Statistical approach for verification of medium range weather forcasts

摘要

The medium range weather forecast weekly bulletins containing elements viz., cloud cover, precipitation, wind (direction and speed) and temperature (maximum and minimum) issued by National Centre for Medium Range Weather Forecast (NCMRWF), New Delhi for Parbhani station (19 deg 16'N and 74 deg 47'E) during June 15, 1993 to September 30, 1994 were used in this study. They were classified on the basis of qualitative signals of self exclusive dichotomous events. The signal of forecast of rainfall was designated as (F_1) and no rainfall as (F_0). Similarly, the realisation of rainfall was designated as (R_1) and its no-realisation as (R_0) Whereas, in temperature (maximum and mininmm), the tendency signals were designated as increase (+), decrease (-) and no change (0). The conditional frequencies of there permutations and combinations of realisation were worked out. The binomial (Munn, 1970) and conditional (Medhi, 1976) probabilities were worked out. The elements were subjected to further mathematical treatments for working priory probabilities, oscillations over stretch of forecasts and their equilibria. The degree of divergence between forecast and realisation was ascertained by Chi-square test.