Comparison of gap filling methodologies of annual historical series of rainfall

The absence of rainfall data for stations due to equipament or operator problems compromise information continuity. In order to apply a statistical treatment on rainfall historical series, gap filling needs to be done. Usual methods for this are: regional means (PR), linear (RM) and potential (RP) multiple regression, regional means using linear regression (RS) and regional vector (VR). Using total annual rainfall historical series with 22 years of observation for six rainfall stations in the State of Goias, a comparison analysis was done between gap filling methods and the differences between observed and estimated values were evaluated. For the regional vector method five possibilities were considered extracting from the series the observed data, replacing the estimated value in the series by the methods of regional means (VR/PR), linear regression (VR/RM), potential regression (VR/RP) and regional means using linear regression (VR/RS). According to results it was concluded, in ascending order, minor the lowest differences in relative desviation were obtained with the methods: RM, VR/RP, PR, VR/RM, VR/RS, RP, RS, VR/PR and VR. According to the regional vector method, an improvement was verified in data values was verified when a combination between some other methods was used.