Statistical Long-Term Predictability of Turkish Surface Temperatures from Global Sea Surface Temperatures

We present the preliminary results of an ongoing, comprehensive analysis; in which monthly and seasonal land surface temperatures in Turkey are statistically related to global SSTs of the preceding seasons. The ultimate aim is to assess the purely statistical, long-term predictability of Turkish temperatures by using appropriate subsets of global SST anomalies. As an initial step, cross correlations between each monthly mean temperature time series (predictands) from 41 selected stations in Turkey and monthly mean SSTs (predictors) from each grid point on the globe were computed for the period 1950-2015. This was done for numerous time averages of both the predictor and predictand time series; with lead times up to 7 months and predictor series averaged over up to 12 months, backwards from the most recent predictor month. The outcome was 9.5 x 10⁸ correlation coefficients showing the degree of time-lagged relationship (if any) between global SSTs and Turkish temperatures. Results indicate that, generally; the greater the number of predictor (SST) months are, the higher the absolute values of the correlations become; hinting at the role of SST persistence in monthly and seasonal climate. The less expected and more interesting results are, 1) some correlations are higher as the lead/lag time increases, implying potential long term predictability, 2) there are signals of moderate relationship between Turkish temperatures and the SSTs of remote regions such as the southern Indian Ocean or the central Pacific. These first results will help determine, for each prediction effort (i.e. each lead time and predictand month/season), the best predictor (SST) regions and time series; which will then be the input for more advanced techniques such as multiple linear regression or canonical correlation analysis.