Imputing large gaps of high-resolution environment temperature

![CDATA[Weather data are widely used in climatology and other environmental studies. One of the key challenges in preprocessing these data is to deal with missing values. Since the measurements are recorded through sensors, missing values are ubiquitous in weather variables such as environmental temperature. Even though there are well-established methods to impute missing values in univariate time series data, the need of developing improved methods to impute large gaps persists. This paper compares the performances of ten existing methods in imputing missing values of hourly temperature data. Among the methods considered, Kalman smoothing on Auto-Regressive Integrated Moving Average model(ARIMA) and Kalman smoothing on structural time series model are the best methods in imputing missing values under MCAR (Missing Completely at Random) mechanism with exponentially distributed missing values. Moreover, this paper proposes a novel method to impute large gaps of hourly temperature data using regularized regression models on deseasonalized data. This method outperforms all the other considered methods in imputing large gaps.]]