Description : The principal aim of this project is to obtain detailed weather forecast information for the region of Galicia, so that it can reflect its climatological diversity (Figure 3). For that purpose, a non-hydrostatic numerical model for prediction (ARPS) with a resolution of 10 Km is executed in CESGA's VPP300E. Thus, it is possible to get a prediction for the following day taking into account different meteorological variables: temperature, wind speed, wind direction, humidity, cloudiness, rain, etc.
Summary of the research project
Due to the climatological diversity of Galicia, it is necessary to apply a non-hydrostatic meteorological model with the appropriate resolution that allows to describe this climatology in a detailed way. The numerical model chosen was the ARPS (Advanced Regional Prediction of Storms), from the University of Oklahoma, for it includes the physical parametrizations needed to describe atmospherical movements within Galicia's complex topography (mountain zones, plain land, estuaries and open water spaces in a really small area).
Besides, this model has gone through different optimization processes so that it can generate a 24-hour prediction with a really small amount of computer waste. This way, there is enough time for meteorologists to make an evaluation of the results and to write each day's prediction. A detailed description will therefore be available early in the morning for the next day's 24 hours.
The calculation domain takes in a mesh of 40' 40 points in horizontal with a resolution of 10 Km (and 25 levels vertically orientated up to 10,000 m high), which covers an area of 400' 400 Km², including not only Galicia, but also a large sea area Northeastwards, the oriental strip of Asturias and the North of Portugal.
The vast domain it covers make meteorologists take into account the influence of the phenomena ocurring at a synoptic scale and which affect directly Galician local meteorology, such as frontal systems entrance or the influence of summer temperature falls that happen in the centre of the Spanish peninsula , affecting Galicia through the East.
The prediction was executed using the following methodology:
The files are received late in the afternoon. They contain both the initial conditions and the environmental conditions needed to solve the systems of numerical weather prediction equations. A larger scale meteorological model (AVN) is started from these data for a resolution of 1º and in 16 vertical levels up to a heigh corresponding to 10mb.
Early in the night, a programme called EXTZARPS is executed, and its function is to generate the starting point for the higher resolution mesh and the environment conditions. With this, a tridimensional initial field is obtained with values of the metereological variables in each of the points of the mesh. Moreover, a new subroutine (add_cloud.f) was added to this programme, making possible the initialization of the cloudiness. This means that, in the initial condition, the existing cloudiness will be reproduced, and there is no need to wait for the ARPS model to generate it, avoiding the time gap.
In the night, the ARPS model is executed. This model is in charge to solve the prediction equation for the different meteorologic variables. In this case, it was thought appropriate to store the results every hour. Consequently, the 24 output files obtained will be processed properly to get to see the results graphically with any of the existing interfaces: Vis5d, GrADS, or even the ARPS, ARPSLT.
Thus, the information needed to get the next day's prediction is available very early in the morning. Such a local information obtained in the ARPS model is completed with the departure of other global models to make the predictions as close to reality as possible.
In Figure 1 there is an example of the superficial temperature field predicted for 4 p.m., June 1, 2000. As it can be seen, the ARPS model can clearly make out marked temperature differences between coast areas (18º C in the Southern Atlantic coastal area of Galicia, Rias Baixas), and inland areas (30º C in Ourense). It is possible to know the temperatures for any other geographical point; for example, 25º C in Santiago de Compostela. Figure 2 shows the departure model generated with Vis5d, thanks to which it is possible to watch at the same time the following items: the wind field at a certain heigh level, a vertical profile of the temperature field in the mountain areas of Southeastern Galicia. Figure 3 is an example (June 1st, 2000) of a day's work for Galicia, which is available in the web page for those who may be interested.
Finally, it is necessary to emphasize that this project is constantly developing and operating. Thus, in addition to this main goal, which consists on getting the weather forecast 36 hours beforehand, other jobs are carried out, such as assimilation of real data collected all over the 17 weather stations of the Consellería de Medioambiente (Environment Regional Management). These stations are spread all over the Galician region, allowing to make a very specific and almost immediate weather prediction. This way, it could be applied to meteorologic alert issues such as storms or forest fires.
Computing Techniques and Applications
a) Vector Supercomputer VPP300E
The computer code of the meteorologic ARPS is in FORTRAN 77. Such code was validated and optimized for its proper functioning in VPP300E. Consequently, the options of compilation of the programme were optimized firstly.
Besides being executed vectorially and applied the maximum degree optimization (-Of) to the code, different options about the inline execution of intrinsic functions (-ilfunc) were also added. On the other hand, the code was modified to optimize vectorization.
The computing of the functions f_esl, f_esi, f_desdtl, f_desdti, f_desdt, f_qvsat, f_es, f_tdewl, f_tdewi, f_tdew, f_mrsat, f_pt2pte and f_tmr, which are inside a loop, becomes quicker when they are made "inline".
The compiler could not make certain loops authomatically, though they could be united. Finally they were merged with the directive !OCL VCT (VOL).
The order of these loops which had not optimized their memory access was interchanged.
The compiler was reported with the temporal variables which are not used when the loop is exited. For that purpose, the directive used was !OCL TEMP(var), where "var" corresponds to temporal variable.
Dependencies on some loops were erased following two methods: using directive !OCL NOVREC and dividing the loop into two parts.
Having included all the improvements, the final computing time needed to complete a day's prediction is 80 minutes. Thus, the initial time is improved almost 3 hours, and the vectorization reaches 82%.
b) Massive Storage System
On the other hand, it is necessary to store daily the data about meteorological model initialization, which can be divided into two types:
- Initialization data coming from less resolution mesoscale models which cover a larger region, i.e. european model departures, AVN...
-Data about the weather stations spread all over Galicia, and which belong to the Environment Regional Management of Galicia. Daily data about radiosonde activities are also collected. These data are useful in order to adjust the initial meteorological camp to the meteorology of Galicia.
It is also necessary to store the departure of meteorological computing models (ARPS, MMS, etc.), which are executed daily in CESGA to be later visualized (i) to be used by predictors and (ii) to be at the public's disposal in the Web.
This massive storage implies the necessity to have a large storage capacity "in situ" to hold constant usage.
Estimation of the Supercomputing time so far
Approximately 1500 CPU-hours per year.
Estimation of Supercomcomputing time until the end of the work
Approximately 1500 CPU-hours per year.
|
cesga.es
:: Telf.: +34 981 569810 - Fax: 981 594616 ::
Avda. de Vigo s/n 15705, Santiago de Compostela.