Autocorrelation Analysis of Meteorological Data from a RASS Sodar

Abstract

Autocorrelation analysis is necessary in persistence studies and identification of cyclical processes. In this paper, autocorrelations of available wind speed and temperature data from a radio acoustic sounding system (RASS) sodar were calculated. This device was placed on flat terrain, and the measuring campaign extended over April 2001. Ten-minute averages were considered from 40 to 500 m in 20-m levels. The direction frequency rose indicated clear, prevailing directions in the east-northeast?west-southwest axis. Analysis of median temperatures revealed that east-northeast advections were 5°C colder than those from the west-southwest. A defined pattern was obtained for both autocorrelations, comprising deterministic and random parts. Noise became more relevant at the higher levels. The deterministic part could be considered as an initial fast-decaying term with the addition of two harmonic functions. The initial decay, linked to fast changes, increased with height for wind speed and decreased for temperature. A diurnal cycle was relevant at intermediate levels for wind speed and at lower temperature levels. The absence of the surface influence added to the horizontal movement associated with the stable night stratification and diurnal convection produced a sharp daily contrast in wind speed at intermediate levels. The influence of the surface decreased with height for temperature. The second cycle was linked to changes in the synoptic pattern and had a 5?6-day period. It was more relevant at lower levels for wind speed, and its amplitude decreased with height. For temperature, this second cycle was less significant. Following these assumptions, a model for the autocorrelation function was proposed and its coefficients are calculated by means of a simple method?a multiple linear regression beyond the first day and a simple linear regression for the first 12-h residuals. This model proved satisfactory, especially below 300 m. A rough height parameterization has been proposed that retained the relevant information and provided satisfactory fits. The influence of the number of observations was investigated. Only extremely high data reduction provided noticeable noise.