Rain pattern analysis and forecast model based on GPS estimated atmospheric water vapor content

Rain is one of the fundamental processes of the hydrologic cycle as it can be the source of wealth or natural hazards. This experiment focuses in the relationship between rain occurrence and atmospheric pressure (Patm) and atmospheric water vapor content (PW), GPS estimated. The available nine years time series of each variable were analyzed. It allowed to state the existence of three rain patterns and monthly differences in the Patm-PW combinations. In spite of rain episodes take place only for some of the Patm-PW combinations, only these variables are unable to explain the rain occurrences because of not always they take place. This because a forecast sliding windows model with neural network was developed, to capture nonlinear relations that can not to be fully reflected by the lineal probabilistic ones based on the observed rains, Patm and PW series. This model stated a good correlation between the observed rains and the forecast, with a positive impact of the PW but negative of Patm. This model was able to predict the rain precipitation with a reasonable precision and reliable accuracy up to a 56 h horizon.