Evaluation of Infilling Methods for Time Series of Daily Temperature Data: Case Study of Limpopo Province, South Africa

Abstract

Incomplete climate records pose a major challenge to decision makers that utilize climate data as one of their main inputs. In this study, different climate data infilling methods (arithmetic averaging, inverse distance weighting, UK traditional, normal ratio and multiple regression) were evaluated against measured daily minimum and maximum temperatures. Eight target stations that are evenly distributed in Limpopo province, South Africa, were used. The objective was to recommend the best approach that results in lowest errors. The optimum number of buddy/neighboring weather stations required for best estimate for each of the approaches was determined. The evaluation indices employed in this study were the correlation coefficient (r), mean absolute error (MAE), root mean square error (RMSE), accuracy rate (AR) and mean bias error (MBE). The results showed high correlation (r > 0.92) for all the stations, different methods and varying number of neighboring stations utilised. The MAE [RMSE] for the best performing methods (multiple regression and UK traditional) of estimating daily minimum temperature and maximum temperature was less than 1.8 °C [2.3 °C] and 1.0 °C [1.6 °C], respectively. The AR technique showed the MR method as the best approach of estimating daily minimum and maximum temperatures. The other recommended methods are the UK traditional and normal ratio. The MBEs for the arithmetic averaging and inverse-distance weighing techniques are large, indicating either over- or underestimating of the air temperature in the province. Based on the low values for the error estimating statistics, these data infilling methods for daily minimum and maximum air temperatures using neighboring stations data can be utilised to complete the datasets that are used in various applications.