Premium
An enquiry into the possibilities and limits of statistical weather forecasting
Author(s) -
Schumann T. E. W.
Publication year - 1944
Publication title -
quarterly journal of the royal meteorological society
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.744
H-Index - 143
eISSN - 1477-870X
pISSN - 0035-9009
DOI - 10.1002/qj.49707030505
Subject(s) - value (mathematics) , limiting , statistics , reliability (semiconductor) , mathematics , correlation coefficient , degree (music) , regression analysis , linear regression , regression , current (fluid) , physics , quantum mechanics , thermodynamics , mechanical engineering , power (physics) , acoustics , engineering
There are three methods of approach to the problem of weather forecasting:–‐ ( a ) the physical method; ( b ) the statistical method; and ( c ) the current method which in effect is a combination of the first two methods. The physical or strictly mathematical method has been tried, notably by L. F. Richardson, but has been found to be impracticable. Everyone is acquainted with the degree of success achieved by the current method, but the possibilities and limitations of the statistical approach to the problem have never been fully investigated, and it is a matter of considerable importance that this should be done. The statistical method essentially amounts to the prediction of the weather element P at a station A by means of the regression equationin which x 1 are the values of a number of weather elements observed at the control stations B 1 ……. B n a specified time before P is observed, and the regression coefficients k 1 …. k n are computed from previous records of the weather elements P, x 1 …… x n . The reliability of the predicted value of P depends upon the numerical value of the multiple correlation coefficient between P and x 1 …. x n , and there exists a limiting maximum value M z of this correlation coefficient. By the suitable selection of a limited number of control stations this maximum value can be ascertained to a very close degree of approximation. Research should be directed towards determining the value of M z for different values of the time lapse z , over which the prediction is made. Once the value of M z has been found, there is available a direct method of judging between the relative merits of statistical and current weather prediction. A considerable amount of labour is involved in the full investigation of the nature of the function M z , but several considerations show that such labour would be fully warranted.