To obtain the regression winds three screening-regression runs were 

 raade resulting in three sets of six wind specification equations. The 

 first run used ship B data only and was used primarily as a test of 

 machine programs. The second run used data from all four weather ships 

 and provided the equations subsequently used for specifjdng the Worth 

 Atlantic wind field. The third run was made to test the usefulness of 

 the measured air- and sea-temperature fiinctions. From the results of 

 all three runs it appeared that the best wind estimate had a rms error 

 of about 8.5 kts. for each wind component. The wind speed was deter- 

 mined mainly by the geostrophic wind speed modified slightly for 

 curvature effects. The wind direction was given by the geostrophic- 

 wind direction, rotated counterclockwise 15.5°- It was found from the 

 third run that the measured air- sea temperature difference did not 

 demonstrably show itself as any aid in specifying the indraft angle. 

 Hence, further studies with computed variables that are merely approxi- 

 mations of this measured quantity were considered unnecessary. 



Based on all the results of the three runs it was decided by 

 Travelers to use the following u- and v- wind component equations com- 

 puted from 4,iH7 samples for specifying the North Atlantic wind field: 



(1) u = 0.266762+0.798937 Ug -0.3^6355 Vg^+ 9.30208 X 10^1 joQ 



(2) V = -3.^5584+0.7829^3 V_. + 0. 241837 U„+l. 23438 X 10^-?,^^^ 



& g 1000 



(3)«C = 15.5490° 



These three equations then specify the regression winds. A test 

 sample wind field was computed with these equations for 17-I8 December 

 1959. The resiilts (Fig. l) show two large extratropical cyclones moving 

 northeastward across the Atlantic Ocean at approximately 20 kts. In the 

 five separate fields computed, the continuity of the cyclones was ex- 

 cellent, the wind pattern well organized, and the wind speeds reasonable. 

 The large grid mesh, however, causes small scale features, such as 

 multiple centers, to disappear and frontal shear lines have been quite 

 blurred for the same reason. 



In view of" the rms error of 8.5 kts. for each of the components of 

 the regression vri.nds it was decided to test the possibility that a 

 better wind specification could be obtained from the direct use of ship 

 wind observations. Objective-analysis winds were prepared for the 

 period 15-27 December 1959 from North Atlantic surface-weather ship 

 reports which had been manually checked for errors. In a large-scale 

 effort a program would have to be developed whereby this could be 

 handled by computer. The initial-guess wind field for the objective- 

 analysis winds was the regression wind. The ship observations were 

 integrated with the regression winds by a conditional relaxation anal- 

 ysis method and then verified by the areal-mean- error method. The grid- 

 point values were determined by extrapolating from or interpolating 

 between locations at which observations of the winds were available. 

 The procedure required that the gridpoint values satisfied Poisson's 

 equation subject to the constraints imposed by the observations and an 



