Lots of damages, losses, and costs have been the major concern, why handling natural disasters of tornados is very important. Several attempts using different approaches have been carried out, but up to now the results are not yet satisfactory. More promising approaches through a kind of artificial intelligent forecaster have been started for a while, but the results are still not satisfactory either. The capability of mHGN as a pattern recognizer has opened up a new possibility of recognizing a pattern of tornado many hours earlier. Therefore, it can be used to forecast a tornado more efficiently. The results taken from a simulated circumstances of a multidimensional pattern recognition have shown, that the 91% of accuracy can be regarded as satisfactory. Though, several modifications related to the data representation within the mHGN architecture need to be implemented. The deployment of mHGN in several risky areas of tornados can then be expected as a tool for reducing those damages, losses, and costs.

HAL is a multi-disciplinary open access archive for the deposit and dissemination of scientific research documents, whether they are published or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers. L'archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.

The architecture of mHGN has been improved and become Single Layer Hierarchical Graph Neuron (SLHGN). The speed of this new architecture for recognizing multidimensional patterns is faster than the one of mHGN. It is therefore more suitable for forecasting multidimensional and complex process of tornado's genesis in real-time. Additionally, two important issues related to data handlings of non-accurate recorded data and data handlings of complex weather data have been solved. These improvements have given significant and positive quality of SLHGN in forecasting tornado. Although the accuracy and the forecasting performance cannot be calculated properly, due to the fact that weather data is not always available, the specific characteristics of the SLHGN experiment results show very promising values. This results suggest that tornado can be forecasted at least 5 hours before it occurs. People in the to-be-hit area will then have adequate time to be evacuated or to escape. The deployment of SLHGN in risky areas of tornados can then be expected as a tool for reducing damages, losses, and costs. Several improvements in weather station distribution still need to be carried out in order to improve the quality of tornado forecasting using SLHGN.

The usage of the mHGN as a pattern recognizer cannot necessarily be used to recognize tornados. Two important issues that need to be solved first are related to data handlings of not-accurately recorded data, and to those of complex weather data. The not-so-appropriate data handlings will produce high false positive and true negative rate of the recognition results. Yet, the latest development of those data handlings has been carried out, and has shown positive and promising results. Such a new approach of data handlings can, therefore, be used to improve the quality and the accuracy of forecasting a tornado. The results taken from a simulated circumstances of a multidimensional pattern recognition have shown, that the tornado can be recognized around 9 hours earlier with 90% of accuracy. However, several improvements related to the data representation within the mHGN architecture need to be implemented. The deployment of mHGN in several risky areas of tornados can then be expected as an alternative way of reducing damages, losses, and costs.

An attempt for an earthquake forecasting has been challenged by the current earth coordinate system. The latest architecture of mHGN which is called SLHGN requires that the observed locations must be spread out regularly, that is within regular grid-like distances. Such a requirement would not be fulfilled, as within the current earth coordinate system a longitude-difference would produce different distances. The extreme examples are locations in equator compared to those on the earth poles. Therefore, an earth coordinate system has been developed to support the ongoing earthquake forecasting technology using SLHGN. Additionally, two important positive issues related to this earth coordinate system have been developed, they are: 1) each location is not represented through twovalue (longitude and latitude), but only a single value. This value does not represent a point but an area; 2) the conversion of this earth coordinate system to the x-y Cartesian System requires no angular formulas, which is therefore fast. These issues have given positive support to the SLHGN in forecasting earthquakes. Although the accuracy and the performance are not yet ready to be analyzed properly, because local weather data at the time of an earthquake occurrence is not always available, the characteristics of the SLHGN experiments show very promising results