Homogenization of Daily Wind Speed and Wind Gust Time Series in the Era of Automatic Observations in Catalonia

The homogeneity of wind speed and wind gust data is very sensitive to changes in the instrumentation or the environment of the sensor. Time series of wind observations over Catalonia have been analysed, and in spite of the relatively short history of automated observations, several significant inhomogeneity biases were detected and removed.

A dense network of observing stations and documentation of station histories (metadata) helped the homogenization. Only time series with observation periods of at least 10 years were used, resulting in the homogenization of 209 time series for both wind speed and wind gust. The data was originated from three sources, i.e. the primary network of the Catalan Meteorological Service (SMC), the agrometeorological observation network of the SMC and the network of the Spanish Meteorological Agency (AEMET) in Catalonia.

The data underwent basic quality control before homogenization. Homogenization was performed with ACMANT, known to be the most accurate homogenization method based on currently available method comparison test results. The used new version ACMANTv5.2 can take the benefit of metadata either in automatic or interactive mode. Prior to homogenization, a specific examination was performed, in which residual standard deviations of two inhomogeneity models were compared, i.e., the additive model and multiplicative model. The additive model showed clear advantages for wind gust homogenization, while the suitability of the two models appeared to be similar for the homogenization of mean wind speed data. Metadata indicated several technical changes during the observation periods of 24 years on average. Most stations experienced changes in sensors and data logger multiple times, while a few time series were affected by station relocations or changes in the sensor height.

The homogenization process revealed that changes in sensors and data logger often did not cause perceptible inhomogeneities. On the other hand, two network-wide changes in data loggers occurred, in 2005 in the red agrometeorological network and in 2007 in the primary network of Meteocat, when the technical changes impacted significantly and almost synchronously the homogeneity of the data. Since all relative homogenization methods, including ACMANT, presume that inhomogeneities are station-specific, metadata played a crucial role in detecting these problems. The final homogenization was performed in a way that neighbour series affected by the same type inhomogeneity bias as that of the candidate series were excluded from the homogenization.