Missing Data Imputation and Meta-analysis on Correlation of Spatio-Temporal Weather Series Data

Even though weather time series are easy to be found, complete and large data sets are almost impossible to be retrieved. In this paper, an assessment of missing weather data in small data sets is introduced utilizing correlation and meta-analysis of different weather parameters like temperature, humidity and wind speed. Auto regressive integrated moving average (ARIMA), a well-known artificial model and widely used for weather prediction, is evaluated on various sets with missing data. The results of an univariate and multivariate ARIMA model are presented to come up to the best model for each feature and parameter. Finally, the most accurate model is tested against real life data, revealing that imputation of missing data increases prediction accuracy for almost 50%.