Analyses & Numerical Prediction

Operations Bulletin No. 51
Upgrade to TCLAPS
12 January 2000

INTRODUCTION

A new high-resolution version of the Limited Area Prediction System (LAPS, Puri et al, 1998), specifically for tropical cyclone forecasting, known as TCLAPS. has been developed by the Regional Meteorology Group in BMRC. It was introduced into operations at the NMOC Melbourne on December 7th, 1999, before the first cyclone of the 1999-2000 Australian tropical cyclone season. TCLAPS incorporates major features of the current operational LAPS and TLAPS models, including 12 hours of data assimilation and 24 hours of dynamical nudging. Additionally the system includes sophisticated vortex specification and initialisation for high resolution prediction of intense circulations.

TCLAPS is configured at a resolution of 0.15^o  in the horizontal and 19 levels in the vertical. In addition to the use of conventional and remotely sensed data, it uses high resolution GMS cloud top temperature and bogus moisture profiling. It incorporates a relocatable 27^ox27^o domain centred upon the tropical cyclone of interest, producing forecasts out to 48 hours. As with all new versions of the LAPS model, TCLAPS is Year 2000 compliant.
 
 

OVERALL SCHEME

The TCLAPS system can be considered to consist of two model runs: a large scale environment (LSE) component and the high resolution (HR) tropical cyclone-centric component, nested in the LSE.

The LSE component is a low resolution configuration consisting of 12 hours of data assimilation and a 48 hours forecast  over the domain 55S-19N, 75E-160W (Figure 1). In the current configuration, observational data are inserted every six hours. After each data insertion, an analysis is performed followed by a mass flux adjustment, initialisation and prediction. The prediction model is integrated forward to either the next analysis time (six hours at T-12 and T-6) or a longer forecast out to 48 hours at T0. A “cold start” is done at the beginning of cycle one (T-12h). In effect, this means that the latest GASP analysis becomes the first-guess for the LSE analysis in the first cycle. After this, the first-guess fields for the analysis in the second and third cycle are provided by six hour forecasts from the LSE prediction model. GASP also provides six-hourly boundary conditions for the LSE with the pre-processor collating and converting GASP data on pressure levels to sigma levels for input into the analysis and prediction parts of the system.

The HR component is initiated by the existence of one or more tropical cyclones within the LSE domain, as indicated by a tropical cyclone advice supplied by the Darwin Regional Specialised Meteorological Centre (RSMC). This advice includes information on the location, size, intensity and past motion of the cyclone. Five high resolution analyses are performed, centred on the tropical cyclone, at six hour periods extending back 24 hours. First guess fields for these analyses are provided by the LSE with the T-24 and T-18 fields from the LSE run 12 hours earlier. Any pre-existing tropical cyclone circulation in the LSE analyses is removed and replaced with an idealised vortex at the observed location. The LSE also provides the three-hourly boundary conditions for the HR model forecasts. Again the pre-processor collates and converts LSE model data from pressure levels onto sigma levels before input into the HR model. The HR model commences with a 24-hour nudging period from T-24, using the HR analyses as target analyses, before producing forecasts out to 48 hours

Further details can be found in Davidson and Weber (2000).
 

DATA ANALYSIS

TCLAPS has two levels of data analysis; the first covering the LSE at T-12, T-6 and T0 hours, while the second covers the HR domain re-extracting and analysing data extending back 24 hours at T-24, T-18, T-12, T-6 and T0 intervals.

Analysis is performed on sigma levels and operates on a latitude-longitude grid. The analysis method used is a multi-Variate Statistical interpolation (MVSI) scheme, which makes simultaneous use of geopotential and wind observations in three dimensions. This is the same analysis scheme as used in the global system, GASP. The MVSI scheme interpolates the observed increments (ie the deviations from the first guess field) of geopotential heights, thicknesses and winds to produce increments of geopotentials and winds at the grid points. Mass and wind increments are adjusted for geostrophic consistency over latitudes beyond 15o from the equator. The moisture analysis is carried out using univariate statistical interpolation. Gross error checking and a comprehensive "cross-validation" is carried out in the analysis. Use is also made of "super-observations" - the combination of closely spaced observations.
 

OBSERVATIONAL DATA USED

The analysis uses a variety of observational data that includes: surface SYNOPs, ship and drifting buoy reports, radiosonde and rawinsonde observations, remotely sensed GTS SATEMs and GMS winds, and winds from aircraft. Note that (i)  mandatory level wind and moisture data, (ii) locally processed satellite sounding data and locally derived cloud drift winds, and (iii) synthetic GMS moisture data are used in the analysis.

Data is extracted as required before each assimilation run, which results in cut of times of 11 (T-12h), five (T-6h) and two hours at T0 hours for the LSE. For the HR analysis, data for the past 24 hours is re-extracted over the high resolution domain at the time of assimilation, about T+2 hours.
 

TROPICAL CYCLONE BOGUS

Tropical cyclone (TC) advisories, prepared by Darwin RSMC when tropical cyclones are in the LSE analysis area, are used by the HR analysis. The bogus program (i) inserts the LSE analyses for the past 24 hours into the corresponding GASP global fields; (ii) locates any circulations in the modified  global fields  near the observed TC; (iii) locally removes these TCs by careful filtering; (iv) inserts an axisymmetric vortex at the observed locations;  (iv) builds large scale wind field asymmetries, similar to ?-gyres, consistent with the past 12-hour motion of the storm, and (v) generates a set of bogus observations based on the new wind fields that are then added to the earlier extracted observation data-sets. These updated data-sets are then entered into the HR objective analyses, which uses the modified LSE inserted global fields for the past 24 hours as their first guess fields.

In the situation where only a T-12 and a T0 TC advisories exist, an interpolation scheme produces  intermediate data at T-6. This scheme is also used to produce T-12 and T-6 advisories when a cyclone is first present in the LSE.  This sets up the first-guess environment making it easier for the model to assimilate the new tropical cyclone circulation, improving the TC structure at T0.

A caution needs to be noted when a sheared TC system is bogused; currently the sheared nature of the system is not taken into account by bogus program and a “normal” circulation is entered at all levels. This may lead to forecast errors in the early part of the model run.
 

CLOUD TOP TEMPERATURE

Cloud top temperatures are extracted from hourly GMS satellite imagery averaged over 0.5 degree boxes in the region 80oE to 180o, 50oS to 48oN. This data is used in the initialisation of synthetic moisture profiles and in the dynamic nudging period of the HR model. This technique preserves the analysed vorticity and surface pressure, while reconstructing the vertical motion field to be consistent with the observed distribution of cloudiness.

MASS FLUX ADJUSTMENT

After completion of the pre-processing and analysis stages and before input into the LSE model prediction component, an adjustment to winds at the boundaries is made, to balance the mass flows. This does not occur in the HR component.
 

INITIALISATION AND PREDICTION MODEL

Initialisation during the LSE data assimilation, based on a digital filtering technique, is incorporated in the prediction model component to control the generation of spurious gravity waves. The forecast component is a hydrostatic primitive equation model formulated on sigma levels for a non-staggered ("Arakawa A") latitude-longitude grid. Higher-order numerics are a feature of the system. Detailed physical parameterisations, in line with those in GASP include: a mass-flux convective scheme (for deep, mid-level and shallow convection), large-scale rain, radiative transfer with a diurnal cycle, diagnostic clouds, stability dependent surface fluxes, and interactive soil moisture. The horizontal grid and vertical level structure of the forecast component is identical with that of the analysis component.

The HR prediction model commences 24 hours before the forecast base-time from a LSE implanted global analyses supplemented with bogus moisture data from GMS imagery and TC vortices as appropriate. The model then uses dynamic nudging as it steps towards target analyses at T-18, T-12, T-6 and T0 hours. While moving towards the target analyses, the rotational wind components are preserved, whereas the divergent winds are  replaced by model-generated divergence. This divergence is forced by an imposed convective heating function defined by the GMS cloud top temperatures (Davidson & Puri, 1992). The resulting “nudged” analysis is not identical to the target analyses but will have generated vertical motion and moisture fields that are consistent with the GMS observed tropical convection. The prediction model is then integrated out to 48 hours.
 

BOUNDARY CONDITIONS

GASP six-hourly forecasts, from T-24 hours out to T+48 hours, are used to define the necessary lateral boundary conditions for large scale environment assimilation. Absolute values of the mean sea level pressure, wind components, temperatures and mixing ratios are used at 6-hourly intervals throughout the nesting procedure. These nesting files are derived from the 1.50 latitude-longitude post-processed files from GASP (currently the T239/29L version).

For the high-resolution component, the boundary conditions are defined by the six-hourly LSE analyses implanted with the TC circulation for the first 24 hours of dynamical nudging. For the last 48 hours, three-hourly output fields from the LSE forecast are used.
 

OPERATIONAL CONFIGURATION

Table 1: Approximate heights corresponding to sigma levels in TCLAPS.
 

EXAMPLES OF REAL-TIME PERFORMANCE

Objective verification using TCLAPS has been undertaken by BMRC on many tropical cyclones in the Darwin RSMC analysis domain (40N- 40S, 70E - 180). Tables 2a and b show the tracking errors and central pressures for Tropical Cyclone Dan that passed over the Philippines in October 1999. These results are displayed in Figures 2a-d, where OBS=observed pressure (hPa), FORC=forecast pressure and TERR=track error (km).

Table 2a: Forecast track errors (km) for TC Dan

Date-Time	Fcst +0h	Fcst +12h	Fcst +24h	Fcst +36h	Fcst +48h
991003 2300	12	32	77	66	35
991004 2300	12	50	60	127	139
991005 1100	5	46	40	66	72

Table 2b: Forecast and observed central pressures (hPa) for TC Dan

Date-Time	Obs / Fcst +0h	Obs / Fcst +12h	Obs / Fcst +24h	Obs / Fcst +36h	Obs / Fcst +48h
991003 2300	985 / 987	970 / 963	960 / 979	970 / 983	960 / 981
991004 2300	960 / 962	970 / 969	960 / 974	960 / 964	962 / 959
991005 1100	970 / 972	960 / 962	962 / 965	962 / 970	960 / 974

Figure 2a: TC Dan Track 3 Oct 1999
Figure 2b: TC Dan Track 4 Oct 1999
Figure 2c: TC Dan Track 5 Oct 1999
Figure 2d: TC Dan Central Pressure 5 Oct 1999

PERFORMANCE ON HISTORICAL EVENTS

The system was developed and tested on a sample of recent tropical cyclone events in the Australian Region. Forecast track errors of recent tropical cyclones are shown in Figures 3a-d, for the official forecast, CLIPER and TCLAPS. For these events TCLAPS produced smaller track errors for all forecast periods. However the large errors for two or three forecasts shown in Fig. 3c are evidence of the fallibility of the system.
Figure 3a: Offical Forecast Track Errors (km)
Figure 3b: CLIPERl Forecast Track Errors (km)
Figure 3c: TCLAPS Forecast Track Errors (km)
Figure 3d: Mean Forecast Track Errors (km)

PRODUCT AVAILABILITY

Cyclone Tracking

An example of the graphical output from TC John is shown in Figure 7.

This cyclone tracking software, which has been developed by Noel Davidson of the BMRC Regional Meteorology Group, will only supply information for the TC on which the model has been centred. It was found through trials using TCs John and Ilsa in December 1999, that erroneos data can be generated when a weakening second system is present. Care also needs to be taken after the TC has made landfall as other weak circulations and pressure troughs are generated by the model which interfere with the track generated. However this problem is usually obvious by refering to the graphical image.

DIFACS

The relevant DIFACS slots are as follows:  

MSLP	286-289
850 hPa Wind	319-322
200 hPa Wind	336-339

Unit Conventions on DIFACS:    Isotachs and wind barbs are displayed in knots.

In this archive, fields are stored in NetCDF form, designated by 'nc', observational data in a sequential box IEEE format by 'ieeeseq', ASCII format by asc, and cray format data by 'cray'. It should be noted that the cray format will be replaced by an IEEE format with the upgrade of the NEC SX-4 to the SX-5 expected to occur early in 2000.

FUTURE DEVELOPMENTS

Current plans include incorporating new types of observational data such as METARs and locally derived hourly water vapour and high resolution visible satellite winds. Upcoming changes to the model will include the semi-Lagrangian semi-implicit (SLSI) and non-hydrostatic formulations and improvements in the parameterisation of moist processes. There will be some minor changes to model data storage and manipulation with the upgrade of the NEC SX-4 to the SX-5. It is also planned to look at the application of the cyclone tracking package to other models available through NMOC rtdb, such as TLAPS, LAPS, MESOLAPS, GASP and ECMWF.
 

REFERENCES

Davidson, N.E. and Weber, H.C.: "The BMRC High Resolution Tropical Cyclone Prediction System: TCLAPS.", Mon.Wea.Rev,. accepted for 2000

Davidson, N.E. and Puri, K.: "Tropical prediction using dynamical nudging, satellite-defined convective heat sources and cyclone bogus.", Mon.Wea.Rev,. 120., 1992

Puri, K., Dietachmayer, G., Mills, G.A., Davidson, N.E., Bowen, R.A., and Logan, L.W.: "The new BMRC Limited Area Prediction System, LAPS." Australian Meteorological Magazine Vol 47, No 3, 203-223, 1998.