Machine learning techniques can be applied to make predictions of significant weather events by utilizing weather model forecast data to predict observed sensor data. In this regard, correlation analysis and grid optimization can be used to automatically select appropriate fields from weather models to be used in the prediction. Open-source software packages can then be applied to evaluate the effectiveness of different learning algorithms. Training such learning systems utilizing the large sets of data found in weather models poses its own set of unique challenges.

This paper will discuss open-source learning systems and describe their value in the area of weather prediction. It will cover pre-processing techniques connected with time lagged ensembles and automatic normalization. This paper will also discuss issues related to large scale learning using online learning systems to automatically select model fields for prediction of sensor values plus the verification of such predictions.