A data mining methodology for tracing convective kernels from cloud-to-ground discharge and other atmospheric datasets.
Jacques
Politi, CAP/INPE, jacques.politi@lac.inpe.br
(Presenting)
Stephan
Stephany, LAC/INPE, stephan@lac.inpe.br
Margarete
Oliveira
Domingues, LAC/INPE, margarete@lac.inpe.br
Odim
Mendes Jr., DGE/INPE, odim@dge.inpe.br
Convective activity is used in many tropical atmospherics studies, and usually requires radar and satellite generated data in order to trace it. However, the availability of such data sets may be very limited according to the sampled time interval and area coverage. An alternative strategy to overcome these issues would be the tracing of convective activity by means of electric discharge data sets. This work presents a data mining based methodology to process and analyze cloud-to-ground discharge and other atmospheric data sets, combining sensor-collected and modelling data. Data mining techniques are used to analyze great amount of data trying to identify frequent correlations, patterns and anomalies in a large domain of commercial or scientific applications. In this context, the current work provides means for monitoring and diagnosis of convective kernels. The data mining cycle requires data pre-processing and filtering. As discharge data is very sparse in both the space and time domains, a spatial analysis method, the kernel estimator, was used to reduce this data aggrupating it in clusters. The tuning of the system parameters allows to choose different meteorological scales for the analysis. Tests aimed at the Brazil Pantanal Sul Matogrossense area using Interdisciplinary Pantanal Experiment datasets. Input data included meteorological data such as precipitation rates, pressure, temperature, wind field and also some eletric properties. Test results are shown and discussed.
Authors would like to acknowledge the support received by projects CNPq 78707/2003-7, CNPq 477819/2003-6 and FAPESP 98/00105-5 and CNPq MSc grant.