality "
with those site and stand characteristi cs which can be clearly
recognized on medium scale erR - photographs? In this sense ,
it i s an analysis of how the interdependent variables con tribute to "pi ne mortality ".
The surveyed area is 426 sqkrn with 130 sqkrn of forest land .
The industrialized val l ey is more than 100 krn long and rela tively narrow , encompassed by alpine mountains rising to an
elevation of some 3000 - 4000 rn above the valley bottom . The
frequent , low and stab l e i nversion layers enclose an atmos pher i c volume of a limi ted di l ution capacity for air pollutants.
Not only air pollution but also periodic drought and other
natural stress factors might be possible causes of the observed
forest damage .
Methods and results
The 781 sample plots , arranged in a systemat i c network covering
a test area of 196 ha , were examined on CIR - aerial photographs
(23x23 ern , f=l53 rnrn , scale 1 : 13000) using an autograph Wild B8 .
Each plot of 0 . 25 ha was rated according to the nine criteria
given in Tab . 1.
BDL.:&:.
Tab. 1
Rat in g criteria
a) site charact er istics
b) stand characteristics
1 elevation above sea level
50 m intervals
(1 = 451 -500 m ... 15 = 1151·-1200 m)
4 stand type
6 classes: composition of pine, dec iduous and
other coniferous species
2 exposure
9 classes: 8 az imuth segments and horizontal
plots without exposure
5 canopy coverage
4 classes: 0.1 -0 .3, 0.3-0.6, 0.6-0.8, >0.8
6 age of stand
4 classes : young, intermediate, old, heterogeneous
3 topography
5 classes : trough, horizonta l, gentle slope,
steep slope, ridge top
7 site index
3 classes: poor, intermediate, good
c) damage characteristics
9 age of the dead pine trees
5 classes: no dead pine trees, young, intermediate,
o ld, heterogeneous
8 pine mortality
4 classes: expressed in% of all pine trees
per plot
The frequency distributions of the 781 observations per criteri o n prov ide a first quantitative informat i on (SCHERRER
et al . 1979). They are illustrated in Fig. 1 for the two
criteria "topography " and "exposure ". 83 % of all sampled
plo ts are located on slopes , 8 % on poor ridge top sites
and on ly 9 % on fairly pr oduct ive sites in d epr ess ions
(troughs , Fig . la). The predominant exposure is NW (63 %,
4 89 plots) which is perpendicular to the valley axis (Fig . lb) .
Frequen t and strong winds , mainly fr om the west , aggravate
the effects of drought periods on W- exposed sites (11 %) •
On N-facing slopes the water regime is more favorable . The
Fig. 2 Sampling plots with and without dead pine trees
in different altitudes
Fig. 1 Absolute freq uencies of stand characteristics
of the sampled plots
% equals number of
D all100
plots without dead
pi ne trees (N 302)
lliiiiJ
m
=
1100
1000
=
E
Q)
I
I
900
>
2
300
ni
number of sampling plots
I
"'>
IIIIII
l
0
"'c:
700
0
·;::;
"'>
b) Exposure
"'
600
I
a;
~
I
I
1111111111111111
I
500
(51
II
I
"'~ BOO
.0
~
I
100% equals number of
all plots wit h dead
pme t rees (N = 479)
I
L
400
(%)
(61
aos.
15
10
5
0
5
10
15 (%)
variable "exposure" contains information about several cli matic factors such as wind, precipitation and ~alation.
Stratification of the sample plots according to one variable ,
i.e. "damaged" vs . "undamaged" stands, provides insight into
mutual interdependences of the correlated variables . Fig . 2
shows the relative occurence of damaged and undamaged stands
at different elevations . The maxima and minima of the two
distributions are complements of eachother. The most severely
damaged stands are located at the average elevation of typi cal pine sites .
A systematic multifactorial sampling network yields a mosaic type map for each variable which is a useful tool for forest
management . This information about the spatial distribution
of the nine characteristics is obtained in a way as by - product
of the analysis (Fig . 3) .
Fig. 3
M~saic-type
map of pine mortality
+
+
+
+
+
t
0
0
@
+
+
0
@
00~
+
+
+
+
0 mortality 0%
0
@mortality <20%
00
@)mortality >20% to <40%
Charrat-Vison
~mortality
+
+
+
+
WRLDSCHRDENKRRTIERUNG CHRRRRT-SRX0N (VSJ, M0RTRLITRET DER F0EHREN
806.
;;;.40%
+
500 m
+
In the context of this study , " pine mortality " is of prime
concern . It might be considered as the dependent variable
Yi expr essed as a funct i on of the other eight variables which
were included in th i s study .
In order to apply a multiple linear regression model (MLR)
two conditions must be met . First , the eight "independent "
variables should be uncorrelated and second , the average
"pine mortality " per criteria class of the individual vari ables should exhibit an either i ncreasing or decreasing trend.
Whenever it can be logically justified the code classes should
be r earranged to satisfy this requirement . For certain cr i teria
such as stand type , topography and exposure one cannot objec tively def i ne the sequence and espec i al l y the ranking codes
because they were arbitrarily chosen and encoded . Four criteria
were partially re - classified . Since for a given pine mortality
the relative frequencies per observation class are unevenly
distributed the following procedure was applied :
The observed frequencies were expressed as percentages per
code class of the re - examined variable . Doing this , the un equa lly represented frequencies are we i ghted . As shown in
Tab . 2a and 2b , and i n F i g . 4a and 4b the original code classes
Tab. 2a Absolute and relat ive frequencies of
different exposures per pine tree mortality class
Tab . 2b Absolute and relative f requencies of different
forest stand types per p ine mortality class
stand type
exposure
w
SN+NW
S+O
SE+N
* *
E+NE
pine
mo rtal ity
pi ne
mortal ity
* * **
0
'<!"
0
Cl
\1
\/
8
8
*
'*'
O?f?.o
OO?f?.
o ?f2, o
VI/\
V /\ u
II\ I /'t.
~om
~mo
0. "0 u
0."0 u
~o<D
?ft.?!?.
00
'<!"Cl
vv
......
?f?.?f?.
0 0
oo?f?.
~<Do
*mo
0
\,/ +-"
o*-*...
oo
0
~
~ ~.
<0
"
'<!"
*
*~
0
Cl
/\. !'\ II
· ~
· ~
0. "0 u
0. "0 u
o.::~
o.c~
1
(0%)
17
(20)
186
(37)
30
(39)
60
(62)
9
(60)
1
(0%)
137
(88)
(72)
21
(36)
22
(17)
23
(12)
3
(3)
2
(< 20%)
16
(19)
11 6
(23)
14
(18)
12
( 12)
3
(20)
2
(< 20%)
8
(5)
6
(4)
22
(36)
29
(22)
65
(32)
31
(30)
( ~ · 20%
3
19
(23)
103
(20)
15
( 19)
15
( 16)
0
(0)
3
5
(3)
7
(5)
8
( 14)
37
(29)
57
(29)
38
(36)
4
(:;;;,40%)
31
(38)
103
(20)
19
(24)
10
( 10)
3
(20)
4
(:;;;,40%)
6
(4)
25
(19)
8
( 14)
41
(32)
53
(27)
33
(3 1)
total
83
( 100)
508
(100)
78
(100)
97
(100)
15
(100)
total
156
(100)
134
(100)
59
(100)
129
(100)
198
(100)
105
( 100)
orig inal
code
8
7+9
6+ 1
5+2
4+3
orig ina l
code
5
1
6
2
3
4
rearranged
code
1
2
3
4
5
rearranged
code
1
2
3
4
5
6
to
< 40%)
(percentages in parentheses)
( ;.. 20% to
< 40%)
96
(percentages in pa rentheses)
807.
p = p ine trees
d = deciduous trees
c = other co ni ferous trees
were rearranged in a way that the relative frequencies either
increase or decrease within each pine mortality class .
The average "pine mortality" of the nine exposure classes
exhibits a max i mum on W- facing , a minimum on E-facing, and
an intermediate value on horizontal sites . Note, that only
the code sequence but not the numerical value of individual
observations was adjusted.
Pi 1%] =number of sample plots per
100
total number of topography
class (s. Tab . 1).
Pi [%] =number of sample plots per total number of elevation class (s. Tab. 1 ).
100
I
pine mortality class 1 (no mortality)
80
-----
60
-
I
pine mortality class 4 (mortality > 40%)
I
o.c.
original code
I
r.c.
rearranged code
pine mortality class 1
(mortality O' ;)
80
- - pine mortality class 4
(mortality : 40 ~,)
o.c. original code
60
(= 6 height from 725 m)
40
40
I
,- _,
20
I
I
'' '
1
20
\
\
\
0
____ ... '
\. ..........
, ...
I
o~~-;---+--,_--r--+---r--+-~r--+--4---+-_,--'~ -~
o.c.
r.c.
Fig. 4a
1
6
475
2
3
5
4
525 575
4
5
6
3
2
1
625 675 725
7
8
9
10
11
12
13
14
15
2
3
4
5
6
7
8
9
10
775 825 875 925 975 1025 1075 1125 11 75 m a.s.l.
o.c. 5
r.c. 1
1
2
2
3
rn
.r:
c:
"'c.0
u;
g"'
::>
2
'§
.r:
~
~
~
"'
pine mortality J:lasses.
Most of the nine variables, each with 781 observations, are
highly correlated (Tab. 3) . Correlation coefficients higher
than 0 . 09 or 0 .1 2 are significant at a 95 % or 99 % level,
respectively . Hence these variables are not "independent"
as required for a regression analysis .
Correlation coefficiants between the rated variables
"'
~
.....
Q)
Q)
Q)
Ol
>
.<=:
E
c:
-~
~
.....
"'>
Q)
"'"'c.
Q;
Q)
0
X
>
..r::::
c.
E
Ol
0
c.
.....
0
Q)
c.
.....>
"t:J
c:
"'
t>
Q)
"t:J
0
-=
>
u
>
c.
0
c:
"'u
c:
.....
"'"'
0
Q)
"'
Cii
X
Q)
"t:J
pine mortality
age of dead pine trees
BOB.
t
0
c:
E
.....
c:
c.
Q)
·;;;
1.00 0.07 - 0.10 -0.26 0.03 -0.02 0.28
exposure
1.00 0.06 -0.20 0.13-0.03 0.17
topography
1.00 0.17 -0.18 -0.05 -0.40
1.00 -0.17 0.08 -0.50
stand type
1.00 0.00 0.4 7
canopy coverage
1.00 0.07
age of stand
1.00
site index
elevation
4
5
"'
c.
0
u;
c.
0
u;
c.
~
"'
ill,
"0
·;:::
Fi~. 4b
Relative frequencies of
different topagraphy classes for two
Relative frequencies of different elevations for two pine mortality classes.
Tab. 3
3
4
Q)
c:
c.
"t:J
"'
Q)
"t:J
0
Q)
Ol
"'
0.37-0.32
0.19 0.21
0.25 0.21
0.54 0.63
0.11 -0.09
0.22 0.33
0.42 -0.38
1.00 0.83
1.00
In order to obtain truly "independent" variables the observations Xij (variable j, plot i) were normalized , Xij=(Xij-x)/sx· r
and then transformed by means of a principal component (PC)
J
analysis which yields the orthogonal , uncorrelated variables
*Xij !/. The multiple linear regression program for "pine
mortality" (yi), run with the *Xij's returns the regression
coefficients for both , the orthogonal and the original coordinates , *Xij and Xij r respectively . This technique was
used by FRITTS et al . (1971) for a s i milar statistical problem
i n the field of tree ring analysis .
The response functions which are the regression coefficients
computed with the *Xij's and projected onto the coordinates
of the Xij ' s reflect the relative importance of the individual
variables with respect to pine mortality (yi) •
k
Yi = '1 + Sy
l:
j=l
a . x ..
J
l J
In a first run (model A) of the PC-MLR- program all variables
(k=8) were included. This model explains 69 . 9 % of the observed
variance in Yi (Fig . Sa) . Those variables wi th coefficients
significantly different from zero were included i n a second
run (model B, k=4) whi ch still accounts for 69 . 9 % of the
variance (Fig . Sb) . With only the two most significant vari ables (k=2) of run A and B ( " age of dead pine trees " and
" elevation " ) 68 . 8 % of the variance i s exp l ained (model c,
Fig. Sc) . Almost no in formation is lost due to the drastically
reduced number of variables . A simple linear regression versus
the most significant variable ( " age of dead p i ne trees " )
accounts for 67 . 7 % of the variance . The role of the remaining site and stand var i ables i s obviously covered up by the
dominant "age of dead pine trees" variable . Without this
var i able but including all the others , model D (Fig . Sd ,
k=7) leads to the interesting conclusion that all remaining
site and stand variables significantly contr i bute to the
pred i ctor of Yi · However, i n this last run only 42 % of the
var i ance is explained .
!/
Program package RESPONS Version 1971 , H. C . FRITTS, Tree
Ring Laboratory, Univ. of Arizona , Tucson .
809.
rT
0.8
a)
c)
M =9
M =2
R2 = 0.688
R2 = 0.699
d)
M =7
R2 = 0.420
0.6
0.4
0.2
0.0
~
ill
Q)
~
~
Q)
Q)
c
"'E
"'>0
Cl
-0.2
c
0
-~
>
"'
-.;
E
iii0
Q.
X
"'
>
.r:::
Q.
E
Cl
"'
Q.
~
0
1J
B
t!
Q.
c:
u
>
Q.
0
c
"'
u
c
·c.
1J
1J
c
t;
"'
0
Q)
Cl
"'
Fig . 5a ·- d
"'"'
-5"'
.....
1J
1J
X
"'
.!':
1J
~
·;;;
0
"'
"'
Cl
·c.
1J
"'"'
1J
c
0
·;:;
>
"'
"'
-.;
>
.c
Q.
E
Cl
0
Q.
B
"'"'
1J
X
"'
1J
.!':
~
·;;;
-5"'
.....0
"'
"'
Cl
c:
0
·;:;
>
"'
"'
-.;
-5"'
.....
0
"'
Cl
"'
"'E
"'>0
Cl
c
0
·;:;
"'>
"'
-.;
:;"'
>
.r:::
Q.
"'
Q.
E
~
<':l
Cl
0
1J
X
B
"'
Q.
"'
Q.
c:
t;
u
1J
c
"'
t;
X
.....
1J
0
0
"'
.!':
"'u
"'
~
·;;;
>
Q.
c:
Cl
"'
Regression coefficients of damage prediction models (M =normalized variables)
Conclusions
This statistical approach overcomes the difficu lty that part
of the i nformation of a s i ngle variable is contained in others ,
t oo . It must be recognized that this analysis is no proof
of cause and effect but it is a strong indication of possible
synergisms between different facto rs . In this sense we might
hypothesize that natural or premature aging is e ither the cause
itself or a prerequis it e for the observed severe forest damage .
Under the influence of adverse site conditions such as drought
and/or shallow soils on steep slopes or eroded ridge tops
this trend appears to be more pronounced .
8:1.0.
References
FRI TTS , H. C ., T . J . BLASING , B. P . HAYDEN , J . E . KUTZBACH (1971) .
Multivariate Techniques for Specifying Tree - Growth and Climate
Relationships and for Reconstructing Anomalies in Paleoclimate .
Appl . Meteorology 10(5) : 845 - 864 .
SCHERRER ,
rating on
Proc . 7th
the Plant
H. U., H. FLUEHLER , B. OESTER (1979) . Pine damage
medium and large scale infrared aerial photographs .
Biennal Workshop on Color Aerial Photography in
Sciences , Davis CA . P . 151- 160 .
SCHERRER , H. U., H. FLUEHLER , F . MAHRER , O. U. BRAEKER (1980) .
Walliser Waldschaden . I . Alternati ve Verfahren flir d i e Inter pretation von Fohrenschaden (P . silvestris L . ) auf mittelmass stabl i chen I nfrarot- Farbaufnahmen . (in prep . ) .
811.