ESTIMATEUR DE REGRESSION
i
uem 2 % de culture
-— A N terram
E y^
fi ^. x DS x culture A
4 IM t vr UE
7 (4
TA |seoments = ne a
E Ye Y
7 ^ validetion ) —
‘ Yoga +
! classement
; ımage satelhte
" 1
VE. 3 cz
= Vision exhaustive X Xe % culture
(100%)
Yrg=a(X Xe) + Ye
Var(Y reg) 7 (1- R*) Var (Ye)
*
surfaces
RESULTATS
> Erreur standart
J
TELEDETECTION
EXTRAPOLATION PAR SURFACE PHYSIQUE
The classified data could reasonably be described by a linear line
of type Y = aX + b + e, where the variable X is the surface
obtained by classification. In these conditions, the line of
regression of Y in X has for equation Y = 0.9788 X + 1.5896.
The sample segments are taken on some blocks of (10x10)
segments of systematic uncertain way with a rate of poll of 4 %.
This rate is chosen as a function of degree of precision. The
positions of pull of the segments were taken at random on the
first block, then repeated on the following blocks.
The surface of the image is 44100 hectares. the one obtained by
classification for the cereals class are of 14238.68 hectares. The
evaluation which could be given from ground data is of the order
of 15267.42 hectares. The difference between the result of the
classification and the evaluation obtained to lcave some ground
data, could be explained by the confusion that we note between
the cereals, which represent a rate of very weak cover, and thc
naked soils. Nevertheless, in a global manner, we can note that on
the 36 drawn segments the difference between the classified data
and the ground data is not meaningful (2,5394). The regression
line calculated is presented like follows:
Données terrain (Ha)
50.00
40.00
30 00 :
Y = 0,9788 X + 1,5896
20.00
10 00
0.00 —$-—a— —e ! 1 X
0.00 20.00 40 00
Résultats classés (Ha)
Fig. 11 : Line of regression obtained by crossing the ground data
and remote sensing data.
Note that the interpretation of the regression line is linked to the
importance of the existing correlation between ground data and
the classified data. In our case is high since it is worth 0.92. The
coefficient of determination r? is worth 0,848 then, this means
that 84.896 of variations between the inventoried segments could
be explained by the linear influence of classified data.
We can, because of to the variance obtained by the regression,
calculate the efficiency of remote sensing |[MEYER ROUX and
al. - 87], which could be calculated by the following formula:
Vary,) | 1
Var(y,. ) x 1-r?
reg
We notice that the efficiency of remote sensing increases r? with
the growth of the coefficient of determination.
æ
> 10 ;
a 80 -
9 60 +
=
u 40:
20 +
0 : -— on
e
"t 9 qo) DD (D O X — ©
OO ~~ (wN (00 «x © ON
OO OO OO OO OC CO
Fig. 12 : Line represent efficiency of remote sensing.
In our case, the coefficient of determination is of 0.848 this gives
us an efficiency of 6,58%, allows us to affirm that the
introduction of remote sensing improves the agricultural
statistics. The efficiency. of remote sensing begins to increase
quickly from a coefficient of determination r? of 0,81.
GENERALE CONCLUSION
Agriculture must be based on some agricultural reliable statistics
for an optimal set up. However, the present statistics do not
answer the needs. It is therefore necessary to attempt to improve
these data, unfortunately little useful in their present state. The
contribution of remote sensing in the improvement of the
agricultural statistics - in terms of surfaces- was studied on the
region of Oum El Bouaghi. The accent was put on the collection
of the surfaces cultivated by cereals. The results of this survey
allowed us to note imprecisions. In order to improve the
evaluations appreciablv, the results of the image processing was
confronted to the results of the ground rcality with regression
estimator. Indeed, this method allowed us to get a result for which .
the margin of errors on the whole of the segments samples are of
2,53%. The image processing combined to ground data permitted
the development of a methodology of the following cultivated
surfaces. Remote sensing is a tool appropriated for the
cartography of agricultural domain to the regional scale. The
evaluation of the agricultural surfaces is reliable, but then some
outputs prove out to be delicate in the absence of measurement on
the segments. Remote sensing permits the constitution of a basis
of area poll also for some terrain investigations. The stratification
of the region in homogeneous zones improve the precision of the
evaluations. The perspectives of this study must first confirm the
results obtained concerning the regression estimator and
especially go toward the improvement of the operational
processing. This will make in a worry of the generalization of this
approach to some agricultural different contexts.
BIBLIOGRAPHIE
[BENHAMOUDA et al. 94]: F. Benhamouda, A. Hassani, T.
Mostephaoui, A. Dif, L. Kebir, Z. Zebbar, A.Z. Saad (1994):
« évaluation des potentialités agricoles de la wilaya d'Oum El
Bouaghi par télédétection ». Rapp. N°1, CNTS/DSA d’Oum El
Bouaghi de Mai 1994; 42 p.
298 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
[FOUL
Approc
de do
l'Essor
[HIED
Guede
surface
sympo:
du 19 a
[MEYI
Sigmai
statisti
agricole
[WILN
Amélio
Rappor
régiona
[D. YO
pour |:
Applic:
CNTS,
