Missing Rainfall Daily Data: A Comparison Among Gap-Filling Approaches

Abstract

Missing data is a frequent problem in meteorological and hydrological temporal observation data sets. Finding effective solutions to this problem is essential because complete time series are required to conduct reliable analyses. This study used daily rainfall data from 60 rain gauges spatially distributed within Portugal's Guadiana River basin over a 30-year reference period (1976–2005). Gap-filling approaches using kriging-based interpolation methods (i.e. ordinary kriging and simple cokriging) are presented and compared to a deterministic approach proposed by the Food and Agriculture Organization (FAO method). The suggested procedure consists of fitting monthly semi-variogram models using the average daily rainfall from all available meteorological stations for each month in a reference period. This approach makes it possible to use only 12 monthly semi-variograms instead of one for each day of the gap period. Ordinary kriging and simple cokriging are used to estimate the missing daily precipitation using the semi-variograms of the month of interest. The cokriging method is applied considering the elevation data as the secondary variable. One year of data were removed from some stations to assess the efficacy of the proposed approaches, and the missing precipitation data were estimated using the three procedures. The methods were validated through a cross-validation process and compared using different performance metrics. The results showed that the geostatistical methods outperformed the FAO method in daily estimation. In the investigated study area, cokriging did not significantly improve the estimates compared to ordinary kriging, which was deemed the best interpolation method for a large majority of the rainfall stations.