Development of Digital Filter Initialization for WRF and its implementation at IUM Xiang-Yu Huang1, Min Chen2, Wei Wang1, Ju-Won Kim3, Bill Skamarock1, Tom Henderson1 1
National Center for Atmospheric Research, P. O. Box 3000, Boulder, CO 80307 2 Institute of Urban Meteorology, Beijing, China 3 Korea Meteorological Administration, Seoul, South Korea
1. A short introduction to DFI Imbalance between the wind and mass in the initial field can be introduced, e.g., by objective analyses like 3D-Var or by simple interpolations. The noise could cause spurious precipitation, lead to numerical instabilities, degrade the forecast, and damage the subsequent data assimilation through noisy first-guesses. The digital filter initialization (DFI) is one of the methods to remove the imbalance (Lynch and Huang, 1992). It has also been shown to construct consistent fields which are not analyzed or do not exist initially, e.g., cloud water content and vertical velocity; and to reduce the spin-up problem (Huang and Lynch, 1993; Huang and Sundqvist, 1994; Chen and Huang, 2006).
A 4-day period, 12 UTC 25 July – 12 UTC 29 July 2006, is chosen for assessing the impact of DFI on the data assimilation and forecasts. The period is characterized by a series of vigorous convection in the area of Beijing and North China. Two parallel data assimilation experiments with and without digital filtering initialization, i.e. DFI and NOI, have been carried. In the DFI experiment, the Dolph filter is selected (Lynch, 1997), the cutoff parameter is 1h (this should filter out oscillations with frequency shorter than 1/1h) and the filter span is 2h.
2. Configuration of the cycling system of IUM A WRF-based rapid updated cycling system with 9km resolution and 3-hour time interval has been established in the Institute of Urban Meteorology, China Meteorological Administration. The model domain and model terrain are shown in Fig.1. The model top is at 50hPa and its 184×142×38 grid points use 9 km horizontal grid spacing. The time step used for the integrations is 60s. The physics package includes WSM6 microphysics, LSM surface scheme, YSU PBL scheme, and the old Kain-Fritch cumulus parameterization. The lateral boundaries come from the 48-h forecasts of a corser 27km domain using one-way nesting technique. The analysis scheme is the WRF 3D-Var. The background for WRF 3D-Var is the 3-h forecast from the previous cycle. In addition to the GTS data that include SYNOP, METAR, TEMP, SHIP, BUOY and PILOT, the local AWS and ground-based GPS precipitable water in Beijing area are assimilated by the WRF 3D-Var.
* Corresponding author: Dr. Xiang-Yu Huang, NCAR/MMM, P.O. Box 3000, Boulder, CO 80307, USA. Email:
[email protected]
Fig. 1. Model domain and model terrain used in all experiments. 3. Results 3.1 Noise control The main motivation of using DFI is to filter out the noise. The noise level during the time integration of a numerical model can be measured by the domain averaged surface pressure tendency, N1. For the WRF model, the dry surface pressure (MU) is used as a prognostic variable and the domain averaged MU tendency, N2, is therefore a better measure of the noise. In Fig. 2, we compare the noise level of forecasts from the parallel experiments.
The initial noise level in NOI is very high, indicating that significant noise can be introduced into the forecast during the analysis procedure, even without the horizontal interpolation between domains with different resolution (Chen and Huang, 2006). Evidently, DFI has efficiently removed the noise. Without initialization, the noise level decreases in NOI runs due to other filtering mechanism of WRF model and its lateral boundary formulation. For a 3-h cycling configuration like the one used by NOI, the average noise level in the background fields (3 h forecast) is still over 10hPa/3h, which is high and may have detrimental effect on the following analysis. The difference between N1 and N2 are due to the non-hydrostatic effect and model physics. The
difference shows clearly only on the first a few time steps. Fig. 3 and Fig. 4 display the initial surface pressure tendency (dpsdt) and the initial dry surface pressure tendency (dmudt) maps of the first cycle initiated from 1200UTC 25 July 2006, which can be regarded as another way to study the initial noise level. From the figures, we can find terrain related noise patterns (in NOI) and analysis increment related noise patterns. After digital filtering, most of the noise is removed. However, both of the initial (dry) surface pressure tendency maps of DFI, the noise patterns can still be identified relating to the narrow north-south terrain to the west side of Beijing area.
Fig 2 The N1 (left) and N2 (right), which are average from the 32 cycling runs during the period from 12 UTC 25 July – 12 UTC 29 July 2006 for both NOI and DFI experiments, are shown as functions of forecast time.
Fig 3 Sea level pressure (contoured) in hPa and the initial surface pressure tendency (dpsdt, shaded) in hPa/3h of the first cycle for NOI (left) and DFI (right).
Fig. 4. Sea level pressure (contoured) in hPa and the initial dry surface pressure tendency (dmudt, shaded) in hPa/3h of the first cycle for NOI (left) and DFI (right). 3.2 Conventional observation verification The impact of DFI on forecast is assessed by conventional observation verification (Fig. 5 and Fig.6). In contrast to the previous results of Lynch and Huang (1992), Huang and Yang (2002), Chen and Huang (2006), DFI does lead to improved model performances for nearly all of the sounding and some of SYNOP variables. Comparing the average numbers of the assimilated observations for both NOI and DFI experiments, we find that the DFI experiment can ingest more data than NOI (Table 1). For TEMP
wind observations, each DFI cycle can assimilate around 3%~5% more data than that of NOI. In addition, for the TEMP and PILOT observations, DFI has smaller O-B RMS than that of NOI (Table 2), i.e. the 3-hr forecasts of previous cycle generated from DFI experiments are closer to observations than the forecasts from NOI. This may account for the reason that DFI lead to a better forecast performance. It should be pointed out that it is only in this 3-h RUC system, we see the forecast improvement. In nearly all of other cycling systems configured with 6-h or 12-h updated interval, we have not seen much improvements due to DFI.
Fig 5 RMS difference between model forecasts and radiosonde observations for NOI and DFI.
Fig 6 RMS difference between model forecasts and surface observations for NOI and DFI.
Table 1 The average numbers of the assimilated observations for NOI and DFI experiments. DATA TYPES
SOUNDING T U V
Q
SYNOP T U
V
Q
P
PILOT U V
METAR U V
T
NOI DFI
404
173
168
310
444
389
357
444
424
113
113
425
389
466
407
182
174
311
444
388
357
444
427
116
114
425
391
466
Table 2 The average RMS of the observation and analysis background (the 3-hr forecast of previous cycle) in the procedure of WRF3DVAR DATA TYPES
SOUNDING T U V
Q
SYNOP T U
V
Q
P
PILOT U V
METAR U V
T
NOI DFI
1.934
5.242
5.774
1.531
1.895
2.137
2.589
1.493
1.861
4.705
5.502
1.983
2.135
2.543
1.863
4.195
5.453
1.478
2.050
2.143
2.639
1.546
1.638
4.550
5.107
2.126
2.093
2.623
3.3 Precipitation verification
3. Conclusions
Precipitation skill scores are calculated and averaged over 4 days and 121 AWS observation points over Beijing area. The scores are defined as following: Hit (H): event forecast to occur AND did occur Miss (M): event forecast not to occur BUT did occur False_alarm (F): event forecast to occur BUT did not Threat Score = H / ( H + M + F ) DFI has a positive impact on the precipitation forecast for almost all of the thresholds (Fig 7), which is consistent to the results of Chen and Huang (2006).
An implementation of DFI for WRFV2.2 has been made successfully. Tests have been conducted over a 4-day period. The results indicate a satisfying noise reduction, a reasonable spin-up reduction, improved conventional observation verification scores, and improved precipitation scores. Acknowledgments This work was supported by KMA, South Korea and the NSFC Project (No. 40505020) of P.R.C. References
0.6
Threat Score
DFI 00-06h NOI 00-06h DFI 00-12h NOI 00-12h DFI 00-24h NOI 00-24h
0.4
0.2
0 0.1
1
5 Threshold (mm)
10
25
Fig 7 Threat scores for 0-6h, 0-12h and 0-24 h accumulated precipitation as functions of precipitation thresholds.
50
Chen, M. and Huang, X.-Y. 2006. Digital Filter Initialization for MM5. Mon. Wea. Rev. 134, 1222-1236. Huang, X.-Y. and P. Lynch, 1993: Diabatic digital filter initialization: Application to the HIRLAM model. Mon. Wea. Rev. 121, 589-603. Huang, X.-Y. and H. Sundqvist, 1993: Initialization of cloud water content and cloud cover for numerical prediction models. Mon. Wea. Rev. 121, 2719-2726. Huang, X.-Y. and Yang, X. 2002. A new implementation of digital filtering initialization schemes for HIRLAM. HIRLAM Technical Report No. 53, 36 pp. Lynch, P. 1997: The Dolph-Chebyshev Window: A Simple Optimal Filter. Mon. Wea. Rev. 125, 655660. Lynch. P. and X.-Y. Huang, 1992: Initialization of the HIRLAM model using a Digital Filter. Mon. Wea. Rev. 120, 1019-1034.
