assimilation to model te Amsterdam urban climate 18 July 2018, S. - - PowerPoint PPT Presentation

Using weather radar observations for data assimilation to model te Amsterdam urban climate

18 July 2018, S. Koopmans, N.E. Theeuwes, R. van Haren, G.J. Steeneveld,

• R. Ronda, R. Uijlenhoet and A.A.M. Holtslag

Motivation

§ Urbanization together with climate change illustrate need of

understanding urban hydrometeorology.

• Occurrence of extreme precipitation
• Heat stress

§ Lack of long term city observations and model simulations

§ Aim: 15 year fine scale (100 m !) urban climatology re-

analysis archive for Amsterdam (and other cities in NL)

Verification and testing: July 2014

§ Case July 2014:

• 16-20 July, heat wave
• 28 July, extreme precipitation

Dimensions WRF model

• 4 Domains with 12500:2500:500:100 m resolutions
• Use ECMWF (Reanalysis) boundaries every 6 hours, 0.5⁰ x 0.5⁰

Methodology

Detailed urban morphology data

Ronda, R.J., G.J. Steeneveld, B.G. Heusinkveld, J. Attema, B. Holtslag, 2017: Urban fine-scale forecasting reveals weather conditions with unprecedented detail, BAMS

• Inner domain subgrid turbulence closure:

(Smagorinsky first order closure 3D)

• Detailed urban morphology NUDAPT

Other WRF model settings

§ 72 vertical (eta) levels § NOAH LSM § PBL scheme YSU in largest 3 domains § Grell-Devenyi cumulus scheme in outer domain § Make use of SLUCM (Single Layer Urban Canopy Model)

Kusaka, 2001

§ Data-assimilation techniques of various sources of observations.

• Step 1: Include WMO stations, (pressure, wind, T2m and TD)
• Step 2: Include radar observations
• Step 3: Include urban stations from weather hobbyists

§ With every step of data assimilation the analysis is updated

Data assimilation (3DVAR)

Analysis

Observations (with predefined errors) (T, TD, P, radar) Data Assimilation (WRFDA) Background error model

(hr) 0

12 12 Run 1 Run 2

Data assimilation cycles (3DVAR)

WRF Forecast

Observations (T, TD, P, radar)

Data Assimilation (WRFDA)

Background error model

ECMWF boundaries

Update boundaries

2 hr 6 hr Radar and wmo stations update SST and soil moisture every 6 hour

Radar data assimilation

Use volumetric radar reflectivity and radial wind velocity from De Bilt radar Interpolation to model resolution. (exclude lowest elevation angle 0.3º)

• Indirect data-assimilation. Assimilate rainwater mixing ratio derived from

volumetric radar data. (Wang et al, 2013)

Reflectivity 28 July 14 UTC Radial velocity error 28 July 14 UTC

Error variance reflectivity 2 DBZ

Urban data assimilation

Weather Underground About 300 stations in NL Direct transfer to wunderground website Various time intervals, instruments, siting...

Urban data assimilation

Bell et al, 2015

Urban data assimilation

§ SLUCM does not interact with WRFDA § Single urban stations are not representative for neighborhood

and city scale due to local variability

Urban data assimilation

§ Urban temperatures are nudged by applying

correction on urban fabric (walls and roads). Urban fabric has storage to preserve the effect of nudging between 2 cycles.

§ 𝑫𝒑

𝑫𝒑𝒔𝒔𝒇𝒅 𝒇𝒅𝒖𝒋𝒑 𝒖𝒋𝒑𝒐 𝒑𝒈 𝒑𝒈 𝑼𝟑𝒏=𝜷∗𝑴↓+𝜸∗𝑽+𝜹

𝑴↓=𝒆𝒑𝒙𝒐𝒙

𝒐𝒙𝒃𝒔𝒆 𝒎𝒑𝒐𝒉 𝒐𝒉𝒙𝒃𝒘𝒇 𝒔𝒃 𝒔𝒃𝒆𝒋 𝒆𝒋𝒃𝒖𝒋𝒑𝒐 𝒑𝒐 (𝑿​ 𝒏↑ 𝒏↑−𝟑 )

𝑽=𝒙𝒋𝒐𝒆

𝒐𝒆 𝒕𝒒 𝒕𝒒𝒇𝒇𝒆 (𝒏​𝒕↑ 𝒕↑−𝟐 )

Example of outer wall temperature correction in the evening. (ºC)

§ Correction T2m is divided over wall and

road layers. Division scaled to amplitude

• material. Large amplitude-> larger

adjustment.

Urban results

WMO+radar RMSE = 2.07 ºC Bias = -1.36 ºC WMO+radar+urban: RMSE= 1.69 ºC Bias= -0.71 ºC

• No data assimilation shows cold bias WRF for cities

NL domain

Results

Urban results

Amsterdam domain

WMO+radar RMSE = 1.86 ºC Bias = -0.84 ºC WMO+radar+urban: RMSE= 1.25 ºC Bias= -0.18 ºC

Case 28 July 2014

• Weak flow in atmosphere with a convergence zone over the

Netherlands

Results

Case 28 July 2014

WMO Observations Radar + urban Noda (no data Assimilation)

Wind 28 July 2014

WMO Radar + urban Noda

Fractional skill score

Roberts and Lean, 2008

Radar + urban

MSE = 0.11

WMO

MSE= 0.23

𝐺𝑇𝑇=1−​𝑁𝑇𝐹/𝑁𝑇​𝐹↓𝑠𝑓𝑔 𝑁𝑇𝐹=​1/​𝑂↓𝑦 ​𝑂↓𝑧 ∑𝑗=1↑​𝑂↓𝑦 ▒∑𝑘=1↑​𝑂↓𝑧 ▒​[​𝑃↓𝑗,𝑘 −​𝑁↓𝑗,𝑘 ]↑2

Error variance

Error variance Error variance

Obs

General verification

Precipitation

• Verification on subset of wmo stations 1

and 2 hour after data assimilation

Conclusions

§ WMO data assimilation improves forecast substantially

in various meteorological metrics.

§ Radar data assimilation is challenging, addition

slightly/moderately better in predicting location precipitation

§ Nudging the urban temperatures with citizen weather

stations reduces the cold biases present in WRF