Recognition of meteorological situations with neural networks
暂无分享,去创建一个
This paper deals with the estimation of the amount of clouds and number of cloud layers above airports. A neural network model has been developed based on heterogeneous measurements from original instruments, to give a global estimate of cloudiness. A heuristic procedure is then suggested to nd the number and heights of the cloud layers, and to divide the cloudiness into its components estimated layer by layer. The Ametis II project is a large project of the Swiss Meteorological Institute, which aims at automatizing weather messages sent to airplanes at the Swiss airports. Aeronautical meteorological observations are performed every thirty minutes, by human observers. These observations are encoded in METAR messages, broadcasted to the International Aeronautical Telecommunication Network. These METAR messages encode all the necessary information that airplanes have to know when they are in the vicinity of the airport. Here is an example of such a METAR: This METAR message is valid for Zurich airport at 6:40 AM GMT. Wind is blowing from 230 at 12 knots with gusts up to 25 knots. Minimum visibility of 1200 m occurs in the south direction, maximum visibility of 6000 m in the northeast. Mist is present. A rst scattered layer of clouds is present at 1500 ft, a second broken layer is present at 5000 ft. Temperature is 5 C, dew point is just below freezing, and the see level reduced pressure is 1012 hPa. The objective of the Ametis II project is to design, realize and put in operation a system providing fully automatically generated METARs during the night. It aims at replacing human observers during the periods of low air traac. In the future, similar systems could also be used at small airports where no human observer is present. Some of the parameters needed in the METAR are very straightforward to obtain. These include temperature, pressure , wind speed and direction. Those parameters are provided by a conventional automatic meteorological station. But others are much harder to get automatically, such as the present weather and the description of the cloud layers. This information is provided by visual observations at the moment. In order to achieve complete automatiza-tion, new instruments have been installed at Zurich airport, and their physical properties have been studied by meteo-rologists at the Swiss Meteorological Institute. In addition to present weather sensors, pyrgeometers and ceilometers are connected to the instruments network. Pyrgeometers when directed to the …
[1] Gerhard Czeplak,et al. Solar and terrestrial radiation dependent on the amount and type of cloud , 1980 .
[2] Geoffrey E. Hinton,et al. Learning representations by back-propagating errors , 1986, Nature.