TRANSNA-M04.indd 151
TRANSNA-M04.indd 151 4/27/2013 1:56:03 PM4/27/2013 1:56:03 PM
This page intentionally left blank This page intentionally left blank
153
Meteorological Aspects and Weather Condition Navigational Problems – Marine Navigation and Safety of Sea Transportation – Weintrit (ed.)
1 INTRODUCTION
The numerical weather prediction (NWP) is an advanced software application that utilizes mathematical models of the atmosphere and the hydrosphere for preparing weather predictions based on current weather data. Due to rapid development of fields of software data processing, remote sensing and parallel computing techniques, numerical modelling became a major tool for studying atmosphere dynamics and is utilized in various industrial applications, such as inland and marine navigation (Krata 2012), sustainable energy forecasting and others (Weintrit 2012).
Numerical models are based on the physical laws that govern the temporal evolution of the flow and express the mass, momentum and energy of the fluid. These computations are performed by algorithms that use independent variables (input), i.e. initial conditions needed for model initialization, along with many parameters that define the physical and numerical conditions. The models’ dependent variables (output) consist of the temporal sequence of meteorological fields created during the integration process. They also include many derived quantities that one may wish to compute from that sequence.
In this context, one of most challenging tasks for operational NWP systems is optimal and efficient observational data assimilation. In the paper, authors present method of enhancing one of the NWP models, namely Weather Research and Forecasting
Environmental Modeling System (WRF EMS), which is based on utilizing satellite meteorological data acquired from Department of Geoinformatics’
imagery data receiver ground station. Proposed improvement is based on real-time updates of initialization and boundary conditions datasets of the NWP system. This paper presents the advantages of proposed solution.
2 NUMERICAL WEATHER PREDICTIONS 2.1 Basic equations
During numerical weather simulations the values of all the meteorological parameters are determined by the initialization data, which describes current weather conditions over the given area. The state of the atmosphere, aforementioned values and their mutual interactions are all reduced to the form of mathematical equations, which are solved using numerical methods. This enables the prediction of future atmospheric states in addition to time and location of various weather phenomenon (e.g.
rainfall, snowfall, storm, fog, tornado, dew).
Majority of these equations are based on the momentum equation (1), the energy conservation (thermodynamic) equation (2), the mass conservation (continuity) equation (3), the water vapour conservation equation (4) and the equation of state (5).
Operational Enhancement of Numerical Weather Prediction with Data from Real-time Satellite Images
à. Markiewicz, A. Chybicki, K. Drypczewski, K. Bruniecki & J. Dąbrowski Gdansk University of Technology, Gdansk, Poland
ABSTRACT: Numerical weather prediction (NWP) is a rapidly expanding field of science, which is related to meteorology, remote sensing and computer science. Authors present methods of enhancing WRF EMS (Weather Research and Forecast Environmental Modeling System) weather prediction system using data from satellites equipped with AMSU sensor (Advanced Microwave Sounding Unit). The data is acquired with Department of Geoinformatics’ ground receiver station (1.5 metre HRPT/MetOp). Aforementioned improvement is based on real-time updates of initialization and boundary conditions dataset of the NWP system. Conclusions and advantages of proposed solution are presented in the paper.
TRANSNA-M04.indd 153
TRANSNA-M04.indd 153 4/27/2013 1:56:03 PM4/27/2013 1:56:03 PM
154
3
3 3 3
2 1 ,
dV V p F
dt : u ) U (1)
,
p p
dT R T dp Q
dt C p dtC (2)
3 3,
d V
dt
U U x (3)
dq ,
dt M (4)
,
p URT (5)
where: V3= wind 3-dimensional velocity vector; ȍ
= Earth angular velocity vector; ȡ = air density; = scalar function gradient of 3-dimensional vector; p = air pressure; ) Earth gravitational field potential;
F = torque force; R = gas constant; Cp = heat capacity (at constant pressure); T = absolute temperature; Q = heat in unit of mass and time; x
= divergence operator of vector field; q = relative humidity; M = water vapour mass.
Additionally, several simplifications can be made taking into account the order of magnitude of various terms to be considered. For instance, if it is assumed that the horizontal speeds are far greater than the vertical and the area of the atmospheric changes’ horizontal scale exceeds 10km the momentum and thermodynamic equations can be reduced to (6) and (7).
dV RT ,
fk V p
dt u p (6)
,
p
dT R T dp
dt C p dt (7)
where: V = horizontal wind velocity; f = Coriolis parameter; k = unit vector (vertical); p = pressure gradient.
The scale requirement makes it impossible to use these expressions to simulate convection - it allows inertia-gravity waves, but not sound waves because of the filtering effect. Nonetheless, the reduced equations are the foundation of most models used for numerical weather prediction.
Moreover, this set of equations can be simplified by considering the atmosphere as a fluid of limited depth in which density and the vertical distribution of horizontal velocity are constant (Jean Coiffier 2011). Owing to the second condition that the wind is also constant in the vertical, the vertical influx of the air masses is no longer existent in a model.
When taking aforementioned assumptions into account the atmospheric state on a given area is described by the horizontal wind vector and the surface height (where the pressure is zero).
Furthermore the pressure can be identified as the
geopotential gradient of the atmosphere’s free surface. This allows obtaining continuity equation from the conservation of mass for a fluid column of constant density, thickness and base area.
Altogether, these equations and presumptions enables the description of change in velocity of a fluid column along with the change in the geopotential the free surface. It is known as the Saint-Venant system, which was originally created for the river water motion research. It is widely used in NWP systems because it permits the easy assessment of the numerical methods’ attributes before employing them.
In many weather prediction models the vertical pressure coordinate, which is strictly surface dependent, is used. It enables the simplification of the continuity equation and the whole model, by describing the atmosphere as a group of layers separated by surfaces of the same pressure. To solve the problem of the lower layers, which intersect the higher areas of the surface (i.e. hills, mountains), the sigma coordinate was proposed (9).
s t,
P S S (8)
( ) t, V S S
P
(9)
where: ʌs = pressure at surface level; ʌ = pressure at given height; ʌt = pressure at the top of the modelled layer; ı = vertical coordinate.
It is also known as the normalized pressure coordinate, because it is equal to 0 at the upper boundary of the domain and equals 1 at the bottom (where the pressure is zero). When using sigma coordinate, the atmosphere can be represented as a group of layers of the equal sigma value, which embrace the surface at its higher areas, no longer intersecting them.
2.2 Numerical weather prediction suite
An atmospheric numerical prediction model is the main component in a much larger system (numerical prediction suite), which comprises of many various implemented processes that make operational weather forecasts possible. Among these processes such elements as the acquisition of meteorological data, the objective analysis, forecasting based on one of the atmospheric models, the determination of model run parameters, and finally, results dissemination and visualization can be distinguished.
Though the quality of weather forecast obtained in a purely deterministic way is mainly related to the accuracy with which the initial state is defined, major effort has been made to make the best of all available assimilated observations. Among many observation techniques two groups, namely in-situ
TRANSNA-M04.indd 154
TRANSNA-M04.indd 154 4/27/2013 1:56:03 PM4/27/2013 1:56:03 PM
155 observations and remote sensing observations, provide basic input for NWP systems.
In-situ observations network comprises of the network of not uniformly distributed fixed-location stations that measure the main weather parameters at given reference time coordinated in Universal Time Clock (UTC). Most of the network stations make measurements every hour, however, some of them (mainly upper air weather stations) make vertical soundings every 12 to 24 hours. This observation network is supplemented by a set of measurements made at fixed time by ships at sea, buoys and on board airliners.
The measurements of the Earth global parameters are made using remote sensing techniques. Apart from images of Earth and cloud cover provided by meteorological satellites, radiation measurements over a wide range of spectral channels are also available. Basically, meteorological satellites can be divided into two groups: sun-synchronous polar orbiting satellites travelling in a low orbit (about 800km over ground) providing relatively high- resolution images in daily routines, and geostationary satellites travelling in equatorial orbit remaining relatively motionless to the Earth. Unlike polar-orbiting satellite, geostationary satellites provide measurements for the same zone of the Earth at high frequencies, i.e. each 15 minutes for MeteoSat.
2.3 Boundary conditions and initialization dataset Due to the fact that forecasts for ranges not exceeding 48 hours (over a limited geographical areas) may be functional on a relatively small domain, compared with the sphere, so called Limited Area Models (LAM) have become significant tools in many weather services. When integrating a limited area model, it is necessary to prescribe the values of the fields on the boundary of the working domain by interpolating in space and time the forecasts from another model operating over a larger domain at lower resolution. Although LAM models omit important atmospheric processes such as the effects of gravity waves, experience shows that the gain from using a greater horizontal resolution exceeds the degradation from perturbations introduced on the lateral boundaries of the LAM model domain.
Since WRF EMS model is running as a LAM model, lateral and boundary conditions datasets need to be imported from larger domain running model.
In this context Global Forecast System (GFS), containing a global computer model and variational analysis, administrated by National Oceanic and Atmospheric Administration (NOAA) is utilized.
The resolution of GFS model varies in each part of the model: horizontally, it divides the surface of the earth into 35 or 70 kilometre grid squares; vertically,
it divides the atmosphere into 64 layers and temporally, it produces a forecast for every 3rd hour for the first 192 hours. The GFS model output is freely available in the public domain for variety of national and commercial weather services.
GFS model data is served as files in General Regularly-distributed Information in Binary form (GRIB, also known as GRIdded Binary) format which is a general purpose, bit-oriented data exchange format, that is well suited for transmitting large volumes of data. It is widely used in meteorological and climate research for storing weather (and forecast) information. GRIB format was approved by World Meteorological Organization (WMO) (Guide To GRIB). There were three versions of the file format: versions 0, 1 and 2.
Currently version 0 is deprecated and version 2 is most commonly used in meteorological or climate projects. GRIB data is also self-describing, meaning that the information needed for file reading is present within the file. GRIB file contains collection of messages (term record is also used), each of them holds the gridded data for one parameter at a given time and at only one level (BADC GRIB Documentation). Messages may contain sub- messages which allows creating hierarchical and ordered structures.
2.4 Data assimilation
In order to take advantage of the various information, acquired during data acquisition process, the quality of obtained information needs to be represented mathematically. Variational data assimilation is a technique that measures quality of the data basing on error statistics obtained by combination of repeating analysis and measurements. In this context, data assimilation is a relatively complex process and the effective implementation has been in the interest of considerable scientific investment and requires very substantial computing capabilities. Basically, data assimilations schemes are based on modifying an earlier forecast as background field at grid points using a weighted sum of differences between background field and measurements at fixed locations or statistical approaches.
Among many approaches for data assimilations, techniques of successive correction method and statistical interpolation by the least squares approach made the most significant improvement in this area of research over past decades.
Enhancement of aforementioned methods relies on the least squares estimation (sometimes referred as Best Linear Unbiased Estimator - BLUE) of the analysed state that aims to minimize the cost function considered as variational problem (Gandin 1963, Lorenc 1981, Daley 1991).
TRANSNA-M04.indd 155
TRANSNA-M04.indd 155 4/27/2013 1:56:04 PM4/27/2013 1:56:04 PM
156 In 1985 Lewis and Derber proposed the new solution by seeking to minimize an objective criterion defined as a function of the problem, later called 3D variational approach (3DVar).
Minimalization relates to a quadratic function quantifying the deviations from the available information combined from observation and the background (previous analysis) weighted by their respective standard deviations. Enhancement of 3DVar approach was presented in four-dimensional assimilation scheme (4DVar), in which the objective function to be minimized, quantifies the distance between the models trajectory and measurements within the assimilation time interval.
3 WRF EMS NUMERICAL WEATHER PREDICTION SYSTEM
The Weather Research and Forecast Environmental Modeling System is a complete, numerical weather prediction package that incorporates dynamical cores from both the National Center for Atmospheric Research (NCAR) Advanced Research WRF (ARW) (Klemp et al. 2007) and the National Center for Environmental Predictions (NCEP) non-hydrostatic mesoscale model (NMM) (Janjic et al. 2001) as one integrated forecast system. All the capabilities of the NCEP and NCAR WRF packages, the installation, configuration, and execution of the cores have been greatly simplified in order to encourage to use of them throughout the operational forecasting by universities, private companies or research communities (Michalakes et al. 2004).
Nearly every element of an operational NWP system has been integrated into the WRF EMS, including the acquisition and processing of initialization data, model execution, output data processing, file migration and archiving. Real-time forecasting operations are enhanced through the use of an automated process that integrates various fail- over options, the synchronous post processing and distribution of forecast files.
The system includes pre-compiled binaries optimized for 32- and 64-bit Linux systems running in shared or distributed memory environments. The MPICH2 executables are also included for running on local clusters across multiple workstations. The WRF EMS is designed to give the users flexibility in configuring and running NWP simulations, whether it is for local “offline” research or real-time forecasting purposes. It also allows for the acquisition of multiple initialization data sets via Network File System (NFS), FTP and HTTP. The post processing software (WRF-post) supports wide variety of display software including Advanced Weather Interactive Processing System (AWIPS), BUFSKIT, NCAR Command Language (NCL), Grid Analysis and Display System (GrADS),
GEMPAK, NCEP Advanced Weather Interactive Processing System (NAWIPS) and Network Common Data Form (netCDF). The WRF-post can process forecast fields on 81 different pressure levels from 10 to 1025mb.
3.1 WRF EMS data processing diagram
Figure 1. WRF EMS system architecture.
WRF EMS system architecture is modular and consists of many software components (Fig. 1).
They were created in joint cooperation by various scientific institutes, which specialize in atmospheric and weather research (e.g. NOAA, NCEP, NCAR).
These segments, by sharing data at various stages of the simulation process, are capable of creating a complete weather prediction using given parameters (temporal, atmospheric, etc.) for a predetermined area.
Figure 2. WPS module.
Initial processing of the input data is made in the WRF Preprocessing System (WPS) module and is divided into 5 phases (Fig. 2). Predefined domain (configured prior to the system run) is copied into the system in a form of static data. It determines the prediction process’ area and its geographical projection. This domain is used to create GEOGRID files. Initialization data is downloaded (in compressed GRIB format) from the worldwide meteorological services (i.e. GFS, North America Mesoscale Model - NAM etc.) and uncompressed by UNGRIB module. The whole process is completed when uncompressed data is copied into METGRID file. It stores all the needed weather parameters.
Finally METGRID file is sent to one of the cores (set of algorithms) that handles the subsequent steps of NWP process.
TRANSNA-M04.indd 156
TRANSNA-M04.indd 156 4/27/2013 1:56:04 PM4/27/2013 1:56:04 PM
157 3.2 WRF EMS configuration and data
dissemination
There are number of variables that determine model initialization and model computing process, including parameters of physical modelling, processing grid parameters, input and output data format and processing environment (processing cluster configuration). The instance of WRF EMS system that operates in the Department of Geoinformatics utilizes multi-node Ubuntu cluster based on 4 multicore workstations running in MPICH architecture. The model generates short- term 48-hour forecasts 4 times a day in 4-km spatial grid.
Figure 3. Spatial domain of the used WRF EMS system instance.
Spatial domain covers the territory of Poland as shown in Figure 3. The results of predictions are presented via Web portal as weather maps, diagrams, icons, descriptions and are available at http://www.weathersense.pl (Chybicki et al. 2011).
4 SATELLITE DATA ASSIMILATION
There is a number of Earth observing meteorological satellites that make their data available by direct broadcast service. Direct broadcast approach demands that the end-user is equipped with satellite ground station capable of receiving signal from satellite transmitter. In this research project authors utilized a ground station (1.5 metre HRPT/MetOp), which is located on the roof of the Electronics, Telecommunication and Informatics faculty building (Fig. 4).
The received signal consists of the High Resolution Picture Transmission (HRPT) stream of digital data where the real time Earth observations are multiplexed.
4.1 HRPT-MetOp groundstation
The Gdansk University of Technology (GUT) is operating 1.5 meter HRPT/MetOp satellite
groundstation from the year 2009. The groundstation is capable of obtaining data from the Advanced Very High Resolution Radiometer (AVHRR) which is a major sensor on board of NOAA-* and MetOp-A/B satellites.
Figure 4. Department’s HRPT-MetOp ground station.
The Advanced Microwave Sounding Unit – A1 (AMSU-A1) sensor is deployed on board the NOAA-* satellites as well as on the European MetOp-A, -B and -C.
The data from the AMSU is not natively supported by the station software, however authors of this paper have developed AMSU data acquisition software module.
The main advantages of using the data from the ground station are:
the negligible time to delivery for the end-user (almost immediately after the reception from HRPT),
accessibility to the Internet and other third-party service providers.
For the future research it is essential that the HRPT/MetOp-A ground station is adapted and upgraded into the National Polar-orbiting Operational Environmental Satellite System Preparatory Project (NPP/NPOESS) compliance mode as well as for receiving the data from the Chinese FY2 meteorological satellite system, which is equipped with sensors essential in meteorology and weather forecasting.
4.2 HRPT stream/AMSU data
The geometry of the AMSU sensor is of the whisk broom type and is presented in the Figure 5. Two versions of the sensor are available: a 15-channel (AMSU-A version) or 5-channel (AMSU-B) microwave radiometer.
These sensors are used for measuring global temperature and humidity profiles (T/Q profiles).
AMSU sensors are also capable of verifying water existence in all forms (i.e. snow/ice coverage, rains and vapour) in the atmospheric layers by microwave radiation analysis (NOAA AMSU-A Guide).
TRANSNA-M04.indd 157
TRANSNA-M04.indd 157 4/27/2013 1:56:04 PM4/27/2013 1:56:04 PM
158
Figure 5. AMSU sensor geometry.
Cloud and ozone properties can be also determined by mutually using AMSU and other sensors. This kind of data is extensively used in numerical weather prediction and climate studies.
Figure 6. NOAA HRPT system diagram.
AMSU data is a part of the HRPT data stream. It is relayed to the end-users using split-phased S-band transmission with time-division multiplexing, with rate of about 660 kbps (NOAA AMSU-B Guide).
Transmission power is 6.35W (38.03 dBm) and it occurs on three frequencies: 1698/1707 MHz (primary), 1702.5 MHz (secondary). Data resolution is 1.1km, which is sufficient for short- and long-term weather prediction (NOAA KLM User Guide).
The HRPT data stream is divided into major frames, each of them incorporates 3 minor frames.
Data from the AMSU sensors is located in the third minor frame. Minor frame’s data is separated using 10-bit words, as shown in Table 1.
Table 1. AMSU frame data description.
__________________________________________________
Data segment Number of bits
__________________________________________________
Frame sync 60 ID 20 Time Code 40 Telemetry 100 Calibration target view 300 Space data 500 Synchronization data 10 Data words 5200 Spare words 1270 Earth data 102400 Auxiliary sync 1000
__________________________________________________
Each of the 520 data words consists of module identification and device flags (temperature values of mixer, amplifier, reflector and board, local oscillator monitor current, bridge voltage, processor and pixel data verification). Last two bits are used for parity and inverted bit checks.
HRPT stream contains many channels with data and headers from various satellite sensors. Earth data, the most significant data segment, stores all the samples from each channel (sample 1 – channel 1, sample 1 – channel 2 etc.).
ID’s first word (second word is empty) determines minor frame number and the source satellite. It also includes the most significant bit (MSB) and least significant bit (LSB) for frame stability check, outcome of which in also located in ID.
Next segment describes time code, which identifies day and millisecond count. Telemetry and calibration target view are used for sensor ramp calibration. All of the frame/auxiliary synchronization and spare words have predefined bit values.
Spare words consist of three sets of data from every major frame; each has information about three minor frames. First minor frame includes five minor TIROS information processor (TIP) frames, second has five frames of spare data and third has five frames of AMSU data from AMSU information processor (AIP).
AMSU digital format synchronization is divided into 8-second pulses. Synchronization data stores information about sensor synchronization and MSB/LSB bits. Not only Earth data is collected - space data segment stores values of the space scan in channels 1 to 5.
AMSU telemetry contains all the radiometric data acquired in each sensor scan, along with the information about device self-adjustments done on orbit (i.e. reflector and scene positions, sensor state, sensor power and result of the cold/worm calibrations).
TRANSNA-M04.indd 158
TRANSNA-M04.indd 158 4/27/2013 1:56:04 PM4/27/2013 1:56:04 PM