Evaluation of NWP rainfall forecasts for 7-day ensemble

Evaluation of NWP rainfall forecasts for 7-day ensemble

Evaluation of NWP rainfall forecasts for 7-day ensemble flow forecasting service Dr. Sophie Zhang Water Forecasting Services Bureau of Meteorology 27 November 2018, Bureau R&D Workshop Prasantha Hapuarachchi Aynul Kabir Mohammed Hasan Mohammed Bari Sophie Zhang Jayaratne Liyanage Nilantha Gamage David Robertsen (CSIRO) James Bennett (CSIRO) Service Development 7day streamflow forecasts Current deterministic forecasts Numbers of Forecast locations: Ensemble forecasts in development http://www.bom.gov.au/water/7daystreamflow Public Regd. users TAS 22 24 WA 25 28 SA 3 4 NT 14

16 VIC 34 43 NSW 38 56 QLD 32 38 Total 168 209 Workflow and Evaluation methodology Data period (overlapping): 2014 2016 (3 years) Rainfall/Flow evaluation: with and without post-processing Raw Flow Temporal resolution: daily Metrics: Bias, CRPS, KS-D Rainfall products: ACCESS-G ACCESS-GE2 (24 ens) PME ECMWF (50 ens) RPP Rain Raw Rain Corrected Flow

-41.6 -41.7 -41.7 -41.8 -41.8 -41.9 -41.9 -41.3 -41.3 -41.4 -41.4 -41.5 -41.5 -41.6 -41.6 -41.7 -41.7 -41.8 -41.8 -41.9 -41.9 -41.3 -41.3 -41.4 -41.4 *Robertson et al. (2013); Shreshtha et al. (2013)

Observations Obs ervation -41.3 Raw NWP -41.5 Pos t-p Post-processed ACCESS-G Raw ACCESS-G -41.4 Day 1 -41.6 Day 3 Rainfall post-processing Day 1 -41.5 Day 3 -41.5 -41.6 -41.7 -41.8 -41.9 -41.6 -41.6 -41.7 -41.7 -41.3 -41.4 -41.5 -41.6 Day 9

-41.5 Day 9 -41.5 -41.7 -41.8 -41.9 -41.3 -41.8 -41.8 -41.9 -41.9 -41.4 -41.5 -41.6 -41.7 -41.8 147.2 147.2 147.4147.2 147.6147.4 147.8147.6 148147.8 148.2148147.2 148.2147.4 147.4 147.6 -41.9 147.2 147.6 147.8 147.4 147.8 148 147.6 147.8 148 147.2 148.2 147.2 148.2

147.4 148 148.2 147.2 147.4 147.6 147.4 147.6 Rainfall (mm/day) Rainfall (mm/day) 1.5 2 2.5 147.6 147.8 147.8 147.8 148 148 148 148.2 148.2 147.2 148.2 147.4 Rainfall (mm/day) 3 3.5 4 4.5 5

147.6 26 Study Catchments No Region Catchment 1 NSW Abercrombie Area (km2) 2 NSW Bega 316 Bemboka River at Morans Crossing 3 NSW; ACT Cotter 130 Cotter River at Gingera 4 NSW Goobarragandra 668 Goobarragandra River at Lacmalac 5 NSW Hunter

1189 Hunter River at Belltrees 6 NSW Mandagery 1697 Mandagery Creek at Eugowra Smithfield 7 NSW; ACT Queanbeyan 906 8 NSW Shaolheaven 4660 9 NT Adelaide 638 Adelaide River at Railway Bridge 10 QLD Brisbane 3875 Brisbane River at Gregor Creek 11 QLD

Gilbert 12 QLD Stanley 13 QLD Tully 14 QLD Warrego 15 SA Onkaparinga 224 Onkaparinga D/S Hahndorf Creek 16 SA Torrens 347 Torrens River at Gorge Weir 17 TAS Duck 347 Duck River at Scotchtown 18

TAS Florentine 445 Florentine River above Derwent River 19 VIC Latrobe 558 Latrobe River at Willow Grove 20 VIC Loddon 1029 Loddon River at Newstead 21 VIC Macalister 1903 Macalister River D/S Lake Glenmaggie 22 VIC Seven Creeks 1519 Seven Creeks at Kialla West 23 VIC Tambo

2676 Tambo River D/S Ramrod Creek 24 WA Blackwood 25 WA Collie 26 WA Mcknoes 2632 11537 246 1388 44786 19201 1328 27 Site Name (Outlet) Abercrombie River at Abercrombie No.2 Queanbeyan River at Wickerslack Shoalhaven River at Fossickers Flat Gilbert River at Rockfields Stanley River at Woodford Alert-B Tully River at Euramo Warrego River at Cunnamulla Weir Blackwood River at Nannup Collie River East at Coolangatta Farm McKnoes Brook at Urquharts Evaluation: bias (%) Raw and post-processed rainfall

post-processed streamflow Evaluation: skill MAE/ CRPS (mm/day) Raw and post-processed rainfall post-processed streamflow Evaluation: reliability (KS-D) Raw and post-processed rainfall post-processed streamflow Investigation of optimal ensemble number 200 ensemble size Summary: rainfall forecasts Metric Raw forecasts Post processed Bias (%) ECMWF forecasts are relatively less biased, followed by PME. Significantly reduced biases across all catchments at all leadtimes. CRPS (mm/day) Increases with the lead-time. Improves ACCESS-G rainfall forecasts, but no notable improvement in ECMWF and ACCESS-GE. Reliability (KS-D) Slightly improves with the leadtime. Significant improvement compared to raw rainfall. Summary: streamflow forecasts Metric Raw forecasts

Post processed Bias (%) Largely variable from catchment to catchment. Post-processed rainfall produces streamflow with slightly lower bias. Significantly reduced biases at short lead-times and the effect declines with lead-time. CRPS (mm/day) Post-processed rainfall driven streamflow is slightly improved and found to be consistent. Improves significantly at shorter lead-times and noticeable improvements up to 6 days. Reliability (KS-D) Improves for all rainfall products with the lead-time. Postprocessed rainfall driven streamflow is slightly better. Improves for all rainfall products with lead-time. Post-processed rainfall driven streamflow is significantly better. Outcomes 1. Raw forecast rainfall forecast performance order (ECMWF, ACCESS-GE, PME, ACCESS-G) 2. Both rainfall and streamflow post processing are required for the operational system 3. 200 ensembles of each NWP rainfall forecasts recommended for operational service 4. Super ensemble comprised of 401 members are now planned for operational implementation o 200 post processed ECMWF rainfall forecasts o 200 post processed ACCESS-GE2 rainfall forecasts o PME deterministic forecasts (mean of NWP models available to Bureau) Thank you Questions?

Recently Viewed Presentations

  • Revolution: Romanticism to Realism

    Revolution: Romanticism to Realism

    Revolution: Romanticism to Realism Agenda Bell Ringer: Who am I? Very quick lecture: Romanticism and Revolution Chart interpretation: Romanticism vs Impressionism vs Realism Literary analysis: Comparing Romanticism to Realism Image analysis: The changing paintings of the era. Review Guide and...
  • [Enter Presentation Name Here]

    [Enter Presentation Name Here]

    Facilities Tasks and Timeline. February 2016. Cave1 electrical work complete. Facilities has issued PR for Cave2 concrete and bracing, out for bids. March. Installation Group willing to secure Cave2 structure, install other utilities. April. Facilities moves IPC outside Cave2. Concrete...
  • MODIS Surface Albedos and Distribution of Global Ecosystem Types

    MODIS Surface Albedos and Distribution of Global Ecosystem Types

    MODIS Surface Albedos and Distribution of Global Ecosystem Types Eric G. Moody1,2, Michael D. King1, Steven Platnick1,3, Crystal Schaff4 1NASA Goddard Space Flight Center 2Emergent IT, Inc. 3University of Maryland Baltimore County 4Boston University Outline Motivation Issues with original albedo...
  • Middle Ages

    Middle Ages

    He led the peasants in the First Crusade, also known as the Peasant's Crusade. The First Crusade set out from France and first had to conquer Antioch before the Crusaders could make their way into Jerusalem. Knights and peasants were...
  • GOTHIC ARCHITECTURE Week 11 Gothic architecture is a

    GOTHIC ARCHITECTURE Week 11 Gothic architecture is a

    Gothic architecture is a style of architecture which flourished during the high and late medieval period. It evolved from Romanesque architecture and was succeeded by Renaissance architecture. Its characteristic features include the pointed arch, the ribbed vault and the flying...
  • Reliability & Validity

    Reliability & Validity

    Basically, humans have the right to use animals as we see fit. Continued research with animals will be necessary if we are to ever conquer diseases such as cancer, heart disease, and AIDS. ... Issue: Need to make sure the...
  • The Care Act Issues likely to lead to

    The Care Act Issues likely to lead to

    Regs - or ignoring the statutory purpose. Failing to allow a person to require the involvement of a nominated person in a decision where this is required, or regarding their relative as their informal supporter without getting the person's ....
  • Friendship With GOD Name Lay Person Kairos -

    Friendship With GOD Name Lay Person Kairos -

    God sent these men and used their friendship with Him to spread His word. He finally sent His son, Jesus, the Christ Friendship With GOD YOU ARE WORTHY TO BE GOD'S FRIEND! Jesus is God. He and the Father are...