There is disclosed a method and a system for creating a weather forecast. The method is implemented on a server. The method comprises: receiving, by the machine learning module, at least one current weather measurement parameter at a measurement time; receiving, by the machine learning module, a first average value of a historical weather parameter for the measurement time; generating, by the machine learning module, a normalized value of the weather measurement parameter based on a difference between the current weather measurement parameter and the first average value of the historical weather parameter for the measurement time; training the machine learning module to create a normalized value of a weather forecasting parameter based on, at least partially, the normalized value of the weather measuring parameter, the normalized value of the weather forecasting parameter being associated with a future forecasting time, the future forecasting time occurring after the measurement time.