Computers& GeosciencesVol. 10. No.4. top.397-410,1984 0098 3004/84 $3.00+ .

00
printedin the U.S.A. PergamonPressLtd.

A PROCEDURE FOR THE CLASSIFICATION OF

SYNOPTIC WEATHER MAPS FROM GRIDDED
ATMOSPHERIC PRESSURE SURFACE DATA

BRENT YARNALt
Department of Geography, The Pennsylvania State University, University Park, PA 16802, U.S.A.

(Received 17 May 1983; revised 29 September 1983)

A~traet--Although synoptic climatological research includes two stages, the classification of daily
weather maps and the evaluation of the relationships of these synoptic type categories with local weather
elements, most expenditures of time and money are on classification procedures. To alleviate this
problem, a series of 6 FORTRAN programs for the classification of synoptic weather maps from gridded
atmospheric pressure surface data is presented. An example illustrating the applicaton of the procedure
is discussed.

Key Words: Synoptic climatology, Map pattern classification, Objective synoptic typing, Kirchhofer
technique, Sums-of-squares classification procedure.

INTRODUCTION vance in synoptic climatology in the last few years

Synoptic climatology is concerned with understanding (Barry, 1980). An objective synoptic climatology takes
local or regional climate by examining the relationship advantage of the large data handling capacity of mod-
between local weather elements and the circulation of ern computers, allowing the machine to determine sta-
the atmosphere over that area. Based on this tistically similar and significant synoptic groupings.
definition, Barry and Perry (1973) have identified two When based on standardized criteria and data, results
stages in any synoptic climatology. The first stage can be replicated and studies can be compared. Stan-
involves the categorization of atmospheric circulation dardized gridded climatological data sets, such as
patterns, usually in the form of synoptic weather those prepared by the National Center for Atmos-
maps. The second stage is the assessment of the re- pheric Research (Jenne, 1975) are ideally suited to
lationship of these categories with local weather ele- objective synoptic typing (Barry, 1980).
ments. Two main types of objective classification pro-
Much of the work in synoptic climatology has been cedure have been developed, the correlation method
directed towards the formulation of effective synoptic (Lund, 1963) and the sums-of-squares technique
type categories. Although there is a wide range of (Kirchhofer, 1973). The Lurid correlation method
research concerned with this topic, the resulting typically classifies 60-80~o of the maps analyzed, al-
classification techniques can be grouped under the though Petzold (1982) has developed a technique to
headings of subjective and objective procedures. improve significantly the percentage of maps classified
There are two major problems with subjective typ- by this method. The sums-of-squares technique is an
ing techniques. First, although some subjectively de- improvement over the correlation techniques, typi-
rived classifications are well established with accepted cally categorizing more than 90~ of the weather maps.
criteria for the determination of synoptic types, such In this paper, a procedure is presented for classi-
as the surface airflow-type catalog for the British Isles fying synoptic weather maps from gridded NCAR
(Lamb, 1972), most subjective classifications are data using the sums-of-squares technique. The origi-
unique, with typing criteria derived only for the re- nal algorithm was developed by Kirchhofer (1973) and
search at hand. Consequently, the results of one study applied to the climate of Switzerland (Kirchhofer,
are not compared easily to the findings of another. 1976). Barry and Eccles (1981, pers. comm.) later
Second, atmospheric features are continuous in time developed a F O R T R A N version of this technique for
and space, so that the identification of synoptic type use with NCAR data sets on CDC computers. Their
boundaries must be arbitrary. Even when clearly approach has been employed by Barry and Keen
defined criteria are applied, replication of type catego- (1978), Moritz (1979), Bradley and England (1979),
ries and type member frequencies is difficult because and Keen (1980) for areas of the North American
the judgement of any two investigators will differ. Arctic and by Barry, Kiladis, and Bradley (1981) for
Due to the problems associated with subjective pro- the American west. I have rewritten the CDC version
cedures, the development and application of objective of the Kirchhofer algorithm for use at IBM installa-
classification techniques is the most important ad- tions. This procedure has been applied to two different
t Visiting Fellow (1984-85), Cooperative Institute for scales of synoptic activity over southwestern Canada
Research in Environmental Sciences, University of Colorado/ (Yarnal, in press), one of which is presented to help
NOAA, Boulder, CO 80309, U.S.A. describe the algorithm.

397
398 BRENT YARNAL

while the study period includes the 9 glaciological yr

from 1 October 1965 to 30 September 1974 of the
International Hydrological Decade (IHD; 1965-74).
The results of the analysis are used to relate the atmos-
pheric circulation over southwestern Canada to gla-
cier mass balance in the region (Yarnal, in press).
Later, the results are discussed in terms of the climate
of southwestern Canada.
To apply the Kirchhofer technique, the daily grids
first are normalized using program N O R M A L :

(xi- ~)
Z, = - - (1)
S

where Zi = normalized value of grid point i, xi = data

value at grid point i, f = mean of the N-point grid,
and s = standard deviation of the grid.
In SCORES, each normalized grid is compared to
all other grids by the sums-of-squares equation:
N
Figure 1. National Meteorological Center (NMC) 500 mb s = y~ ( Z o , - z ~ , ) ~ (2)
grid points used in the analysis. i=1

THE PROCEDURE where S = Kirchhofer score, Zai = normalized grid

The classification procedure uses prepared sets of value of point i on day a, Zbi = normalized grid value
gridded atmospheric data. Reasonably priced data are of point i on day b, and N = number of data points.
available on magnetic tape from N C A R (Jenne, 1975), It is possible for a comparison of any two grids to
where several atmospheric pressure surfaces have been generate a low S value, denoting overall statistical
fit to the 1977-point National Meteorological Center similarity, and yet have widely differing patterns in
(NMC) grid of the Northern Hemisphere. The data specific sectors of the map. Therefore, to insure
come in packed binary form and are unpacked using pattern similarity in all areas of the grids, subscore
subroutines provided by N C A R (Jenne and Joseph, values for each row and column of the 5 by 6 matrices
1974), although modifications to the F O R T R A N AS- are calculated using equation (2).
SEMBLER language are probably necessary. The Grids are considered similar if S < 15 (in other
programs presented here (Appendix 1) are designed words, 0.5N) and row and column scores SR and Sc
for rectangular matrices, but can be changed to ac- 1.0NR and l.ONc, where N R and N c equals the
commodate uneven row and column lengths. number of points in the row or column, respectively.
In the present example, a daily 1200 G M T synoptic- Mean sealevel pressure surfaces using the Kirchhofer
scale weather map is created by extracting every other method (such as Moritz, 1979) have used thresholds
grid point from an 11 by l0 section of the N M C for S of 1.0N and for row and column scores of 1.8NR
500 mb grid to produce a 30-point, 5 by 6 matrix (Fig. and 1.8No However, to provide a comparably high
1). The mapped area covers the northeastern North percentage of classified days and number of patterns,
Pacific Ocean and adjacent areas of North America, a lower threshold score must be used for upper air

Table 1. Summary of synoptic type classification

FREQUENCY MEAN KIRCHHOFER KIRCHHOFER SCORE

TYPE (%) SCORE STANDARD DEVIATION

1 60.0 5.5 2.5

2 12.0 7.0 2.6

3 7.2 7.3 2.8

4 7.8 7.6 2.7

5 3.3 8.4 2.4

6 2.9 8.4 2.6

7 0.6 10.5 3.7

8 0.7 9.7 3.5

Unclassified 2.9 . . . .

Missing data 2.6 . . . .

 Procedure for the classification of synoptic weather maps 399

maps (Barry, Kiladis, and Bradley, 1981), as in the size. This study uses 4 sample years in which it was
present example. assumed the widest range of synoptic patterns occurred
Due to computer storage limitations and cost for the 9 glaciological years of the study period.
considerations, it is probably necessary to use a Thirty-nine of the 1463 sample days were eliminated
sample of the days. Sample size may be determined prior to analysis because of missing or questionable
by hardware limitations and/or restrictions imposed data in the NCAR 500 mb data set, leaving 1424
by a computer facility. The algorithm used to deter- sample grids. Although just over one megabyte was
mine Kirchhofer scores (S) requires: needed for SCORES, approximately 56 rain of CPU
time were used by this stage in the analysis.
n(n - 1) In SCORES, S, SR, and Sc values are calculated for
Space= 2 (3) every pair of grids in the sample. If the previously
mentioned threshold requirements are met, the pair
where Space = number of bytes in central memory. of grids is considered significantly similar and the S
For example, the computer facility used for the value is entered into memory. All significant pairs of
present analysis imposes a 1.6megabyte memory daily grids from SCORES are entered into program
limit per user, thereby restricting the sample size to KEYDAY. In this section of the procedure, the daily
less than 1800 days. Costs also may limit sample grid with the most S values associated with it is

KEYDAY 2, 30 MAR 67

Fig. 2(a).
400 BRENT YARNAL

KEYDAY 5, 18

KEYDAY 6, 29 NOV 66 ~"/~ f L~ " /

Figure 2. (a) 500 mb pressure distributions on keydays 1-4; elevations in meters above sealevel. (b) 500
mb pressure distributions on keydays 5-8; elevations in meters above sealevel.

designated Keyday 1. That keyday is then removed possible for any day to be related significantly to
from the analysis along with all grids associated with more than 1 keyday, days misclassified by early
the keyday and all days associated with those days. removal during the keyday determination procedure
This process is repeated to determine subsequent described above are reclassified. The results of SYN-
keydays until all days are classified into m groups of TYP are first printed, to produce a synoptic type
5 days or more. Remaining days are termed catalog for the entire study period, and then entered
"unclassified" and a listing of the m keydays is into program SORT, to produce a summary of the
printed by KEYDAY. If these results are satis- classification.
factory, they are entered into KEYMAP, which
prints the keyday grids. These may be entered into a RESULTS
computer mapping program. Table 1 is adapted from the output of SORT and
In program SYNTYP, S, S~, and S c values are summarizes the results of the synoptic type
calculated again, this time for each of the m keydays classification. The 8 synoptic types recognized account
with each day of the total population of daily grids. for more than 94% of the days in the study period. The
The lowest significant Kirchhofer score is recorded first 4 types account for 87% of the days, with higher
for each daily grid, with the associated keyday de- order types being much less important in terms of
noting the synoptic type of the day. Because it is frequency. Mean Kirchhofer scores are well below the
 Procedure for the classification of synoptic weather maps 401

threshold value of 15.0 and Kirchhofer score standard With the dissemination of objective synoptic type
deviations are relatively small, suggesting the classification procedures based on standardized data
classification fit is good. Kirchhofer scores are related sets, synoptic climatologists can spend less time on
inversely to frequency, indicating greater pressure pat- classification and more time on the evaluaton of the
tern diversity in less frequent types (Bradley and En- relationships of synoptic type categories with local
gland, 1979). Keyday maps for the 8 synoptic type weather elements.
categories are shown in Figures 2(a) and 2(b).
Acknowledgments--I wish to acknowledge the following
DISCUSSION persons, Brian Sagar, Simon Fraser University, for super-
Inspection of the keyday maps reveals much about vising this research, John Knox, University of British Col-
umbia, for providing the "clean" NCAR 500mb data,
the climate of southwestern Canada, especially in
Mark Roseberry for consultation on many aspects of the
terms of the regional airflow patterns and associated programming, Roger Barry and Margaret Eccles, Univer-
temperature and precipitation characteristics. Type 1, sity of Colorado, for providing their version of the sums-of-
which occurs on 60% of all days and therefore domi- squares program, Careen Mackay Yarnal for producing the
nates the climate of the study period, displays strong figures, and the reviewers of this article for suggesting
improvements to the text. Research expenses were covered
westerly flow. Types 2 and 3 together account for by the Department of Geography, Simon Fraser University.
nearly 20% of the days and represent increased oscil-
lation in the westerlies. The less common Types 4-8 REFERENCES
illustrate north-south flow with meridional exchange Barry, R. G., 1980, Synoptic and dynamic climatology:
of cold and warm air. Prog. Phys. Geogr., v. 4, no. 1, p. 88-96.
Clear relationships between the airflow patterns of Barry, R. G., and Keen, R. A., 1978, Regional climatic
the synoptic types and regional temperatures are evi- setting, in Barry, R. G., and Jacobs, J. D., eds., Energy
budget studies in relation to fast-ice break-up processes in
dent. Type 1 days bring moderate temperatures to Davis Strait: climatological overview: Inst. Arctic and
southwestern Canada. This type is associated closely Alpine Res. Occas. Paper no. 26, Univ. Colorado,
with seasonal and annual mean temperatures because Boulder, p. 8-67.
of its high frequency and its westerly maritime flow. A Barry, R. G., Kiladis, G., and Bradley, R. S., 1981, Synoptic
climatology of the western United States in relation to
northwesterly trajectory maintains cooler conditions
climatic fluctuations during the Twentieth Century: Jour.
in the region during all Type 2 situations, while south- Climatol., v. I, no. 2, p. 97-113.
westerly flow keeps Type 3 temperatures above aver- Barry, R. G., and Perry, A. H., 1973, Synoptic Climatology,
age. Because southwestern Canada is positioned be- Methods and Applications: Methuen, London, 555 p.
tween two connecting ridges and experiences Bradley, R. S., and England, J., 1979, Synoptic climatology
of the Canadian High Arctic: Geogr. Annaler, v. 61A, no.
southerly airflow on Type 4 days, temperatures are 34, p. 187-201.
relatively high. Of the less frequently occurring Jenne, R., 1975, Data sets for meteorological research: Nat.
500 mb patterns, Types 5 and 6 are usually warm, Center for Atmos. Res. Tech. Note NCAR-TN/IA-111,
Type 7 generally advects cold air into the region, and Boulder, Colorado, 194 p.
Type 8 is associated with outbreaks of Arctic air. Jenne, R. L., and Joseph, D. H., 1974, Techniques for the
processing, storage, and exchange of data: Nat. Center for
Regional precipitation usually can be explained by Atmos. Res. Teeh. Note NCAR-TN/IA-93, Boulder, Col-
the 500 mb wave patterns. For example, Type 1 days orado, 46 p.
in southwestern Canada are associated with moist Keen, R. A., 1980, Temperature and circulation anomalies in
westerly cyclonic flows, upward vertical motion, and the eastern Canadian Arctic, summer 1946--76:Inst. Arctic
maximum vorticity advection ahead of the main and Alpine Res. Occas. Paper no. 34, Univ. Colorado,
Boulder, 159 p.
trough axis. On the other hand, Type 2 days are Kirchhofer, W., 1973, Classification of 500mb patterns:
associated with divergence and subsidence of rela- Arbeitsbericht Schweiz. Meteorol. Zentralanstalt Nr. 43,
tively less moist northwesterly flows behind the main Zurich, 16 p.
trough axis. Type 1 patterns are highly efficient bearers Kirchhofer, W., 1976, Stationsberzogene Wetterklassifik-
of precipitation, delivering approximately 70% of the ation: Veroffentlichungen Schweiz. Meteorol. Zentralan-
stalt Nr. 34, 50 p.
annual precipitation of the region while occurring on
Lamb, H. H., 1972, British Isles weather types and a register
only 60~o of the days. Type 2 is associated with lower of the daily sequence of circulation patterns, 1861-1971:
annual precipitation totals (approximately 99/0) than Geogr. Mem. no. 16, 85 p.
its relative frequency of occurrance (12%). Lund, I. A., 1963, Map-pattern classification by statistical
A comparison of the 8 keyday maps to a smaller, methods: Jour. Appl. Meteorol., v. 2, no. 1, p. 56--65.
more detailed grid with 18 types classified by the sums- Moritz, R. E., 1979, Synoptic climatology of the Beaufort
Sea coast: Inst. Arctic and Alpine Res. Occas. Paper no.
of-squares procedure is presented by Yarnal (in press). 30, Univ. Colorado, Boulder, 176p.
Relationships between the synoptic types and glacier Petzold, D. E., 1982, The summer weather types of Quebec-
mass balance in southwestern Canada also have been Labrador: McGill Subarctic Res. Paper no. 34, McGill
established (Yarnal, in press). Univ., Montreal, 160p.
Yarnal, B., The effect of weather map scale on the results
SUMMARY AND CONCLUSIONS of a synoptic climatology:Jour. Climatol., v. 4 (in press).
Yarnal, B., Relationships between synoptic-scale atmos-
A series of 6 F O R T R A N programs for the pheric circulation and glacier mass balance in southwest-
classification of synoptic weather maps from gridded ern Canada during the International Hydrological De-
atmospheric pressure surface data was presented. cade, 1965-74: Jour. Glaciol. (in press).
402 BRENT YARNAL

APPENDIX

Program listing
CJJJJJJiJJJJJJJJJJJJJJJJJJJJJJJJJJJJJ~JJJJJJJJJJJeJJJJJJJJJJJJJJJJJJJJJJ
C•
C • PREFACE: BEFORE USING THE PROCEDURE, THE GRIDDED DATA SHOULD
C* BE CHECKED FOR CONFORMITY AND CLEANSED OF ANY GRIDS
Ci THAT HAVE BAD OR MISSING DATA. THE TOTAL NUMBER OF
C• GRIDS TO BE ANALYSED MUST BE ENTERED INTO PROGRAMS
C• SYNTYP AND SORT. THE NUMBER OF SYNOPTIC TYPES
C• RESULTING F R ~ THE ANALYSIS IS DETERMINED BY THE
C• THRESHOLD VALHES ASSIGNED TO SCORE, RSCORE, AND
C• CSCORE WHICH ARE FOUND IN PROGRAMS SCORES AND SYNTYP.
C• FOR AN EXCELLENT DESCRIPTION OF THE THRESHOLD PROBLEM
C* AND FOR WAYS TO USE THRESHOLD VALUES TO CONTROL THE
C• RESULTS OF THE ANALYSIS, SEE MORITZ (1979).
Ci
Cl
C• NORMAL, A PROGRAM TO:
C• I) READ A PREPARED DAILY GRID FROM TAPE 8 (GRID);
C• 2) CALCULATE THE MEAN AND STANDARD DEVIATION OF THE GRID;
C• 3) NORMALIZE THE GRID;
C* 4) WRITE THE YEAR, MONTH, DAY, HOUR, MEAN, STANDARD
C• DEVIATION AND NORMALIZED VALUES ONTO A NEW GRID
C• (GRIDR);
C* 5) CONTINUE TO THE NEXT DAY'S GRID.
C•
C* NOTE: SIZE OF THE ARRAY MUST BE DIMENSIONED; FIRST ROW OF THE
C* ARRAY CONTAINS INFORMATION ABOUT THAT GRID.
C•
CJJJJJNJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJ
C*
C* DIMENSION THE ARRAYS AND READ A DAILY GRID.
C•
DIMENSION GRIDR(6,6)
INTEGER•2 GRID(6,6),LEN
11 CALL READ (GRID,LEN,0,LNUM, 8,&99)
C•
C* CALCULATE THE MEAN AND STANDARD DEVIATION.
C•
XMEAN=0.0
SD = 0.0
COUNT = 0.0
DO 12 I = 2,6
DO 13 I = 1,6
X = (GRID(I,J) + 50000)/10.0
XMEAN = XMEAN + X
SD = SD + (X • X)
COUNT = COUNT + 1.0
13 CONTINUE
12 CONTINUE
XMEAN = XMEAN/COUNT
SD = SQRT ((SD - (COUNT • XMEAN • XMEAN)) / (COUNT - 1.0))
C•
C • NORMALIZE EACH GRID POINT VALUE.
C•
DO 14 I = 2,6
DO 15 J = 1,6
X = (GRID(I,J) + 50000)/10.0
GRIDR(I,J) = (X - XMEAN)/SD
15 CONTINUE
14 CONTINUE
C•
C* WRITE THE YEAR, MONTH, DAY, HOUR, MEAN, STANDARD DEVIATION AND
CO NORMALIZED GRID VALUES ONTO A NEW GRID (GRIDE).
Ci
GRIDR(I,I) = GRID(I,I)
GRIDR(I,2) = GRID(I,2)
GRIDR(I,3) = GRID(I,3)
GRIDR(I,4) = 0
GRIDR(I,5) = XMEAN
GEIDR(I,6) = SD
 Procedure for the classification of synoptic weather m a p s 403
LEN = 14~
CALL WRITE (GRIDR,LEN,0,LNUM, 9)
Cm
C| CONTINUE TO THE NEXT DAY'S GRID.
Cs
GO TO 11
99 WRITE (6,101)
101 FORMAT (, END OF DATA REACHED')
STOP
END

CJJJJJJJJJJ JJJJJJ JiiJ JJJ JJ JJiJJ JJilJiJJJJJJ JJJJJJJ JJJ JJJi JJJJJJJ JJJJJJJJ
C• •
C a SCORES, A PROGRAM TO: m
Cm I) CALCULATE THE KIRCHOFER SCORE, ROW SCORE I
Cm AND COLUMN SCORE FOR ALL POSSIBLE PAIRS OF I
Ci NORMALIZED GRIDS; m
Cm 2) DETERMINE THE SIGNIFICANCE OF ALL POSSIBLE •
Cm PAIRS OF GRIDS; i
Ci 3) RECORD THE SIGNIFICANT PAIRS ON DISK FOR m
Cm ENTRY INTO PROGRAM KEYDAY. •
Cm •
C• NOTE: I) THE FOLLOWING PARAMETERS MUST BE SET: •
Ci N: NUMBER OF DAYS IN THE SAMPLE •
C• NS (CALLED "SPACE" IN TEXT): (N(N-I)/2); t
Ci 2) THRESHOLDS FOR SCORE, RSCORE, AND CSCORE MUST BE SET •
Cm (SEE TEXT) ; t
C• 3) ARRAYS MUST BE DIMENSIONED. •
Cs •
CJJJIJJJIJJJJIJJJIJIJJJJJJJJJJIJJJJ JJJJJiJJJiJJJJi JJJJJJJJJJJJJJJJJJ J JJJ
Ci
C• DIMENSION THE ARRAYS AND SPECIFY NUMBER OF DAYS TO BE TESTED (N) AND
C• NECESSARY SPACE (NS) FOR STORING SIGNIFICANT KIRCHHOFER SCORES.
C•
DIMENSION GRIDR (6,6) ,XGRID(6,6 ), YGRID(6,6) ,D(6,6 )
LOGICAL•I NS(IOI3176)/IOI3176i.FALSE./,LSCORE(4)
INTEGERS2 LEN
DATA N/I ~24/
EQUIVALENCE (KK,LSCORE(I) )
KK= 0
Cs
C • SET THE STORAGE ARRAYS TO ZERO AND INITIALIZE THE XGRID COUNTER.
C•
NSCORE = 0
STORE = 0.0
KOUNTX = 1
C•
C • CREATE XGRID, RECORD THE DATE AND INITIALIZE THE YGRID COUNTER.
Cs
DO I K = KOUNTX,N
NMATCH = 0
LNUM = K • 1000
CALL READ (GRIDR,LEN,2,LNUM, 8,&99)
IXDATE = (GRIDR(I,I) • 10000).+ (GRIDR(I,2) • 100) + GRIDR(I,3)
WRITE (9,1001) IXDATE
IF (K .EQ. N) GO TO 99
DO 11 I = 1,6
DO 21 J = 1 , 6
XGRID(I,J) = GRIDR(I,J)
21 CONTINUE
11 CONTINUE
KOUNTY = K + I
C•
C • CREATE YGRID AND RECORD THE DATE.
C•
DO 2 L = KOUNTY,N
LNUM = L • 1000
CALL READ (GRIDR ,LEN,2 ,LNUM, 8, &99)
IYDATE = GRIDR(I,I) • 10000 + GR~DR(I,2) • 100 + GRIDR(I,3)
DO 12 I = 1,6
DO 22 J = I ,6
YGRID(I,J)-= GRIDR(I,J)
22 CONTINUE
12 CONTINUE
404 BRENT YARNAL

Cg
C m CALCULATE THE TOTAL KIRCHHOFER SCORE (SCORE) AND EACH ROW
C m SCORE (RSCORE); IF NOT SIGNIFICANT, GO ON TO THE NEXT DAY.
C•
SCORE = 0.0
DO 13 1 : 2,6
RSCORE = 0.0
DO 23 J = 1,6
D(I,J) = (XGRID(I,J) - YGRID(I,J))•m2
SCORE = SCORE + D(I,J)
RSCORE = RSCORE + D ( I , J )
23 CONTINUE
IF (RSCORE .GT. 6.0) GO TO 2
13 CONTINUE
IF (SCORE .GT. 15.0) GO TO 2
Cm
Cm CALCULATE THE COLUMN SCORES (CSCORE); IF NOT SIGNIFICANT,
Cm GO TO THE NEXT DAY.
C•
DO 14 J = 1,6
CSCORE = 0.0
DO 24 I = 2,6
CSCORE = CSCORE + D(I,J)
24 CONTINUE
IF (CSCORE .GT. 5.0) GO TO 2
14 CONTINUE
C~
C• RECORD SIGNIFICANT KIRCHHOFER SCORES AND CONTINUE TO THE NEXT DAY.
CW
KK = IFIX(SCORE • 8.0 + 0.5)
NS((L-2) • (L-I)/2+K) = LSCORE(4)
2 CONTINUE
C•
C • MAKE THE NEXT PASS THROUGH THE DATA SET.
Ct
I CONTINUE
C•
C • CONCLUDING STATEMENTS.
C•
99 DO 33 I = 2,N
LEN = I - I
CALL WRITE (NS((I-2) • (I-I)/2+I),LEN,0,LNR,10)
33 CONTINUE
WRITE (6,1004)
1001 FORMAT (I8)
1004 FORMAT (' END OF DATA REACHED')
STOP
END

CJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJJ
C| •
Cm KEYDAY, A PROGRAM TO ESTABLISH THE KEYDAYS OF THE SYNOPTIC i
C• CLIMATOLOGY BY: •
C• I ) FINDING WHICH DAY HAS THE GREATEST NUMBER OF •
C• SIGNIFICANT DAYS ASSOCIATED WITH IT; •
Cm 2) RECORDING THE DATE, NUMBER OF ASSOCIATED SIGNIFICANT •
Ci DAYS, MEAN KIRCHHOFER SCORE AND STANDARD DEVIATION •
Cm OF THE KIRCHHOFER SCORE FOR THAT DAY; •
C• 3) ELIMINATING THAT DAY AND ALL ASSOCIATED DAYS FROM •
C• THE DATA RECORD; •
Cm 4) CONTINUING THIS PROCESS UNTIL ALL DAYS ARE CLUSTERED •
C• INT0 GROUPS OF FIVE OR MORE SIGNIFINCANT DAYS PER •
C• KEYDAY. •
Cm •
Ci NOTE: I) THE FOLLOWING PARAMETERS MUST BE SET: •
Cm N, IDATE, NUSTOR: NUMBER OF DAYS IN THE SAMPLE •
Cm STORE: CALLED "SPACE" IN TEXT AND "NS" IN •
Cm PROGRAM SCORES; •
Ci 2) ARRAYS MUST BE DIMENSIONED. •
Cm •
CllJJlliJJlJJllllJlJiJJ JlJmilJllJilliliJllllllllllillllliillllllJlillli•
Cm
C • DIMENSION THE ARRAYS, DATA AND COMMON STATEMENTS, AND READ
Cm THE INPUT DATA.
 Procedure for the classification of synoptic weather maps 405
C*
DIMENSION IDATE ( 1424 ) ,NUSTOR (1424)
LOGICALll STORE(1013176) ,LS(4)
COMMON STORE
DATA N, KEYDAY/I ~24,0/
READ (8,101) (IDATE(I),I = 1,N)
DO 7 I = 2,N
LEN = I - 1
CALL READ (STORE((I-2) I (I-I)/2 + I), LEN,O,LNUM,9,&99)
7 CONTINUE
Cm
C m START THE KEYDAY SEARCH.
Cm
DO I I = I ,50
KEYNUM = 0
KEYROW = 0
C•
C • FIND THE TOTAL NUMBER OF SIGNIFICANT SCORES FOR EACH ROW.
C•
D011 J = I , N
NUSTOR(J) = 0
DO 12 K = I,N
IF (J .EQ. K) GOTO 12
IF (J .GT. K) GOTO 8
J2 = K
K2 = J
GOTO 9
8J2 =J
K2=K
9NS=O
IF (STORE((J2-2) • (J2-I)/2 + K2)) NS = I
NUSTOR(J) = NUSTOR(J) ÷ NS
12 CONTINUE
11 CONTINUE
C*
C* FIND THE ROW WITH THE MOST SIGNIFICANT SCORES, NOTE HOW MANY
C m SIGNIFICANT SCORES THERE ARE AND WHERE THE CORRESPONDING DATE IS
C* STORED.
C*
DO 21 J = I ,N
IF (NUSTOR(J) .LE. KEYNUM) GO TO 21
KEYNUM = NUSTOR (J)
KEYPOW = J
21 CONTINUE
C•
C* CHECK ON THE CLUSTER SIZE: IF IT IS LESS THAN 5,
C • NO MORE KEYDAYS ARE DETERMINED.
C•
IF (KEYNUM .LT. 5) GO TO 99
C•
C • DETERMINE KEYDAY (J).
C*
KEYDAY = IDATE(KEYROW)
C*
C* CALL SUBROUTINE KEYS; WRITE THE SYNOPTIC TYPE
C* NUMBER, KEYDAY, NUMBER OF SIGNIFICANT SCORES
C* ASSOCIATED WITH THAT DAY, MEAN KIRCHHOFER SCORE
C* AND STANDARD DEVIATION OF THE KIRCHOFFER SCORE.
C•
CALL KEYS (N,KEYNUM,KEYROW,XMEAN,SD)
WRITE ( 11,1001 ) I,KEYDAY,KEYB-JM,KEYROW, XMEAN,SD
C•
C* CALL SUBROUTINE DELETE.
C*
CALL DELETE (N,KEYROW)
C*
C • GO ON TO DETERMINE THE NEXT KEYDAY.
C*
I CONTINUE
C*
C* CONCLUDING STATEMENTS.
C*
101 FORMAT (I8)
102 FORMAT (F5.2)
406 BRENT YARNAL

103 FORMAT (II)

1001 FORMAT (' ',I2,I8,IS,F7.2,F6.2)
99 CONTINUE
STOP
END

C•
C • SUBROUTINE KEYS: A SUBROUTINE TO CALCULATE THE MEAN AND STANDARD
C* DEVIATION FOR EACH DAY.
C*
SUBROUTINE KEYS (N,NUM,IROW,XMEAN,SD)
LOGICAL•I STORE(1013176) ,L(4)
COMMON STORE
EQUIVALENCE (K, L( 1 ) )
K=0
XMEAN = 0.0
SD = 0.0
DO 10 J = 1 ,N
IF (J .EQ. IROW) GOTO 10
IF (IROW .LT. J) GOTO 5
J2 = IROW
I2 = J
GOTO 7
5J2=J
I2 = IROW
7 L(4) = STORE((J2 - 2) • (J2 - I)/2 + I2)
X = FLOAT (K)/8.0
XMEAN = XMEAN + NUM
SD = SQRT ((SD - (NUM • XMEAN • XMEAN))/(NUM - 1.0))
RETURN
END
Cllililiilliliillilllll•llllilllllll•lliill•lilillliill•lililliiliilili
C•
C i SUBROUTINE DELETE: A SUBROUTINE TO DELETE ALL DAYS ASSOCIATED WITH
Cm THE KEYDAY AND ALL DAYS ASSOCIATED WITH THOSE
C* DAYS.
C*
SUBROUTINE DELETE (N,IROW)
LOGICAL*I STORE(1013176) ,KK(4)
COMMON STORE
EQUIVALENCE (KK( I ) ,K)
K=0
DO 10 J = I ,N
IF (IROW .EQ. J) GOTO 10
IF (IROW .LT. J) G O T O 2
J2 = IROW
I2 = J
GOTO 3
2J2 =J
I2 =IROW
3 IF (.NOT. STORE((J2 - 2) * (J2 - I) / 2 + I2)) GOTO 10
D06I= I,N
IF (I .EQ. J) GOTO 6
IF (I .LT. J) GOTO 4
J2 = I
I2=J
GOTO 5
4J2 =J
I2 = I
5 STORE((J2 - 2) • (g2 - I)/2 + I2) = KK(4)
6 CONTINUE
I 0 CONTINUE
RETURN
END

C•••|••••••|••••••••••••|••••••••••••••••••••••••••••••••t•••••••••••l••
C• •
C* KEYMAP: A PROGRAM TO PRINT OUT THE KEYDAY GRIDS *
C* RESULTING FROM PROGRAM KEYDAY. INPUT ON •
C• CHANNEL 8 IS FROM KEYDAY. OUTPUT FROM KEYMAP *
C• CAN BE ROUTED TO A COMPUTER MAPPING PROGRAM. •
C• •

C•
 Procedure for the classificastion of synoptic weather maps 407

INTEGER.2 GRID(6,6),LEN
1 READ (8,101, END q 98) 1,IROW
WRITE (6,lOOl) 1,IRGW
WRITE (6,1002)
LNUM = IRGU l 1000
CALL READ (CRID,LEN,2,LNUM,9,&99)
DO 5 J = 2,6
DO 5 K E 1,6
GRID(J,K) = 5000 + GRID (J,K)/lO.O + 0.5
5 CONTINUE
WRITE (6,1003) (GRID(~,L),L = 1,3)
WRITE (10,1003) (GRID(I,L),L = 1,3)
WRITE (6,1003) ((GRID(K,L),L q 1,6),K = 2,6)
WRITE (10,1004) ((GRID(K,L),L q 1,6),K = 2,6)
GOT0 1
98 WRITE (6,100)
STOP
99 WRITE (6,102) IROW
STOP
100 FORMAT (' END OF DATA REACHED')
101 FORMAT (13,13X,15)
102 FORMAT (' INVALID SAMPLE DAY INDEX ENCOUNTERED:',151
1001 FORMAT ('1KEYDAY ',12,' SAMPLE DAY #',I$)
1002 FORMAT ('0')
1003 FORMAT (' ',6110)
1004 FORMAT (6110)
END

cttttttttttttttttttttttttttttttttttttttttttttttttttttttttttttttttttttttt
C’ t
Ct SYNTYPE, A PROGRAM TO: t
ct 1) CALCULATE THE KIRCHHOFER SCORE FOR EACH t
ct DAILY PRESSURE MAP WITH EACH KEYDAY; t
ct 2) SELECT THE LOWEST SIGNIFICANT SCORE AND t
ct ASSIGN EACH DAY TO A SYNOPTIC TYPE; t
ct 3) WRITE A CATALOG OF DAILY SYNOPTIC TYPES AND t
ct RELEVANT INFORMATION; KEYDAYS ARE IDENTIFIED t
ct IN THE CATALOG BY A MINUS SIGN; NONSIGNIFICANT t
ct SCORES AND UNCLASSIFIED DAYS ARE IDENTIFIED BY t
ct 999.9 AND 0, RESPECTIVELY. t
ct t
ct NOTE: IN THE MAIN PROCRAM t
ct 1) THE FOLLOWING PARAMETERS MUST BE SET: t
ct I, SKORE, VALUE, UTYPE, IDATE, YMEAN, STD: t
ct TOTAL NUMBER OF DAYS IN THE STUDY PERIOD t
ct (EXCLUDING MISSING GRIDS) t
ct M, KEYDAY, SKORE: NUMBER OF KEYDAYS t
ct IDENTIFIED IN PROGRAM KBYDAY; t
ct 2) IN SUBROUTINE SKORES, THE FOLLOWING PARAMETERS t
ct MUST BE SET: t
ct IDATE, TMEAN, STD: TOTAL NUMBER OF DAYS IN t
ct THE STUDY PERIOD; t
ct 3) ARRAYS MUST BE SET IN BOTH MAIN PROGRAM AND t
ct SUBROUTINES KGRID, SKORES, AND SCORES. 0
ct 4) IN SUBROUTINE SKORES, THRESHOLDS FOR SCORE, t
ct RSCORE AND CSCORE MUST BE SET USING THE VALUES t
ct USED IN PROGRAM SCORES. t
ct t
cttttttttttttttttttttttttttttttttttttttttttttttttttttttttttttttttttttttt
ct
Ct DIMENSION THE ARRAYS, DATA AND COMMON STATEMENTS.
C@
REAL KGRID
DIMENSION GRIDR(6,6),KGRID(6,6),KEYDAY(8),
SKORE(1424,8),VALUE(1424),NTYPE(l424)
'COMMON/ETC/IDATE(1424),TMEAN(1424),STD(1424)
COMMON N,M
N = 1424
M=8
c*
C* ESTABLISH THE DATE (IDAY) OF THE KEYDAY FOR A GIVEN
c* SYNOPTIC TYPE (ITYPE).
c*
408 BRENT YARNAL

DO 10 1 = 1,H
READ (10,101) KEYDAY(1)
IDAY = KEYDAY(1)
ITYPE = I
C•
C• FIND THE MATCHING PRESSURE GRID FOR THE KEYDAY.
C•
CALL KGRIDS (IDAY,KGRID)
C•
C• CALCULATE A KIRCHHOFER SCORE FOR EACH POSSIBLE
C• DAY WITH THE KEYDAY.
C•
CALL SKORES (IDAY,ITYPE,KGRID,SKORE)
C•
C• GO ON TO THE NEXT KEYDAY.
C•
10 CONTINUE
C•
C• DETERMINE THE SYNOPTIC TYPE AND KIRCHHOFER SCORE
C• FOR EACH DAY; CREATE A CATALOG THAT INCLUDES THE
C• DATE, SYNOPTIC TYPE, KIRCHHOFER SCORE FOR THAT DATE,
C• THE MEAN PRESSURE VALUE FOR THAT GRID, AND THE
C• STANDARD DEVIATION OF THAT PRESSURE FIELD.
C•
WRITE (6,1001)
DO 11 1 = I,N
NTYPE(I) = 0
VALUE(I) = 999.99
DO 2 1 J = 1,M
SKR = SKORE(I,J)
IF (SKR .GE. 0.0) GO TO 15
NTYPE(I) = SKR
VALUE(I) = 0.0
GOTO 21
15 IF (SKR .GE. VALUE(I)) GO TO 21
VALUE(I) = SKR
NTYPE(1) = J
21 CONTINUE
WRITE (6,1002) IDATE(I),NTYPE(I),VALUE(I),TMEAN(I),STD(I)
WRITE (9,1003) IDATE(I),NTYPE(I),VALUE(I),TMEAN(I),STD(I)
11 CONTINUE
C•
C m CONCLUDING STATEMENTS.
C•
101 FORMAT (3X,I8)
1001 FORMAT (' DATE TYPE SCORE MEAN SD')
1002 FORMAT (' ',I8,14,4X,F8.2,2X,F8.2,2X,F6.2)
1003 FORMAT (18,14,4X,FS. 2,2X,F8.2,2X,F6.2)
STOP
END
Ci||||i||I|i||II||I||||||||t||||||||tl||||I|iI|i||||||i||I|li||||||i||||
Cm
C m SUBROUTINE KGRIDS: FINDS THE KEYDAY GRID FOR EACH
C• SYNOPTIC TYPE KEYDAY.
C•
SUBROUTINE KGRIDS (IDAY,KGRID)
REAL KGRID
DIMENSION KGRID(6,6),GRIDR(6,6)
COMMON N,M
INTEGER•2 LEN
LEN = 144
IND = 0
3 IND = IND + I
LNUM = IND • 1000
CALL READ (GRIDR,LEN 2,LNUM,8,&99)
IDATE = GRIDR(I,I) • 10000 + GRIDR(I,2) • 100 + GRIDR(I,3)
IF (IDATE .NE. IDAY) GOTO 3
DO 1 1 K = 1,6
DO 2 1 L = 1,6
KGRID(K,L) = GRIDR(K,L)
21 CONTINUE
11 CONTINUE
GO TO 5
99 WRITE (6,2001)
 Procedu~ ~ r the classificastion of synodic weather maps 409

2001 FORMAT (' END OF DATA REACHED---I ')

RETURN
END
C|||m|m,,g,R|JQ||Jl,|Rlm|RRmm|,Jm|lll|RRN|J||tJ|%|m|m|,RgmJJm|Qimm|m|g|
Cm
Cm SUBROUTINE SKORES: CALLS SUBROUTINE SCORES TO CALCULATE
Cm KIRCHHOFER SCORES.
Cm
SUBROUTINE SKORES (IDAY,ITYPE,KGRID,SKORE)
DIMENSION SKORE(N,M)
REAL EGRID(6,6),GRIDR(6,6)
COMMON/ETC/IDATE(1424),TMEAN(1424),STD(1424)
COMMON N,M
INTEGER*2 LEN
LEN = 144
DO 10 I = I,N
LNUM = Im1000
CALL READ (GRIDR,LEN,2,LNUM,8,&99)
IDATE(I) = GRIDR(I,I) • 10000 + GRIDR(I,2) • 100 + GRIDR(I,3)
TMEAN(I) = GRIDR(I,5)
STD(I) = GRIDR(I,6)
IF (IDATE(I) .NE. IDAY) GO TO 98
SCORE = -1.0 • ITYPE
GO TO 97
98 CALL SCORES (SCORE,KGRID,GRIDR)
97 SKORE (I,ITYPE) = SCORE
10 CONTINUE
GO TO 5
99 WRITE (6,3001)
3001 FORMAT (' END OF DATA REACHED')
5 RETURN
END

CJm~llJJi||JiiJmi~m~I|om|oom|~I||g|imJ|||Jou|RgRa|g||~|l~I|~I||J||m|~o|
C*
C m SUBROUTINE SCORES. A SUBROUTINE TO CALCULATE THE SCORE,
Cm RSCORE, AND CSCORE FOR EACH DAY AND KEYDAY;
Ce A SIGNIFICANT SCORE (.LE. 15.0) OR
C• NONSIGNIFICANT SCORE (999.99) IS RETURNED
Cm TO SUBPROGRAM 3KORES.
C•
SUBROUTINE SCORES (SCORE,KGRID,GRIDR)
REAL KGRID(6,6),GRIDR(6,6)
DIMENSION D(6,6)
SCORE = 0.0
DO 10 I = 2,6
RSCORE = 0.0
DO 20 J = 1 , 6
D(I,J) = (GRIDR(I,J) - KGRID(I,J))**2
RSCORE = RSCORE + D(I,J)
20 CONTINUE
IF (RSCORE .GT. 6.0) GO TO 98
10 CONTINUE
IF (SCORE .GT. 15.0) GO TO 98
DO 1 1 J = 1,6
CSCORE = 0.0
DO 2 1 1 = 1,5
CSCORE = CSCORE = D(I,J)
21 CONTINUE
IF (CSCORE .GT. 5.0) O0 TO 98
11 CONTINUE
GO TO 97
SCORE = 999.99
98 CONTINUE
RETURN
END

C* *
C* SORT, A PROGRAM TO: *
C* I) SORT THE SYNOPTIC CATALOG FROM SYNTYP INTO *
C* M SYNOPTIC TYPES; *
C~ 2) COUNT THE NUMBER OF DAYS PER TYPE; *
C* 3) DETERMINE THE PERCENT FREQUENCY OF EACH TYPE; *
410 BRENT YARNAL

C* 4) CALCULATE THE MEAN KIRCHHOFER SCORE FOR EACH *

C* SYNOPTIC TYPE ; *
C* 5) CALCULATE THE STANDARD DEVIATION OF THE MEAN *
C* KIRCHHCFER SCORE FOR EACH SYNOPTIC TYPE. *
C* *
C• NOTE: THE FOLLOWING PARAMETERS MUST BE SET: *
C* N, IDATE, NTYPE, VALUE: TOTAL NUMBER OF DAYS •
C* IN THE STUDY PERIOD. •
C* M: NUMBER OF KEYDAYS IDENTIFIED IN PROGRAM *
C* KEYDAY. *
C• •
C•••••••no•••m•••m•,m••••••,••*********•**•**•**,•********••••••***•,•**••
COO
C • DIMENSION THE ARRAYS AND DATA STATEMENTS, AND READ IN
C* THE INPUT DATA. CREATE A COLUMN HEADER.
C*
DIMENSION IDATE(1424) ,NTYPE(1424) ,VALUE(1424)
REAL•8 SD
DATA M,N/8,1424/
I = 999
WRITE (6,66) I
66 FORMAT (I2)
READ (8,101) (IDATE(K),NTYPE(K),VALUE(E), K = I,N)
WRITE (6,1001)
C*
C* SET THE COUNTERS. DETERMINE AND RECORD THE NUMBER,
C * % FREQUENCY, MEAN KIRKHOFFER SCORE AND STANDARD
C* DEVIATION OF EACH OF THE M SYNOPTIC TYPES.
C•
MM=M+I
DO 10 I = I,MM
II = I - I
COUNT = 0.0
]O~EAN = 0 . 0
SD = 0.0
DO 20 J = I ,N
IF (IABS(NTYPE(J)) .NE. II) GO TO 20
COUNT = COUNT + 1.0
XMEAN = XMEAN + VALUE(J)
SD = SD + (VALUE(J) • VALUE(J))
20 CONTINUE
XMEAN = XMEAN / COUNT
SD = DSQRT ( ( S D - (COUNT • XMEAN * XMEAN))/(COUNT - 1.0)
IF (XMEAN .LT. 500.0) GOTO 25
XMEAN = 999999.99
SD = 999999.99
25 FREQ = COUNT / FLOAT(N) • 100.0
NCOUNT = COUNT
WRITE (6,1002) II, NCOUNT, FREQ, XMEAN, SD
WRITE (9,1003) II, NCOUNT, FREQ, XMEAN, SD
I 0 CONTINUE
C*
C* CONCLUDING STATEMENTS.
C*
101 FORMAT (IS,I4,4X,F8.2)
1001 FORMAT ('ITYPE NUMBER %FREQ MEAN SD')
1002 FORMAT (' ',I3,3X,IS,3X,F4.1,3X,F6.1,2X,F4~1)
1003 FORMAT (I3,3X,IS,3X,F4.1,3X,F6.1,2X,F4.1)
STOP
END