The University of Washington Pacific Northwest Mesoscale Analysis

The University of Washington Pacific Northwest Mesoscale Analysis System Brian Ancell, Cliff Mass, Gregory J. Hakim University of Washington

Motivation n High-resolution analyses are important for: • Operational forecasting (fire weather, air quality. . )

Motivation n High-resolution analyses are important for: • • Operational forecasting (fire weather, air quality. . ) Studying the mesoscale effects of climate change

Motivation n High-resolution analyses are important for: • • • Operational forecasting (fire weather, air quality. . ) Studying the mesoscale effects of climate change Alternative energy development

Motivation n High-resolution analyses are important for: • • • n Operational forecasting (fire weather, air quality. . ) Studying the mesoscale effects of climate change Alternative energy development Pacific Northwest complex terrain presents a challenge to creating good analyses • Flow-dependence during data assimilation may

An Attractive Option: En. KF n An ensemble Kalman filter (En. KF) has strong potential for mesoscale analysis: • Observational information is spread spatially using flow-dependent statistics

An Attractive Option: En. KF Temperature observation 3 DVAR En. KF

An Attractive Option: En. KF n An ensemble Kalman filter (En. KF) has strong potential for mesoscale analysis: • • Observational information is spread spatially using flow-dependent statistics Analysis and forecast uncertainty is easily calculated and is also flow-dependent

An Attractive Option: En. KF n An ensemble Kalman filter (En. KF) has strong potential for mesoscale analysis: • • • Observational information is spread spatially using flow-dependent statistics Analysis and forecast uncertainty is easily calculated and is also flow-dependent Computational resources can handle En. KF demand

How the En. KF Works n An analysis is created from: 1) An ensemble of short-term forecasts (Background) 2) Observations For a single observation: Observation (T 1) Mean Forecast (T 2) Observation Variance (V 1) Forecast Variance (V 2) Analysis (T 3, V 3)

How the En. KF Works n An analysis is created from: 1) An ensemble of short-term forecasts (Background) 2) Observations For a single observation: Observation (T 1) Mean Forecast (T 2) Observation Variance (V 1) Forecast Variance (V 2) Analysis (T 3, V 3) Analysis increment then spread spatially using covariance statistics of

En. KF Configuration n Large, coarse domain En. KF already tested (Torn and Hakim 2008) - En. KF competitive with global models

En. KF Configuration D 3 (4 km) D 2 (12 km) D 1 (36 km)

En. KF Configuration n n WRF model V 2. 1. 2 38 vertical levels 80 ensemble members 6 -hour update cycle Observations: • Surface temperature, wind, altimeter • ACARS aircraft winds, temperature • Cloud-track winds • Radiosonde wind, temperature, relative humidity Half of surface obs used for assimilation, other half for

36 -km vs. 12 -km En. KF 36 -km 12 -km SLP, 925 -mb temperature, surface winds

36 -km vs. 12 -km En. KF 36 -km 12 -km SLP, 925 -mb temperature, surface winds

36 -km vs. 12 -km En. KF 36 -km 12 -km SLP, 925 -mb temperature, surface winds

En. KF 36 -km vs. 12 -km Wind Temperature Improvement of 12 -km En. KF Analysis 10% 13% Forecast 10%

Issue #1 – Representative Error n Model terrain = Actual terrain at and near observation sites Model terrain Actual terrain

Surface Observations Model grid points (12 -km resolution)

Surface Observations Model grid points (12 -km resolution) Observation location Model grid points (12 -km resolution)

Surface Observations Model grid points (12 -km resolution) High-resolution terrain data (1. 33 km resolution) Observation location Model grid points (12 -km resolution)

Issue #1 – Representative Error n Using representative observations only, we can reduce observation uncertainty: Observation Standard Deviations Temp: 1. 8 K (36 -km) 1. 0 K (12 -km) Wind: 2. 5 m/s (36 -km) 1. 5 m/s (12 km)

Issue #1 – Representative Error n Using representative observations only, we can reduce observation uncertainty: Observation Standard Deviations Temp: 1. 8 K (36 -km) 1. 0 K (12 -km) Wind: 2. 5 m/s (36 -km) 1. 5 m/s (12 km) Drawback: Lose ~75% of available surface obs

Issue #1 – Representative Error Wind Temperature Improvement using reduced observation uncertainty Analysis 5% 10%

Issue #2 – Lack of Background Surface Variance n Too little background variance exists in model surface fields

Issue #2 – Lack of Background Surface Variance n Too little background variance exists in model surface fields Solution: Inflate surface variance with variance aloft

Issue #3 – Model Surface Bias n Significant biases exist in the model surface wind and temperature fields Temperature Bias Light Wind Speed (<3 knots) Bias

Further Improvement After Variance Inflation, Bias Removal Wind Temperature Improvement using inflation and bias removal Analysis 9% 3%

En. KF 12 -km vs. GFS, NAM, RUC Wind Temperature RMS analysis errors GFS NAM RUC En. KF 12 -km 2. 38 m/s 2. 30 m/s 2. 13 m/s 1. 85 m/s 2. 28 K 2. 54 K 2. 35 K 1. 67 K

12 -km vs 4 -km En. KF 12 -km SLP, 925 -mb temperature, surface winds 4 -km

Summary n A multi-scale, nested WRF En. KF (36 km, 12 km, 4 km) is being tested over the Pacific Northwest to produce quality analyses and short-term forecasts n Three obstacles to accurate surface analyses were discovered and dealt with using the 12 -km En. KF: • Poor model terrain height profile (representative check) • Lack of model surface forecast variance (variance inflation from aloft) • Model surface wind and temperature bias (pre-assimilation bias removal) n Resulting WRF 12 -km En. KF surface analyses were better than the WRF 36 -km En. KF, GFS, NAM, and RUC n Future direction: • Better bias removal techniques • Tuning of data assimilation parameters • Testing of 4 -km nested domain • Evaluation of analysis fields aloft • Short-range forecast verification • Comparison with current NWS mesoscale analysis techniques (RTMA, MOA)