A logistic model for climatological series in a region of the Venezuelan Andes

The objective of this paper was to evaluate a series of monthly rainfall by a multinomial logistic regression model in order to compare the trend, seasonality and presence of outliers of monthly precipitation. For which were used data from San Cristobal weather station in Tachira state and simulated series of models of extreme events; Pearson Type III, Type I Gumbel, Log-normal and log-Pearson Type III. Also, for analysis of trend and seasonal were used saturation graphics of the variance of the series, to detect outliers was used Mahalanobis distance (D2). To adjust the patterns of extreme events the maximum likelihood estimation and graphic density adjustment was used. Thus, the results showed a skewed distribution of rainfall with a discontinuity in the trend of the series in the period 1973-1983, associated with high variability (75.75 %) due to the presence of outliers caused by errors in the records. Likewise, the presence of outliers distributed mainly in the rainy season was detected, associated to August 1960, June 1984, July 1985 and July 1989. Moreover, monthly precipitation data were adjusted a Pearson type III distribution. Logistic regression suggested that the only variable significantly related to the type of theoretical distribution of the series was the precipitation. Thus, the Monte Carlo simulation showed consistency of maximum likelihood estimators of logistic model in the analysis of monthly precipitation series. Finally, the results obtained in this research showed that the methodologies considered; saturation of the variance, Mahalanobis distances (D2) and logistic regression are a powerful tool for studying the trend and homogeneity of the monthly precipitation, multivariate outlier detection and comparison of series of monthly precipitation respectively.