Site Map - Computational Intelligence - Artificial Neural Networks - Application - Meteorology

 

Meteorology is characterized by intrinsic nonlinearities and to a certain extent, a lack of complete understanding of underlying physical processes. However, there exists a good level of qualitative understanding and tangible evidence of most aspects of the science. These features make meteorology a suitable candidate field for application of artificial neural network techniques both for the better understanding of the dynamics inherent in the discipline and for producing suitable weather predictions. 

Artificial Neural Networks have recently begun to emerge as a novel approach for the modeling of various complex, non-linear phenomena in the field of meteorology. It is apparent from the rather limited documentation of applications of artificial neural networks to meteorological issues, that the scientific community involved in this endeavour have already produced quite promising results.



Research Areas

Forecasting Minimum Temperature

Estimation of Missing Rainfall Records

Generation of Isohyets Involving Land Configuration

Climatic Classification of Rainfall and Temperature Regimes

Satellite Cloud Imagery Classification

Synoptic Classification of Rainfall and Temperature


Key People








Forecasting Minimum Temperature

Related Papers

Authors: Schizas C.N., Michaelides S.C., Pattichis C.S and Livesay R.R. (1991)
Title: Artificial neural networks in forecasting minimum temperature.
Presented: Institution of Electrical Engineers, Publ.No.349, 112-114.
Abstract: The usefulness of Artificial Neural Networks in forecasting minimum temperature was investigated. In this study the back propagation method was used in order to develop suitable forecasting models based on hourly meteorological records at the meteorological site of Larnaka Airport in Cyprus.

Although the results in this study are preliminary, it is inferred that the introduced method can claim a place in practical operations. The present method requires a commonly available meteorological database without any additional experimental requirements. Also, once the models are developed, the operational application of the method requires limited computational resources. The above render the method advantageous over other methods that demand specific experimental designs to generate the data sets and require extensive computational resources.

Top

 

Authors: Schizas, C.N., Pattichis, C.S and Michaelides, S.C. (1994)
Title: Artificial Neural Networks in Weather Forecasting.
Presented: Neural Networks, 219-230.Pergamon Press.
Abstract: Artificial Neural Networks (ANN) are used for developing models that are fed with weather data and forecast the minimum temperature (Tmin) of the day. The advantage of the ANN approach compared to other existing methods is demonstrated by training the neural network models with short time-length data. The best neural network forecasting models had a success rate of the order of 65% when they were tested with unknown data. It is also shown how difficult is the problem of forecasting Tmin in spite of the recent technological advances and improved numerical techniques currently available.

The artificial neural network models developed in this way, can make use of almost all the available data in a quantitative manner. This is considered to be an advantage of the approach since it allows the use of meteorological elements that are difficult to handle statistically, as for example the present weather concept that is of great importance. This is quite promising for the enlargement of the database by considering elements of more qualitative nature and which have a profound effect on nocturnal cooling, such as the description of the prevailing weather patterns, topographical characteristics etc.


Top





Estimation of Missing Rainfall Records

Related Papers

Authors: Michaelides, S.C., Neocleous C., and Schizas, C.N. (1995)
Title: Artificial Neural Networks and multiple linear regression in estimating missing rainfall records.
Presented: International Conference on Digital Signal Processing, 26-28 June 1995, Limassol, Cyprus.
Abstract: The Meteorological Service of Cyprus operates a quite dense network of meteorological stations for recording rainfall. Some stations have an uninterrupted operation for many decades. However, through the years, some stations have discontinued their operation due to various reasons (e.g. force majeure, maintenance and accessibility problems). Some of these observation sites have been abandoned altogether, whereas, some others have been relocated. Also, new observation sites are established to fill in gaps in the observation network.

It is obvious from the above that for those stations that have a long archive of rainfall data but do not operate anymore, their respective rainfall time series is discontinued; also for those stations that became operational relatively recently, their respective rainfall time series is limited.

A sufficiently long archive of rainfall data is commonly required for most applications. A technique is therefore proposed that can be put forward in order to generate a sufficiently long time series of rainfall records for those locations for which the existing time series is either discontinued (forward extension) or where the archives have a relatively recent start (backward extension). The method uses artificial neural networks for the estimation of daily rainfall at particular observation sites in Cyprus (termed target stations) using as input daily rainfall observations from neighbouring sites that have a sufficiently long and complete archive of data (termed control stations). In this way, the technique can be used to fill in missing data from the rainfall observation network but also for checking suspected data by using the records from surrounding stations.

The above technique of using neural networks is contrasted to the traditional multiple linear regression method. The target station is considered as the dependent variable and the control stations as the independent variables.


Top

 

Authors: Kalogirou, S., Neocleous, C., Michaelides S.C. and Schizas, C.N. (1997)
Title: A time series reconstruction of precipitation records using artificial neural networks.
Presented: Fifth European Congress on Intelligent Techniques and Soft Computing, EUFIT 97, Aaachen, Germany, September 8-11, 1997.
Abstract:

Feedforward multilayer neural networks have been used for the estimation of precipitation in selected rainfall collecting stations in Cyprus. Archived data collected for nine years and six control stations distributed around a target station have been used for training a suitable artificial neural network. Different neural network architectures and learning rates were tested, aiming at establishing a network that resulted in the best reconstruction of missing rainfall records. A multiple hidden layer architecture was chosen. This kind of architecture has been adopted for solving problems with similar requirements. The parameters used for the training of the network were collected at each control station. These are the Julian day, height, distance between target and control stations, and the precipitation. The correlation coefficient obtained for the training data set was 0.933. The verification of the network was done by using unknown data for the target station. This was done for a year, whose data were excluded from the training set. The correlation coefficient for the unknown case was 0.961. The prediction error was confined to less than 17.1mm of precipitation which is considered as acceptable.

Top






Generation of Isohyets involving Land Configuration

Related Papers

Authors: Kalogirou, S., Neocleous, C., Michaelides, S.C., and Schizas, C.N. (1998)
Title: Artificial Neural Networks for the generation of isohyets by considering land configuration.
Presented: Engineering Application of Neural Networks, EANN98. Gibraltar, June 1998.
Abstract:

Feedforward multilayer neural networks have been used for the generation of isohyets of the mean annual rainfall in Cyprus (i.e. contours of mean annual rainfall). Archived data of precipitation recorded over a period of sixteen years at 51 meteorological stations have been used for training a suitable artificial neural network. Various neural network architectures and learning rates were tested, aiming at establishing a network which can yield the best possible estimation of precipitation at any arbitrary location on the island. Such a neural network was subsequently used for drawing isohyets. A multiple hidden layer architecture was chosen. This kind of architecture has been adopted for solving problems with similar requirements.

Initially, five stations were excluded from the training data set for validation purposes. Precipitation data for the remaining 46 stations were used for training and testing the network. The parameters used for the training of the network were a) the X and Y coordinates of each station measured from a random reference point, b) the station elevation and c) the mean annual precipitation for the respective station. The correlation coefficient obtained for the training data set was 0.921. The validation of the network was performed by using unknown data for the five stations which were not included in the training phase. The correlation coefficient for the unknown cases was 0.916. The prediction error of the mean annual rainfall was confined to less than 8.8% which is considered quite adequate. A sensitivity analysis was then carried out to investigate the effect of the station elevation on the estimated precipitation. This showed that there are stations for which the elevation is very important whereas for some others is insignificant.

In order to broaden the data base, the five stations used for the validation of the technique were embedded into the training data set and a new training of the network was performed. The architecture of the network, the momentum, the learning rate and the initial weight values were the same as in the validation phase. The correlation coefficient value for the training was increased to 0.962 which was expected due to the increase in the number of data used. It is believed that the accuracy of prediction increases too.

Subsequently, a grid with a grid distance of 10 km was drawn over a detailed topographic map of Cyprus and the X and Y coordinates and the elevation of each grid-point were recorded. This information was then supplied to the network which produced an estimate of the mean annual precipitation at each grid-point. The X and Y coordinates and the mean annual precipitation at both the original meteorological stations and the grid-points were then used as input to a specialised contour drawing software in order to draw the isohyets. It should be noted that the recorded data inherently include "elevation" information, which is part of the site specificity, whereas the estimated precipitation at the grid-points include the elevation information because the network was trained by considering "elevation" as one of the input parameters.

The effect of orography upon rainfall is widely known. Therefore, it is believed that the proposed method for implicitly involving the station elevation in isohyet drawing is more realistic than the traditional methods which make use of only the X and Y station coordinates.

Top




Climatic Classification of Rainfall and Temperature Regimes

Related Papers

Authors: Michaelides, S.C., Pattichis, C. and Kleovoulou, G. (1998)
Title: Classification of Climatic Variability using Artificial Neural Networks.
Presented: Conference on Applied Mathematical Modeling, Limassol, Cyprus, March 11-13 1998.
Abstract: The variability of the weather has always been a major concern to mankind, because it critically affects almost all of the human activities and especially the planning ahead of weather sensitive operations. For the better understanding of this variability in weather, scientists have sought for the factors which play an important role and several theories have evolved in this respect. The use of sophisticated climatic models running on computers is becoming an increasingly established tool in the effort to understand climatic cause and effect relationships. Through simulated cases, these models offer the opportunity to estimate the impact that various physical processes (natural or man-made) have on possible climatic changes and trends. However, a first step towards a more definite approach for understanding and explaining the variability in the various components of the weather, is to quantify and document this variability and the present research is an endeavour in this respect.

Statistical techniques have traditionally been used in the study of climatic variables. In the present article, a first attempt is been made to identify and thereby classify weather patterns in Cyprus by using artificial neural networks (ANN). Two weather elements have been used, namely the temperature and rainfall. The raw data consist of the daily measurements of these two elements at a number of meteorological sites in Cyprus. The raw data used in this study are the hourly temperature readings recorded on the hour (in degrees Celsius, C) and the daily accumulated rainfall (in millimetres, mm) recorded at 8 a.m. local time. The temperature records cover the period from 1976 to 1995 and the rainfall records the period from 1917 to 1995. The above raw data base was used to derive two time series that have subsequently been used in the present analysis. The temperature time series for the 1976-1995 period, consists of each month's average daily temperature; similarly, the rainfall time series for the 1917-1995 period, consists of each month's average daily rainfall. To obtain comparable values in both time series have been normalised on the basis of the actual number of days in each month.

In the above time series, each one-year subset is considered as one "element". Each "element" represents the temporal distribution of the appropriate meteorological parameter for the respective year. Using artificial neural networks, these "elements" were grouped on the basis of their temporal characteristics. A finite number of classes were derived, each class representing a characteristic one-year temporal pattern for the respective meteorological parameter.

From the examination of the results of a series of preliminary runs with various class numbers, a number of 6 classes for the temperature regime, and 16 classes for the rainfall regime appeared to be the optimum for further detailed analysis and discussion.

With this technique, years belonging to the same class are considered to be "similar" (with regard to the temporal distribution of a particular meteorological element), with no reference to the statistical characteristics of the distribution. Rather, the temporal distribution is treated by the artificial neural network as a single "element".

The same classification procedure was adopted for the two-year subsets in the time series, with one year overlapping in two consecutive "elements". In this respect, the two-year temporal distributions of temperature and rainfall were also classified by the artificial neural network in 6 and 16 classes, respectively.

Top

 

Authors: Michaelides,S. and Pattichis,C. (1998)
Title: Classification of rainfall distributions with the use of Artificial Neural Networks.
Presented: Proceedings: 4th Panhellenic Conference on Meteorology, climatology and Atmospheric Physics. Athens, Greece, September 1998. pp. 251-256.
Abstract: This research is an attempt to utilize and apply Artificial Neural Networks in the study of the long term climatic changes. In particular, a methodology for investigating the classification of the climatic variability of rainfall distibutions. The data used in the formulation of the respective models are the daily rainfall records at a climatological station in Cyprus for a long time period. The elements which are derived by the Artificial Neural Network models comprise the respective climatic classes. The formulation of the models are based on Kohonen's method which allows for the creation of the respective models without supervision. Prototype models are also presented on the basis of the classification that was made.

Top







Satellite Cloud Imagery Classification

Related Papers

Authors: Kyriakou, K., Michaelides, S.C., Pattichis, C., Christodoulou, C. (1999)
Title: Cloud classification from satellite imagery with artificial neural networks
Presented: Fifth International Conference on Engineering Applications of Neural Networks, EANN99, Warsaw, Poland, September 1999.
Abstract: The purpose of this pilot project is to develop a computer aided system based on artificial neural networks and texture analysis that will facilitate the automated interpretation of cloud images. This would speed up the interpretation process and provide continuity in the application of satellite imagery in the process of weather forecasting. A series of 98 (370 cloud cases) infrared satellite images from the Geostationary satellite METEOSAT7 covering the period from 6 to 31 January 1999 were processed in this study. Seven different texture features were extracted from the cloud images. Subsequently, these six classes were further grouped into three major cloud classes. For each feature set an SOFM classifier was trained with 210 cases and evaluated with 160 cases. The percentage of correct classifications score for the evaluation set for the best feature sets was in the region of 76 to 74%.

Top

Synoptic Classification of Rainfall and Temperature

Related Papers

Authors: F. Liassidou, S.C. Michaelides, C.C. Neocleous, C.N. Schizas (1999)
Title: Identification of synoptic patterns on weather charts by artificial neural networks
Presented: Fifth International Conference on Engineering Applications of Neural Networks, EANN99, Warsaw, Poland, September 1999.
Abstract: The purpose of the present research is to investigate whether identification of synoptic patterns on weather charts can be made objectively by training artificial neural networks. In order to achieve this, the daily weather analyses at 0000 UTC for 1996 were employed. The respective data consist of the grid point values of the geopotential height of the 500hPa isobaric surface. A uniform grid-point spacing of 2.5o is used and the geographical area covered by the investigation lies between 25oN and 65oN and between 20oW and 50oE, covering Europe, the Middle East and the Northern African coast. An unsupervised learning self-organizing feature map algorithm (Kohonen) was used. The results referred to in this study employ a generation of 15 synoptic classes. The main conclusion from this endeavor is that the present technique produced a satisfactory classification of the synoptic patterns over the geographical region mentioned above. This classification exhibits a strong seasonal relationship.

Top


Key People

 

Kalogirou S. (Higher Technical Institute, Cyprus) skalogir@spidernet.com.cy
Michaelides S.Chr. (Meteorological Service, Cyprus) cssilas@ucy.ac.cy
Pattichis C. S. (University of Cyprus, Cyprus) pattichi@ucy.ac.cy
Neocleous C.C. (Higher Technical Institute, Cyprus) costas@ucy.ac.cy
Schizas C.N. (Univerisity of Cyprus, Cyprus)  schizas@ucy.ac.cy

Top