WSRC-RP-93-269 July 1993 / COMPARISON OF SAVANNAH RIVER SITE'S METEOROLOGICAL DATABASES(U) A. H. Weber Approved by A.L. Dbnl Research Manager Environmental Technology Section ..,_: S SAVANNAH RIVER SITE iii i i 'Westinghouse Savannah River Company Savannah River Technology Center Aikenr SC 29808 :,, 93zo$?oJIwo DtSTRIBU]'ION OF THIS DOC:LJMEN[ IS UI"4LIMIT'C'_ .j
126
Embed
COMPARISON OF SAVANNAH RIVER SITE'S METEOROLOGICAL ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
WSRC-RP-93-269July 1993
/
COMPARISON OF SAVANNAH RIVER SITE'SMETEOROLOGICAL DATABASES(U)
'Westinghouse Savannah River CompanySavannah River Technology CenterAikenr SC 29808
:,, 93zo$?oJIwo
DtSTRIBU]'ION OF THIS DOC:LJMEN[ IS UI"4LIMIT'C'_ .j
ill
Thisreportwaspreparedby theWestinghouseSavannahRiverCompany(WSRC)for theUnitedStatesDeparunentof EnergyunderConnct No. DE-AC09-89SR18035and is an accountof workperformedunder thatCOltIII_L Neither the United States Deparunent of Energy, norWSRC, nor any Of theiremployeesmakeany warranty,expressedorimplied,norassumeanylegal liabilityorresponsibilityfortheaccuracy,completeness,orusefulness,of anyinformation,apparatus,orproductorprocessdisclosedhereinor representsthatitsusewill not_ge privatelyownedrighL_,Referencehereintoanyspecificcommercial product,prtx:ess,or service by trademark,name, manufacturer,or otherwise does notnecessarilyconstituteor implyendorsement,recommendation,orfavoringof thesame by WSRCor bythe UnitedStatesGovernmentor anyagency thereof.The views andopinions of the authorsexpressedherein do notnecessarily stateor reflect thoseof the U.S. Governmentor any agency thereof.
Comparison of Savannah River Site's Five.Year Meteorological Databases
Executive Summary.
A five-year meteorological database from the 61-meter, H-Area tower for theperiod 1987-91 was compared to an earlier database for the period 1982-86. Theamount of invalid data for the newer 87-91 database was one third that for theearlier database. The data recovery percentage for the last four years of the 87-91database was well above 90%. As described in the background sections,considerable effort was necessary to fiU in for missing data periods for the newerdatabase for the H-Area tower. Therefore, additional databases that have beenprepared for the remaining SRS meteorological towers have had missing anderroneous data flagged, but not replaced. The F-Area tower's database was usedfor cross-comparison purposes because of its proximity to H Area.
Statistical tests were used to compare the frequency of occurrence and the meanwind speeds for the 82-86 and 87-91 databases. These tests are extremelysensitive because of the large number (-35,000-45,000) of hourly observationsin the data sets. The tests show statistically significant differences in thefrequency of occurrence of winds for the H-Area tower for all seven Pasquillstability classes, two of sixteen wind directions sectors (IVand S), and four of sixwind speed categories. The mean wind speeds between the two time periodscompared more favorably, with only G stability category showing a statisticallysignificant difference.
The differences in frequency of occurrence for stability classes may be caused byclimatic changes, such as a 0.6°C temperature increase between 1982 and 1991,the period during which data was collected. The differences in the north windsector were probably due to a coding fault and the instrument maker's design forthe wind direction instruments. The differences found for the south wind sector
Comparison of Savannah River Site's Five-Year Meteorological Databases
and all wind speed classes are probably either due to natural climatic variation,statistical variability, or differences in the data screening methods used for thetwo time.periods. The fact that the mean wind speeds are significantly differentfor the G stah;lity class may be due to the use of different data screening methodsfor the two timeperiods.
A baa's was provided for future comparisons by examining wind speed powerspectra and by stratifying the data by season and time of day. The wind speedpower spectra showed expected yearly, 3-5 day, 24 and 12 hour peaks. Peaksoccurring at periods of 8 hours and 32-26 days have not as yet been explained.The stratification by season showed a trimodal spring pattern with dominantwinds from the S and SSE. During summer, winds from the N become veryinfrequent and the S to W wind sectors are dominant. A dramatic shift occurs inthe autumn to winds from the NE and ENE. In the winter there is a very narrowpeak of windsfrom the NE and a broad quadrant of windsfrom the S-WNW witha local maxima of winds from the W. Winds from the N and NNW are aboutequallyfrequent for all seasons except summer.
The evening and midnight patterns during the spring and summer are verysimilar, all showing minima out of the N quadrant and maxima out of the Squadrant. Spring and summer mornings and middays resemble one another withmaxima out of the WSW and NE or ENE. Pre-dawn and morning periods in theautumn are similar with local maxima from the NE. Spring and summerafternoon patterns are similar with maxima in the S to W quadrant. Duringspring and summer the pre-dawn periods have winds which are fairly uniformlydistributed with minimafrom the WNW toN sectors.
The primary purpose of this report is to compare the H-Tower databases for 82-86 and 87-91. Statistical methods enable the use of probability statements to bemade concerning the hypothesis of no chfferences between the distributions of thetwo time periods, assuming each database is a random samplefrom its respectivedistribution. This assumption is required for the statistical tests to be valid. Anumber of statistical comparisons can be made between the two data sets, eventhough the 82-86 database exists only as distributions of frequency and meanspeed.
2 93XOJTO.MWO
Comparison of Savannah River Site'sFive.YearMeteorologicalDatabases
Background
In early 1992 the Environmental Technology filled fromnearbytowers, backupdatacollectedSection of the Savannah River Technology at the site (called SUM-X data), Plant Vogtle'sCenter (SRTC) completeda quality-assuredfive- mek:orological tower, or fromNational Weatheryearmeteorologicaldatabasefor the 61-meter,H- Service observations. The result of these effortsArea towerfor the period 1987-91 (Parker,et al., was a high quality database with no missing1992). The database was created for the New observations. A total of only 8.3% of the dataProduction Reactor's (NPR) safety and had to be substituted in these ways for the H-environmentalstudies. Several onsite and offsite Area tower 87-91 database. The amount ofgroups have requestedsimilar data sets since the invalid data was one third that foran earlierfiveNPR database was created. The Parker, et al. yeardatabase (L,aurinat,1987).(1992) report, provides descriptions of the datacollection, processing, and validation used in Laurinat(1987) produceda quality assuredfive-producing the database. The Savannah River year database covering the period 1982-86 forSite's (SRS) 87-91 H-Area database is unique each of the SRS Area towers. The 82-86since there are no missing data in the five year databases differfromthe 87-91 H-Towerdatabaseperiod. The data set includes a total of 43,824 in that the quality assurance tests were appliedhours of data for the five-year time period only duringcomputerprocessing usingstatistical(including one leap year). Each data entry screening algorithms. Visual data inspectioncontains wind speed, wind direction, and lateral techniques were not used. Time periods withturbulence intensity. The data entries for the missing data were not filled, consequently, theyears 1989-91 contain temperatures and dew 1982-86 data records contain missing periodspoints as well. ' whenever the data failed the computer screening
algorithms or the tower was out of service forOne.procedureused to qualityassure the 87-91 H- maintenance,damagedby lightning, etc.Towerdatabase involved plottinga time seriesoffifteen-minute records by tower for one-week The 82-86 computer screening algorithmsperiods. These one-week plots, showing all eight rejected data for failure to pass a statistical testof SRS's meteorological towers simultaneously, knownas the Dixon ratio test (Dixon, 1957), andallowed comparativechecks to be madeof the H- for large differences between lateral andArea tower both spatially and temporally, longitudina! turbulence intensities. The DixonMalfunctioning sensors and missing data could ratio test is used to detect outliers based onbe readilydistinguishedfrom validdataand time statistical similarities of the towers' distributionperiodsnoted.Invaliddam were also identifiedby (orspatial "agreement")of windspeed, direction,inspection of log books and records which or turbulence intensity during a given hour'sidentified calibration periods, maintenance time). Data was rejectedfromboth 82-86 and87-activities, and other downtimes. In addition, 91 data sets whenever voltages exceededduringthe initial, real-time computeracquisition acceptablerangesor when more than 90% of theof the data,software algorithmsare used to check possible data within a ffl'teen-minuteperiod wasdata ranges and to test the consistency of data missing. The amount of data rejected due toentries. Additional data consistency checks were application of these tests in the 82-86 databasealso used in the software which preparedthe 87- amounted to about 21% for each of the Area91 database, towers. Missing or unreadable tape archive files
accounted for an additional 2% of data beingTime periods were flaggedwhen the H-Area data incomplete. The quality assurance proceduresdid not pass the visual or computer quality used to producethe dataset resultedin about23%assurancetests, or had missing data blocks. If the missing data for each tower. However, at thelength of the flagged time interval was 1-4 hours lime this database was generated, there was no(1.1% of the total), the data was filled by linear requirementforpercentageof data recoverability.interpolation. If the time period was from 5-13 The 87-91 database, by contrast,was required tohours (0.4% of the total), data was either filled meet the DOE'scurrentguidance of 90% annualfroma nearbytower'srecords,or by interpolating data recovery for meteorological data. Newand accounting for the diurnalcycle. If the length proceduresand meteorological instrumentswereof the flagged time intervalwas from 13 hours to installed in 1987 to improve the quality and toseveral days (6.8% of the total), the data was capturea largerpercentageof meteorologicaldata.
9Jxo57oJ_4rwo 3
Comparison of Savannah River Site'sFive.YearMeteorologicalDatabases
The new proceduresand instrumentsresultedin a from later years in the table below (extractedsignificant gain in the percentage of valid data from Parker, et al., 1992). The large increasecaptured in annual data sets. This can be between 1987 and succeeding years in Table 1substantiatedby comparingthe datarecoveryrate shows the successful nature of the qualitypercentages for the 1987 wind data with those improvementprograms.
Table 1. Data recovery percentages for each year of the period 1987-91 for the H-Area meteorologicaltower.
WindData 1987 1988 1989 1990 1991 5-Yr. Ave.(Yr.)
H-Area (%) ;'/8.15 96.21 92.84 96.58 94.55 91.71
The 82-86 databaseswere convertedto frequency Additionaldatabaseswere createdlaterin 1992 fordistributionsshortly after theircreation for use in the remainingArea towers A, C, D, F, K, L, anddosimetry models. The frequency distributions P for the 87-91 time period except that missingconsist of frequenciesof occurrenceswithin each data was not filled. The applications of theseof six speed ranges (0-2, 2-4, 4-6, 6-8, 8-12, and databases did not require that the missing time12-50 m/dec), seven Pasquill-Gifford stability periods be filled, and in some instances this iscategories (A-G), and sixteen wind direction not desired. In addition, considerable effort issectors (I-N -- 16-NNW). The ,data sets also needed to provideoata for a tower when the datacontain mean wind _7_eeds(andmean wind speeds records are incomplete. The 87-91 data sets forby inverse averaging) for the same speed ranges, the remaining towers contain hourly-averagedPasquill.Gfffordcategories, and direction sectors. . data and frequency distributions. These databases(Pasquill-Gifford stability categories were exist primarily for special studies or researchdetermined from the measured latend turbulence where having a completely Idled database for theintensity according to the U. S. Nuclear entire five years is unimportant. One of these 87-Regulatory Commission Guide 1.23.) The 87-91 91 data sets for the F-Area tower was usedH-Tower database exists both as a frequency extensively in this report for cross comparisondistribution (as for the 82-86 database) and as a purposes in some of the statistical studiespermanent on-line data set, containing the described later. The earlier 82-86 database for theoriginal hourly observations. F-Area tower by Laurinat was also used for cross
comparison purposes.
Statistical Comparisons Between the 82-86 and 87-91Databases
The statistical comparisons are described in this A test was applied to the differences between thesection but more mathematical details are observed andexpected numberof observations forincludedin AppendixA. The (null) hypothesis thesixteensectors,sevenstabilitycategories,andbeing tested is: There is no difference six windspeedcategories.The squareddifferencebetween the distributions for the two between the number observed and the number
//mLJ/g£_p.dE In order to test this hypothesis, expected for both data sets can be used tothe frequencyof occurrenceandmean windspeed determine if the differences are statisticallydistributionsby sector, wind speed category, and significant.If this squareddifference is divided bystability for both periods were compared. A the expectednumberof observations,the result isprimary question to be answered -- are the the adjusted residual squared, or the cellproportional differences that exist between the contributionto the chi square. The sum of thesenumberof observations in each data set within contributions is assumed to form a chi-squarethe range of expected random statistical distributionwith degreesof freedom equal to onevariations?And similarly -- are the mean wind less thanthenumberof classes being compared.speed differences within the range of expectedrandomstatisticalvariations? The probabilityof gettinga largerchi square than
the one observed (assumingthere is no difference
4 93XOJ70.MWO
Comparison of Savannah River Site's Five.Year Meteorolol_ical Databases
between the two time periods) and the value of of getting a largerchi square is less than 0.001the chi squarestatistic are shown in Table 2 for forstability, sector,andwind speed. If one took aF-Area tower (used for comparison)and H-Area large numberof samplesfroma distributionwithtower databy stability category,wind speed, and the same proportions in each category andsector. The results in Table 2 show there are randomlyassigned them to the two time periodssignificant differences between time periods for in the same proportions as observed, fewer thanboth F and H Area towers, since the probability one sample in 1000 would produce a larger chi
square.
Table 2. Differences between the frequencyof occurrenceof the 82-86 and 87-91 databases determinedfrom the chi squaretest.
The chi square statistic indicates there are statistically significant difference for the H-Areadifferences between the data sets by stability, tower.For the 16 wind direction sectors, five ofsector, and wind speed; but it is not able to the sixteen were significantlydifferent forF-Areaindicate whether the differences are due to large towerand two (winds fromN and S) for H-Areadifferences in only one or two categories, or . tower.Fourout of six windspeed categoriesworewhether all categories have differences. It is differentforboth towers.important to understand that the chi-squarestatistic and hence whether there arestatistically Figure 1shows the percentage of observations insignificant differences depends not only on the each of the sixteen wind directionsectors for thepercentagedifferences buton the total numberof two time periods for the H-Area tower. One' sobservations being compared. As the total visual perceptionmay be that the differences arenumber of observations increases, smaller small enough to indicate that the two data setspercentage differences will be statistically come from the same distribution, even thoughsignificant, the chi-square statistic indicates two statistically
significant differences. Visual impression is notTo determine which categories had significant reliable in this case because of the large numberdifferences between the old andnew data sets, the of observationsin each of the data sets (H-Area:contributionforeach category was determined by 35,252 for 82-86 and 43,824 for 87-91; F-Area:dividing the adjusted residual by its standard 33,331 for 82-86 and 41,818 for 87-91). If theredeviation. This randomvariableis approximately had been only 200 observations over the twonormally distributedwith mean0 and variance 1. time periods, the percentage differences andIf the null hypothesis is true, then all category dis_butions observed in Fig. 1 would not resultcompari_'-as should be within acceptable limits in a statisticallysignificantchi-square,and visualwith pr_,oability 0.95. The Bonferroni method impressionwouldbe vindicated.(The ordinatesin(Miller, 1981) was used to determine the Figures 1-12 are connected by a straight lineindividual probabilities for acceptance in order rather than plotting separatepoints to enable thethat the simultaneous probability was 0.95. eye to gain an overall impression of theAppendix A describes this methodology. If the differencebetween the curves.)standardized adjusted residual is less than thecomparison value given in Table 3 forstability, Similar small but statistically significantsector, and wind direction categories, then we differences in the percentage of observations foraccept the (null) hypothesis of no differences five of the sixteen wind direction sectors for F-between time periods. Table 3 shows which Area tower are shown in the plot in Figure 2.categories have the biggest contributions to the Figures3 and4 show the percentagedifferencesover-all chi square. Figs. 1-6 show which by stability categories for H-Area tower and F-categories have the biggest differences. Only Area tower, respectively. Figures 5 and 6 showstability categories B and G were not the percentage differences by wind speedsignificantly different for the F-Area tower, categories for H-Area and F-Area towers,however all stability categories showed a respectively.Again the probabilitiesofaccepting
93XO570_CWO 5
Comparison of Savannah River Site's Five.Year Meteorological Databases
Table 3. Differences between the frequencyof occurrenceof the 82-86 and 87-91 databases determinedfrom the chi square test for individual contributions by stability category, sector, and speedcategory.
Stability Results of the AdjustedResiduals Dividedby the Standard--Deviation,Category F Area H Area
Valuesgreaterthan2.955 indicatesignificantdifferencesandaremarkedby an asterisk.
Wind Speed{m/sec) F Area H Arm0-2 19.298" 1.9032-4 3.430* 16.076"4-6 11.687" 8.284*6-8 5.195' 15.160"8-12 1.829 7.521"12-50 2.204 2.004
Values greater than 2.638 indicate significant differences and are markedby an asterisk.
6 93XO570MWO
, Comparison of Savannah River Site's Five-Year Meteorological Databases
Table 4. Differences between the 82-86 and 87-91 databases for mean wind speed determinedfrom thet-testfor individualcontributionsby stabilitycategory, sector,andspeed category.
Stability F Area H AreaCategory t-Ratio Prob >ltl t-Ratio Prob >ltli
Ratios withprobabilitiesless than 0.003125 indicate significant differences.
the individualcomparisonshavebeendetermined there areno truedifferences. Figs. 7-8 show theby the Bonferroni method, so that the averagewind speeds forF andH Area towersforsimultaneous probability of accepting no the sixteen wind direction sectors, Figs 9-10 aredifference across the seven stability categories, average windspeeds forstability classes A-G, andsixteen sectors, and six wind speed categories is Figs. 11-12 are average wind speeds for the six0.95 foreach category (stability, sector,andwind wind speed classes. There appears to be aspeed).To determine whether the mean speed statistically significant difference in the resultsdistributionsarestatistically differentbetween the for stability category G of Table 4 and Figs. 9-82-86 and 87-91 data sets, the mean speedswere 10. Fig. 9 shows the difference was 1.24 m/secaveraged within each stability, sector, and wind for stabilitycategory G on the H-Areatower (thespeed category for the two data sets, and largest observeddifferencebetween time periodscomparedusing the two _mple t-test.This test for any of the stability categories or windassumes the differences divided by the standard directionsectors).However, the largestdifferencedeviation of the difference is from a student'st- between the two time periods for the F-Areadistribution.Tests for equality of variance were tower was 1.36 m/sec in sector9 (winds fromS,appliedbefore computing the t-ratios. The results Fig. 8) which was not statistically significant.
the t-tests are given in Table 4. The table The reasonfor this apparentinconsistency is thatshows thatdifferences betweenmeanwind speeds stabilitycategory G for the H-Towerhad one ofare insignificant except for stability category G the smallest estimates of variability for time
the H-Areatower. The probabilityrequiredto period,stability, or sectorfor either tower. Smalldeclare a significant difference has been variabilitywithin timeperiod, stability,or sectordetermined so that all (within category) may cause statistically significant differencescomparisonsforstability, sector, and wind speed even when these differences have no practicalwill be insignificant with 0.95 probabilitywhen significance.
i
9JXO570.MWO 7
Comparison oi"Savannah River Site'sFive.YearMeteorologicalDatabases
Sources of Differences Between the 1982-86 and the 1987-91 Databases
(a) Climatic variation
Meteorologicaldatacould be expectedto vary the absolutevalues was 13.2%, a considerablebetween two contiguous five year periods due to amounLThe amountof this shift from Exhibit 1worldwide climatic variations which are caused for the F-Area tower, summed similarlyover theby increased concentrations of optically active stabilitycategories, is equal to 7.92%.constituentsin the atmosphere.The timescale ofthese variations is generally quite long, but A numberof books andpapershave been writtenvariations can occur on shorter time scales, on the connection between climatic variationsMeasuredchanges in the mean temperature for and sunspots, solar flares, and other solarthe Southeast region of the U.S. show a mean activity. Some of the relationships have beentemperature increaseof about 0.6° C from 1982 shown to be statistically significant, but oftento 1991 (Michaels, et al. 1991). A temperature the effect is obscured by other equally importantchange of this amount might be large enough to changes in atmospheric constituents. Sunspotscause changes in the wind frequencydistribution occur over a period of 10-12 yearsand the 82-86and/ormean values, and 87-91 databases are about half the required
period, so sunspot effects on the database,Itis worth noting thatthe statistically significant although slight, mightalso appearin the data.differences in stability categories in Table 3might be a result of the mean temperature 'Climatic variations can also be caused byincrease. Fig. 3 shows a shift between 82-86 and significant atmosphericforcing functionssuch as87-91 from the more stable categories E-G to volcanic activity, El Niflo, deforestation, etc.;more unstable categories A-C, which could be these phenomenonarewell documented.Randomattributedto the temperatureincrease. This shift statistical fluctuations can also occur in anyin stability categories is also found in F-Area turbulentfluid such as the atmosphere. Severaltower data shown in Fig. 4, Exhibit 4 was used causes for statistical differences between the twoto sum the absolute values of the percentage databases exist if we acknowledge thesedifferences over stability categoriesA-C and E-G mechanismsforclimatic variations.for the H-Area tower during 87-91. The sum of
(b) Instrument and coding faults
D. M. Hamby (1992) used the 82-86 and 87-91 Area towers prior to the end of 1987; there isH-Areadata sets in dosimetrycodes to determine also a bivane resolver gap of about one degreethe relative two-hour and annual average x/Q width which occurs at 359.50-0.5° (north)with(concentrationsdivided by sourcestrength) from improvedwind directioninstrumentationinstalledboth data sets. One of Hamby's plots showed a on theArea towersat the end of 1987.noticeabledip for one wind directionsector in anotherwise smooth curve. Examining the curve The algebraicsign faultexcluded wind directionsclosely showed the 87.91 dataset hada relative within the range0 to +5° foran interval of timeminimum in 7JQ for northwinds, after a software maintenance modification was
implemented.The code faultwas carefully tracedAn extensive investigation led to two possible using computer archive tapes and was found tocauses for the north wind minimum have originatedon February24, 1989. The faultconcentrations which have been identified since was corrected on July 29, 1992. Data for 34 ofHamby'sreport:(1) the algebraic sign in an "IF" 60 months of the 87-91 database were affected.
' statement of an archiving module of VAX-8550 The percentage of missing data caused by thissoftware was found to have a positive value factor can be estimated by assuming the windwhere a negative value was required; (2) there is a directions are uniformly distributed.A five degreebivane potentiometer gap of about 3-5° width exclusion zone would havecaused the deletion ofwhich occurs at (north) v,ith bivanes used on the 1.11%of the azimuths in the north sector in the
i i,i1|1
8 9JxoJ7oMwo
, Comparison of Savannah RiverSite'sFive.YearMeteorologicalDatabases
1987-91 databasefor the H-Area tower. It should component translatingmechanical displacementbe noted that the effect of the excluded wind into electrical signals.)Replacement of the olderdirections would actually be spreadsomewhatto bivanes was also highly desirable since it led toadjoining sectors because of turbulent more accurateand reliable wind directions overfluctuationsduringa fifteen minute period, the compass range 359.50-0.5.
' The north wind direction was one of two sectors The effect of the potentiometer gap operatingwhich showed statistically significant differences over a 60 month period for the 1982-86 &_tabase
I in frequency of occurrence and is flagged inTable can be estimated by assumingthe wind directions' 3. The dip in Hamby's _Q curve is caused by are uniformly distributed. A 3-50gap would haveI the wind speeds being biased to lower values caused 0.83-1.39% of the azimuth data around1 and/or the percentage frequency of occurrence north to be missing. For the 12 months in 19871 being lower for the 87-91 data. Both effects are includedin the new database(1987-91), a totalof1 present and can be verified in the tables and 0,17.0.28% of the data would have been affected.i figures. One can determine from Table 4 and theI software used that the negative t-ratio means the The effect of the resolver gap can be estimated1 mean speeds for 4-6 m/see category are biased similarly. If the resolver gap is one degree wideI toward lower values in the 87-91 H-Area and the wind directions are distributeduniformly,1 database. Fig. 1 shows the percentage frequency then about 0.22% of the total azimuths would( of occurrence for North winds is lower for the have been affected in a 48 month period. Wind1 new database for H-Area. tunnel experiments have shown the effect on the
voltage signal caused by the resolver gap is aThe blvanes used prior to the end of 1987 had spurious voltage, i.e., the wind is indicated astwo-potentiometer systems designed to correct . coming from an incorrect sector. This means thatthe problem of bivanes with a single a secondary effect of the reso_ver gap ispotentiometer. The two-potentiometer system excessively large standard deviations of azimuthalways has a wiper in contact with potentiometer angle. This would produce a slightly larger thanwindings and thus generates a valid signal even expected frequency of A-D stability categorieswhen one of the potentiometer wipers is in the when winds blow from the north sector. Thegap. In spite of the existence of the two percentage of data affected is estimated to be thepotentiometer system, it was apparently not same as the percentage of azimuths affectedused, because of complexity of the software that (0.22%).would have been needed. Rather, the system wasoperated with a single potentiometer. This caused The potentiometer andresolver gap effects arenota gap of 3-5* in the wind directions about north independent of the software "sign" omission,prior to the end of 1987. Improved wind because of overlapping azimuth ranges.instrumentswere installed on the Area towers at Consequendy, subjective estimations of the exactthe end of 1987. (Resolvers replaced amount ofdataaffectedinthe 1987-91 databasepotentiometers as the mechanical-electronic have ahigh uncertainty.
I (c) Sources used for filling missing data
1 Data were filled in for the H-Area tower 87-91 0.92% of the wind data for the 87-91 H-Areadatabaseas described earlier. Linear interpolation, tower were substituted from the Plant Vogtledata from backup computers at the towers, and tower.data from nearby Area towers were the preferredchoices for fill-in data. Data from the Augusta The wind instruments used by Plant Vogtle were
1 National Weather Service (NWS) tower and Plant less sensitive than those used on SRS's AreaVogtle were also used when absolutely necessary, towers. The Pasquill-Gifford stability categories
' The NWS service tower was needed for only determined from these instruments utilizedI 0.54% of 1988's wind data or 0.11% of the total fifteen-minute averages of turbulence intensityt five years. The Plant Vogtle tower instruments during the one-hour period rather than true one-
are mounted at 60 meters, about the same height hour averages which include the effect of wind1 as the 61-meter Area tower instruments. Only meandering (as is done at the SRS).
,i
9JxosTo_nvo 9
Comparison of Savannah River Site'sFive.YearMaeoroloRical,Databases,
Providing a Basis for Future Comparisons
(a) Spectra for the 86-91 Database
Spectral analysis is used to determine layer. There is no way of ascribing absoluteperiodicities in statisticaldata.Boundarylayer causalityusing spectralmethods,howeverthewindspectrafor shortperiods(0.1 secondsto a mostreasonablephysicalmechanismis beingfew hours) often show broad maxima proposedforthepeaks.corresponding to thermal convection andmechanical turbulence and a broad minimum Statisticalsignificance of spectraaremeasuredbycorrespondingto a mesoscale spectralgap. These determining95% confidence limits. These spectraspectralpropertiesareusuallypresentedin "equal- were smoothed by applyinga 21 pointtriangulararea"plots in which the abscissa is the logarithm smoothing to the raw spectral density estimatesof frequency and the ordinate is frequency from the SAS® Proc Spectraprocedure.The 21multiplied by spectral density. Equal-area plots point triangular smoothing results in aare used because the total energy is the same confidence limit multiplier in the range 0.59 tobetween frequencies bounding equal areas. 2.4 for spectral peaks. There are 21 degrees ofUnfortunately,equal-area presentationsobscure freedomcorrespondingto the confidence intervalperiodicities and were not used in the plots of (Jenkins and Watts, 1968). Another rule ofspectraforthefive-yeardatabase, thumb for determining if a spectral peak is
significant is if it contains at least threeThe 87-91 five year database presents an computed points which deviate from the noiseopportunity to determine spectral density for , and has a maximum two or three times greaterperiods between two hours (the Nyquist than the surroundingnoise level (Stuart,et al.,frequency)andfive years(the rangeof thedam).It 1971). All peaks described earlier fit thesewas decided to present the wind speed spectra in criteria.linear plots (spectraldensity vs. period)using the87-91 database to determine where periodicities In orderto increase confidence thatthe spectraloccur, peaks discussed above arereal, a similarspectral
analysis was performed on one of the fifteenSpectral densities were determined from the minute data records from H-Area tower whichhourly-averaged wind speeds at H-Area. The were used to construct the one.hour averages.spectral density estimates are shown in Figs. This fifteen minute data set contained13a-d where the power (m2/sec 2) per unit substitution data from other towers on plant butfrequency(cycles/hr)areplottedagainstthe period lacked the one-hour interpolations and(hrs/cycle). The longest period possible to substitutionsfrom PlantVogtle and the Nationalrepresenton the plot correspondsto the lengthof WeatherService thatwere discussedearlier.(Therecord (five years). The majorpeak in Fig. 13a fact that this data set was not a continuous fivecorrespondsto (8760 hrs) one year-- the annual year record was adjusted by removing timecycle. The annual peak is separatedfromseveral periods corresponding to missing data andsmaller spectral maxima near the origin by a concatenating the remaining time periods.) Thisrelative minimum, or spectral gap at 1300 to process resulted in a data set that was 955 hours1400 hrs (about2 months). The smaller spectral (40 days) short of the full five years. The factmaxima all occur at periods less than 1000 hrs thatthe data set contains missing periods and the(one month or less) and areshown in more detail resulting time series are joined together in ain Figs. 13b-d. The peaks at 760, 590, and 300 discontinuous manner means that the periodhrs correspond to 31.6, 24.5, and 12.5 days, corresponding to a given peak will be shorterrespectively. These long periodicities may be than expected. However, if the peaks are real,related to shifts in the Rossby wave patterns in they should be observed in the same relativethe upperatmosphere.The peaksat 140, 100, positionas theoriginalspectrum.and70 hrsprobablycorrespondto the periodofextratropical cyclones (about3-6 days). The spectral analysis of the fifteen minute data
set is shown in Figs. 14a-d. These dataThe peaks at 24 and 12 hours are caused by substantiatethe peaks at 8, 12, and 24 hours, butatmospheric tides which follow the sun as the do notshow peaks at 6, 4.8 and 300 hours. Tableearth turns.The peak at 8 hrs may correspond to 5 summarizesthe peaks found in both data sets.theperiod of convective activity in the boundary
l0 ojxo57ol_rwo
Comparison o,fSavannah River Site'sFive.YearMeteorologicalDatabases
Table 5. Periods of the major spectral density peaks measured in days and hours for the H-Area towerhourly and fifteen-minute data sets (1987-91).
The third row of Table 5 shows the peaks that are but are simply due to the inherent uncertainty ofnot present in the fifteen-minute data set as power spectrum estimation techniques. Anothermissing values (.). The fact that three of the possibility is that the peaks which disappear arepeaks are not present in the fifteen-minute data caused by the substitution techniques for flaggedsets is evidence that the peaks may not be real, and missing data.
(b) Stratification by Season and Time of Day
An interesting basis for es_blishing future compare quantitatively the amounts in variouscomparisons of five-year databases can be stability categories. Fig. 15a could be describedobtained by stratifying the data by season and as a trimodal distribution with maxima at thetime of day. Figures 15a-b show the H-Area NE, WSW, and S sectors. The 82-86 database intower 82-86 & 87-91 hourly averaged wind data Fig. 15b is bimodal and lacks the S sectoras percentage of occurrence for the sixteen wind maximum. The plots for the F-Area tower (Figs.sectors presented in bar charts. Similar 15c-d) both show the trimodal distribution butinformation was produced for the Laurinat's 82- the 82-86 South sector maximum is relatively86 database except wind "roses" were used. The . small. The H-Area tower sectors with the greatestbar charts were used here instead of wind roses difference between 82-86 and 87-91 are the S andsince they providean easier method for the eye to N sectors.
When the H-Area tower data is stratified by season and time-of-day some interesting patterns begin toemerge. The following definitionswere applied.
Spring: March, April, and MaySummer:. June, July, and AugustAutumn: September, October, and NovemberWinter: December,January, and FebruaryMorning: Between sunrise and true solar noon minus 1/6of the daylight hoursMidday: Time period between end of morning and beginning of afternoonAfternoon: Between true solar noon and sunset plus 1/6 of the daylight hoursEvening: Between sunset and true solar midnight minus 1/6 of the nighttime hoursMidnight: Time period between end of evening and beginningof pre-dawnPredawn: Between true solar midnight and sunrise plus 1/6of the nighttime hours
The spring pattern for the H-Area tower 1987-91 patterns are very similar and all show (Figs. 17a-database in Fig. 16a shows the trimodal d) minima out of the north quadrant and maximadistribution but the peak for winds from S and out of the south quadrant. Spring and summerSSE is very sharp and dominant. Fig. 16b shows morningsand middays resemble one another withthe northerly winds during summer becoming maxima out of the WSW and NE or ENE (Figs.very infrequent and the S to W wind sectors 18a-d). Autumn pre-dawn and mornings aredominate. Fig. 16c shows a dramatic shift in the similar with local maxima from the NE (Figs.autumn to winds from the NE and ENE. In the 19a-b). Spring and summer afternoons havewinter Fig. 16d shows there is a very narrow maxima in the S to W quadrant (Figs 20a-b).peak of winds from the NE and a broad quadrant Spring and summer pre-dawns are fairlyof winds from the S-WNW with a local maxima uniformly distributed with minima from theof winds from the west. Winds from the north WNW to N sectors (Figs. 21a-b). The remainingand north-northwest are about equally frequent for plots have varying degrees of similarity and areall seasons except summer, included for completeness (autumn--Figs 22a-d,
and winter--23a-f).
Spring and summer evening and midnight
93xo57oa_wo 11
Comparison of Savannah River Site's Five-Year Meteorological Databases ,,
Conclusions
New procedures and meteorological instruments differences between the new and old data sets.resulted in a significant gain in the percentageof Only the G stability category for H-Area towervalid datacapturedin annualdatasets forming a indicateda significantdifferenceoutof the sevenbasis for the 198%91 database.The datarecovery stabilitycategories and sixteen direction sectors.percentage for the last four years of the 87-91 This difference was only 1.24 m/see for the Gdatabase was well above 90%. The missing and stability class. This gives a measure ofbad data decreasedby a factor of three from the confidence that dosimetry calculations will not1982-86 database. This fact and the considerable change significantly when using the 87-91 datawork expended in filling in missing periods for set.the new database permitteddetailed analyses thatcould not have been performed on the earlier The fact that the two data sets have differentdatabase, mean wind speed distributionsfor theG stability
class may be due to the fact that the 82-86The chi-squarestatisticwas used to compare the database was created using an outlier screeningfrequencyof occurrencefor two time periods for test. This could have eliminatedvalid dataduringF-Area and H-Area towers. The chi-square frontal passages, thunderstorms, periods of lightstatisticshowed therewere significantdifferences winds, etc., when there were valid reasons toby stability, wind direction sector, and wind expect a given Area tower to show a differentspeedclass. The chi-squarestatisticdoes not have . wind direction than the others. If these cases ofthe ability to show wherethe differences exist, so screening out valid data was not uniformlythe standardizedadjustedresidual together with distributed,thena bias could have beenc_ inthe Bonferronimethod was used. These methods the mean wind speeds (or frequency ofshowed that only two of sixteen wind direction occurrence).It would seem reasonableto assumesectors (north and south) had statistical that application of the Dixon ratio test duringadifferencesbetween the 82-86 and87-91 datasets period of light winds (nighttime, stablefor the H-Area tower. Reasons were found to conditions)probablyresultedin morecases beingexplain the northsector difference but the south rejectedfor the 82-86 database.sectorremainsunexplained. The differences in thenorth wind sector were probably due to a The statistical tests used to detect differencescombination of a coding faultand the instrument between the two data sets is quite sensitivemaker's design for the wind direction because of the large number of observationsinstruments. The F-Area tower showed five available.A summaryof the ability of statisticalsectors with differences but these may reflect a tests to determinedifferencesis as follows:combinationof naturalmeteorologicalvariabilityand differences in the methods used to createthe • Largesamplesizes providethe abilitydatabases" to detectsmalldifferencesbetween
categories.Tests on the differencebetween the two data sets • Small variabilityin meanwind speedsusing the standardizedadjustedresidual together providesthe ability to detect smallwith the Bonferroni method showed that a differencesbetweencategories.majority of stability categories and wind speed • Small sample sizes only provide theclasses weredifferent forboth F-AreaandH-Area abilityto detectlargedifferencestowers. The reasons for the differences in betweencategories.stability categories may be related to an increase • Largevariabilityin meanwindof 0.6° C in temperatureover the ten yearperiod, only providesthe ability to detectClimatic variations resulting from volcanic largedifferencesbetweencategories.activity, El Niflo, and deforestation could causedifferences both in stabilitycategories and speed Power spectrashow expected peaks at periods ofclassesbetween the datasets. one yeardue to the earth'srotationabout the sun,
3-6 days due to extratropicalcyclones, 24 and 12hours due to solar tides. The probable cause is
The majority of mean wind speeddifferences for not apparentforpeaks occurringat periods of 32-both H-Area and F-Area towers when examined 26 days and8 hours.by sector, and stability category showed no
12 93X057oaiwo
.- Comparison of Savannah River Site's Five-Year Meteorological Databases
B Bar charts were used to compare the distribution Spring and summer evening and midnightol of stability categories within wind direction patterns are very similar and all show minimas4 sectors. The frequency distribution by wind out of the north quadrant and maxima out of thedJ direction sector could be described as a trimodal south quadrant. Spring and summer mornings anddJ distribution with maxima at the NE, WSW, and middays resemble one another with maxima outS S sectors, of the WSW and NE or ENE. Autumn pre-dawn
and mornings are similar with local maxima
When the H-Area tower data is stratified by from the NE. Spring and summer afternoons ares¢ season and time-of-day some interesting patterns similar with maxima in the S to W quadrant.el emerge. The spring pattern shows a trimodal During spring and summer pre-dawn winds are
distribution with dominant winds from the S and fairly uniformly distributed with minima fromS; SSE. Northerly winds become very infrequent the WNW to N sectors.dl during summer and the south through west windsd_ dominate. In the autumn a dramatic shift occurs The new 87-91 H-Tower data set provides a goodtc to winds from the NE and ENE. In the winter basis for future comparisons since the additional
there is a very narrow peak of winds from the NE time and effort spent on quality assuranceal and a broad quadrant of winds from the S to provides more confidence that new information
WNW with a local maxima from the west. determined by inference and by statisticalWinds from the N and NNW are about equally methods is valid.
fr frequent for all seasons except summer.
m
9J 93xoszoawo 13
Comparison of Savannah River Site's Five.Year Meteorolo[_ical Databases
References
Dixon, W. J., and F. J. Massey, Jr., 1957: Michaels, P. J., 1991: Frosts and Freezes,Introduction to Statistical Analysis, 2nd Ed., Southeastern Climate Review, ISSN 1050-1428,McGraw-Hill, New York, pp. 275-278, and 412. Vol. 2/No. 4, Spring 1991. Southeast Regional
Climate Center, 1201 Main St. Columbia, SC
Hamby, D. M. ,1992: Atmospheric Transport 29201.31 pp.Modeling Utilizing Meteorological Data setsgenerated by Two Quality Assurance Methods. Parker, M. J., R. P. Addis, C. H. Hunter, R. J.Inter-Office Memorandum, SRT-ETS-920191. Kurzeja, C. P. Tatum, and A. H. Weber, 1992:Westinghouse Savannah River Company, Aiken, The 1987-1991 Savannah River SiteS.C. Meteorological Database (U). WSRC-RP-598.
Westinghouse Savannah River Company, Aiken,
Jenkins, (3. M. and D. G. Watts, 1968: Spectral S.C.Analysis and Its Applications. Holden-Day, 525pp. Stuart, W. F., Sherwood, V. and Macintosh, S.
M., 1971: The power spectral density technique
Laurinat, J. E., 1987: Average Wind Statistics applied to micropulsation analysis. Pure Appi.for SRP Area Meteorological Towers. DPST-87- Geophys., 92, pp 150-164.341. Savannah River Laboratory, E. I. DuPont,Aiken, S. C.
Comparison of Savannah River Site's Five.Year Meteorological Databases ,
Ha: Therearcno differencesbetweenthe distributionsin the old andnew datasets.
Differences were determinedin the numberof observations for eachof seven stabilitycategories, 16 winddirection sectors, and six wir.d speed categories by comparing the number observed with the numberexpectedm_ingthechi-squaredistribution.
Dcf'mitionof terms:
Let nij = numberof actualobservationsfor data set i (i=l for the 87-91 data set andi=2 for the 82-86data set) and class j, (j=l to 7 for stability categories,j=l to 6 for wind speed categories, andj=l to 16 forwind directionsectors.)
Ni+ = _nij = total numberobservations for data set i summedover all classes (j=l to 7 forstabilitycategories,j=l to 6 forwind _ categories,andj=l to 16 forwind directionsectors.)
N+j = gn U= total numberof observations in categoryj summed overboth data sets.
N = N1 + N2 = _N+j = totalnumberof observations in both data sets.
Under thenull hypothesis, the same proportionin each category shouldcome fromeach data seLThus, if pjis theproportionobserved in category-j overboth datasets, then the same proportionof each data set shouldalso be in category-j or
pj = N+j/N.
The expected numberof observationsin each class is given by
Fij =pj * Ni+ =Ni+ * N+j/N.
The squared difference between the observed and expected numberdivided by the expected numberofobservations is termed the adjusted residual squared.The adjusted residual squared is summed over allcategoriesfor bothdata sets.
rij2 = (nij- Fij)2/Fij, and
X2 = g,_.,rij2.
presented in Exhibit 1 for stability category andThe _2 ratio is distributed asa chi square random F-Tower data, Exhibit 2 for wind speed categoryvariablewith (I-1)0-1) degrees of freedomwhere I and F-Tower data, Exhibit 3 for wind direction= 2 and J = no. of categories, 6, 7 or 16. Since sectors for F-Tower data, Exhibit 4 for stabilitythere are only two data sets, the number of category and H-Tower data, Exhibit 5 for winddegrees of freedom is equal to the number of speed category and H-Tower data, Exhibit 6 forcategories minus one. wind direction sectors and H-Tower data. In
addition to the _2 defined above, two additionalSAS@ (Proc Freq) software was used to analyze chi-square variables are computed by [email protected] data. The results of the SAS® procedure are These are the likelihood ratio _2 and the Mantel-
93xo57o._veo 15
, Comparison of Savannah River Site'sFive.YearMeteorologicalDatabases
Haenszel Chi-Square.Probabilities of accepting periods for the jth category. The simultaneousthe null hypothesis based on these two additional probabilityof accepting all J comparisons whentests were approximatelythe same as for the chi- the null hypothesis is true was set to be 0.95squaretest discussed. These additionaltests will (95%confidence of accepting the null hypothesisnot be described. SAS® (Proc Freq)also when it is true.) In order to have thecomputes three tests forassociation between the simultaneous probability equal to 0.95, theclasses and the time periods. These are the Phi individualcomparisonprobabilitiesof acceptanceCoefficient, Contingency Coefficient, and were determinedusing the following BonferroniCramer'sV. Since no probabilitiesof acceptance method.nor rejection arecomputed, these were not usedandarenotdescribed. 0_2 ffi(1-0.95)/(72).
A frequency table is also included with the two For the seven stability categories, o,/2 ffi0.0036time periods (NEW for 87-91 data set and OLD and z(1-a/2) = 2.69. For the six wind speedfor 82-86 data set) as the two rows, and stability, categories, or/2 = 0.0042 and z(1-ot/'2) = 2.638.sector, or wind speed category as the column. For the sixteen sectors, a/'2 = 0.00156 andThe numbersin eachcell are labeled underperiod, z(1-cff2) = 2.955.The first value is the number observed, the
second is the number expected under the null Additionalstatistical tests were performedfor thehypothesis, the third is the cell chi-square null hypothesis that there is no difference in thecontribution or the adjusted residual squared mean wind speeds between the two data sets. To(rij2), the fourth is the percent of the total of test this null hypothesis, an analysis of varianceboth data sets in the cell and the last numberis (ANOVA) was done which modeled the meanthe percent of each data set in the cell. The .wind speedas a function of time period, stabilitymarginal totals for the rows include the total per category, and wind direction sector. The meanrow or data set and the percentof the total in each wind speedwas determinedfromthe originaldatadata set. The marginal totals for columns gives by grouping all the observations into 672the total per column or category and the percent categories (seven stability by six wind speed byof the total in each category which is the 16 sectors) and calculating the average windexpectedproportionunderthe null hypothesis, speed. These 672 meanwind speeds were the set
of observations analyzed by the analysis ofSAS® Proc Freq does not compute the adjusted variance.The wind speed class was not includedresidual divided by its standarddeviation. The in the model since there is a correlationbetweenresidual squaredis computed and labeled as the mean wind speed and wind speed category. Thecell chi-squarevalue. The standarddeviation(Sij) model also included the two-factor interactionsof theadjustedresidualis given by which are the interactionbetween time periods
and category and between stability and sectorcategories. The results aregiven in Exhibit 7 for
I Sij--"[(1-N+j/N)(1-Ni+/N)] 1/2. F Tower and Exhibit 8 for H Tower. Onlystability was a significant factor in the analysis
' The random variable, eij = rij/Sij, is distributed of variance.i approximately as a normal variate with mean1 zero and variance one. Since only two time An analysis of variance was also done withI periods are being compared, elj = - e2j for all j stability as a blocking factor (this compares onlyI categories. Thus, residuals need to be computed the differences between time periods within eachI foronly one data set anda two-tailed comparison stability category and not the averages betweenI basedon the normal distributionis made. stability category). When this was done, there
was no significant difference between timeI Since each individualcomparisonis distributedas periods. If there are significant differencesI a normal (0,1) variate, the probabilityof getting between time periods, the F-test on the period
a larger absolute value can be determined by within stability factor will be inflated and theI Prob{leijl< z(l-cx/2)} = (l-a) if the null probability of getting a larger F-value will bet hypothesis is true where z(1-a/2) is the small. This was not the case when both period1 100(1-_r2) percentile for the normaldistribution, within stability category and period within sector
category were compared.The F-test is similar to
1 Thus, if the calculated leijl ._z(l-o_f2),accept the the chi-square in that it only indicates significantI null hypothesis of no difference between time differences acrossall subeategories butnot which
subcategory(ies) are different. The results for
i 16 9_XO_7OMWO
, Comparison of Savannah Rive r $1tels Five.Year MeteoroloMcal Databases
differences between data sets within stability either anequal or unequalvariance t-test.SAS®category for F-Tower data is shown in Exhibit 9 ANOVA proceduredoes not test for equality ofand for sector in Exhibit 10. The results for variances. For these reasons, the two sample t-stability category for H-Tower data is shown in testswere used.Exhibit 11 and for sectors in Exhibit 12 for H-
Tower. SAS ® procedure"VI'ESTwas usedto computethe t-test or t.ratio between the two time periods
The 672 mean wind speeds for each time period foreach of the 7 stability categories, and 16 windwere averaged for each of the seven stability directioncategories. The mean windspeeds werecategories, sixteen sectors, and six wind speed averaged within each time period. TTESTcalegories within each time periodand compareA calculates the average, variance, standardusinga two samplet-test. There is one difference deviation,F-ratio, probabilityof getting a largerbetweenthe two sample t.test done separatelyfor F-ratio,t-ratio,and probabilityof getting a largereach category and the ANOVA. In the ANOVA Itl ratio assuming both equal and unequalthe variance estimates within time periods and variances. The ratio of the largest variance tocategories are pooled, while in the t-test, a smallest variance is compared with thepreliminaryF-test is done to determineif the true correspondingpercentilesof the F-distributionsovariances between time periods within each that two-tailed percentiles are used (folded Fcategory can be consideredequal, and depending statirtic).on the results of the F-test, SAS® computes
Let xijk = the kth meanwind speed for timeperiod i in categoryj, then
iij .= Y.Xijk/nijand
sij2 = E(Xijk - iij. )2/(nij- 1),
Then for each category -j, form the following ratio
F = largest of Slj 2 and s2j2/smallest of Slj2 and s2j2.
The probability computed by SAS® is a the probability of getting a larger ratio under theassumption
Again the probabilityof simultaneousacceptance However, SAS has already computed theof all J comparisonswhen the null hypothesis is probability that Itl > t(1-cx/2)(dfj) or Prob.true was set at 0.95 and the probability of {Itl>t(l-¢x/2)(dfj)} = a. So the Bonferroniaccepting each individual comparison was adjustmentfor a for the SAS two-sample t-testsdewrmined using the Bonferroni method. Each is given byindividual t-ratio is distributedas a student's t-
distributionwithdfj degrees of freedom.Thus, if ¢x= (1-0.95)/J.t(1-a/2)(dfj) is the 100(1-_'2) percentile of thestudent's t-distribution with dfj degrees of For the 7 stability categories ¢x= 0.00714. Forfreedom, then Prob.{Itl_ t(l.ct/2)(dfj)} = (1 - ¢x), the sixteen wind directions, cz = 0.003125 andwhere dfj is the degrees of freedom for the jth ¢x= 0.00833 for the six wind speedcategories.category. The Bonferronimethod requiressolvingthe following for ¢x Results from the SAS® TTEST procedure are
shown in Exhibits 13, 14, and 15 for stability,¢Z/2= (1-0.95)/(2"J). sector, and wind speed, respectively for the
F-Tower data and in Exhibits 16, 17, and 18respectively,for the H-Towerdata.
m
18 _Jxo57ouwo
Comparison of Savannah RiverSite'sFlve.YearMeteorologicalDatabases
References for Appendix A:
SAS@ User's Guide: Version 6.03, SAS® Satterwaite. F. W., 1946: "An ApproximateInstituteInc. Box 8000, Cary, N C. Distribution of Estimates of Variance
Components,"Biometrics Bulletin, 2, 110-114.Reynolds, H. T., 1977: The Analysis of Cross-Classifications, TheFree Press,N'Y. Miller, R. G., Jr., 1981: Simultaneous
Statistical Inference, Springer-Verlag,NY.
93XO570J_fWO
,, Comparison of Savannah RiverSite'sFlve.YearMeteorologicalDatabases ,
AppendiX B .................... ...................
Ap AppendixB containsa numberofplots forF and H.Areatowersforboth the 82-86 & 87-91 datasets whichare are too numerousto commenton individuallywithin thetext, nevertheless,occasions will arise when theseplo plots areneededforvariouspurposes.Theirbreakdownis as follows:
(1) (1) plots showinghow the stabilitycategoriesaredistributedover the speed classes (Figs. 24a-d)
(2) (2) plots showing how thespeedclasses aredistributedoverthe stability categories(Figs. 25a-d)
O) O) plots showinghow the speedclasses aredistributedoverthe directionsectors(Figs. 26a-d)
(4) (4) plots showing how the directionsectorsaredistributedover the stabilitycategories (Figs. 27a-d)
9Jxo 9Jxo_7oJ,_wo 21
, comp,a,r!sonof 5avannob River Site'S Five.Year MeteorolomtcalDatabases ,-
Exhibit 1. Tables from SAS® Proc Frequency for F.Area Tower. Comparisonof New versus Old databasesby the seven stabilityclasses, "New" r_fersto 1987-91 data,"Old" is 1982.86 data.
Exhibit2. As in Exhibit I except forcomparisonsby the six windspeed classes.
Exhibit 3, As in Exhibit 1 except forcomparisonsby the sixteen sectors,
Exhibit4, As in Exhibit 1 except for H-Areatower,
Exhibit5, As in Exhibit I except forH-Areatowerandcomparisonsby the six wind speed classes.
Exhibit6. As in Exhibit 1except for H-Areatower andcomparisonsby the sixteen sectors,
Exhibit 7. Tables fromSAS® Pro(:GLM(GeneralLinearModels) for the average speed from the F-Areatower, "STAB" refers to the seven stability classes, "SEC" refers to the sixteen sectors, and"PERIOD" refersto thetwo datasets,
Exhibit8. As in Exhibit 7 except for the H.Area tower.
Exhibit9, As in Exhibit 7 except thatthe two datasets"PERIOD" aremodeled as nested within the sevenstability"STAB" classes,
Exhibit 10, As in Exhibit 9 except for the two data.setsare modeled as nested within the sixteen sectors"SEC".
Exhibit 11, As in Exhibit 9 except for the H-Area tower,
Exhibit 12, As in Exhibit 10except for the H-Areatower,
Exhibit 13, Tables fromSAS@ Proc 'I'TESTforcomparing theaverage wind speed from the F.Area rowerbetween the twodatasets,"NEW" versus"OLD" within each of the seven stabilityclasses.
Exhibit 14. As in Exhibit 13except forcomparisons within each of the sixteen sectors.
Exhibit 15. As in Exhibit 13 except forcomparisonswithin each of the six wind speed classes.
Exhibit 16. As in Exhibit 13except for H.Area tower.
Exhibit 17. As in Exhibit 14 except for H.Area tower.
Exhibit 18, As in Exhibit 15 except for H.Area tower,
_ __m
9_xo_7oJ#_'o 23
......,Comoarlsonof Savannah River Site'sFive.year MeteoroloRlcal'Data#a,s,es,,,
Thispageintentionallyleftblank
2_ 93xoJ70MWO
Comparison of Savannah River Site's Five.Year Meteorological Databases
Figures
Figure 1. Percentage of H-Area tower Figure 13a. Spectral density estimates of one-observations from the 82-86 (solid) and 87-91 hour mean wind speeds for the 87-91 H-Area(dashed)databasesforeachwinddirectionsector towerdatabase.Spectraldensityestimateshave(1-16). units of power (m2/sec 2) per unit frequency
(cycles/hr) and are plotted against the periodFigure 2. Percentage of F-Area tower (hrs/cycle).observations from the 82-86 (solid) and 87-91(dashed)databases foreach wind directionsector Figure 13b. Same as Fig. 13a except for a(1-16). limitedrangeof theperiod.
Figure 3. Percentage of H-Area tower Figure 13c. Same as Fig. 13b except for aobservations from the 82-86 (solid) and 87-91 limitedrangeof theperiod.(dashed) databases for each stability class(A-G). Figure 13d. Same as Fig. 13c except for a
limitedrangeof the period.Figure 4. Percentage of F-Area towerobservations from the 82-86 (solid) and 87-91 Figure 14a. Spectral density estimates of fifteen-(dashed) databases for each stability class minute mean wind speeds for the 87-91 H-Area(A-G). towerdatabase.Spectral densityestimateshave
units of power (m2/sec2) per unit frequencyFigure 5. Percentage of H-Area tower • (cycles/hr)and arc plottedagainstthe periodobservationsfrom the 82-86 (solid) and87-91 (his/cycle).(dashed)databasesforeachwindspeedclass (O-
2- >12). Figure 14b. Same as Fig. 14a except for alimited range of the period.
Figure 6. Percentage of F-Area towerobservations from the 82-86 (solid) and 87-91
Figure 14c. Same as Fig. 14b except for a(dashed)databas_ foreach windspeedclass (0-2 -- _12). limitedrangeof the period.
Figure 7. Mean wind speeds for the H-Area Figure 14d. Same as Fig. 14c except for atower from the 82-86 (solid) and 87-91 (dashed) limitedrangeof theperiod.databasesforeach winddirectionsector (1-16).
Figure 15a. Percentage of H-Area towerobservationsfrom the 87-91 databases for each
Figure8. Mean wind_ for the F-Areatowerwind direction sector (1-15) broken down byfrom the 82-86 (solid) and 87-91 (dashed)
databasesfor each winddirectionsector (1-16). stabilityclass.
Figure9. Meanwind _ for the H-Areatower Figure 15b. As for 15aexcept for82-86.from the 82-86 (solid) and 87-91 (dashed)databasesforeach stabilityclass (A-F & G). Figure 15c. As for 15a except for the F-Area
tower.
Figure 10. Mean wind speeds for the F-Areatower from the 82-86 (solid) and 87-91 (dashed) Figure150. As for 15c except for82-86.da_ -basesfor each stability class (A-F & G).
Figure 16a. Frequencyof occurrence of H-AreaFigure 11. Mean wind speeds for the H-Area towerobservations from the 87-91 databases fortower from the 82-86 (solid) and 87-91 (dashed) each wind directionsector (1-16) brokendown bydatabasesfor eachwindspeedclass(1-6), stability class for the spring season (March,
April, May).
Figure 12. Mean wind speeds for the F-Areatower from the 82-86 (solid) and 87-91 (dashed) Figure 16b.As for 16aexcept for summer(June,databasesforeachwind_ class (I-6). July, August).
93xo57o._wo 25
Comparison of Savannah River Site's Five.Year Meteorological Databases
Figure 16c. As for 16a except for autumn Figure 22b. As for 17a except for autumn and(September,October,November. evening.
Figure 16d. As for 16a except for winter Figure 22c. As for 17a except for autumn and(l_ember, January,February). afternoon.
Figure 17a. As for 16a except for evening Figure 22d. As for 17a except for autumn and(between sunset and true solar midnight minus midday.1/6 of the nighttimehours).
Figure 23a. As for 17a except for winter andFigure 17"o.As for 17aexcept for midnight(time morning.periodbetween end of evening and beginning ofp_-dawn). Figure 23b. As for 17a except for winter and
midday.Figure 17c. As for 17a except for summerandevening. Figure 23c. As for 17a except for winter and
afternoon.Figure 17d. As for 1To except for summerandmidnight. Figure 23d. As for 17a except for winter and
evening., , Figure 18a. As for 17a except for spring and
morning (between sunrise and true solar noon Figure 23e. As for 17a except for winter andminus 1/6 of the daylighthours). .midnight.
Figure 18b. As for 17a except for spring and Figure23t".As for 17aexcept for winterand pre-midday (time period betweenend of morningand dawn.beginning of afternoon).
Figure 24a. Percentage of H-Area towerFigure 18(:.As for 17a except for summer and observations from the 87-91 database for eachmorning, speed class (0-2 m >12) broken down by
stabilityclass (A -- F-G).Figure 18d. As for 17a except for summer andmidday. Figure24b. Same as Fig. 24a except for 82-86.
Figure 19a. As for 17a except for autumn and Figure 24c. Same as Fig. 24a except for the F-pre-dawn (between sunrise and true solar Areatower.midnight plus 1/6 of the nighttimehours).
Figure 24d. Same as Fig. 24c except for82-86.Figure 19b. As for 17a except for autumnand
morning. Figure 25a. Percentage of H-Area towerobservations from the 87-91 database for each
Figure 20a. As for 17a except for spring and stabilityclass (A -- F-G) broken down by speedafternoon (between sunset and true solar noon class (0-2 -- >12).plus 1/6 of the daylight hours).
Figure25b. Same as 25a except for 82-86.Figure 20b. As for 17a except for summer and
afternoon. Figure 25c. Same as 25a except for the F-Areatower.
Figure21a. As for 17aexcept forspringand pre-
dawn. Figure 25d. Same as 25c except for82-86.
Figure 2lb. As for 17a except for summer and Figure 26a. Percentage of H-Area towerpie.dawn, observations from the 87-91 database for each
direction sector (S -- SSE) broken down byFigure 22a. As for 17a except for autumn and speed class (0-2- >-12).midnight.
26 9Jxo.f70Mwo
.. Comparison of Savannah River Site's Five.Year Meteorological Databases ,
E Figure26b. Same as Fig. 26a except for82-86.
Ft Figure 26c. Same as Fig. 26a except for theF, F-Are,a tower.
Ft Figure 26d. Same as Fig. 26c except for82-86.
FJ Figure 27a. Percentage of H-Area towerol observations from the 87-91 database for eachst stability category (A-G) broken down by sector(S (S -- SSE).
Ft Figure 27b. Same as 27a except for 82-86.
Ft Figure 27c. Same as 27a except for the F-Areatower.
Fl Figure27d. Same as 27c except for 82-86.
,, , Comparison of Savannah River Site'sFive.Year MeteorologicalDatabases ,
This page intentionally left blank
93XOJTOMWO
DATA FROM _f AREA TOWER 09:25 Friday, January 22, 1993 7
TABLE OF PERIOD BY STAB
PERIOD STAB
Frequency JExpected lCell Chi-SquarelPercent IRow Pct l iI 21 31 41 51 61 71 Total............... +........ +........ +........ +........ +........ +........ +........ +
NEW l 8092 l 4083 1 7092 l 12095 1 8661 1 1638 I 157 1 41818J 7798.3 j 4007.7 J 6835.1 j 11876 j 9256.3 J 1886.4 J 158.59 JI 11.058 I 1.4155 I 9.6561 J 4.0541 I 38.284 I 32.716 I 0.016 ll 10.77 J, 5.43 I 9.44 J 16.09 1 11.53 I 2.18 I 0.21 I 55.65I 19.35 I 9.76 I 16.96 I 28.92 I 20.71 1 3.92 I 0.38 I
OLD J 5922 j 3_19 j 5191 J 9246 I 7973 J 1752 I 128 I 33331I 6215.7 I 319_.3 I 5447.9 I 9465.4 l 7377.7 I 1503.6 i 126.41 jI 13.874 l 1.7759 I 12.115 l 5.0864 I 48.032 l 41.046 I 0.0201 II 7.88 j 4.15 I 6.91 I 12.30 I 10.61 I 2.33 I 0.17 j 44.35I 17.77 I 9.36 I 15.57 I 27.74 I 23.92 I 5.26 I 0.38 I
Frequency TExpected ICell Chl-SquarelPercent IRow Pct l II 21 31 41 5l 61 TotAl
+ + + + . . +
NEW J 3261 j 16978 J 17234 J 3562 I 759 J 24 j 41818I 4037,2 j 16749 I 16456 I 3764.5 I 792,97 I 17.807 lI 149,22 I 3,1276 J 36,741 J 10.893 i 1,455 i 2.1539 il 4.3_ I 22.59 I 22,93 I 4.74 I 1.01 I 0,03 I 55.65l 7,80 I 40,60 I 41,21 I 8,52 I 1,82 I 0,06 I+ + + + + + +
OLD ! 3994 i 13121 I 12339 I 3203 I 666 I 8 I 33331) 3217.8 I 13350 I 13117 I 3000°5 I 632,03 I 14,193 II 187.22 I 3.9239 I 46,096 I 13,667 I 1,8254 I 2,7023 II 5.31 I 17,46 I 16.42 I 4.26 I 0.89 i 0,01 I 44.35I 11.98 I 39.37 I 37.02 I 9.61 I 2,00 I 0.02 I
Chi-SquAre 5 459.032 0.000Likelihood Ratio Chi-SqruAre 5 456.744 0.000Mantel-Haenszel Chl-SquAre 1 83.583 0,000Phi Coefflolent 0,078Contingency Coefficient 0,078CrAmer'a V 0,078
Sample SiZe - 75149
EXHIBIT 2
DATA FROM _F_ _,qY_ATOWER 09:25 Friday, January 22, 1993 9
TABLE OF PERIOD BY $EC
PERIOD SEC
Frequency IExpected iCell Chi-Square IPercent iRow Pet I 11 21 3f 41 51 61 71 81 Total............... +........ +........ +........ +........ +........ +........ +........ +........ +NEW I 1025 I 2362 1 4302 1 3608 1 2111 1 1725 1 1951 1 2950 1 41818
) 1084 i 2249.2 0 4179.6 I 3725 I 2262.6 I 1731.2 I 1967.7 I 2825.7 1i 3,2112I 5.6526i 3.5827i 3,6746i 10.157I 0.02210.1412i 5.4639It 1.36I 3,_4 I 5.72 I 4.80 I 2.81f 2.301 2.60 _ 3.93 I 55.65I 2.451 5 _5 i 10,29t 8.63 I 5,051 4.131 4.67I 7,05 I
I 664I 1792.8I 3331.4a 2969_ 18034 i 1379., , 15683 , 2252.3,i 4,0288 1 7.092 I 4.495 I 4.6103 I 12.744 I 0.0276 I 0.1772 I 6.8552 II 1.23 i 2,24 I 4.27 I 4.11 I 2.60 I 1,84 I 2,11 i 2.83 t 44.35I 2.771 5.04 i 9.631 9.261 5.671 4.161 476a 6.38i
NEW I 2995 I 2607 I 3329 I 3697 I 4065 I 2658 I 1402 I 1031 I 41818I 2920.3 I 2657.1 I 3288.2 I 3748.4 I 3999.9 I 2755.1 I 1419.5 I 1004.4 II 1.9086 I 0.9459 I 0.507 i 0.7039 I 1.0598 I 3,4202 i 0.217 I 0.7032 II 3.99 I 3.47 I 4.43 I 4.92 I 5.41 I 3.54 I 1.87 I 1.37 I 55.65J 7.16 I 6.23 I 7,96 J 8.84 I 9.72 I 6.36 I 3.35 I 2,47 I
............... +.... .... +........ +........ +...... _.+ ........ +.... ....+........ +........ .OLD I 2253 I 2168 I 2580 I 3039 I 3123 I 2293 I 1149 I 774 I 33331
I 2327.7 I 2117.9 I 2620.8 I 2987.6 I 3!88.1 I 2195.9 i 1131.5 I 800.58 iI 2.3945 i 1.1868 I 0.6361 I 0.8832 I 1.3297 I 4.2911 I 0.2722 I 0.8822 II 3.00 I 2.88 I 3.43 I 4.04 I 4.16 I 3.05 I 1.53 I 1.03 I 44.35I 6.76 I 6,50 I 7,74 I ._,12 I 9,37 I 6,88 i 3,45 I 2,32 I+ + + + + + + + +
I 4087,8 I 19570 14914 2744,4 523,17 J 4,8870 II 1.4430 83,741 18,794 94.044 24,913 I 1,?$98 II S.2A 26.14 20,68 2.13 0.52 I 0,00 I 55.42I 9.48 I 47.20 37,31 5.09 0.93 I 0,00 I
oLD , 31,si 14e25 141,9 2721, s3s, ? , 3525_, 3272,2, 15742 13,0_ 220?.4, 420.83 , 4.0122 ,, I.,I,7 , 79.286 23.352 119.4, 30.971 , 2.225 ,l 4,04 i 18.4' 17.92 3.44 i 0.48 I 0,01 i 44,59, 9.01, 41.49 40.I0 ?.?_, 1.52 , 0.02 ,
i 10,100 0.3107 2.135t 0.1112 2.7143 0,3143 I 0.245 I 1.S480 I1.ii 2.11 4,2i 3.4_ _.S2 2.11 I 2,21 I 2,i_ I 44.5iI 3,?? 1,07 I,St ?.iS S,15 4.55 I 4,1| I 4.03 I
PERIOD N Mean Std Dev 8td Error Minimum Maximum Variances T DF Prob>lT].........--............................................................................... ----.......-.......----------.----------------
Pit.DO N _an |_d Dev Std Error NLnLnun HaxLnus VsrLinoes T Dr Prob)(T/mmaeo_e.mmuaamm_mmmmm_ _ O_e_mmaammeeo_uu_mm_eemm_meabem_emauammom_ammmmom_m*U_m_ _mem_ee_om_mmem_muo.o*mmu.mommmuem_mo
_I_ZOD N Heart I_d Dev |_d |_tor HLnLmum NaxLm_a VirLinaem Y Dr Prob)lTI_e__e_me_m_m_mem_mem_s__m_mmmmamemmeee_eee_m_m_ Bmmm_ee_me_mmmwm_mm_me_mm_m_e_me_e.
i_XOD JI_tZOD N I_lln Itd Dev itd |trot NinLmum HauLmue VsrLinael ? Dr Prob)lTI_m_mmm_m_mmmum_m_u_Imum_m_m_ammmmm6_Im_mmmmm_m_am_m_mI_mmm_mmmmmmum_mummmmu_m_m8_m_mmm mmm_O_mmm_mmmmmmII._Dmmammm._m.wmw_I.
_._OD _q_OD N i_ln Jtd Dev 8_d |rror HLntmue Haximum Varlan0e8 ? Dr Pr0b)lTI_e_m_me_m_mm_m__j_m__a__j_u_e_m_m_e__m_s_j uaJeaoeeoOsJt_Ooemae_6a_u_euJmaoeeeeme
PrJUOD N ICeln ltd D¢i¥ ltd irro_ NLnLuum NaxLaum VirLanoe| T Dr Prob>lTI_e_O_om_me_u_m_Bum_a_mm_Iu_u*D_ea_a_m_m_d_mu_a_nm_u_ eeee_n_eeeeou.eeoe.e_emuaeeo_umeeeeu.e
PERIOD N Mean Std Dev Std Error Minimum Maximum Variances T DF Prob>ITi............ . ............ . ........................................................................................... . ...........
PERIOD H Mean Std Dev Std Error Minimum Maximum Variances T DP Prob>IT#... ................................................ ... ....... .... .... ... ........................................................
,r H0:Ir H0: Varlances are equal, F' - 1.17 DF - (90,101) Prob>F' - 0.4496
EXIilI EXIIIBIT 18
DATA FROM H AREA TOWER 13:23 Friday, January 22, 1993 14
TTEST PROCEDURE
Vs: SPDm: SPD
P| ! N Mean Std Dev 8td Error Minimum Maximum Variances T DF Prob>ITl._ .... .,-........-............_.__.__.. ............. .........-..-........................ ..... ....... . ........... .... ......... ..