NO. 5 LONG-RANGE TEMPERATURE FORECAST — ABBOT 9 



In the use of tables of periods in weather for forecasting beyond 

 1957, it is necessary to make extrapolations from the preceding tables 

 of dates. This is done (of course with marginal uncertainty) by aver- 

 aging the intervals (given above) in months SS>20 and SS<20, 

 and assuming that future intervals will be approximately the same as 

 these averages. The uncertainty will usually not lead to important 

 errors of forecasts, for generally the curves representing SS > 20 

 and SS < 20 for the periods are similar for a given period and differ 

 but a few months, or even not at all, in phases. I use for future dates : 

 for SS > 20, 84 months, for SS < 20, 52 months. 



To avoid being misled by sudden large jumps between consecutive 

 monthly departures, the departures from these new normals are 

 smoothed by 3-month consecutive means. Owing to unpredictable lags 

 in the phases of the harmonic periods, it is necessary, as explained in 

 the paper just cited, to divide the record data into special groups, de- 

 pending on the time of the year, the prevalence of sunspots, and the 

 secular march of time. This requires that 220 tables should be com- 

 puted for the forecast of temperature at each station, just as in the 

 precipitation forecasts of Publication 4390. 



This large task requires electronic computation. It was done for 

 the 10 temperature stations and for the 32 precipitation stations by 

 Jonathan Wexler of Tempe, Ariz. In order to avoid new procedure 

 in programming the electronic computer for temperature, I directed 

 him to convert recorded temperatures, published in Fahrenheit de- 

 grees, into absolute temperatures Fahrenheit by adding 491. 7 minus 

 32 , or 459-7°. Departures from the monthly normals were then 

 computed in percentages, as was done with the precipitation values. 



For instance, the normal temperature for June at Detroit for years 

 of sunspots greater than 20 Wolf numbers is 67.8 F. Subtracting 32 

 and adding 491. 7 it becomes 527. 5 ° abs. F. The observed tempera- 

 ture of Detroit for June 1950 was 68.1 ° F. Similar steps make it 

 527.8 abs. F. Its ratio to normal June temperature is 527.8 -=-527.5 = 

 1.0006. All this rearranging was done with the electronic com- 

 puter. In the subsequent computations we used the differences from 

 1. 0000 as far as the fourth decimal place. These differences ranged 

 for the most part between +150 and —80, corresponding to a range 

 of 2.3 percent of the absolute temperature, or about 11. 5° F. 



Readers interested in further details of the method are referred to 

 my paper "A Long-Range Forecast of United States Precipitation" 

 (Publication 4390) and other references cited therein. Temperature 

 forecasts are somewhat less satisfactory than the precipitation fore- 

 casts. For while the range of percentages of normal precipitation goes 



