Thresholding Process
The premise is that an “irrigated” crop needs to meet two con-
ditions: low temperature and high vegetation, as compared to
other land covers. The two conditions were investigated using
the measures of temperature (
Ts
−
Ta
) and vegetation (
NDVI
).
As a process, both measures were segmented into two classes
each. Temperature classes referred to “irrigated” and “non-
irrigated,” whereas vegetation classes were “crop” and “non-
crop.” An iterative thresholding method was used to achieve
the binary classification by minimizing within-class variance
(
σ
2
Within
)(Otsu, 1979):
T
T
σ
ω σ
ω σ
Within i
i
i
T
2
0 0
2
1 1
2
( )
=
( )
+
( )
(5)
where
T
i
is the threshold which varies by iteration
i
,
ω
0
and
ω
1
are the weights of the two classes; denote proportion of
pixels in respectively class, and
σ
2
0
and
σ
2
1
are the variance of
the two classes.
For operational purposes, thresholding procedure utilized
the relationship of
σ
2
Within
with between-class variance (
σ
2
Between
)
and total variance (
σ
2
Total
)(Otsu 1979):
T
T
σ
σ
σ
Within i
Total
Between i
2
2
2
( )
= −
( )
(6)
Initial
NDVI
threshold (
α
) was taken as 0.4. All pixels >
α
were considered as “crop.” Initial
Ts
−
Ta
threshold (
β
) was
the median value of
Ts
−
Ta
. All pixels <
β
were taken as “ir-
rigated.” Iteration interval was set at 0.005 within the limit of
±0.025 of the initial thresholds. Altogether 11 iterations each
for
NDVI
and
Ts
−
Ta
were performed for each of Landsat-7
and
ASTER
images. Thresholds with minimum
σ
2
Within
were
used for binary classification.
Binary classes were assigned to each pixel with four
combinations as shown in Figure 4. S1 denotes dry condition
with no or low vegetation, indicating non-irrigated areas with
no crop. S2 denotes wet condition with low or no vegetation,
indicating the possibility of irrigation without any crop. S3
indicates some vegetation, possibly crop, without irrigation
(dry condition). S4 denotes vegetation with wet condition,
i.e., “irrigated crop/pasture.” Pixels with assigned classes
were compiled into seasonal matrices.
Figure 4. Identification of pixels based on
ndvi
threshold (
α
) and
Ts − Ta
threshold (
β
). S4 denotes irrigated crop / pasture, where-
as S3 denotes possible crop or pasture which is not currently
irrigated. S2 denotes wet condition with no or negligible vegeta-
tion. S1 denotes dry condition with no or negligible vegetation.
A validation was carried out on “irrigated crop/pasture”
(S4) for each of the three seasons separately. The results of
mapped S4 areas were compared with the actual irrigation
deliveries to farms. Information on irrigation water supplies
at farm level, for the three seasons, was sourced from the
Victorian Water Register (
VWR
), a statewide irrigation water
database (
waterregister.vic.gov.au/
).
(a)
(b)
(c)
Figure 5. Schematic showing the profiling of irrigated farm-
lands to determine land cover classes, using seasonal surface
temperature and vegetation status. Examples of irrigated crops
shown include: (a) Summer crops, (b) Annual crops or pasture,
and (c) Perennial crops or pasture.
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
March 2015
233
