Abstract

Venezuelan monthly precipitation data is geocoded, discontinuous in space and time. Missing data, summarized records, installation and removal of stations is common. Second order properties were evaluated (spatio-temporal autocorrelation), but data is not normally distributed, then a Box-Cox transformation was applied because it is known that non normal distributions tend to hide spatio-temporal autocorrelation. Intensity of spatio-temporal autocorrelation was estimated using Moran's I and its cluster variant LISA. Both with their spatio-temporal version. Cluster analysis allowed to characterize precipitation stations using k-means with five groups of monthly precipitation (low, low-medium, medium, medium-high and high). Results show an increase of the amount of clusters mapped and better representation of precipitation.