220 likes | 363 Views
NWS Hydrology Forecast Verification Team: 17 th Meeting. 06/29/2009 –2 pm EDT. Outline. Verification in CHPS: Demo Next steps Final team report: consensus on Recommended verification metrics and products RFC verification case studies Impact of QPF horizon
E N D
NWS Hydrology Forecast Verification Team:17th Meeting 06/29/2009 –2 pm EDT
Outline • Verification in CHPS: • Demo • Next steps • Final team report: consensus on • Recommended verification metrics and products • RFC verification case studies • Impact of QPF horizon • Impact of run-time mods made on the fly • New Verification Team charter
Verification in CHPS • 2 classes of verification with different user requirements • Diagnostic verification to know/improve model performance; done off-line with archived forecasts or hindcasts to compute verification statistics given different conditions (e.g., time periods, above threshold) • Real-time verification to help forecasters make decision in real-time; done in real-time by • querying and displaying historical analogs using multiple criteria • displaying summary of past diagnostic verification statistics • checking for forecast anomalies • if necessary, run bias-correction program and display result
Verification in CHPS: demo • Demonstrate analog capability display • Select analogs from a pre-defined set of historical events and display with ‘live’ forecast in FEWS Time Series Display • Demonstrate summary diagnostic verification displays • Display summary diagnostic verification products in FEWS Map Display: map with 1 metric (e.g. Relative RMSE)
Verification in CHPS: analog demo • Demonstrate analog capability display w/ FEWS Time Series Display • Select analogs from a pre-defined set of historical events and display with ‘live’ forecast Live forecast Observations
Analog 2 Analog 3 Verification in CHPS: analog demo • Demonstrate analog capability display w/ FEWS Time Series Display • Select analogs from a pre-defined set of historical events and display with ‘live’ forecast Analog 1
Verification in CHPS: analog demo • Demonstrate analog capability display w/ FEWS Time Series Display • Select analogs from a pre-defined set of historical events and display with ‘live’ forecast Analog 3 Live forecast What happened
Verification in CHPS: analogs • Build analog query prototype using multiple criteria • Queries from forecast and/or observed attributes (e.g. forecast amount) on different time and/or space coordinates (e.g. same forecast lead time and location). Could include auxiliary variables (e.g. select flow analogs based on precipitation forcing). • Need to identify appropriate auxiliary information for selecting and interpreting the analogs (e.g. initial soil moisture conditions). • Should we just focus on a restricted range of flows, such as high flows? Of course, potential problems with sample size and uniqueness of major events.
Examples of queries for analogs • Seeking analogs for precipitation: “Give me past forecasts for the 10 largest events relative to hurricanes for this basin.” • Seeking analogs for flow: “Give me all past forecasts for which the forecast value at lead hour 6 was within an interval [current 6-hr forecast value ± 20%] and the forecast peak occurred within the next 48 hours”.
More complex queries for analogs • Seeking analogs for temperature: “Give me all past forecasts with lead time 12 hours whose ensemble mean was within 5% of the live ensemble mean.” • Seeking analogs for precipitation: “Give me all past forecasts with lead time 6 hours whose PoP was >=0.95, whose ensemble mean was >= 0.5 inches and for which lightening strikes were observed within a 50km radius of the forecast point at the forecast issue time.” • Seeking analogs for flow: “Give me all past forecasts with lead times of 12-48 hours whose probability of flooding was >=0.95, where the basin-averaged soil-moisture was > x and the immediately prior observed flow exceeded y at the forecast issue time”.
January April Verification in CHPS: map demo • Demonstrate summary diagnostic verification displays w/ FEWS Spatial Display • Display verification map with monthly results of Relative RMSE October
Lead Day 2 Lead Day 1 Lead Day 3 Verification in CHPS: map demo • Demonstrate summary diagnostic verification displays w/ FEWS Spatial Display • Display verification map with monthly results of Relative RMSE
Verification in CHPS: next steps • Perform user analysis to identify functional requirements and meaningful verification products • To be done by NWS Verification Team, CAT/CAT2 Teams, Graphics Generator Requirements Team, and SCHs • New RFC verification case studies to work with proposed standards
Verification in CHPS: time line • Summer ’09 - Winter ’10 • define user requirements for CHPS VS (HEP w/ RFCs & teams) • develop analog query prototype (wo/ database querying) (HEP w/ some RFCs) • Winter ’10 - ’11 • prepare functional/technical software design (HSEB w/ HEP) • Continue to enhance verification science for CHPS VS (HEP)
Final team report: consensus? • Recommendations on • sets of verification metrics and products to be used at all RFCs • sensitivity analyses on impact of QPF horizon and impact of run-time mods made on the fly • New future team activities • Team work to be presented at the HIC meeting on 07/10 • approval of second team charter?
Sensitivity analysis: QPF horizon • Goal: what is the optimized QPF horizon for hydrologic forecasts? • QPF horizon to test: • 0 (no QPF), 6-hr, 12-hr, 18-hr, 24-hr, 30-hr, 36-hr, 48-hr, 72-hr, 96-hr • Model states to use: • Similar to operational mods except mods that impact future states • Metadata to store which mods were used in these runs • What forecast to verify • 6-hr stage forecasts for 7-day window (longer for slow response basins)
Sensitivity analysis: run-time MODs • Goal: do run-time mods made on the fly improve forecasts? • 4 scenarios • Operational forecasts (w/ all mods) • Forecasts w/ best available obs. and fcst. inputs wo/ on-the-fly mods • Forecasts w/ best available obs. inputs (no fcst) w/ all mods • Forecasts w/ best available obs. inputs (no fcst) wo/ on-the-fly mods • What forecast to verify • 6-hr stage forecasts for same window as in operations • Model states: • Carryover from 5 days ago (w/ past mods) + a priori mods (known before producing any forecast)
Next meeting • 18th meeting: early September • Update on comments for team report • Discussion on • RFC verification case studies • CHPS Verification Service
Final team report: Metrics • 4 different levels of information
Final team report: Analyses • Use different temporal aggregations • 6-hr instantaneous flow vs. weekly minimum flow • Avoid data pooling across different lead times • Quality strongly depends on lead time • Plot verification statistic as function of lead time • Perform spatial aggregation carefully • Aggregate verification results across basins with similar hydrologic processes • Analyze results for each individual basin and analyze results plotted on verification maps
Final team report: Analyses • Analyze forecast performance • w/ time conditioning: by month, by season • w/ atmospheric/hydrologic conditioning: • low/high probability threshold • absolute thresholds (e.g., PoP, Flood Stage) • Plot also sample size (in future confidence intervals) • Analyze sources of uncertainty and error • Verify forcing input forecasts and output forecasts • For extreme events, verify both stage and flow • Sensitivity analysis to be set up at all RFCs: 1) impact of QPF horizon; 2) impact of run-time mods made on the fly