Laserfiche WebLink
except 1990, and so was included in the analysis – treating 1990 as among the “seeded” <br />years, even though no seeding actually occurred. Slope County was out of the program in <br />1990 and again in 1999, and since 2000 only the eastern part of the county has <br />participated. Consequently only the eastern part, containing two climatic gages, was <br />included in the analysis, on the same basis as was Bowman County. <br />2.1Adjustment for missing data <br />Beyond the difficulties already noted, the climatic gage records include three <br />types of entries in addition to the actual gage readings. These are notations concerning <br />“missing”, “incomplete”, and “estimated” data. The “incomplete” values represent cases <br />where data from a few days during the month, evidently a maximum of nine, are lacking; <br />those values were incorporated as is into the analysis, there being no reason to anticipate <br />any systematic bias resulting from their occurrence. “Incomplete” values affect some data <br />from about one-third of the stations, but only one station had more than four <br />“incomplete” months (out of the total of 159 months included in the analysis). In that <br />instance, all nine “incomplete” months were August data, a period of generally light <br />rainfall in the region and possibly when the observers tended to take vacation. Overall, <br />there were 13 “incomplete” gage-months in the target area, out of a possible 1,590 gage- <br />months (10 gages over 3 months for 53 years) – that is, less than 1%. For the control area <br />the comparable numbers are 15 incomplete gage-months out of a possible total of 3,339 – <br />less than 0.5%. <br />The process for arriving at the “estimated” values is not known to us, but as it is <br />evidently satisfactory to the National Weather Service we also used those values as is in <br />the analysis. All the “estimated” values were for control-area gages, with 22 gage-months <br />affected (less than 0.7%). Once again a single station was responsible for more than half <br />of those cases. <br />“Missing” monthly values appeared in the records for most of the stations, with <br />up to six monthly values missing for any one station. For the target area there were 20 <br />“missing” gage-months out of 1,590 possible, or 1.26%, while there were 31 “missing” <br />out of 3,339, or 0.93%, for the control area. Whereas the “incomplete” and “estimated” <br />values were used directly in the analysis, some means was needed to replace the <br />“missing” entries with values that represented plausible estimates of missing data. <br />For that purpose, area by area, station by station, and month by month multiple <br />LAD regression relationships were used. That is, for missing June 19xx data for Station <br />A in the target area, June data for all target stations and for all other years with June data <br />were used to develop a regression relationship to predict the June values for Station A. <br />Then the existing data values for the other stations for June 19xx were used as predictors <br />to obtain an estimate for the missing value for Station A. In two, out of the 51 total, <br />instances the predictand value turned out to be negative, so the entry in the database was <br />set to 0. <br />4 <br />