This paper describes a simple but efficient method, based on classical techniques, which enables one to estimate missing data and to extrapolate short instrumental climate series from other instrumental or proxy ones. By mutually exclusive digital filters, the variations of the data are decomposed into a rough spectrum. A canonical analysis is performed on each of these parts, providing relationships for each principal time-scale. A global equation is then computed by inverse transformation. Generally, the predictands are better fitted by this method than by a standard regression. The improvement is shown for both instrumental and proxy data. The method can easily be applied in many topics where relationships between two groups of variables have to be estimated: agroclimatology, ecology and geophysics in general. It can run on a microcomputer with a BASIC program available from the author.
By this method, a reconstruction of June-July temperature at Marseille (France) is proposed showing a period of weak variability of the temperature during 140CL1850, that is to say a period of rare warm summers with very cold summers only in the early 16th and late 17th centuries. An extrapolation of the instrumental series at Nice from corresponding series of Marseille and Gap is also developed. KEY WORDS Palaeoclimatic reconstructions Missing data Digital filter Canonical analysis Cross-spectral analysis Principal components analysis Transfer functions Tree rings