We will develop machine learning algorithms that are able to learn from typical estimation errors in the prediction of the precipitation using merged radar and satellite products over the Continental United States. We aim to find a collection of independent variables that, regionally or globally, jointly explain the occurrence and magnitude of the precipitation process, its estimation biases and their temporal evolution. Results from this proposal will help the National Weather Service and NOAA improve their precipitation retrieval algorithms adding support variables to their prediction system and therefore reducing prediction errors.