[PPT] - Dynamical Downscaling of ECMWF Motivation Past work with Seasonal PowerPoint Presentation

SLIDE 1

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Dynamical Downscaling of ECMWF Seasonal Forecasts:

case studies over Eastern and QWeCI southern Africa domains

Gulilat Tefera Diro and Adrian Tompkins

Earth System Physics Section

The Abdus Salam ICTP

Thanks to RegCM Team — October 23, 2012 ILRI, Nairobi

SLIDE 2

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Outline

1

Motivation

2

Past work with ECMWF cycle 29r2 over East Africa

3

ECMWF Sys 3 Ensemble data and RegCM set up

4

Perfect boundary condition run

5

ECMWF ENSEMBLE forcing runs

6

Summary and Future work

SLIDE 3

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Motivation

Limited area models(eg. MM5, WRF , RegCM,...) have been used for downscaling short range weather forecasts and/or climate change studies

For example, RegCM has been extensively validated on various domain for its ability to have an "added" value compared to the global climate models

There is a growing demand from impact models for a detailed (localized) seasonal forecasts Do regional climate models improve seasonal forecasts and able to reproduce the year to year fluctuation?

SLIDE 4

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

ECMWF and RegCM3 JJAS hindcast climatology

Longitude Latitude

ERA-Interim JJAS total rainfall [mm] for 1991-2000 25 30 35 40 45 50 55

5

5 10 15 20 200 400 600 800 1000 1200 1400 1600 1800

Longitude Latitude

RegCM-ERA-Interim JJAS total rainfall [mm] for 1991-2000 25 30 35 40 45 50 55

5

5 10 15 20 200 400 600 800 1000 1200 1400 1600 1800 200 400 600 800 1000 1200 1400 1600 1800 Latitude Longitude ECMWF Ensemble mean JJAS hindcast (1991-2000)

5

5 10 15 20 25 30 35 40 45 50 55

Longitude Latitude RegCM3 ENSEMBLE-mean JJAS hindcast (1991-2000) 25 30 35 40 45 50 55

5

5 10 15 20

200 400 600 800 1000 1200 1400 1600 1800

Longitude Latitude

GPCP JJAS total rainfall [mm] for 1991-2000 25 30 35 40 45 50 55

5

5 10 15 20 200 400 600 800 1000 1200 1400 1600 1800

Longitude Latitude TRMM JJAS total rainfall [mm] for 1998 − 2007 25 30 35 40 45 50 55 −5 5 10 15 20 25

200 400 600 800 1000 1200 1400 1600 1800

ECMWF ERAIN RegCM3-ERAIM ECMWF ensemble mean RegCM3 ensemble mean GPCP TRMM

SLIDE 5

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Perfect boundary run: Inter-annual variability

1990 1992 1994 1996 1998 2000

2.5
2
1.5
1
0.5

0.5 1 1.5 2 2.5

year JJAS standard rainfall Average over all five zones

Gauge GPCP ECMWF-ERAIM RegCM-ERAIM

Table: Correlation of ERA-Interim and RegCM3 with observations Gauge GPCP ECMWF-ERA Interim 0.63 0.77 RegCM-ERA Int. forced 0.83 0.88

5

5 10 15 20 25 30 35 40 45 50 55 Latitude Longitude

35 40 45 4 6 8 10 12 14 JJAS total rainfall (mm) Longitude Latitude 200 400 600 800 1000 1200 1400 1600 1800

IIa Dry I III IIb IV

SLIDE 6

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

ECMWF cycle 29r2 with RegCM: 1998 - 1997

Propagation of error from GCM to RCM

Longitude Latitude

GPCP JJAS total rainfall [mm] for 1998 -1997 25 30 35 40 45 50 55

5

5 10 15 20

600
400
200

200 400 600

Longitude Latitude

RegCM-ERA-Interim forced JJAS total rainfall [mm] for 1998 - 1997 25 30 35 40 45 50 55

5

5 10 15 20

600
400
200

200 400 600

Longitude Latitude

ECMWF ENSEMBLE mean JJAS total rainfall [mm]: 1998 - 1997 25 30 35 40 45 50 55

5

5 10 15 20

600
400
200

200 400 600

Longitude Latitude

RegCM3 ENSEMBLE-mean JJAS total rainfall [mm]: 1998 - 1997 25 30 35 40 45 50 55

5

5 10 15 20

600
400
200

200 400 600

SLIDE 7

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Propagation of error: ENSO vs western Indian ocean SST teleconnection

Longitude Latitude

ERA−Interim JJAS Skin temperature [1998 − 1997]

150° W 100° W 50° W 0° 50° E 100° E 150° E 60° S 30° S 0° 30° N 60° N

−5 5

Latitude

ECMWF Ensemble mean hindcast for JJAS skin temperature [1998 − 1997]

150° W 100° W 50° W 0° 50° E 100° E 150° E 60° S 30° S 0° 30° N 60° N

5

SLIDE 8

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Probabilistic verification: ROCS

Area under ROC curve (ROCA) Compares against a random forecast A skillful forecast → ROCA > 0.5

r →2* (ROC area -0.5) > 0

ROC Curve

1 False Alarm Rate Hit Rate 1

Table: ROC Area (Area under the roc curve) over Ethiopia

ROCs Dry Normal Wet Gauge GPCP Gauge GPCP Gauge GPCP ECMWF-ENSEMBLE 0.67 0.86 0.54 0.75 0.55 0.90 RegCM3 0.86 0.64 0.69 0.38 0.71 0.86

Longitude Latitude RegCM ROCS for dry category of JJAS 25 30 35 40 45 50 55

5

5 10 15 20

1
0.8
0.6
0.4
0.2

0.2 0.4 0.6 0.8 1

Longitude Latitude ECMWF ROCS for dry category of JJAS 25 30 35 40 45 50 55

5

5 10 15 20

1
0.8
0.6
0.4
0.2

0.2 0.4 0.6 0.8 1

SLIDE 9

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

ECMWF cycle 29r2 with RegCM: RPSS

RPS =

1 K−1[

K

i=1(CDFforc,i − CDFobs,i)2]

RPSS−D = 1 −

RPSforc (RPSclim+RPSclim∗ 1

M )

A skillful forecast → RPSS > 0

Forc

1 1 CDF

Obs

Longitude Latitude Mean debiased RPSS with respect to climatology for ECMWF 25 30 35 40 45 50 55

5

5 10 15 20

1
0.8
0.6
0.4
0.2

0.2 0.4 0.6 0.8 1

Longitude Latitude Mean debiased RPSS with respect to climatology for RegCM3 25 30 35 40 45 50 55

5

5 10 15 20

1
0.8
0.6
0.4
0.2

0.2 0.4 0.6 0.8 1

RegCM3_wrt_GPCP ECMWF_wrt_GPCP RegCM3_wrt_gage ECMWF_wrt_gage RPSS 0.0 0.5 1.0

SLIDE 10

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

ECMWF ensemble hindcasts from 33r1

Resolution : 1.1250×1.1250 in horizontal and 62L in vertical Hindcast period: 1991-2001 9 member ensembles addressing forecast uncertainty

uncertainty in initial condition: Perturbed initial conditions model error: Perturbed physics

Two start dates (May and November): here we used the Nov 1st start for Malawi domain 6 month hindcasts starting from November 1st

SLIDE 11

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Model setup and experimental design

Resolution : 25km (288x200) in horizontal and 18 levels in the vertical Convection scheme: Grell over land and Emanuel over the ocean Three experiments were carried out:

’Perfect’ boundary condition run

with climate mode: continuous run with seasonal forecast mode: initialize every year from Nov 1st

seasonal hindcasts boundary condition

For the ’perfect’ boundary simulation:

ERA-Interim re-analysis,OI-weekly SST simulation period: Jan 1990 to May 2002

For hindcasts simulation

The 9 ECMWF ensemble members from 33r1 cycle are downscaled independently simulation period: Nov 1st to May 1st every year between 1991 and 2002

SLIDE 12

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Model domain and Topography

RegCM ECMWF

Longitude Latitude

10 20 30 40 50 60

35
30
25
20
15
10
5

5

500 1000 1500 2000 2500 3000

Longitude Latitude

10 20 30 40 50 60 −35 −30 −25 −20 −15 −10 −5 5

500 1000 1500 2000 2500 3000

RegCM has a better realistic surface features due to its high resolution.

SLIDE 13

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Perfect boundary run: mean DJFMA Climatology

Longitude Latitude

ERA-Interim DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5

2 4 6 8 10 12 14 16 18 20

Longitude Latitude

RegCM4-Cont DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5

2 4 6 8 10 12 14 16 18 20

Longitude Latitude

CRU DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5

2 4 6 8 10 12 14 16 18 20

Longitude Latitude

GPCP DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5

2 4 6 8 10 12 14 16 18 20

CRU GPCP ERA-Interim RegCM4

SLIDE 14

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Impact of land surface initialization: mean DJFMA Climatology

Longitude Latitude

RegCM4-reinit DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5

2 4 6 8 10 12 14 16 18 20

Longitude Latitude

RegCM4-Cont DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5

2 4 6 8 10 12 14 16 18 20

Longitude Latitude

RegCM4-reinit - CRU DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5
5
4
3
2
1

1 2 3 4 5

Longitude Latitude

RegCM4-cont - CRU DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5
5
4
3
2
1

1 2 3 4 5

Bias with respect to CRU Re-initialized run Continuous run

SLIDE 15

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Impact of land surface initialization: interannual variability

Longitude Latitude

Correlation of CRU with RegCM4-reinit for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5
0.6
0.4
0.2

0.2 0.4 0.6

Longitude Latitude

Correlation of CRU with RegCM4-contrun for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5
0.6
0.4
0.2

0.2 0.4 0.6

Re-initialized run Continuous run

In both integrations (i.e. the climate and seasonal forecasting modes), the bias and correlation pattern is similar This suggests that a one month spin up is a good compromise

SLIDE 16

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Mean Climate (DJFMA) : ENSEMBLE forcing

Longitude Latitude

ECMWF ensemble mean DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5

2 4 6 8 10 12 14 16 18 20

Longitude Latitude

ECMWF ensemble spread DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

Longitude Latitude

Ensemble RegCM4 DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5

2 4 6 8 10 12 14 16 18 20

Longitude Latitude

Ensemble spread of RegCM4 DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5

0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

RegCM ensemble mean RegCM ensemble spread ECMWF ensemble mean ECMWF ensemble spread For RegCM, the ensemble spread over land (except over Madagascar) is smaller

SLIDE 17

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Mean Climate (DJFMA) : Biases compared to CRU

Longitude Latitude

ECMWF-ens mean - CRU DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5
5
4
3
2
1

1 2 3 4 5

Longitude Latitude

RegCM4-ensmean - CRU DJFMA rainfall [mm/day] for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5
5
4
3
2
1

1 2 3 4 5

ECMWF RegCM

RegCM is able to reduce most of the biases of the ECMWF GCM

SLIDE 18

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Inter-annual variability (DJFMA): ensemble mean correlation with CRU

Longitude Latitude

Correlation of CRU with ECMWF-ens mean for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5
0.6
0.4
0.2

0.2 0.4 0.6

Longitude Latitude

Correlation of CRU with RegCM4-ensmean for [1991:2001]

10 20 30 40 50

35
30
25
20
15
10
5
0.6
0.4
0.2

0.2 0.4 0.6

ECMWF RegCM

RegCM improves the DJFMA (1 to 5 months lead time) mean correlation over most part of southern Africa compared to the driving GCM but still not good over Malawi

SLIDE 19

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Skill as a function of lead time

Longitude Latitude

DJF correlation (ECMWF−ENS−mean vs CRU) 10 20 30 40 50 −35 −30 −25 −20 −15 −10 −5 −0.8 −0.6 −0.4 −0.2 0.2 0.4 0.6 0.8

Longitude Latitude

DJF correlation (RegCM4−ENS−mean vs CRU) 10 20 30 40 50 −35 −30 −25 −20 −15 −10 −5 −0.8 −0.6 −0.4 −0.2 0.2 0.4 0.6 0.8

Longitude Latitude

JFM correlation (ECMWF−ENS−mean vs CRU) 10 20 30 40 50 −35 −30 −25 −20 −15 −10 −5 −0.8 −0.6 −0.4 −0.2 0.2 0.4 0.6 0.8

Longitude Latitude

JFM correlation (RegCM4−ENS−mean vs CRU) 10 20 30 40 50 −35 −30 −25 −20 −15 −10 −5 −0.8 −0.6 −0.4 −0.2 0.2 0.4 0.6 0.8

Longitude Latitude

FMA correlation (ECMWF−ENS−mean vs CRU) 10 20 30 40 50 −35 −30 −25 −20 −15 −10 −5 −0.8 −0.6 −0.4 −0.2 0.2 0.4 0.6 0.8

Longitude Latitude

FMA correlation (RegCM4−ENS−mean vs CRU) 10 20 30 40 50 −35 −30 −25 −20 −15 −10 −5 −0.8 −0.6 −0.4 −0.2 0.2 0.4 0.6 0.8

3months lead time ECMWF RegCM 1month lead time 2months lead time

SLIDE 20

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work

Summary

For East Africa, RegCM is performing better than ECMWF only when compared over a large area-average and with high resolution gage dataset Errors could propagate from GCM to RegCM and affect the result of the simulation e.g.WIO SST RegCM4 reproduced the mean seasonal climate over the southern Africa when forced by ERA-interim the impact of land surface initialization on skill of the RegCM forecast is smaller for malawi domain RegCM4 has an added value i.e. reduced the bias and increased the temporal correlation of ECMWF ensemble seasonal hindcast Further analysis should be done w.r.t probabilistic assessment

SLIDE 21

Motivation Past work with ECMWF cycle 29r2 over East Africa ECMWF Sys 3 Ensemble data and RegCM set up Perfect boundary condition run ECMWF ENSEMBLE forcing runs Summary and Future work