On the Improvement of Numerical Weather Prediction by
On the Improvement of Numerical Weather Prediction by Assimilation of Wind Power data Stefan Declair*, Klaus Stephan, Roland Potthast Erstellung innovativer Wetter- und Leistungsprognosemodelle für die Netzintegration wetterabhängiger Energieträger - Eine Kooperation von Meteorologie und Energiewirtschaft - 79. DPG-Jahrestagung, Arbeitskreis Energie Berlin, March 18 th 2015
Source: Andrea Streiner, DWD
Who is EWe. Li. NE?
Agenda 1. Data Assimilation 2. Impact-Study
Agenda 1. Data Assimilation 2. Impact-Study
Forecast: Can I cross the street without getting hit? Information used: Forecast errors due to: • Observations • Observation (estimation) errors • Knowledge about cars, street, etc • Model errors (icy street) • Experience statistics • Case does not match statistics
Weather forecast Numerical model Data assimilation tool Observations Improved initial conditions for next integration step
Agenda 1. Data Assimilation 2. Impact-Study
OSSE Ø What: Observation System Simulation Experiment Ø Goal: Test the impact of newly available observations in the data assimilation Ø Method: assimilate artificial observations in slightly perturbed truth Ø Advantages: Ø Truth is known exactly Ø All generated athmospheric fields can be used as observations Ø Observation system can be altered easily Ø Observation errors Ø Observation densities Ø Temporal resolution/delay
OSSE Ø What: Observation System Simulation Experiment Ø Goal: Test the impact of newly available observations in the data assimilation Ø Method: assimilate artificial observations in slightly perturbed truth create artificial obs * truth assimilate perturb * obs: all conventional obs ervations PLUS wind observations at average park hub height control free forecast
OSSE – Settings Ø Artificial wind observations Ø 68 wind farm sites Ø Average hub height, farm point of mass Ø 15 min resolution/10 min delay Ø Observation error: N(0, 2 ms-1) Ø Control Ø 2 perturbations @ physics Ø 2 perturbations @ dynamical core
OSSE – Settings Ø Cycling over N-day evaluation period Ø Hourly assimilation of artificial wind observations Ø Hourly free forecast over 21 h days 1 21 h forecast analysis 2 21 h forecast analysis 3 21 h forecast analysis N-1 21 h forecast analysis N 21 h forecast analysis UTC time 12 18 00 06 12 18
OSSE – Results Test Period Ø Results for 2013062100 - 2013062918, mean over all 00 UTC free forecasts Computational domain evaluation region
OSSE – Results Test Period Ø Results for 2013062100 – 2013062918 Ø How many observations have been assimilated? Ø Conventional observations (AIREP, TEMP, etc): ~4000 -5000 / h Ø Artificial wind information: <300 / h Ø New observations have small weight compared to conventional obs! Ø 3 possibilities: Ø Reduce amount of conventional observations Ø Evaluate locally around station / along wind path Ø Rerun with higher artificial wind observation density (work in progress)
OSSE – Evaluation 1 Ø Results for 2013062100 - 2013062918, mean over all 00 UTC free forecasts Computational domain evaluation region
OSSE – Evaluation 2 Ø Evaluate locally : Ø at reference wind park Ø propagate evaluation point with wind field x x x
OSSE – Evaluation 2 Ø Results for 2013062100 - 2013062918, mean over all 00 UTC free forecasts Ø RMSE between NTR analysis and ctl (marks) / exp Ø 68 stations Ø Positive local impact Ø Horizon: Ø Stat: up to 12 h Ø Dyn: up to 17 h Ø Diurnal error: slightly…
Conclusion Ø Data assimilation Ø NWP is a (boundary and) inital value problem: you need accurate initial fields Ø Task: create a best-fit atmospheric state according to first guess and observations Ø Impact study: OSSE Ø Visible positive impact of artificial hub height wind speeds Ø Regional: Ø Fierce competition with conventional observation networks: neutral Ø Unrivaled: strongly positive over 8 hours Ø Local: Ø positive effect for more tha half a day even with conventional observation networks included
Thank you for your attention! Now: Q & A
- Slides: 19