Data Mining Techniques for Interpretation of Weather Forecast

Abstract

In this paper we have selected the output products of European Center for Medium-range Weather Forecasting
(ECMWF) and Weather Research and Forecasting (WRF) models with reference to occurrences of Cloudbursts
and Tornado in India. The NWP model output products have been pre-processed using Multidimensional Data
Model which considerably improved the storage and retrieval of required weather variables on a selected time
and space scale. The important ingredients to tornado and cloudburst formation have been derived using
primary output forecasts provided by the models. Clustering technique of data mining has been applied first on
the synoptic scale data of well formed Low Pressure Systems (LPS) which form near head Bay of Bengal and
move west, west north-west to north-west and produce rainfall in their wake. Same technique has been applied
on sub-grid scale weather processes. Clusters of ensemble (forecast valid for a particular time based on
different initial conditions) of derived weather variable viz. convergence at various atmospheric pressure levels
for a real life case of tornado and four real life cases of cloudbursts have been generated and analyzed using kmeans clustering technique. This approach resulted in locating patterns conducive to formation of cloudbursts
four days in advance. The data availability was a limitation but promising advanced signals of formation of
patterns have been shown in the study.