I 



for time-steps of about 15 minutes. 

 Each iteration starts from the last 

 prediction, and the forecast is carried 

 forward for many days. 



Numerical predictions of this kind 

 form the basis for the extended (5- 

 day) forecasts made by the National 

 Weather Service, an additional com- 

 ponent being supplied by the experi- 

 ence of the forecaster. How accurate 

 are they? 



The skill of the final pressure- 

 pattern predictions made by the pres- 

 ent "man-machine mix" from two to 

 six days in advance is shown in 



Figure IV-9 by the curve marked 

 "Present." Without going into de- 

 tails, 1.00 on the vertical scale implies 

 perfect forecasts, and indicates fore- 

 casts that are no better than maps 

 randomly selected from the same 

 month of past years. As can be seen 

 in Figure IV-9, extended forecasts de- 

 teriorate rapidly from day to day; by 

 the sixth day, one might as well use 

 the initial day's map as a forecast 

 ("persistence"). Even at the fourth 

 day, the skill is low enough to be of 

 marginal economic value. Assuming, 

 however, that the present accuracy of 

 forecasts for the fourth day are eco- 

 nomically valuable, we might ask 



Figure IV-9 — FORECASTING SKILL 



1.00 



BASED ON 156 CASES 



The graph shows the accuracy — and limitations — of National Weather Service 

 forecasts of the pressure pattern for North America for the period March 1968 to 

 February 1969. 



how good the two- to six-day predic- 

 tions would have to be to give us 

 accuracy equal to the four-day figure 

 at two weeks, or 14 days in advance. 

 These computed values are shown in 

 the upper curve marked "Future." 

 Thus, we see that a six-day forecast 

 will have to be about as good as a 

 two-day forecast is now. A six-day 

 forecast will have to be about 25 

 times better than at present (ratio of 

 the squares of the six-day correlations 

 for "Present" and "Future"), a four- 

 day forecast about 10 times better. 



Prospects — These are tremendous 

 strides that will have to be made, 

 especially if one considers the frus- 

 tratingly slow rate of progress in 

 improving short-range weather fore- 

 casts over the past twenty years. The 

 situation suggests that some major 

 breakthrough in understanding, and 

 in the density and quality of observa- 

 tions, must come about before de- 

 tailed predictions in time and space 

 out to two weeks or more will be 

 realized. There is controversy in the 

 meteorological community as to 

 whether forecasts of this type will 

 ever be possible. 



Yet the potential for economically 

 valuable long-range predictions is not 

 as bleak as might be gathered from 

 this discussion. While the forecast 

 for a given day well in advance may 

 be greatly in error using the above 

 method, the general weather charac- 

 teristics of a period — say, the aver- 

 age of computerized forecasts for the 

 second or third week in advance — 

 may turn out to contain economically 

 valuable information. There is still 

 no evidence this is so, but the hope 

 exists that better and more observa- 

 tions combined with more knowledge 

 of atmospheric modeling will result 

 in this advance. Numerical modeling 

 may make a major spurt forward be- 

 cause of the development of a first 

 model aimed at coupling air and sea. 



What Needs To Be Done 



In order to bring about this prog- 

 ress and raise the level of the "Pres- 



99 



