Seamless water forecasting for Australia Narendra Tuteja, Dasarath Jayasuriya and Jeff Perkins 2 December 2015
Built on extensive research partnerships WIRADA
What we do Perspective Situational awareness Foresight PAST PRESENT FUTURE DECADES YEARS WEEKS DAYS DAYS WEEKS YEARS DECADES National water account Long-term trends Water balance reporting Water market website Seasonal Forecasts Flood and short-range forecasts
Water information services (http://www.bom.gov.au/water/)
Short range forecasts Flood and 7-day forecasts
Flood forecasting and warnings Collect and publish river height and rainfall data Develop event based flood forecasting models Prepare watches, warnings and predictions Deliver briefings to government, emergency services and media Maintain historical flood intelligence
7 Day streamflow forecast service http://www.bom.gov.au/water/7daystreamflow
Service coverage 7-days ahead, updated every day, since 2013 104 locations in all States and Territories Targeted at river and storage operation and environmental flow management Useful for flood guidance Deterministic approach gradually moving to ensemble mode during 2016-18
Forecast performance at each site Rigorous forecast verification Mean Absolute Error (MAE) Skill Score Forecasts evaluated during 2011-14 Performance assessed relative to historical data Mean Absolute Error (MAE) evaluated over each lead time MAE skill score assessed with lead time
Forecast performance across Australia (MAE skill score) 1 day 2 days 3 days 5 days 7 days 1 day 2 days 3 days 5 days 7 days
NWP models in Australia future developments APS-2 APS-3 (Operational in 2018) APS-4 (Operational in 2020) ACCESS-G 25 km 12 km 12 km ACCESS-R 12 km 8 km 4.5 km ACCESS-TC 12 km 4.5 km 4.5 km ACCESS-GE 60 km 30 km 30 km ACCESS-C 1.5 km 1.5 km 1.5 km ACCESS-CE - 2.2 km 1.5 km ACCESS-X - 1.5 km 1.5 km ACCESS-XE - - 1.5 km ACCESS roadmap (APS - Australian Parallel Suite)
Seasonal forecasts 1-month and 3-month forecasts Ѷ
Seasonal Streamflow Forecasting Products http://www.bom.gov.au/water/ssf/forecasts.shtml Forecast summary Educational video Outlook video Dynamic Map of Forecast sites
Seasonal forecast service continues to evolve... 2009 2010 2011 2012 2013 2014 2015 2016 2017 Statistical forecasts 11 sites 21 sites 70 sites 200 sites Dynamic forecasts Experimental evaluation using POAMA 38 sites 44 sites 100 sites Release to registered users using POAMA
Seasonal Streamflow Forecasting Service 140 forecast locations ŠѼ
Seasonal Streamflow Forecasting Service 140 forecast locations 3-month outlook of streamflow volumes Issued every month 5000 ensemble members 1600 subscribers to monthly email 88 % satisfaction from stakeholder survey
Forecasting models statistical POAMA Global Climate Model Rainfall Downscaling Rainfall, Streamflow Data Atmosphere, Sea Surface Monitoring Data Statistical model (BJP) Publicly Available since Dec 2010 Dynamic Models Statistical Models Merging Probabilistic Forecast of Streamflow Monthly and 3 Monthly
Forecasting models dynamic Dynamic model Registered User (released Nov 14) POAMA Global Climate Model Rainfall, Streamflow Data Atmosphere, Sea Surface Monitoring Data Rainfall Downscaling Dynamic Models 匐 ҏ Statistical Models Merging Probabilistic Forecast of Streamflow Monthly and 3 Monthly
Forecasting models merged POAMA Global Climate Model Rainfall, Streamflow Data Atmosphere, Sea Surface Monitoring Data Rainfall Downscaling Merged forecasts Future work (scheduled 2017) Dynamic Models 匐 ҏ Merging Statistical Models Probabilistic Forecast of Streamflow Monthly and 3 Monthly
Forecast verification - Metrics Skill scores: CRPS, RMSE, RMSEP Hit Rates: Tercile hit rates for low / high flows Precision: Inter quantile range (10%-90%) Reliability: PIT plots 匐 ҏ
Aggregated Forecast Performance Index (AFPI) Reliability rating Accuracy rating CRPS RMSEP RMSE Tercile hit rate Precision AFPI is a composite metric of reliability and accuracy ratings
Looking forward seasonal forecasts Transition to ACCESS-S Grid/catchment scale post processing of rainfall and other variables of interest Merging statistical and dynamic forecasts Maintain rigorous forecast verification and refine release criteria Extend forecast lead time beyond 3 months Servicing more than 350 sites every month
Catchment forecasts to marine models Great Barrier Reef lagoon 첰 Ҋ
Great Barrier Reef forecasts Streamflow forecast inputs from catchments to: Operational coastal forecast model (ROMS) Scenario coastal marine model (SHOC) Re-locatable high resolution coastal model (RECOM) Water quality forecast inputs from catchments to the marine Biogeochemical Model (BGC) Aiming to link land management practices to water quality in the rivers in forecast mode
Where are we heading Prototype Forecasts Website Water Quality ROMS Water Volume SHOC Statistical Model Simple daily, lumped Detailed hourly, semidistributed Distributed hourly, fullydistributed
Challenges and lessons learnt Building water forecast services require a national effort using a cooperative approach An extensive user needs analysis and ongoing stakeholder engagement is essential Prioritization is important because not everything can be achieved Data challenges for real time national scale applications is non-trivial User needs guides research priorities and not the other way around Research tools need to be differentiated from operational tools but technology choices in research phase must be cognisant of how the research will be operationalised
Challenges and lessons learnt (contd.) End user confidence is important for adoption of forecasts for decision making Many competing rainfall forecast products could be available which one to use? Operational water forecasting approaches must be simple, and only as complex as necessary objective assessments are warranted 諰 Maintaining currency with research developments for continual improvements in operational services is very demanding Forecasts are probabilistic and communication of these concepts in plain English remains a challenge.
Thank you Questions? 諰