Wales real time forecasting for fast responding rivers Andy Wall Flood Forecasting Team Leader Environment Agency Wales
Flood risk & forecasting in Wales
Fast responding rivers 220,000 properties at risk from rivers & sea 3 million population 1 in 6 live or work on the floodplain 8 billion worth of assets at risk
Typical river response From normal to severe event in less than 2 hours
Typical flood timeline Prepare = what if = how bad could it be? = informing professional partners Forecasting Gather real time data = what is happening? Interpret data, make decisions = what could happen? Warn public & professional partners Respond on the ground
Up to 5 days 24 to 48 hours More timely More accurate 6 to 24 hours Community ~0 to 1 hour? River basin <<6 hours
Radar quality High risk area Densely populated, fast responding rivers
Warning without a forecast Flooding takes place (threshold) River Level LEAD TIME Traditional trigger = issue warning when observed level reaches it Time
Flood Forecast at 11:00 Flood Warning threshold will be exceeded at 12:30
What actually happened Flood Warning threshold exceeded just before 12:30
Recent developments
Wales flood forecasting project 1.8m incl. 500k EU grant New forecasting system 755k New models 650k Expanded measurement network
National Flood Forecasting System Observed rainfall, river & tide data Radar/Weather Forecast (Met. Office) Regional Telemetry System NFFS Weather & tidal forecasts (Met Office & STFS) Web Browser Reports
River forecasting models: 2006 to 2010
Typical model structure LF CR Craig Y Nos N Legend Craig Y Nos G.S. PDM gauged U Update location U PDM ungauged ISIS node and forecasting point ISIS Node Telemetere d raingauges LF: Llyn Y Fan DL CR: Crai Reservoir N: Nantyrwydd Yg: Ystradgynlais Yo: Ystradowen T: Trebanos S: Spite LF N Gurnos Yg Yo Teddy Bear Bridge G.S. U Teddy Bear Bridge LF C N Yg Gurnos G.S. U Yo T Yg Ynystanglws S Ynystanglws G.S. U Downstream boundary
Performance testing
Why? Q How well does the model predict what happened during previous floods - does it match what was observed? Q How accurate & timely are model forecasts when would we first forecast threshold exceedance, what is the effective lead time? Q If raingauge(s) fail how well does the model perform? Q Where would we used observed data to update model output and what updating method do we use?
Standard model calibration
Performance testing using NFFS Identify suitable flood events Gather rainfall & river level data Change NFFS configuration files Run historical batch forecasts Run event hourly batch forecasts Analyse and report results
Forecast 5 hours before peak (with perfect rainfall knowledge) Hourly forecast runs Performance testing using observed data up to time of forecast (T0) Hour by hour model runs tell you your lead time & accuracy only 1 hour here
Benefits of performance testing Determine forecast lead times for flood risk locations throughout a river catchment BEFORE a model goes operational Identify areas where we need to improve lead time Quantifies performance of our forecasting service Duty officers understand strengths and limitations of model better decision making during a flood Helps plan future model development (lessons learnt)
Gaining lead time?
Tp vs Lead Time 5 Lead Time (hrs) 4 3 2 1 0 0 2 4 6 8 10 12 Tp (hrs) Tp<3hrs = no effective lead time (if you only use observed rainfall)
How can we gain more lead time? Speed up transfer from telemetry system to NFFS but really only accuracy Run models in fast catchments every ½ hour on NFFS (only hourly in testing) Fixed rainfall rate/duration scenarios?
Fixed rainfall rate/duration scenarios Scenarios: Average of 2.5, 5 & 10 mm for next hour Average of 2.5, 5 & 10 mm for next 2 hours Others? Configured to run on NFFS as an ensemble
Fixed rainfall rate/duration plots Observed level Forecast to T0 T0 2.5 mm/hr 5 mm/hr 10 mm/hr No more rain Flood w atch Flood w arning Severe flood warning 3.5 3 Observed Rainfall up to T0 and after T0 2.5 Stage (mald) 3 2.5 2 1.5 1 Rainfall mm 2 1.5 1 0.5 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 Time T0 Observed rainfall 0.5 0 3 2.5 2 16/07/2009 22:59 3 2.5 2 17/07/2009 22:59 3 2.5 2 Rainfall mm 1.5 Rainfall mm 1.5 Date/time Rainfall mm 1.5 1 1 1 0.5 0.5 0.5 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 Time 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 Time 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 Time T0 Rainfall to T0 then 2 mm rate T0 Rainfall to T0 then 5 mm rate T0 Rainfall to T0 then 10 mm rate
3 2.5 2 1.5 1 0.5 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 Time T0 Rainfall to T0 then 2 mm rate 3 2.5 2 1.5 1 0.5 0 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 Ti me T0 Rainfall to T0 t hen 5 mm rate 3 2.5 2 1.5 1 0.5 0 O bserved Rainfall up to T0 and after T0 1 3 5 7 9 11 13 15 17 19 21 2 3 25 27 29 31 33 35 37 39 41 43 4 5 47 49 51 53 55 57 59 61 63 65 67 Time 3 2.5 2 1.5 1 0.5 0 T0 O bserved rainfall 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63 65 67 Ti me T0 Rainfall to T0 th en 10 mm ra te = Fixed rainfall rate/duration plots Observed level Forecast to T0 T0 2.5 mm/hr 5 mm/hr 10 mm/hr No more rain Flood w atch Flood w arning Severe f lood w arning 3.5 3 2.5 Rainfall mm Stage (m ALD) 2 1.5 1 0.5 0 Rainfall mm 16/07/2009 22:59 Rainfall mm Date/time 17/07/2009 22:59 Rainfall mm best available + nowcast +
What we provide now Operational forecasts use observed rainfall up to T0 (what has fallen) 36 hour ahead forecast based on long term rainfall predication (based on MO NWP model) What we will soon provide Fixed rainfall rate/duration forecasts predetermined rainfall profiles used to give set of scenarios (~ 4). Use best judgement, Hyrad & MO forecaster to pick the best estimate
Closing thoughts
Opportunities & challenges in Wales Expanding forecasting model coverage Accurate short term rainfall forecasts (next few hours) = holy grail of forecasting A better professional partner advisory service Probabilistic forecasting