Daily precipitation data are often useful for running climatological models; nowadays these models make frequent use of computational and algorithmic approaches that require no missing values. Four straightforward methods to reconstruct gaps in precipitation databases have been considered and compared through a series of statistical indexes and applications to some practical issues using the daily precipitation database of the Veneto Region (Italy). The methods are compared from many points of view: estimating extreme errors of reconstruction; pairing observed rainfall values and respective reconstructing errors of each method; ability to predict monthly and annual accumulations, and monthly and annual rainy days; varying the network density. In the first case, a modified Normal Ratio method seems to have the best behaviour; in the second case, the modified Normal Ratio gives the same results as Linear Regression and Inverse Distance Weighting methods, while in the two last cases, Linear Regression seems to be the best performer, showing also a greater robustness when reducing the density of the network. The results highlight the inherent difficulty of dealing with data characterised by a strong spatial and temporal variability such as rainfall. The choice of the reconstruction method should be done considering both the purpose of the analysis (e.g. reconstruction of extreme events or identification of averages of subperiods) and the characteristics of the network and/or the climatic traits of the environment studied.

Keywords:Gap filling; weather networks; precipitation data; reconstruction methods.

• Full Text: PDF(640.90KB)
• Full Text: HTML