Hourly surface wind monitor consistency checking over an extended observation period

A consistency checking methodology is presented to aid in identifying biased values in extended historical records of hourly surface wind measurements obtained from a single station. The method is intended for screening extended observation periods for values which do not fail physical consistency checks (i.e., standard or complex quality assurance methods), yet nonetheless exhibit statistical properties which differ from the bulk of the record. Several specific types of inconsistencies common in surface wind monitoring datasets are considered: annual biases. unexpected values, and discontinuities. The purely empirical method checks for self-consistency in the temporal distribution of the wind measurements by explicitly modeling the diurnal variability. Each year of data is modeled using principal component analysis (PCA) (or empirical orthogonal functions, EOF), then hierarchical clustering with nearest neighbor linkage is used to visualize any annual biases existing in the measurements. The diurnal distributions for wind speed and direction are additionally estimated and visualized to determine any periods of time which are inconsistent with the typical diurnal cycle for a given monitor. The robust consistency checking method is applied to a set of 44 monitors operating in the San Joaquin Valley (SJV) of Central California over a 9-year period. Monitors from the SLAMS, CIMIS, and RAWS networks are considered. Similar inconsistencies are detected in all three networks; however, network-specific types of inconsistencies are found as well. Copyright (C) 2008 John Wiley & Sons, Ltd.