Yu Zhang1 and Haksu Lee2,3
Transcription
Yu Zhang1 and Haksu Lee2,3
Assimilation of Satellite QPE in A Distributed Hydrologic Model for Flood Predictions Yu Zhang1 and Haksu Lee2,3 1Office of Hydrologic Development NOAA National Weather Service 2Office of Climate, Weather and Water NOAA National Weather Service 3Len Technologies, Oak Hil, VA 1 Outline Ø NWS flood forecast in the US Ø Products and services Ø Elements of forecast operation Ø Hydrologic l l l experiment Objectives Experiment Layout Observations Ø Summary and implications for global flood monitoring 2 Flood Forecast in the US Ø NOAA National Weather Service responsible for forecasts for riverine floods in the US Ø Primary products Ø Stage Forecast Flood Warnings Ø Flash Flood Warnings Ø Inundation extent Ø Ø Flood outlook Minor-moderate Flooding 3 Elements of NWS Forecast Hydrologic Models Ø Forcing inputs Ø Precipitation, temperature, etc Ø Reservoir inflow Ø Observed river stages from USGS Ø Forecaster interventions (assimilation) Ø 4 NWS Hydrologic Models Ø Lumped Model Ø Prediction only at designated outlets w/ uniform forcing Ø SAC-SMA (runoff) and Unit Hydrograph (routing) Ø Carefully calibrated at each outlet Ø Distributed Ø Model Ø Gridded (1,2 and 4 km mesh) Ø Multiple runoff mechanisms (SAC-SMA, SAC-HTET, API, etc) Ø Kinematic routing Ø A priori parameters from physiographic data Both models among best performers in the Distributed Model Inter-comparison Projects (DMIPs) 5 Precipitation Inputs Ø Requirements: Ø Accurate, Ø Existing Ø Fused Quantitative Precipitation Estimates (QPEs) multi-sensor data Ø Primarily Ø Limited Ø Issues Ø Satellite low latency, and spatially contiguous radar and gauge observations satellite input (Hydro-estimator) with gaps and quality in radar/gauge data QPEs Ø In particular Global Precipitation Measurement (GPM) Ø Expected to help fill gaps and improve the accuracy in radar/gauge sparse areas 6 Radar Coverage Gaps Summertime Wintertime • More severe radar coverage gaps during winter time 7 Data Assimilation Ø Data Assimilation : Ø Systematic ways of adjusting model using observations Ø Mimics forecast interventions Ø 4D Variational Assimilation (Var) techniques developed at OHD Ø 3D + time dimension = 4D Ø Assimilates Ø Streamflow,soil moisture, etc Ø Forcings (precip, PET) Ø Experimental at River Forecast Centers 8 Simulation Experiment Ø Motivations Ø Understand the accuracy of flood forecast driven by satellite QPEs Ø Assess the impacts of streamflow assimilation on forecast accuracy Ø Experiment Layout Ø Simulate flood events using distributed model Ø Ingest satellite QPE (CMORPH) Ø Use 4D Var to assimilate Ø Streamflow, QPEs and Potential ET Ø Evaluate the simulation at different forecast horizons 9 Study Setting and Data sets Ø TIFM7 Ø TIFM7 Ø Elk River near Tiff City, Mo Ø Drainage Area: 2258 km2 Ø LANAG: subbasin (619 km2) QPEs Ø TIFM7 LANAG Ø Ø CPC-MORPH (CMORPH) Ø 8km, 30-min Multisensor QPE (MQPE) Ø 4km, hourly HL-RDHM (a research prototype of NWS DHM) Ø Ø Ø 4-km grid mesh Hourly time step A priori parameters 10 Hydrologic Results Ø Flood of 3-10, March, 2004 Ø Baseline Simulations (w/o assimilation) Ø DA results Ø Ø From assimilation of streamflow at outlet and QPEs Performance at outlet and interior points Ø Raw and adjusted soil moisture (UZFWC) 11 Baseline Simulations Mar 3-10, 2004 Flood Pos. bias from CMORPH TIFM7 Neg. bias from MPE LANAG Pos. bias more severe for CMORPH over the interior point Time (hours from 0Z, 2004/03/03) Using satellite QPE alone may lead to large bias in predicted streamflow 12 Assimilation Results Mar 3-10, 2004 Flood CMORPH: nearly identical traces after assimilation TIFM7 MQPE: negative bias remains LANAG Both CMORPH and MQPE yielded flood peak higher than observed CMORPH bias is improved but remains positive Assimilation at outlet markedly improves the prediction at interior points 13 Model State Adjustments UZFWM (upper zone free water) CMORPH: Note large differences before and after assimilation – heavy adjustment due to poor performance MQPE: Minor adjustments – baseline results relatively good Baseline Assimilation Major adjustments to the CMORPH-based model states to compensate for precip errors 08/14/10 14 Summary and Findings Ø Hydrologic experiment Ø Ø Ø Ø Baseline simulations: Ø Ø CMORPH vs. operational MQPE Assimilate streamflow at outlet Examine results at interior points Large positive discharge bias in CMORPH-based results Assimilation results: Ø Ø Ø Appreciable improvements after assimilation at interior point Better results from CMORPH Heavy adjustments for CMORPH-driven soil moisture states 15 Implications for Global Flood Monitoring and Forecast Ø Satellite QPEs Ø Ø Ø Can be subject to substantial errors Bias magnified in streamflow predictions Data assimilation Ø Ø Ø Infusing in situ/remotely sensed observations Ø Streamflow/soil moisture/inundation extent Shown to help suppress large positive bias Useful for monitoring floods in drainages where limited observations are available 16