An Observational Study of Tropical Large-Scale Fields: Part I: Statistical Analyses of the Wind Components, Geopotential Height and Temperature

Abstract

It is pointed out in a critical review of works on tropical data analyses that strong discrepancies between the statistics from the observed fields and those obtained from the standard grid-point data sets over the tropics lead to notable ambiguities in the results obtained from the analyses. It appears that these weaknesses are due principally to the model dependency of these grid-point data sets, which are produced by using objective analysis schemes designed without special attention to the tropics. The necessity of fundamental statistical analyses is stressed in the outline of the new analysis approach devised in order to avoid or reduce these shortcomings.
Using a 5-year (1979-1983) tropical upper-air data set, we performed statistical analyses for the wind components (u, ν), the geopotential height (Z) and the temperature (T) over the tropical belt between 25N and 25S of latitude, and 20E and 140W of longitude. Spectral and cross-spectral analyses show the dominance of four large-scale and long-period oscillations: the 2-4 years oscillation, the annual, the semi-annual and the intra-seasonal (40-60 days) ones. Short-period oscillations are shown to be of small scale in space and do not contribute significantly to the horizontal transport processes of heat and momentum. Low-pass filtering is shown to improve the estimates of the spatial correlation of the four variables and to reduce their observational errors by a factor 2-3 as compared to the estimates from the non-filtered data. It is then expected that low-pass filtered data would describe reliably the tropical large-scale phenomena. The correlation field is shown to be homogeneous for all the fields, and anisotropic not only for the wind components, but for Z and T as well.
Most of these findings should be taken into account in statistical objective analysis schemes, such as the Optimum Interpolation Method, for better grid-point estimates over sparse observational network areas in the tropics.