Short abstract on the theme "Automatic system of monitoring and predicting meteorological parameters of atmosphere in real time using simulating tools"
Made by Povzlo Sergey

Weather forecast is a scientifically proved assumption of forthcoming changes of the weather, made on the basis of the analysis of development of large-scale atmospheric processes.

Weather forecasts are classified as short-term (from several hours up to 1-2 days), long-term of small lead-time (3-10 days), long-term of big lead-time (up to a month and more).

Forecasts' preparation using computer models begins with the description of a condition of the atmosphere based on the previous and current observations, in the form of a process called data mastering. For generalization and extrapolation of information, which was extracted from previous observations, numerical weather prediction (NWP) model is usually used.

Data mastering is very effective with insufficiency of the information from the various sources used with the purpose of creation a logically coordinated estimation of a condition of the atmosphere. However, similarly to the forecast, data mastering is based on NWP model and cannot directly use observations of such scales and processes which are not presented in the models. Forecasts, with a lead time more than several hours, are almost always based completely on NWP.

Only weather systems which in exceed the grid step can be predicted accurately therefore the phenomena of smaller scales should be represented approximately using statistical and other methods.

These limitations in NWP models especially affect on detailed forecasts of local elements of weather, such, as overcast and a fog, and also the extreme phenomena, such, as intensive precipitation and the maximal impulses of a wind.

Now for weather forecasting use following basic NWP models: ADAS, ETA, Aviation, Ensemble, MM5, MRF/GFS, NGM, Meso-ETA. Model NGM (Nested Grid Model) is One of the most commonly used model for short-term forecasts (by "short term" I mean less than 48 hours into the future) is the NGM. This model generally does a very good job at describing the weather over the next two days, but especially within the next 24 hours. The numerical output from the NGM breaks down each parameter into 3 or 6 hour segments, so that the forecaster can capture trends in the weather and and have more detail for different times of the day. The output from the NGM is released twice a day. An NWP model ETA is another excellent forecast model for the shorter term, extending out to 48-84 hours. The ETA numerical output provides the same amount of detail as the NGM and is released four times a day. Numerical output is released only twice a day. An NWP model GFS (The Global Forecast System ) provides another excellent look at the short-term conditions. It is especially useful in the 48-72 hour range, where the NGM leaves off. The GFSX (and Extended version of the GFS) provides the same basic information as the GFS out to 72 hours and then continues with more limited, though very valuable, information out to two weeks into the future but quality decreases significantly past about 7 days.

The specified models are large-scale, therefore they can predict only elementary cell area averaged values of prediction models. Using these values it is possible to characterize the "basic" status of weather, or its "background".

But in an atmosphere also there are processes smaller, than an elementary cell, scales which are not considered in global model. Therefore to predict these processes and corresponding weather at a level of city or area are developed special local prediction models.

Another way of creation an effective weather forecast is use neural networks for forecasting. The idea of use of neural networks in forecasting has appeared and has began to use at last 10 years of XX century.

They are used for short-term and intermediate-term forecast. Standard procedure of use of a neural network consists in "training" a network using a large number of available data. During training, using the block of entrance data, the network determines dependence on which entrance data are based, and in the further for forecasting the network will use the received dependence. Advantage of neural networks is the opportunity of their further dynamic training during reception of new data. It is considered, that the neural network in comparison with other methods most precisely determines dependences of behavior of data.

Sources:

1. Отчет 54 сессии совета ВМО (Всемирной метеорологической организации). Дополнение V.

2. Белов П.Н., Борисенков Е.П., Панин Б.Д. Численные методы прогноза погоды – Л.: Гидрометеоиздат, 1989. – 375с

3. Neural Network Load Forecasting with Weather Ensemble Predictions. James W. Taylor and Roberto Buizza IEEE Trans. on Power Systems, 2002, Vol. 17, pp. 626-632.

4. http://www.memphisweather.net/modeldata.html