The robust detection of outlying rainfall observations.

The detection of outliers is important for telemetric rainfall observations because outliers significantly distort the results of most hydrological models. In this paper a three-step robust statistical method combining robust statistical theory with distribution features of precipitation for the detection of outliers in a telemetry system is described. The proposed robust statistical method adopts the Tukey fence insensitive to outliers as identification bounds, and presents a three-step pattern to adapt the distribution of rainfall data. Moreover, the modified method based on dividing precipitation data into several groups further improves detection efficiency. Synthetic data are given to test the performance of the proposed method. The results illustrate that the proposed method produces reliable detection results. It is shown that the new method, based on distribution of the statistical features of rainfall, is suitable for hydrological needs.