Investigation of the efficiency of common methods of infilling missing data in the arid regions of Iran

One of the most common and effective problems in long-term climate studies is the presence of gaps in the time series of various climatic and hydrological data. Therefore, the present study evaluates the accuracy of methods for infilling missing data of daily, monthly and annual temperature time series in the arid regions of Iran. For this purpose, the observed daily minimum, average and maximum temperature data for the period 1987-2014 measured at 73 synoptic stations distributed all over arid regions of Iran were used. Methods of readjustment used include: Normal ratio method, linear regression, multivariate regression and Inverse Distance Weighting (IDW). In this study, the capability of each mentioned methods for infilling missing data of daily, monthly and annual precipitation time series in the arid regions of the Iran was investigated, while the proportion of missing data varies from 5 to 50% of total data. In order to compare and evaluate the accuracy of the four mentioned methods three statistical indicators, namely the correlation coefficient (R), the Root Mean Square Error (RMSE) and Nash coefficient were used. The results showed that in general, each of the methods mentioned had different functionalities at a special level of readjustment and time scale. On annual and monthly scales, linear regression and normal ratio methods are the most accurate method in readjustment temperature data in the arid region of Iran. The correlation value between the readjustment and observational data at different levels reaches more than 0.95 using these methods. On the daily scale, there is no significant difference between the accuracy of the methods used in the readjustment of temperature data, and almost all four of these methods have appropriate accuracy because in all methods the correlation between readjustment and observed data is more than 90%. However, multivariate regression methods with an average correlation of 0.99 showed the most accurate performance in readjustment daily data at different levels of readjustment. Generally, each method should be used in accordance with the conditions, and therefore it is recommended to develop a software package for infilling missing data.