Multisite rainfall downscaling and disaggregation in a tropical … · 2020. 3. 7. · Figure 4: Comparison of downscaled results at site S46 from (a1-d1) single-site GLM combined

This document is downloaded from DR‑NTU (https://dr.ntu.edu.sg)Nanyang Technological University, Singapore.

Multisite rainfall downscaling and disaggregationin a tropical urban area

Qin, Xiaosheng; Lu, Yan

2013

Lu, Y., & Qin, X. (2014). Multisite rainfall downscaling and disaggregation in a tropical urbanarea. Journal of Hydrology, 509, 55‑65.

https://hdl.handle.net/10356/80072

https://doi.org/10.1016/j.jhydrol.2013.11.027

© 2013 Elsevier. This is the author created version of a work that has been peer reviewedand accepted for publication by Journal of Hydrology, Elsevier. It incorporates referee’scomments but changes resulting from the publishing process, such as copyediting,structural formatting, may not be reflected in this document. The published version isavailable at: [DOI: http://dx.doi.org/10.1016/j.jhydrol.2013.11.027].

Downloaded on 31 May 2021 17:15:59 SGT

Figure List

Figure 1: Study area and location of rain gauges

Figure 2: The flowchart of study framework

Figure 3: MAPE values of objective function at various K values for single-site

disaggregation at S24

Figure 4: Comparison of downscaled results at site S46 from (a1-d1) single-site GLM

combined with KNN (S-G-K), and (a2-d2) multisite GLM. MAPE is based on the

observed and average downscaled data

Figure 5: Statistical properties of disaggregated hourly rainfall data using KNN and HYETOS

during the verification period at the stations S24 (denoted as 1) and S46 (denoted

as 2). MAPEK = MAPE value based on KNN; MAPEH = MAPE level based on

HYETOS.

Figure 6: Quantile-quantile plot between the observed and disaggregated data

Figure 7: Statistical properties of disaggregated rainfall at satellite station S24 from KNN and

MuDRain. MAPEK = MAPE value based on KNN; MAPEM = MAPE level based

on MuDRain.

Figure 8: Statistical properties of downscaled rainfall at S46 from GLM based on HadCM3

A2 scenario during 1980-2010. Solid line represents the observed data; dash line is

the downscaled envelop based on 20 ensembles

Figure 9: Cross-correlation coefficients vs. distance for observed and downscaled rainfall.

OBS = observed data; SIM = downscaled data; CC = correlation coefficient

between observed data and average value of downscaled data.

Figure 10: Statistical properties of disaggregated rainfall at master station S46 using KNN.

The plot shows a comparison of the observed time series with envelop curves

generated from 20 ensembles from GLM downscaled daily data

Figure 11: Statistical properties of disaggregated rainfall from multi-site disaggregation using

MuDRain at satellite station S24

Figure 12: Cross-correlation coefficients of observed and disaggregated hourly data against

distance in (a) February, (b) June, (c) September, and (d) December. OBS =

observed data; SIM = disaggregated data; CC = correlation coefficient between

observed data and average value of disaggregated data.

Figure 13: The mean and maximum hourly rainfall for the study region during baseline

period (1980-2010) and future periods, including (a1 & a2) 2030s (2011-2040), (b1

& b2) 2050s (2041-2070) and (c1 & c2) 2080s (2071-2100).

S66

Kranji

S44

NTU

S40

Orchids

S69

Upp Peirce

S46

Golf Club

S60

Sentosa

S55

Woodbridge S24

Changi

Figure 1: Study area and location of rain gauges.

GLM

Spatial downscaling

(Generate multisite

daily data)

Single-site

Temporal

disaggregation

(Generate hourly data at

master station)

Multi-site

Temporal

disaggregation


satellite stations)

Project to future period


multiple sites)

KNN

HYETOS

KNN

MuDRain

Selected

Method

Selected

Method

HadCM3 Predictors

(Evaluate data quality;

predict future condition)

NCEP Predictors

(Training & Verify

GLM model)

Single-site GLM +

KNN

Multisite GLM

Selected

Method

Part 2: Model Application Part 1: Model Verification

Figure 2: The flowchart of study framework.

1 2 3 4 5 6 7 8 9 100.10

0.15

0.20

0.25

0.30

0.35

MA

PE

va

lue

K number

MAPE

Figure 3: MAPE values of objective function at various K values for single-site

disaggregation at S24.

J F M A M J J A S O N D4

5

6

7

8

9

10

11

Mean

d (

mm

/day)

OBS

SIM average

SIM boundary

(a1)

S-G-K

MAPE = 0.10

J F M A M J J A S O N D

4

5

6

7

8

9

10

11

Me

an

d (

mm

/day

)

MAPE = 0.07(a2)

M-G


10

12

14

16

18

20

ST

Dd (

mm

/day)

(b1)

S-G-K

MAPE = 0.07


10

12

14

16

18

20

ST

Dd (

mm

/da

y)

MAPE = 0.07(b2)

M-G


0.4

0.5

0.6

0.7

0.8

Pw

et d

(c1)

S-G-KMAPE = 0.10


0.4

0.5

0.6

0.7

0.8

Pw

et d

MAPE = 0.06(c2)

M-G


60

90

120

150

180

210

240

Max

d (

mm

/day)

(d1)

S-G-KMAPE = 0.17


60

90

120

150

180

210

240

Ma

xd (

mm

/da

y)

MAPE = 0.19(d2)

M-G

Figure 4: Comparison of downscaled results at site S46 from (a1-d1) single-site GLM

combined with KNN (S-G-K), and (a2-d2) multisite GLM. MAPE is based on the observed

and average downscaled data.


2

3

4

ST

Dh

(m

m/h

ou

r)

OBS KNN

HYETOS

(a1)

MAPEK = 0.10

MAPEH = 0.14


1

2

3

4

ST

Dh (

mm

/ho

ur)

(a2) MAPEK = 0.09

MAPEH = 0.14

J F M A M J J A S O N D0.1

0.2

0.3

0.4

0.5

AC

1h

(b1) MAPEK = 0.11

MAPEH = 0.26


0.1

0.2

0.3

0.4

0.5

AC

1h

(b2) MAPEK = 0.20

MAPEH = 0.36


0.1

0.2

Pw

et h

(c1) MAPEK = 0.09

MAPEH = 0.36


0.1

0.2

Pw

et h

(c2) MAPEK = 0.15

MAPEH = 0.38


10

15

20

25

30

35

Sk

ew

ne

ss

h

(d1) MAPEK = 0.16

MAPEH = 0.17


10

15

20

25

30

35

Sk

ew

ne

ss

h

(d2) MAPEK = 0.21

MAPEH = 0.15

Figure 5: Statistical properties of disaggregated hourly rainfall data using KNN and HYETOS

during the verification period at the stations S24 (denoted as 1) and S46 (denoted as 2).

MAPEK = MAPE value based on KNN; MAPEH = MAPE level based on HYETOS.

Figure 6: Quantile-quantile plot between the observed and disaggregated data.


1.5

2.0

2.5

3.0

3.5

4.0 OBS

KNN

MuDRain

ST

Dh (

mm

/ho

ur)

(a) MAPEK = 0.09

MAPEM = 0.15


0.1

0.2

0.3

0.4

0.5

AC

1h

(b) MAPEK = 0.22

MAPEM = 0.08


0.03

0.06

0.09

0.12

0.15

Pw

et h

(c) MAPEK = 0.10

MAPEM = 0.09


10

15

20

25

30

Skew

ness

h

(d) MAPEK = 0.19

MAPEM = 0.20

Figure 7: Statistical properties of disaggregated rainfall at satellite station S24 from KNN and

MuDRain. MAPEK = MAPE value based on KNN; MAPEM = MAPE level based on

MuDRain.


5

6

7

8

9

10

11 OBS

SIM envelope

Me

an

d (

mm

/da

y)

(a)


10

12

14

16

18

20

22

ST

Dd (

mm

/day

)

(b)


0.4

0.5

0.6

0.7

0.8

Pw

et d

(c)


80

120

160

200

240

280

Ma

xd (

mm

/da

y)

(d)

Figure 8: Statistical properties of downscaled rainfall at S46 from GLM based on HadCM3

A2 scenario during 1980-2010. Solid line represents the observed data; dash line is the

downscaled envelop based on 20 ensembles.

5 10 15 20 25 30 35

0.4

0.5

0.6

0.7

0.8 OBS Average SIM SIM

Fitted curve for OBS

Fitted curve for average SIM

Co

rre

lati

on

Distance (km)

CC = 0.99

Figure 9: Cross-correlation coefficients vs. distance for observed and downscaled rainfall.

OBS = observed data; SIM = downscaled data; CC = correlation coefficient between

observed data and average value of downscaled data.


2.0

2.4

2.8

3.2S

TD

h (

mm

/ho

ur)

Observed

Disaggregated

envelope

(a)


0.16

0.20

0.24

0.28

0.32

0.36

0.40

AC

1h

(b)


0.06

0.08

0.10

0.12

0.14

Pw

et h

(c)


12

14

16

18

20

22

Sk

ew

nes

sh

(d)

Figure 10: Statistical properties of disaggregated rainfall at master station S46 using KNN.

The plot shows a comparison of the observed time series with envelop curves generated from

20 ensembles from GLM downscaled daily data.


1.6

2.0

2.4

2.8S

TD

h (

mm

/ho

ur)

Observed

Disaggregated

envelope

(a)


0.18

0.24

0.30

0.36

0.42

0.48

AC

1h

(b)


0.04

0.06

0.08

0.10

0.12

0.14

0.16

Pw

et h

(c)


12

16

20

24

28

32

36

Sk

ew

nes

sh

(d)

Figure 11: Statistical properties of disaggregated rainfall from multi-site disaggregation using

MuDRain at satellite station S24.

5 10 15 20 25 30 35

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8 OBS SIM Ave

SIM

Fitted OBS

Fitted SIM Ave

Co

rre

lati

on

Distance (km)

CC = 0.98

(a)

5 10 15 20 25 30 35

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Co

rre

lati

on

Distance (km)

CC = 0.98(b)

5 10 15 20 25 30 35

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Co

rre

lati

on

Distance (km)

(c) CC = 0.97

5 10 15 20 25 30 35

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8C

orr

ela

tio

n

Distance (km)

(d) CC = 0.98

Figure 12: Cross-correlation coefficients of observed and disaggregated hourly data against

distance in (a) February, (b) June, (c) September, and (d) December. OBS = observed data;

SIM = disaggregated data; CC = correlation coefficient between observed data and average

value of disaggregated data.


0.1

0.2

0.3

0.4

0.5

0.6

Ra

infa

ll (

mm

/ho

ur)

Baseline

H3A2 average H3B2 average

H3A2 boundary H3B2 boundary

(a1)

Period: 2011-2040


0

40

80

120

160

200

Ma

xim

um

(m

m/h

ou

r)

(a2)

Period: 2011-2040


0.1

0.2

0.3

0.4

0.5

0.6

Ra

infa

ll (

mm

/ho

ur)

(b1)

Period: 2041-2070


0

40

80

120

160

200M

ax

imu

m (

mm

/ho

ur)

Period: 2041-2070

(b2)


0.1

0.2

0.3

0.4

0.5

0.6

Ra

infa

ll (

mm

/ho

ur)

(c1)

Period: 2071-2099


0

40

80

120

160

200

Ma

xim

um

(m

m/h

ou

r)

Period: 2071-2099

(c2)

Figure 13: The mean and maximum hourly rainfall for the study region during baseline

period (1980-2010) and future periods, including (a1 & a2) 2030s (2011-2040), (b1 & b2)

2050s (2041-2070) and (c1 & c2) 2080s (2071-2100).

Documents

Multisite rainfall downscaling and disaggregation in a tropical … · 2020. 3. 7. · Figure 4: Comparison of downscaled results at site S46 from (a1-d1) single-site GLM combined