NO. 5 LONG-RANGE TEMPERATURE FORECAST — ABBOT II 



cipal depressions. These ratios, in terms of event divided by forecast, 

 were for most cities between 0.70 and 0.75. Using these average ratios 

 of amplitude of features, I reduced all the forecasts to approximately 

 the same scale of amplitude as the events. 



There remained still another correction, but one not puzzling like 

 that for scale. As many have pointed out, temperatures have gradu- 

 ally risen in parts of the United States for a great many years. It 

 would have been quite wrong to make forecasts of temperature for 

 1950 through 1967 without allowing for this well-established change 

 of level. Hence I took the ratio of the sum of monthly departures 

 from normal after correcting for scale, as forecasted from 1950 

 through 1959, and divided by the corresponding sum for the event. 

 This gave a correction to lift the forecasts bodily by amounts ranging 

 from 0.1 1 to 0.37 percent for the different cities. Expressed in de- 

 grees of temperature, these corrections of level range from 0.5 to 

 1. 9 F. 



Having by these two necessary corrections adjusted the forecasts to 

 terms justly comparable with the events, I computed the correlation 

 coefficients mentioned above. 



RESULTS OF THE INVESTIGATION 



First of all, to inspire confidence in the method, which, as I have 

 said, is substantially the same for temperature as for precipitation, I 

 give in figure 4 for 14 of the 32 stations of Publication 4390 a com- 

 parison of forecasts and events on precipitation for the 24 months of 

 1959 and i960. This interval is several years beyond the latest month, 

 November 1956, used in the basis of the forecasts. Table 1, above, 

 gives for 1959 and i960 an analysis of the monthly values of forecasts 

 and events in percentage departures from the normal monthly pre- 

 cipitation to be found in column B, table 9, of Publication 4390. 



Twelve other precipitation stations gave almost as good correlation 

 as these fourteen stations, except that more cases of displacement of 

 features by one, two, or three months occurred between forecasts and 

 events. Such displacements, frequently noted in my former papers, 

 are as yet impossible to forecast. This is the main defect of my fore- 

 casts. It is true that the amplitudes of features frequently differ be- 

 tween forecasts and events, but if the main features occur zvhen pre- 

 dicted, the moderate difference of amplitudes is not a very serious de- 

 fect. If one could predict when phases of prominent features would 

 be displaced, the correlation between forecasts and events in precipita- 

 tion would rise from being 40 to 70 percent to lie between 70 and 90 

 percent. 



