A data-driven approach to weather forecast using convolutional neural networks

The Weather Research and Forecasting (WRF) Model is a numerical weather prediction system designed for both atmospheric research and operational forecasting applications. WRF can produce simulations based on actual atmospheric conditions or idealized conditions and provide forecasts such as precipitation and geopotential height among others. This paper addresses the analysis of the influence of these variables on rain condition in order to complement the WRF model precipitation forecast. Convolutional neural networks (CNN), used in the field of image processing, were applied to develop a machine learning model that takes the data generated by the WRF model as an input and capture the relationship with measured precipitation data. This process resulted in two models that achieved a rain prediction with an accuracy of about 80%. This results prove the capacity of the model to learn the existing correlation between the input variables and actual data. However, they also revealed the need for data that span a time interval that includes the different existing cycles (e.g. solar cycles) in order to make further reliable predictions.