Online ISSN: 2382-9931
Journal of Water and Irrigation Management
Homepage: https://jwim.ut.ac.ir/
University of Tehran Press
Dissolved Oxygen Modeling Using Deep Learning and Pre-Processor Methods
Kiyoumars Roushangar1 | Sina Davoudi2
1. Corresponding Author, Department of Water Engineering, Faculty of Civil Engineering, University of Tabriz, Tabriz, Iran. E-mail: [email protected]
2. Department of Water Engineering, Faculty of Civil Engineering, University of Tabriz, Tabriz, Iran. E-mail:
Article Info ABSTRACT
Article type:
Research Article
Article history:
Received: 17 July 2022 Received in revised form:
20 September 2022 Accepted: 04 October 2022 Published online:
25 December 2022
Keywords:
Dissolved Oxygen,
Empirical Mode Decomposition, Long Short-Term Memory, Wavelet Transform.
Water pollution is a major global problem that requires constant evaluation and revision of water resources policy at all levels. Dissolved oxygen (DO) is one of the most important indicators of water quality. In the present study, the water quality parameter of dissolved oxygen using intelligent Long Short-Term Memory (LSTM) method based on discrete wavelet transform (DWT) and Complementary Ensemble Empirical Mode Decomposition (CEEMD) pre-processor methods in both temporal and spatial modes. It was investigated in five consecutive stations on the Savannah River. The results of analysis of models showed the ability and high efficiency of the method used in estimating the amount of dissolved oxygen in water. On the other hand, pre-processor methods improved the results. It was also observed in the investigations that the results of analysis based on wavelet transformation in spatial modeling reduced the RMSE error by two percent and also the empirical mode decomposition in temporal modeling by 15 percent. The best evaluation for test data was obtained using the empirical mode decomposition in temporal modeling corresponding to the previous day with values of DC=0.977, R=0.988 and RMSE=0.017. Also, in the spatial modeling to estimate dissolved oxygen in the third station, it was found that the results obtained from the inputs of the dissolved oxygen parameter one day before the second station and two days before the first station have the best results.
Cite this article: Roushangar, K., & Davoudi, S. (2022). Dissolved Oxygen Modeling Using Deep Learning and Pre- Processor Methods. Journal of Water and Irrigation Management, 12 (4), 873-890.
DOI: http://doi.org/10.22059/jwim.2022.345864.1005
© The Author(s). Publisher: University of Tehran Press.
DOI: http://doi.org/10.22059/jwim.2022.345864.1005
نا هﺎﮕﺸ اد تارﺎﺸ ا
Homepage: https://jwim.ut.ac.ir/
! "
# "
1 2 |
1 . !
"
!
"
.
:$ %
2 . !
"
!
"
.
:$ %
: !
' () *+ ,
"
#
$
#
%&
26: / 04 / 1401
"
#
$
!' ( 29 : / 06 / 1401
"
#
$ )*
# + 12 : / 07 / 1401
"
#
$ ,-!
04 : / 10 / 1401
. / : 0
(3 4 56
7 ! 89
$ ":!
; :!
< = $>? @
! 4 A
+
8 7 B%";
9 C
$4
$;
; D%
E
> :!
% C 3 F;
;
%
$ ' G H I 3J@
% . (3 4 ) 56 (DO
D M N ! OP Q R ' C3S34 C
.
%
$T+ H U @ J % ) S34 (3 4 56
% ; SJ D V % Q '
$>? @
< =
! 4 A )
; (LSTM
$ ) V % R ' W3) D ) 7 !
) $J 89
(DWT
V %
$ ":!
; :!
) 7 4
CEEMD
% ( C+ @ D
% X ) J + J ; R %
$ P %
%
% ;
% Y C?
. X J 7Z @ D 7356!
' C35; Y
% 4 R< ; V %
$;
$J?%% 4 %
% N3 [!
"3 (3 4 56
%
$;
; P .
D
? = V %
R '
W3) D )
\ ; ; X J Q . M' N3 ]
%
% ; R ' E : Q ' Q
$4 X J
7Z @ D
$ ":!
^ ;
7 ! 89
% RD Z% "3 $;
M' N3 ]
$ ":!
; :!
% RD "3 $; D 15
R HP "3 Z%
W' 4 %RMSE
. N J ;
C+ @
; D%
R ; R ' D ; SJ D
$ ":!
; :!
% C+ @ RD
D
` ;
$;
8 D % 7 Y ; , 977 / 0
= 988 DC
/ 0
= 017 R
/ 0
=
$; RMSE
C . M' N3 ]
% RD C 9
N3 [!
(3 4 56
% J E
"3 O[
Q X J 7Z @
D R % R ' J % ) (3 4 56
% 8 D % 7 Y J E D % 7 Y J
N J ;
$:3J % % . Q ;
- Q %: .c ^ R ) 1401 .(
D R 4 3(
56 ; SJ D V % 3 R 3 d V % R ' ) 3 W D ) C $ .
R% 3;
12 ) 4 ( 873 - 890 . DOI:http//doi.org/10.22059/jwim.2022.345864.1005
: Q
! A % J $ g
. . ©
1 .
& '
W3) 3;
h + % 3' '
; 3i3)
< ; ) E
J
$ 3 ' 8
W+ ] R ;
C h + % 3' '
C
$4 W, %
% C j36 C D F;
M' N3 ]
% W' 4 R k;
` ;
$;
O3[ ! '
.
% R ' 3P W "?
7; Y 9 !
% T!
% R ' 5 Q
%
D R W3) 3;
8 h + % 3' D
$5 9
R ' %
"
J ;
"
"
% 6
"
' Q . N3 ]
R ' %
D R 7 Q A< T U % C
$4
$ D
?
"3?
D @
$ 5;
D
$ ":!
7356!
R R '
D lg
% )
$;
R % 9 P M' D
$J?
Q C )
Nourani et al., 2014
(.
R ; c%
m36Z
D R ' 6H NJP Q R ' J % ) C3S34
$4 j ! R '
$ Q R '
!% >
$n % Q R
! 3@
C .
5 8
$ Q A% >
;
$9 !
$;
R '% 3T o5J[
)
$;
p D ;
$ "' (
! ;
; ; N3 [!
R ' J % ) q P
D ' J % ) 73 ! Q )
Csábrági et al., 2017
( . R ' C3S34 C3S34
$ P % 9
)
QUAL2K
(
$ ;
$3 Q RD
$ ":!
7356!
C3S34 )
(WASP
% 3 ; 3i3) D 3
$;
A k=
R % ? D MJ 3
$ P % J ' )
Pelletier et al., 2006; Wool et al., 2020
(.
(3 4 56
) (DO
$; 1
8
% 3T R 354 M
% j36 R '
; r5s R ;
; D%
C k
MJ 3 4 R '
;
$ P % '
$] % '
^ 3Y ' SJ Q .
$;
N3 '
% >
W' () R ' R TJ
% N
$ 3 D ; SJ D V % R ' R 3
N3Q E : Q C .
Soyupak et al.
) ( 2003
% 8
$T+ H D
$ Q t R ; W3) 3;
C>5s (3 4 56
% C+ @ C; l
$J ;
$;
R u?
$ Q
% )
%
$
"[
o5J[
SJ 4 . r U J ' M'
R ; R ' 3vJ 9 P
D 3 R Q M' R ; W3;
D
95 / 0 . ;
Elhatip and Kömür
) (2008
R ' J % ) C3S34
%
"[
N3 % D
R 4 .
X J !
2ANN
% R ; W3) 3;
(3 4 56
% d=
h% Q
$35[!
% C N3
.
Palani et al.
) (2009
%
$T+ H R
%
% ) R ; N3 [!
JS'
"3 (3 4 56
%
% D
$ Q t Jw ) ]
$ <
; C35; Y x)
% J HP
$ Q t
3 %
M3 T!
$J?
SJ 4 .
$:3J
$ Q t
3 %
M3 T!
$J?
X J R J ; .CQ
Niroobakhsh
) (2012
C 9
$ 6
C>5s 74 9 56
% ) (TDS
D
$ Q Jw ) ANN
]
$ <
) (MLP
5 3
$ ) TQ
) (RBF
R ;4
$ P % 9 9 SJ
4
$4 X J
MLP
% Y RBF
$;
$3 Q RD R ' 3vJ
C3S34
$ P % 9 9 ; CY W3;
D 90 Z%
% . J '
% R 3' R '
Liu et al.
(2013)
8
MJ % +
IPSO-LSSVR
R ; 5
CP 8 W3)
3;
HP 3s y [!
Q .
IPSO-LSSVR
W D %
% Y C?
; SJ D MJ % + R ' R ;
W3) 3;
R J6 (3 4 56
% R ' [J C 4
z ] P
$ P %
$ , Q . X J W D '
$4
R 3)
IPSO-LSSVR
! CY
W3) 3;
% W "?
' 5 M3 T!
% J ; D 3 %
% ; 3J )
% J ) (SVR
$ Q 6
t
7BP
E : . '
Ji et al.
) (2017
$;
% >
W3) 3;
C>5s
% DO
8
$ P % M4 (3 4
% 9 Y Q
N3]
D N3Q
% ; 3J ) )
(SVM
SJ 4 .
% ] 3 3+ 4
o5J[
$;
% = q P
3 %
HP
$ ]
$ Q t
% J r,
$ Q t
3 %
: SVM
Q . X J
$4
% SVM
W3) 3;
C>5s
% DO
$ P % { - R % z ! C
$;
R '
! ; .C
Ahmed
) ( 2017
"3 (3 4 56
%
$ P %
% % ; SJ D
$ Q t
% [ 3) F; !
$ ) TQ
% Y $T+ H % .
X J A S!
N3;
$ Q t
% ' C+ @ 8!
3vJ 3vJ
$;
;
) r U J '
$;
r3! ! 79 / 0 88 / 0 Q ( .
Shi et al.
) (2019
M' N3 ]
% 8 V % 9 J ;
$Q P
)
CSELM
8( R ; W3) 3;
d3Y
% 4 33v!
(3 4 56
D R ' R D 4 SJ
$:3J
$4
CSELM
D
>
CY
% 4
% W3) 3;
R J6 (3 4 56
% R 3 TY X J W3) -
3;
R J ; C
$;
' R
PLS-ELM
$; ELM
C . %
Hosseinpanahi et al.
) (2021
5
V % ' % TQ $ ) F; ! t $ Q 3J ) % ; N3Q R W3)
3;
$ , % $ 3 3 C3S34
54 M |9 C R ' J % ) % > N R ; . 4
$;
OP Q ' }% t % C3S34 R
. Q $J? > % RD% 4
^ ;
TQ $ ) F; ! t $ Q $; C 3J ) % ; N3Q X J
R ; R J ; 5 CY D W3)
3;
$ 3 3 $ P % C3S34 R ' J % ) . ; % % P ; %
F; % ; E :
Q V % R< ; ;% 4 D
W3) $ 3 D % t t $ Q Q ' R ' 3;
% 56 (3 4 N ; . Q ;
% @
% A 3;
?
;% 4 ; $H; % % ] A T+ H $5 R 3
d3
% W3) 3;
"3 (3 4 56
% % ? = D .C $J? A% Z V % D SJ 3P R '
R '
3 $ ":!
$;
3 7356! $ 3 D % lg % 3 ; N Q % '
.C $J? % Y $9 ! % D R '
N ] $; 5Z 3 V % N % Q $ ":! R D
VD ; r A< T ; $ ":! D 7Z @ R '
V % N $5 9 D . Q % % $4 C 7 4 ; :! $ ":! 9 V % 89 7 ! V % '
$ ":!
RD
$ 9 D SJ ; Q 5Z 3 % 3S " D R 6 M:@
S " R% C p R ' F D ! $4 3
}|@ $ ":! % D J " l % x 4 ? D @ % D J . Q
|+
% N W' () T
' P Q ;
SJ D V % N
$ Q R ' J D ; 3
d V % ' R W3) D ) (3 4 56
%
% ; D
R
Q .
2 .
2 . 1 . ) *+
,
&*- . &'./
$ P % D 9
$ P % R ' A<
6J C
$4
$;
^ 3Y x5=
" % . N
$ P % B%";
% 9 Y Q A<
6J C
$4 W3;
! D N3;
C+
R ' R 3+ % 4
; 9 39% 9 % 73 ! '
.
$P Q D
$ P % + !
$ P %
! ] + Q N ! C Y D
% 73 ! '
.
$U @ 'D
$ P %
$;
C 9 Y Q 4 R ' ]< )
% 7P R 3+ % 4 + Q
$4 j ! } Q
% Y Y Q
6 Q
V J
; .
= N
$ P %
@ 301 7 ) 484 J 534 ( C . D
Yk!
$ P %
+ !
$ P % 73 !
Q C . D N 76 Yk!
% D
$] %
!% '
$= s
% C .
$ !
< + !
$4 ; R %
$ P %
< + ! FY Q C
$P Q R D
$ P % + ! C
$4
$P Q Q
; s
$ P % %
73 ! '
. R ' SJ %
% N W' ()
` ;
$;
X ) J h + % 3' + J
%
= CS'
) 2015 ! 2021 (
; R %
$ P % Q ;
D j ! $4 N3 D
Q )
( USGS
R%
% .C Q 7 Q
) 1 (
$,H
$T+ H % C3TY
J '
"3 A 33v!
(3 4 56
% X ) J
CS' = % + J
$;
A% Z
Q $ D %
M' N3 ]
% 9 ) 1 ( A t[
J ' W
Q C .
Figure 1. Study area and location of stations
Table 1. Stations specifications
Hydrometric station Station no. Station Specifications Dissolved Oxygen (mg/l) Longitude Latitude Distance to previous station (km) Min. Max. Ave.
Station 1 02198840 81°09'03.9" 32°14'07.7" - 3.656 11.814 7.697 Station 2 02198920 81°09'17.7" 32°09'55.2" 7.807 3.202 11.286 6.527 Station 3 021989715 81°07'42.0" 32°06'57.9" 6.021 1.522 10.602 5.631 Station 4 021989773 81°04'52.9" 32°04'49.5" 5.942 1.083 10.445 5.527 Station 5 0219897993 81°00'25" 32°06'11" 7.449 2.454 10.350 5.993
2 . 2 . 01 2 34 )
(WT
10
7 ! 89 R% ";
lg R ; I [J A k=
;
$ ":!
3 5Z
$;
D @ R ' x 4 ? N3 ) x 4 ?
< ; C )
Yadav and Eliza, 2017
( 89 . )
Wavelet
( R D F; ! U % C
$4 C 9
$ ":!
3
$J 3)
$;
$S+g R ' 4 ?
$;
% 4
%
$4 mH 83 S!
'
$S+g
; ; ;
^ 3, C
. 89 ' )
$4
$;
89
%
$JP Q Q
(
$ R ' ,J
$J?
^ 3, Q 8 F; ! ) 89
% (
;
= ' J
$;
A Q 3 J ' . F; ! 89 T; ! C
$4 (
M ;
! 4 A ; % %
)
Roushangar and Ghasempour, 2020
(.
T; !Ѱ(x)
89 C j,?
7 !
$ % ?
` QѰ(ω)
D % U%
4 )
Lau and Weng, 1995
(:
$H; % 1 (
| |
| | ≺ ∞
N
` Q ;
` Q J? |) R ;
89
$JP Q Ѱ(x)
Q . F; ! Ѱ(x)
89
% C )
$H; % 2 (
$4
r U
a
% b
W, ,J
^ 3, 3
> % % % )
Lau and Weng, 1995
(.
$H; % 2 (
, √
F; ! 89 R %
~ R% 3 ; J '
$4 M N ! N ! ;% 4 ) '
7 Q
% ' C+%
k4
"
C5 3 N3
C . 7 ! 89
$J 3)
$;
V % N " 9 7 !
$ % ? D -
! 4
$n % Q } '
dn ? ;
Ak
$; ` ; 83 S!
R |)
% 7 !
$ % ? D -
! 4 C .
% 7 p4 R R ' D
h + % 3'
$;
A% Z
$J
% D ' O[
D 3 R Q . N ; ;
;% 4 89
$J
!
r ! R J ; ; N
~ R D
$JQ Q ; . 7 ! 89
$J
% 7 ' 7 ! 89
$J 3) C
)
$[
$J Q (
$4
% 8 R 89 D 7 ! 89
$J 3)
$
$J?
C MJ 3 R ' R ! 3w 4
C35; Y
$ 6 %
;
$ "' D M4 ! C % . Z 7 ! 89
$J
$;
Q % C6!
z 4
; D
;
$4
% 1976 z R ;
$3+
% | Q
% o3Z !
D -
^ 3, D 3
$J ;
SJ D R ' J53?
J3:
$n % Q . 7 ! 89
7Z @ C' Q :
) N 3+ % 4 (
N3;
R J6 4 ?
) 3, ( 3 F; ! 89
%
^ 3, R ' o5J[
C )
Adamowski et al., 2009
( .
$ ":!
4 R
R '
D ; SJ D R ' J53?
O[
) 89 F; !
^ 3, R% | Q (
~
R D
%
$;
9
• % R '% ;
,!
3Y ; R '% ;
A 3n"9 . N r U 7Z @ D M'%
3i3) R
D 5Z ; J53?
N3 )
%|
'
F; !
^ 3, R% | C J53?
%| < ; ' F; ! 89 C . F; ! 7 ! 89 C35; Y
$ ":!
R D
$;
N ]
R D
D ;
^ 3, R ' o5J[
%
% ;
% ; D R R ' D
$:J D R D 54
% J?%
8] 4
^ 3, B%";
^ 3, 8 ? h + % 3' %
"3+
4 . H; %
$4 W3;
! R ; N3 [!
mH
$ ":!
SJ %
% Y 3
$;
A% Z D C :
$H; % 3 (
= !"# $ % , =&'(&'( -*+,)
$4
% mH L
$ ":!
T!N
` ,
% R D
; ;p
; D 6 Q R ; 89
% [J
Q C )
Chou, 2014
(.
2 . 3 . ) , 94 & :94 (EMD
11
V %
$ ":!
; :!
Q % R ;
$ ":!
3 R ' C
$4 N 7
%
? R C6!
; s 4
A% Z 3
.
= N ? 3 5Z
$;
R T!
$S+g ; R J6
R ;
A SJ
$ ":!
Q . V %
d =EMD
$H; % ) 4 ( 3 5Z %x(n)
$;
R T!
! y ) (IMF
$ ":!12
4 )
Amirat et al., 2018
(:
$H; % 4 (
= ∑ /1 0
02 31
$4
rn(x)
'
$S+g Y ; T;
D T! n IMFs ci(x)
C . 8 N
C
% 8 D R %
N ]
! y Q ; . N R ' F; !
R '
! y ) (IMF
E
% R % j Q J '
• 1 -
% 74
' T!
` , N ? ) M J 4 (
` , SZ ; M'
; ; p4 @
R % 8
@ }kJP J ' 2 -
% '
$H,
N3 3 V ) VD ; Q
;
` ,
$ 3 3;
56 V ) VD ; Q
;
` ,
$ 3 4 56 ; SZ Q ; .
$;
C5
9 !
"
% 3 '
% P ;
%
$;
C5
`kJP
• ' F D ! D @ D - x 4 ?
% ]
~ H, Q
5
% ] EMD
€, Q ) ] N3 3 'IMF
J+ @ O[
$;
P 3
.(
C 9 F?%
N
Q % A SJ ; V %
$ ":!
; :!
$]% w ) (EEMD
3) 13
Q .
% %
$ ":!
RD M:@
R 6 D
"
3S
% 3 5Z Q .
; SJ D
$ 9 R ' C p R%
"
3S
$4 F D ! D J
%
D @ x 4 ?
% l
"
J D %
$ ":!
}|@
Q .
% V %
$ ":!
; :!
$]% w 7 4
)
CEEMD
14(
$?
3S
$;
A% Z
! ) C p R
S (
$;
5Z
$? U Q ! R
IMF
~ :
$JP Q . N ; ; 34 ! 7 J D 5Z
$?
? U M %
$4
~ : 'IMF
; ; 3 5Z
Q
$?
? U
$ 9 D
$ ! ' Ppm
; )
Wu and Huang, 2004
(.
2 . 4 . ,#< = &>? @ 4
A ) (LSTM
15
' % t $ Q R ; $4 C J D ; t $ Q R ' % D RNN
D R R '
$+
% SJ 5Z ( . Q C D ; RNN
3P•!
r9 $4 C $ Q 5 o3Z ! $; $ Q Q
C D ; . Q ; % Y MJ 3 R ) 3P•!
\ ; 3
D % $ Q 9 P Q
$ t
D % R % ; ! ; t
3 D D 7 Y J D ; R ' VD ) % ! N 9 ; . Q j ! t
$+ R '
! 4 R .C A
oTU J ; R 3 E ' % RNN
J ; w = P $; ;% R '
< = A
$ ; % R '
W3) 3;
C D R
)
Sagheer and Kotb, 2019
(.
< = $>? @ R ' 5
! 4 A % 1997 R ' 5 $4 Q ? T ; R 3 ] D = $; %RNN
[;
)
Hochreiter and Schmidhuber, 1997
( < = $>? @ .
! 4 A ) ~ 8 (LSTM
RNN
$>? @ 5 D $4 C Q $JP LSTM
$;
% >
J ; R '
< = A
% ' D R R
SJ N ; k . 4 Q ) $5 7@ R ; LSTM
% Z % RNN
$ 3 D ‚S@ 9 A
< = A
C D 3 % )
Bengio et al., 1994
( 33v! . 5 $; C LSTM
V % + T RNN
$ 6 T?
D .C R T?
D $5@ % R
^ ; t
$ 6 D % ~ % ] D % A k= D % $4 Q
$JP Q 5 D % 9 P D % Q ? Q
)
Fang and Yuan, 2019
(.
$5@ % ) A k= D % R % $ 6 D $H; % ; (t
: Q
$H; % 5 (
4= 5 60 . ℎ4 60 . 4
% $4 .C 3 F; ! 7p HP 3s F; !σ
W9
x ! W
' J ' $4
$;
r3! !
h−1
; %
h
; % 7tJ h
. 4
$;
% = 9 P D % R % Q ? D % R % $;
$;
A% Z $ 6 D Q
:
$H; % 6 (
:4= 5 6; . ℎ4 6; . 4
.
"4= 5 6' . ℎ4 6' . 4
+ 5 D % R %
$;
A% Z $ 6 D : Q
$H; % 7 (
<4 = :4. <4 4. =4
% $4
<−1
7 Y $5@ D 5 C3TU A k=
>
$;
A% Z $ 6 D : Q
$H; % 8 (
=4 = 9 ℎ 6? . ℎ4 6? . 4
% $4 P % .C u ^ F; !tanh
T?
D $5@ % R
$; t
A% Z $ 6 D : Q
$H; % 9 (
ℎ4= "4. 9 ℎ <4
2 . 5 . : - #
D 7@
5Z
% SJ D V % R ' Q ' W3) VD ) '
Q ; .
;
$9 !
$;
$ F3 R '
R %
$;
% >
W "?
CY C
D R R ' SJ %
% N W' () ; SJ D
$H; % ) 10 (
%
D ; ) 1 / 0 ! 1 ( Q
% C
~ : R '
ƒ ? R ; D
R
$;
% 4
$J?
Q .
$H; % 10 (
@0A4B = 0.1 0.9 × GGH GHIHJ
HIKL GHIHJ
$4
%
ximin
M4 ! N
ximax
W3;
! N D
>
% , C )
Roushangar and Shahnazi, 2019
(.
2 . 6 .
, *
$;
% >
; D%
% 4 V % SJ Q
% N3 [!
"3 (3 4 56
$ D % D
$ J % ) R%
SJ Q
$4
!%
D r U J ' N3;
,
! '
! 6 ) (R
r U 17
N33 !
$4 J ' N3;
, W3) 3;
Q TY %
' D
R '% 3T M
; D%
C35; Y 8 C
) (DC
$ % 18
N3 3 A T;
' HP
$4 3;
"3
} 6 N3;
, W3) 3;
Q C ) (RMSE
.19
$] '
% ,
R
R ;DC
8
$;
8 8 "
!
% ,
RMSE
8] 4 ! Q ;
$;
T 5H ; C
. j; % N R ' J % ) R%
$;
A% Z D C :
$H; % 11 (
M< = 1 −∑∑)HS,OPOPQ P*R
Q P*R )HS,
$H; % 12 (
T = ∑)HS,OPQ PQR×OP* P*R
U∑ OP)HS, Q PQR×OP* P*R
$H; % 13 (
TVWX = U∑Y02 OPQYP*R
< ; j; % % $4
Z'
Z[
$;
r3! ! , '
Q W3) 3;
Z' Q
Z[
, j J '
Q
W3) 3;
Q T!N
' . Q ;
3 . D E #
3 . 1 . F *4
%
$T+ H U @ C 9 D
R (3 4 56
%
$;
A% Z D D J % ) (3 4 56
%
% X ) D %
+ J SJ Q C .
$;
N r3! !
$4 (3 4 56
% 8 D % F; ! (3 4 56
% 8
$
% ] X )
D % 7 Y D P Q ; . N D
R
% X ) J + J
%
; D%
% Y
$J?
C . D } =
% D R
$;
; D%
:
` !%
(3 4 56
%
$ J 8 J % 3' + J
$JP ) Q C .
$;
N
r3! !
$4 R ; W3) 3;
(3 4 56
% J N3 ) C D ' R (3 4 56
% C < ; SJ
Q
C .
$;
N
% >
R ; W3) 3;
(3 4 56
% J E D (3 4 56
% D % + J
% J
E SJ Q C .
%
! D
R ' x) D
"3+
4 '
J;
R ' R D j ! 7 !
89
$ ":!
$]% w
; :!
7 4
$;
R D
'
$ ":!
Q xw N D R '
$;
R %
SJ LSTM
Q .
% 7 Q ) 2 ( 83! Q D
R A% Z
$J?
% W' () Q
C . M' N3 ] 9
) 2 ( 3;
' R o T!
Q C . ED<
$;
m3U ! C
$4 R ' R D (3 4 56 D 2015 -
2019
$;
R ' VD
$3,;
' ) 2020 2021 (
$;
R ' D SJ %
% Y
$J?
.
Table 2. Defined models
Temporal Modelling Spatial Modelling
Model Input Target Model Input Target
T(I) DO(t-1) DO(t) S(I) DO1(t-1) DO3(t)
T(II) DO (t-1, t-2) DO(t) S(II) DO1(t-2) DO3(t)
T(III) DO (t-1, t-2, t-3) DO(t) S(III) DO2(t-1) DO3(t)
T(IV) DO (t-1, t-2, t-3, t-4) DO(t) S(IV) DO2(t-2) DO3(t)
T(V) DO (t-1, t-2, t-3, t-4, t-5) DO(t) S(V) DO1(t-1), DO1(t-2) DO3(t) S(VI) DO2(t-1), DO2(t-2) DO3(t) S(VII) DO2(t-1), DO1(t-2) DO3(t)
Figure 2. The schematic view of considered modeling in the study
3 . 2 . E #
&
) G & :94 LSTM
3 . 2 . 1 . ,#
C 9 N3 [!
"3 (3 4 56
%
^ ; R
D 5Z ;
$ ":!
$;
N ]
R D
X )
^ ; R '
$ D % (3 4 56
` ;
$;
D % R ' 7 Y o T!
Q
% 4 V %
$>? @
< =
! 4 A
% ; %
% Y C?
. N ; ; N
' ; V %
"3+ LSTM
Q X J
$5Z @ D 7356!
'
% 9 ) 3 (
7 Q ) 3 ( Q
.;
$9 !
$;
X J
$;
C D
$
% 3T
; D%
R ; R '
VD D '
Q
$4 X J R ;
$ ' '
%
$ ' J '
„ ,!
C . N ; ;
$:3J Q (3 4 56
% 8
D % !
@ R D F; ! (3 4 56
% D % 7 Y Q ; . '
% =
$4 ' Q W "?
T!
'D % 33v!
]
% X J
$;
C ' .
$;
73+
$;
; X J
% X ) J
% ; Q X J
` ;
$;
J R ;
R R ' VD D
% 7 Q ) 4 ( Q
C .
Table 3. Results of temporal modeling evaluation without data decomposition
Model Station Train Test
R DC RMSE R DC RMSE
T(I)
St1 0.991 0.981 0.019 0.979 0.961 0.020
St2 0.991 0.982 0.031 0.980 0.963 0.036
St3 0.995 0.985 0.032 0.981 0.965 0.034
St4 0.996 0.986 0.035 0.981 0.965 0.038
St5 0.996 0.988 0.024 0.982 0.971 0.029
T(II)
St1 0.981 0.972 0.022 0.969 0.952 0.023
St2 0.983 0.973 0.031 0.971 0.958 0.035
St3 0.991 0.988 0.032 0.978 0.969 0.034
St4 0.988 0.979 0.033 0.975 0.961 0.033
St5 0.994 0.979 0.036 0.981 0.964 0.031
T(III)
St1 0.990 0.972 0.022 0.976 0.958 0.022
St2 0.988 0.976 0.041 0.969 0.947 0.044
St3 0.988 0.977 0.035 0.978 0.954 0.039
St4 0.991 0.989 0.031 0.980 0.975 0.035
St5 0.996 0.979 0.031 0.984 0.965 0.036
T(IV)
St1 0.992 0.975 0.024 0.978 0.949 0.023
St2 0.991 0.981 0.034 0.979 0.969 0.036
St3 0.989 0.977 0.033 0.979 0.959 0.038
St4 0.993 0.978 0.033 0.979 0.965 0.035
St5 0.995 0.989 0.025 0.981 0.974 0.031
T(V)
St1 0.990 0.976 0.020 0.975 0.958 0.021
St2 0.983 0.978 0.033 0.972 0.958 0.035
St3 0.991 0.981 0.034 0.979 0.968 0.036
St4 0.989 0.979 0.035 0.974 0.967 0.036
St5 0.995 0.979 0.038 0.981 0.969 0.039
Figure 3. Results of training and testing steps of all models for the station 1 in temporal modeling without data decomposition
Figure 4. Results of the training and testing steps of model T(I) for the station 1 in temporal modeling without data decomposition
3 . 2 . 2 . ,# H
C 9 W3) 3;
3 4 ( 56
% ;
` !%
N3;
' J CS'
^ ; J % ) (3 4 56
%
J 7 Y D %
+ J o T!
Q
% 4 V %
$>? @ < =
! 4 A
% ; Q . X J R '% 3T
; D%
% 9 ) 4 (
% Q C . '
% =
$4 ' Q X J 7Z @ D R % D %
7 Y (3 4 56
%
J E J 7 Y D % 8 N J ;
$:3J %
% )
$JQ C . X J
! ; C 9 W3) 3;
(3 4
56 J
E R ; R R ' VD D
% 7 Q ) 5 ( Q
C .
Table 4. Results of spatial modeling evaluation without data decomposition
Model Train Test
R DC RMSE R DC RMSE
S(I) 0.888 0.780 0.103 0.870 0.355 0.132
S(II) 0.887 0.785 0.102 0.866 0.350 0.132
S(III) 0.986 0.949 0.049 0.985 0.752 0.082
S(IV) 0.980 0.958 0.044 0.979 0.924 0.045
S(V) 0.883 0.777 0.104 0.867 0.359 0.131
S(VI) 0.988 0.972 0.036 0.987 0.885 0.055
S(VII) 0.990 0.977 0.033 0.985 0.939 0.040
Figure 5. Results of the training and testing steps of superior model (S(VII)) in spatial modeling without data decomposition
3 . 3 . E #
&
) G & :94 I"LSTM
% N C Y ; SJ D V %
$ ":!
$]% w
; :!
7 4 7 ! 89
$J J;
R R ' D
$ ":!
Q xw
R D
R '
$;
C
$;
R %
$;
Q LSTM
. D R
"3
% C+ @
D E :
Q . N3+
E Y R ;
$ ":!
3 ; 7 ! 89 [J 89
% C .
% N W' () D
89
% ;db2
$ mH
$ ":!
SJ Q .
^ V %
CEEMD
"3
$ ":!
3
$;
R 'IMF
o5J[
8 3 3Y ; C
$4
% C
~ : N ' ' 3 5Z ' P ; . V % 73 ! 'IMF
^ ;
d S!
F; !
$ ) D 3 5Z C . N
$ ) ! JY
$
;
$4
„ ,!
3 Y ; C; l
Q .
% C
R D
R '
$;
C D ' V %
$;
R %
$;
Q LSTM
.
$;
p W3) VD )
R R ' D J
%
^ 3,
$ D % j ! 7 ! 89 M' N3 ]
$ ":!
; :!
% 7 Q R ' ) 6 ( ) 7 (
Q C .
Figure 6. The subheading of the first station decomposes by discrete wavelet transform
Figure 7. The subheading of the first station decomposes by complementary ensemble empirical mode decomposition
3 . 3 . 1 . ,#
X J 7Z @ D D R D x) D VD ) '
j ! 7 ! 89
$J
$ ":!
; :!
% 9 ) 5 (
% Q C . '
% =
$4
% X J
"3 ' Q
$ ":!
R R ' D ! R @
\ ; ; X J Q
3l•!
$ ":!
; SJ D
$ ":!
$]% w
; :!
7 4 W3;
! D V % 7 ! 89
$J C . d; H X J
$;
C
„ ,!
%
$ ' J ' X J C
d = X J ; W3) VD ) W "?
T!
'D % 3l•!
] ; X J
% . N ; ;
$:3J Q (3 4 56
% 8 D % F; ! (3 4 56
% D % 7 Y ' P ; .
|+
C 9
$ , X J X J J M' ; W3) VD ) '
M' x) D VD ) '
$;
V % Q 4y
% 7 Q R ' ) 8 ( ) 9 (
% Q C . % ; :! $ ":! X J $; $9 ! ; W "? \ ; J % T(I)
% % 3T % R Z% 8 D
; W' 4 DC
15 R HP "3 % R Z%
. QRMSE
Table 5. Results of temporal modeling evaluation after data decomposition
Model Station Train Test
R DC RMSE R DC RMSE
T(I)
St1 DWT 0.993 0.974 0.017 0.987 0.965 0.018
CEEMD 0.986 0.981 0.015 0.988 0.977 0.017
St2 DWT 0.974 0.965 0.026 0.973 0.955 0.025
CEEMD 0.977 0.969 0.015 0.975 0.955 0.017
St3 DWT 0.977 0.965 0.014 0.976 0.957 0.015
CEEMD 0.978 0.967 0.018 0.975 0.959 0.020
St4 DWT 0.977 0.968 0.014 0.976 0.957 0.016
CEEMD 0.975 0.970 0.016 0.978 0.960 0.016
St5 DWT 0.977 0.967 0.013 0.976 0.963 0.017
CEEMD 0.973 0.972 0.015 0.978 0.968 0.017
T(II)
St1 DWT 0.973 0.966 0.015 0.970 0.961 0.016
CEEMD 0.982 0.978 0.013 0.978 0.976 0.015
St2 DWT 0.975 0.970 0.026 0.973 0.965 0.024
CEEMD 0.978 0.978 0.019 0.975 0.968 0.020
St3 DWT 0.973 0.966 0.023 0.970 0.958 0.025
CEEMD 0.979 0.974 0.017 0.977 0.961 0.017
St4 DWT 0.974 0.968 0.020 0.972 0.965 0.022
CEEMD 0.977 0.976 0.013 0.971 0.968 0.015
St5 DWT 0.974 0.964 0.019 0.972 0.961 0.022
CEEMD 0.979 0.967 0.015 0.978 0.964 0.016
T(III)
St1 DWT 0.972 0.963 0.019 0.968 0.952 0.020
CEEMD 0.975 0.966 0.016 0.971 0.954 0.018
St2 DWT 0.973 0.966 0.023 0.969 0.958 0.022
CEEMD 0.978 0.969 0.020 0.978 0.964 0.021
St3 DWT 0.971 0.963 0.025 0.967 0.954 0.028
CEEMD 0.978 0.967 0.021 0.978 0.958 0.023
St4 DWT 0.972 0.970 0.023 0.970 0.969 0.026
CEEMD 0.979 0.972 0.015 0.978 0.968 0.016
St5 DWT 0.977 0.964 0.022 0.971 0.963 0.026
CEEMD 0.977 0.968 0.015 0.976 0.966 0.015
T(IV)
St1 DWT 0.965 0.949 0.023 0.959 0.937 0.025
CEEMD 0.978 0.976 0.014 0.978 0.975 0.016
St2 DWT 0.969 0.959 0.031 0.964 0.951 0.032
CEEMD 0.978 0.969 0.020 0.978 0.961 0.025
St3 DWT 0.972 0.965 0.025 0.968 0.955 0.027
CEEMD 0.978 0.972 0.022 0.978 0.968 0.023
St4 DWT 0.973 0.966 0.021 0.971 0.962 0.024
CEEMD 0.978 0.969 0.015 0.977 0.965 0.016
St5 DWT 0.971 0.962 0.023 0.970 0.960 0.028
CEEMD 0.979 0.969 0.017 0.978 0.964 0.018
T(V)
St1 DWT 0.974 0.963 0.018 0.971 0.955 0.019
CEEMD 0.971 0.967 0.015 0.978 0.966 0.019
St2 DWT 0.967 0.953 0.030 0.961 0.938 0.031
CEEMD 0.978 0.964 0.020 0.978 0.958 0.022
St3 DWT 0.974 0.969 0.021 0.971 0.962 0.023
CEEMD 0.976 0.973 0.013 0.975 0.965 0.014
St4 DWT 0.972 0.964 0.024 0.970 0.958 0.026
CEEMD 0.977 0.968 0.018 0.977 0.961 0.020
St5 DWT 0.972 0.963 0.022 0.972 0.960 0.026
CEEMD 0.977 0.974 0.015 0.973 0.964 0.017
Figure 8. Results of training and testing steps of model T(I) for the first station in temporal modeling after data decomposition
Figure 9. Results of training and testing steps of model T(I) for the station 1 in temporal modeling after data decomposition
3 . 3 . 2 . ,# H
R ; W3) 3;
(3 4 56
% J E ;
$9 !
$;
X J 9 ) 6 ( ' Q
$4 N J ;
S(VII)
Q ;
$4 ; J % ) (3 4 56 8 D % 7 Y J E (3 4 56 D %
7 Y J
$:3J 7; Y + Y C
. '
% =
$4 ' Q
% :
"3 X J 7Z @ D W3) VD ) '
! R @
\ ; ; X J Q C M' N3 ] X J 7Z @ D 7 ! 89
$J ; D% % 3T % R Z% 8 W "? ;
R HP % R Z% W' 4 DC
J ;RMSE
D X J 7Z @ D
$ ":!
; :!
7 4 Q ; . X J VD
D R ; N J ;
% 7 Q ) 10 (
% Q C .
Table 6. Results of spatial modeling evaluation after data decomposition
Model Train Test
R DC RMSE R DC RMSE
S(I) DWT 0.883 0.739 0.112 0.866 0.592 0.105
CEEMD 0.937 0.872 0.078 0.901 0.546 0.111
S(II) DWT 0.883 0.778 0.104 0.868 0.382 0.129
CEEMD 0.923 0.831 0.090 0.913 0.372 0.130
S(III) DWT 0.988 0.968 0.039 0.987 0.844 0.065
CEEMD 0.990 0.978 0.032 0.989 0.774 0.078
S(IV) DWT 0.983 0.966 0.040 0.981 0.929 0.043
CEEMD 0.986 0.964 0.041 0.985 0.852 0.063
S(V) DWT 0.885 0.776 0.104 0.868 0.436 0.123
CEEMD 0.922 0.849 0.085 0.921 0.437 0.123
S(VI) DWT 0.989 0.977 0.033 0.988 0.920 0.046
CEEMD 0.992 0.981 0.030 0.991 0.814 0.071
S(VII) DWT 0.991 0.981 0.030 0.990 0.953 0.039
CEEMD 0.992 0.979 0.031 0.987 0.941 0.045
Figure 10. Results of the training and testing steps of superior model in spatial modeling after data decomposition by wavelet transform
3 . 4 . & ' E #
% N
$T+ H ; SJ D V % R ' J ; V ' t V %
$>? @
! 4 < = A
R ' 34 !
89 -
$>? @
! 4 < = A
$ ":!
; :!
-
$>? @
! 4 < = A
$;
W3) 3;
(3 4 56
$;
A% Z
D
$JP ) Q . X J R '% 3T
; D%
R ;
$5@
D R '
! ; ; SJ D V % R '
% 4|
% 7 Q
) 12 (
% Q C . V % V % $; C R J ; X J LSTM
$; R ' ) N3 3) $J?%% 4
$; (ANN
C '
)
Dabrowski et al., 2018
( d; H . X J ' Q
% 7 Q ) 11
% ( RD
D
% T(I)
RD
S(VII)
; NJQ W3;
N ! M4 DC
N !
$; RMSE
N ! ;
; D%
Q .
Figure 11. Comparison of evaluation criteria of superior models in two types of temporal and spatial modeling
$;
% = n"9 ] ]% ] 54 W3) CY ! WT
3;
[ ; ; R @ ! % 7,J R ' + @%
$4
% R% Q C " …3' M3,J ]% ] '
)
Xu et al., 2021
(.
M' N3 ] V % ; 34 ! R '
EEMD
R HP ! D
R M4 !
$; R ' ; C )
Eze, 2021
(.
$]
% '
~ D R VD ) '
R
\ ;
; X J Q C
% D
R D
$ ":!
; :!
\ ; Q J ; X J
D
? =
% D R
7 ! 89
\ ; J ; Q X J % , W "?
QDC
C .
4 .
&9 #
% N W' () D
$>? @ < =
! 4 A R ; D
R (3 4 56
% ) (DO
%
$ P %
%
% CS'
$+
$;
A% Z R '
$ D % SJ Q .
~ D R D M'
; W3) VD ) '
M' x) D
$ ":!
' R R D ; SJ D 7 ! 89
$ ":!
; :!
E : Q .
;
$9 !
$;
X J
%
D R D
%
$ ' J '
$:3J Q
$4 W "?
T!
'D % 3l•!
R
% ; 5 CQ
x)
!
$:3J C?
(3 4 56
% 8 J
% 8 D % ; CY R D F; ! (3 4 56
% D % 7 Y
Q ; . M' N3 ]
% D R '
Q
$4 X J 7Z @ D 8 D % 7 Y J E D % 7 Y J
% W3) 3;
(3 4 56 J
E
% 4 ' P ; . M' N3 ]
% '
~ D R W3) D ) ' 3l•!
; P
% ; 5 JQ
"3 W "? \ ; QDC
M' N3 ] 7 ! 89
% D R
$ ":!
; :!
% D R D W3;
! N 3l•!
%
% ; 5 CQ LSTM
.
5 . ,"
)J # 1. Dissolved Oxygen
2. Artificial Neural Network 3. Multi-Layer Perceptron 4. Radial Basis Function
5. Least Squares Support Vector Machine-Improved Particle Swarm Optimization 6. Support Vector Machine
7. Back Propagation
8. Constrained Extreme Learning Machines
9. Savannah River 10. Wavelet Transform
11. Empirical Mode Decomposition 12. Intrinsic Mode Functions
13. Ensemble Empirical Mode Decomposition
14. Complementary Ensemble Empirical Mode Decomposition 15. Long Short-Term Memory
16. Recurrent Neural Network 17. Correlation Coefficient 18. Coefficient of Determination 19. Root Mean Square Error
6 . K? / L *4
…3' . % 9 N3; T? †% T! $
7 . K /
Adamowski, K., Prokoph, A., & Adamowski, J. (2009). Development of a new method of wavelet aided trend detection and estimation. Hydrological Processes, 23, 2686-2696.
Ahmed, A. A. M. (2017). Prediction of dissolved oxygen in Surma River by biochemical oxygen demand and chemical oxygen demand using the artificial neural networks (ANNs).
Journal of King Saud University - Engineering Sciences, 29(2), 151-158.
Amirat, Y., Benbouzid, M. E. H., Wang, T., Bacha, K., & Feld, G. (2018). EEMD-based notch filter for induction machine bearing faults detection. Applied Acoustics, 133, 202-209.
Bengio, Y., Simard, P., & Frasconi, P. (1994). Learning long-term dependencies with gradient descent is difficult. IEEE Transactions on Neural Networks, 5(2), 157-166.
Chou, C.-M. (2014). Complexity analysis of rainfall and runoff time series based on sample entropy in different temporal scales. Stochastic Environmental Research and Risk Assessment, 28(6), 1401-1408.
Csábrági, A., Molnár, S., Tanos, P., & Kovács, J. (2017). Application of artificial neural networks to the forecasting of dissolved oxygen content in the Hungarian section
