The use and evaluation of GloFAS for operational flood forecasting - - PowerPoint PPT Presentation

The use and evaluation of GloFAS for

• perational flood forecasting

GFP Annual Meeting, Guangzhou, China 11th – 13th June 2019

Shaun Harrigan, Ervin Zsoter, Lorenzo Alfieri, Christel Prudhomme, Hannah Cloke, Peter Salamon, Elisabeth Stephens, and Florian Pappenberger

GloFAS Map Viewer for TC IDAI 16th March 2019 Web: http://www.globalfloods.eu

October 29, 2014

GloFAS operational chain (supported 24/7 since April 2018)

2 EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Meteorological Forecast Forcing

• ECMWF ENS (46r1):
• Daily 00UTC
• 51 members
• Extended to 30d LT

Hydrological Modelling

• HTESSEL, Lisflood

routing (~10 km)

Post-processing

• probability of

exceedance at different lead time, graphs, maps

Web Interface

• visualisation,

background information

Hydro-met. Initial Conditions

• ERA5-T, ENS-Crtl.

Input Observations

• Satellite and in-situ

Input (static) Datasets

• topography, soil, river

network etc.

Flood Thresholds

• reference discharge

reanalysis (GloFASv2-ERA5)

October 29, 2014 3 EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Anomaly correlation of 500 hPa Geopotential reaching 85%

New meteorological model cycle (46r1) operational today!

• Major data assimilation

upgrade

• Many new products

available (e.g. EFI Water Vapour Flux)

• Watch webinars here:

day

October 29, 2014

GloFAS upgraded to v2 on 14 November 2018

• A.) Scientific developments:

– 1.) Global calibration of Lisflood routing & GW (Hirpa et al., 2018, JoH) – 2.) GloFASv2-ERA5 discharge reanalysis (1981 to NRT) – 3.) Improved initialisation of real-time forecasts (ERA5-T, 2-5 day latency)

• B.) User enhancements:

– 1.) Version numbering system – 2.) Availability of datasets (reforecasts and reanalysis) ahead of launch date – 3.) More comprehensive documentation (see: http://www.globalfloods.eu/)

4 EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

October 29, 2014

GloFASv2 Evaluation Datasets

Datasets GloFASv2 GloFASv1

Reforecasts (ECMWF IFS 43r1/43r3)

• 1997-2016
• Initialised 2 per wk.

from ERA5

• 11 ens. members
• Global 0.1° grid
• Calibrated
• 1997-2016
• Initialised 2 per wk. from

ERA-Interim/L

• 11 ens. members
• Global 0.1° grid
• Not calibrated

Discharge reanalysis GloFASv2-ERA5 (1981-2017) Not used Discharge observations Yes Yes

5 EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

October 29, 2014

Forecast performance GloFAS v2 versus v1, w.r.t. obs.

6 EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

v2 performance at 10 day LT as good as v1 at 5 day LT

Lead time (days) Pearson Correlation

Correlation(ensMean, obs.) 0=None; 1=Perfect

October 29, 2014

Overall ensemble forecast skill of GloFASv2

7 EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

CRPSS = 1 – (𝐝𝐬𝐪𝐭_𝐠𝐝)/𝐝𝐬𝐪𝐭_𝐜𝐟𝐨𝐝𝐢) (w.r.t. Reanalysis) 0=No Skill; 1=Perfect skill

Lead time (days) CRPSS

October 29, 2014

… spatial distribution of skill

8 EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

1 day lead 5 day lead 10 day lead 30 day lead CRPSS = 1 – (𝐝𝐬𝐪𝐭_𝐠𝐝)/𝐝𝐬𝐪𝐭_𝐪𝐟𝐬𝐭𝐪𝐬𝐩𝐜) (w.r.t. Reanalysis) 0=No Skill; 1=Perfect skill

CRPSS CRPSS CRPSS CRPSS

October 29, 2014

Ensemble skill for high flow events (Q10 threshold)

9 EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Perfect discrimination

ROC Area Skill Score

ROCSS = 2×ROCA–1 (w.r.t. obs.) 0=No Skill; 1=Perfect skill 5 day lead

October 29, 2014

CEMS in action during Malawi & Mozambique floods

10

Flood Forecast from GloFAS CEMS Rapping Mapping

October 29, 2014

Using GloFAS forecasts to aid decision makers

11 EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

• DfID (UK) requested emergency reports to

aid response to TC Idai humanitarian disaster

• ECMWF, Uni. Reading, Uni. Bristol
• TC Idai (7 reports between 21st and 1st

April):

• TC Kenneth (5 reports between 24th April

and 3rd May)

October 29, 2014

Global flood forecast guidance & impact going operational!

Likelihood

• f impact

High Medium Low

Required Resources

Sub- national National Inter- national

12 EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

monitor monitor monitor prepare prepare act nil nil nil

X

Aristotle multi-hazard activation

Summary

Shaun Harrigan

shaun.harrigan@ecmwf.int

GloFAS Map Viewer: http://www.globalfloods.eu/

Questions?

1.) GloFASv2 operational since 14 November 2018 (Datasets freely available 😁) 2.) GloFASv2 more skilful than v1 in majority of catchments (extend lead time) 3.) GloFASv2 ensemble forecast evaluation, skilful against persistence + climatology 4.) GloFASv2 skilful during high flow events 5.) Flood forecast guidance & impact going operational at the global scale 6.) The road towards GloFAS 3.0 – Lorenzo Alfieri’s talk tomorrow!

Extra slides…

October 29, 2014

Performance of GloFASv2-ERA5 discharge reanalysis

15 EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

Hydrological performance with KGEmod (Gupta et al., 2009; Kling et al., 2012)

correlation bias variability KGEmod(sim., obs.) 0=No Skill; 1=Perfect skill

KGEmod

See Zsoter et al. (2019) JHM!

October 29, 2014

Improvement KGE GloFAS v2 compared with GloFAS v1 (Hirpa et al., 2018, Journal of Hydrology)

• Runoff from 18-km H-TESSEL
• 10-km resolution routing module
• Calibrated at ~1300 locations using KGE

GloFAS Hydrological Model – HTESSEL + LISFLOOD routing

October 29, 2014

Tougher to beat 1 day lead 10 day lead 30 day lead

Q3: GloFASv2 Evaluation: Best benchmark to use?

CRPSS = 1 – (𝐝𝐬𝐪𝐭_𝐠𝐝)/𝐝𝐬𝐪𝐭_𝐜𝐟𝐨𝐝𝐢) (w.r.t. Reanalysis) 0=No Skill; 1=Perfect skill

October 29, 2014

GloFAS discharge observations network

Observations maintained by Joint Research Centre (JRC) [2041 stations with at least 1 data entry] Sources include: GRDC, National Hydro. Met. Services, GloFAS partners]

20 yr reforecast period (1997-2016)

October 29, 2014

Forecast Bias (reanalysis)

19 EUROPEAN CENTRE FOR MEDIUM-RANGE WEATHER FORECASTS

1 day lead 5 day lead 10 day lead 30 day lead