Analysis of temperature series: estimation of missing data and homogeneity test

Abstract

ABSTRACT:In this study, missing value analysis and homogeneity tests were applied on the 267 meteorological stations having temperature records throughout Turkey. The monthly and annual mean temperature data of stations operated by the Turkish State Meteorological Service (DMI) for the period 1968-1998 were considered. For each station, each month was analysed separately and the stations with more than 5 years missing values were eliminated. The missing values of the stations were extrapolated by the Expectation Maximization (EM) method using the data of the nearest gauging station (reference station). In consequence of the analysis, annual mean temperature data are obtained by using the monthly values. These data have to be hydrologically/statistically reliable if they are to be used in later hydrological, meteorological, climate change and estimation studies. For this reason, the Standard Normal Homogeneity Test (SNHT), the (Swed-Eisenhart) Runs Test and the Pettitt homogeneity test were applied to detect inhomogeneities in the annual mean temperature series. Each test was evaluated separately and inhomogeneous stations were determined.

Reconstruction of surface air temperature in a glaciated region in the western Qilian Mountains, Tibetan Plateau, 1957–2013 and its variation characteristics

“…To estimate or reconstruct the missing data of temperature, many techniques have been adopted, including within-station, between-stations, and regression-based methods (Kemp et al, 1983;Thornton et al, 1997;Allen and DeGaetano, 2001;Firat et al, 2012;Thevakaran and Sonnadara, 2013). The within-station method is one of the simplest ways to estimate missing data, and the observations on the previous and following days or months are used to calculate the average (Kashani and Dinpashoh, 2012).…”

Estimation of Air Surface Temperature From Remote Sensing Images and Pixelwise Modeling of the Estimation Uncertainty Through Support Vector Machines

“…In case (1), all estimates are already available from step 1). For case (2), no estimate is available, and no missing data reconstruction is feasible, i.e., the gap in air temperature estimates due to cloud cover remains. For case (3), denoting the sets of cloud-covered and cloudfree days for the ith pixel, mth month, and hth daily time as…”

“…It has been found effective in several climatological applications [2], [3], [59]- [61] and in multispectral image classification with missing data [62]. We model the spatio-temporal random field {y it } i∈I,t∈T of air temperatures as a Gaussian process with mean μ t = E{y it } and autocovariance function…”

“…Recalling that SVM-based regression operates under the assumption of independent samples [19], an equivalent formulation of the described pixelwise model is that the air temperatures y it of cloud-free samples ((i, t) / ∈ C), when conditioned to the corresponding feature vectors x it and to the training set, are independent and Gaussian-distributed with means equal to the estimatesŷ it in (1) and with variancesσ 2 it expressed by (2). This is similar to analogous conditional independence assumptions, which are commonly accepted in Bayesian image analysis to favor tractability and through which accurate classification and estimation results have been obtained in multiple applications (e.g., through Markov random fields [39], [58]).…”

Changes and trends of seasonal total rainfall in the province of Istanbul, Turkey

“…If there are not significant changes in these fluctuations, this means the series is homogenous (Firat et al, 2012). If the fluctuations do not happen in a standard area, but showing significant changes, the data set is defined to be heterogeneous.…”