Journal of Environmental Management (1984) 19, 239-248

Testing an Ecosystem Regionalization

Robert G. Bailey*

Washington Office Land Management Planning Staff, USDA Forest Service,Fort Collins, Colorado 80524, U.S.A.

Received 24 ,April 1984

As a means of developing reliable estimates of ecosystem productivity, landscapesneed to be stratified into homogeneous geographic regions. Such ecosystem regionsare hypothesized to be productively different in important ways. One measure ofthe difference is hydrologic productivity (i.e. runoff per unit area). To test thehypothesis that hydrologic productivity is significantly different from region toregion, data from 53 hydrologic bench-mark stations within major ecosystemregions of the conterminous United States were subjected to discriminant analysis.The ecosystem regions tested in this study exhibit a high degree of ability tocircumscribe stations with similar hydrologic productivity. Discriminant analysisseem~ to be a particularly appropriate method for testing the validity of land unitsat all scales of interest.

Keywords: ecosystem, region, ecological land classification, landscape ecology,A;scn"..,;n ant an"lv s; s PrnA11...t;"I"tv~. ~. J -, -~ OJ"

1. Introduction

Land management deals with productivity systems, that is ecosystems, from which itattempts to extract efficiently and continuously a product, such as wood or water. Theseproducts are commonly referred to as renewable resources.

Estimates of ecosystem productivity are required for the purpose of assessment andmanagement of these resources. In order to make such estimates, relationships betweenthe needed production information and ecosystem classes must be developed. Theserelationships are called rules (Davis, 1980). Rules take on many forms, from simpleexperience-based judgments to multivariate regression models and complexmathematical simulations. Application of these rules is based upon concepts of transferby analogy. Thus, the rules necessary for estimation of productivity are extrapolatedfrom experimental sites or from management experience to analogous areas defined byclassification.

* Formerly with the Resources Evaluation Techniques Program, Rocky Mountain Forest and RangeExperiment Station, USDA Forest Service, Fort Collins, Colorado, U.S.A.

239

0301-4797/84/070239 + 10 103.00/0 (!;;) 1984 Academic Press Inc. (London) Limited

240 Testing an ecosystem regionalization

Such methods are based on the hypothesis that all replications of a particularecosystem class will have fairly similar productivity. This hypothesis has been questionedby a number of workers on the grounds that correlations between ecosystem classes andproduction are generally low.

One way to establish reliable ecosystem-production relationships is to stratify thelandscape into "relatively homogeneous" geographic regions where similar ecosystemshave developed on sites having similar properties (Rowe, 1962). For example, similarsites (i.e. those having the same landform, slope, parent material and drainagecharacteristics) may be found in several climatic regions. Within a region, these sites willsupport the same vegetation communities, but in other regions vegetation on the siteswill be different. Thus, beach ridges in the tu~dra climatic region support low-growingshrubs and forbs, whereas beaches in the subarctic region usually have dense growth ofblack spruce or jack pine. Soils display similar trends, as the kind and development ofsoil properties vary from region to region on similar sites.

Such ecosystem regions, or ecoregions, have at least two important functions formanagement. First, a map of such regions suggests over what area the productivityrelationships derived from experiments and e:xperience can be applied without too muchadjustment. Second, they provide the necessary geographical stratification for designingcost-efficient sampling programs to estimate ecosystem productivity.

The mapping of these regions has been worked out in a preliminary way by Bailey(1976) from concepts advanced by Crowley (1967). Subdivisions on the map show unitsthat (theoretically) should be relatively homogeneous. This homogeneity has beenpartially tested using biophysical parameters (Olson et al., 1982), aquatic ecosystems(Omernik et al., 1982) and wildlife communities (Inkley and Anderson, 1982). However,the degree of correlation between these regions and productivity has not beendetermined. The objective of this study was to use actual data to determine the degree towhich ecosystem regions represent productivity regions. Effective resource managementbased on ecosystem-production relationships requires that productivity be distributed ina manner similar to the ecosystem regionalization used.

2. Approach to testing

As Rowe and Sheard (1981) point out, maps are hypotheses to be tested and improved.The units express a sense of what is the,orized to be important in the landscape. Regionalboundaries on Bailey's map are based on climatic and vegetational features (Bailey,1983). The approach rests upon the hypothesis of a relationship between the features ofthe environment used to delimit the region and the properties of the region. In this case,regions bounded by changes in macro-features of the climate and vegetation arehypothesized to be productively different in important ways. If actual data onproductivity are assembled for the regions, this hypothesis can be tested statistically andthe validity of the regional structure (map) can be evaluated objectively. This test isindependent inasmuch as productivity did not enter into the initial regionalization.

The use of statistical tests which analyse or test previously de";eioped regionalstructures was pioneered by Zobler (1957). These tests involve the use of a significancetest--:chi square or variance analysis-to ascertain if an association exists between thespatial frame, which is the way an area has been par,titioned, and the areal dispersion ofspecific properties of this frame. Recently, the use of multivariate discriminant analysisto test the validity of land units is attracting attention (Casetti, 1964; Steiner, 1965;Pavlik and Hole, 1977; ami et aZ., 1979; Rowe and Sheard, 1981). This method ofanalysis was viewed as the statistical method best suited for this studv.

R. G. Bailey 241

There are different approaches for l1Jeasuring ecosystel1J productivity. The questionis then: what l1Jeasure of groductivity is to be used for testing?

The traditional measure of productivity is net primary productivity. It is defined asthe total amount of organic matter produced annually by an ecosystem (Sharpe, 1975).It is an expression of a combination of vegetation and environmental site characteristics.Because it can be applied to a wide range of ecosystems, at many levels, it is extremelyuseful in comparing structure and dynamics of ecosystems which differ in floristics andphysiognomy. Analysis of regional structure necessitates productivity data that are notonly accurate, but cover large geographic areas. The existence of such data, however, islimited. Actual measurements of primary productivity are difficult to make and areusually taken from small areas. Different methods of measurement make datacomparison difficult, if not impossible. Thus, there does not exist a comprehensive database on primary productivity that is suitable for nationwide testing.

Another measure of ecosystem productivity is the normal runoff-the amount ofwater runoff, on average, produced during a period from a unit area of land surface.Because runoff data are readily available for the entire country, this measure ofproductivity was used in this study, and is referred to here as hydrologic productivity.

3. Data and methods

To make unbiased comparisons of hydrologic productivity, data are needed on therunoff characteristics of "natural" basins which are unaffected by urbanization, man-made storage, diversion, or ground water pumping. The data used were the streamflowobservations collected by the U.S. Geological Survey at 57 hydrologic bench-markstations (Cobb and Biesecker, 1971). These data for stations from all over the UnitedStates are available in a computerized data library called WATSTORE at the USGScomputer in Reston, Virginia (Hutchison, 1975). Locations of the stations are shown inFigure 1 and listed in Table 1.

;::%42 ("'"\~ ,-/) 21I \:~:~ ,~\ '-'-"'-:::::::-:-:-:-:::-:-:-:- :-"-- 36

\ 22~ ~ro " \ ,--r(".. \:-:-- -( ( -\"-:-:- ) ,,55 ~--:-- 6 238\ 33---

-;--- -1 , -J \\ \

-:-:- \ 16" -, \L J I

::j: --I;. "'-, I 17\ 32 t'-:f- ." I _l'-~:i:: 'I l 1,..1 '\ r ..L- J::: -_J --t-'--48-",3

':-:- r -51,/47- £-~::;: 8 \. J T13~-:::; 3~-4 / I -2 \ "

-.:,: --,,/ I \1-2 4--~ , I--"::~+~f )~~j~~~~(\ 19/ 24 _1' L.- --:-1 ;

--50 J 11-D HumId temperate --::-:: ::

rn Dry

r

500 mile:\,),( 0I ,.0 500km

Figure Locations of hydrologic bench-mark stations and ecodomains of the conterminous United States ofAmerica.

242 Testing an ecosystem regionalization

TABLE 1. Locations of hydrologic bench-mark stations

31. Mogollon Creek near Cliff, New Mexico32. Rio Mora near Tererro, New Mexico33. Esopus Creek at Shandaken, New York34. Cataloochee Creek near Cataloochee,

North Carolina35. Bear Den Creek near Mandaree, North

Dakota36. Beaver Creek near Finley, North Dakota37. Upper Twin Creek at McGaw, Ohio38. Blue Beaver Creek near Cache, Oklahoma39. Kiamichi River near Big Cedar, Oklahoma40. Crater Lake near Crater Lake, Oregon41. Minam River at Minam, Oregon42. Young Woman's Creek near Renova,

Pennsylvania43. Scape Ore Swamp near Bishopville, South

Carolina44. Upper Three Runs near New Ellenton,

South Carolina45. Castle Creek near Hill City, South Dakota46. Little Vermillion River near Salen, South

Dakota47. Buffalo River near Flat Woods, Tennessee48. Little River above Townsend, Tennessee49. Limpia Creek above Fort Davis, Texas50. South Fork Rocky Creek near Briggs,

Texas51. Red Butte Creek near Salt Lake City, Utah52. Holiday Creek near Andersonville,

Virginia53. Andrews Creek near Mazama,

Washington54. North Fork Quinault River near Amanda

Park, Washington55. Popple River near Fence, Wisconsin56. Cache Creek near Jackson, Wyoming57. Encampment River near Encampment,

Wyoming

1. Blackwater River near Bradley, Alabama2. Sipsey Fork near Grayson, Alabama3. Wet Bottom Creek near Childs, Arizona4. Cossatot River near VandervoortArkansas '

5. North Sylamore Creek near Fifty Six,Arkansas

6. Elder Creek near Branscomb, California7. Merced River near Yosemite, California8. Wi1drose Creek near Wi1drose StationCalifornia '

9. Halfmoon Creek near Malta, Colorado10. Vallecito Creek near Bayfield, ColoradoII. Sopchoppy River near Sopchoppy, Florida12. Falling Creek near Juliette, Georgia13. Tallulah River near Clayton, Georgia14. Honolii Stream near Papaikou, Hawaii15. Hayden Creek below North Fork, near

Hayden Lake, Idaho16. Wickahoney Creek near Bruneau, Idaho17. South Hogan Creek near Dillsboro,

Indiana18. Elk Creek near Decatur City,. Iowa19. Big Creek at Pollock, Louisiana20. Wild River at Gilead, Maine21. Washington Creek at Windigo, Isle

Royale, Michigan22. Kawishiwi River near Ely" Minnesota23. North Fork Whitewater River near Elba,

Minnesota24. Cypress Creek near Janice, Mississippi25. Beauvais Creek near St. Xavier, Montana26. Swiftcurrent Creek at Many Glacier,

Montana27. Dismal River near Thedford, Nebraska28. South Twin River near Round Mountain,

Nevada29. Steptoe Creek near Ely, Nevada30. McDonalds Branch in Lebanon State

Forest, New Jersey

For this study, data from 53 stations were analysed. Station Nos 8, 25, and 40 werenot included in the analysis because the station records were incomplete. Station No. 14(Hawaii) was not included because it is outside the study area. Mean monthly runoffvalues, in inches per square mile, were calculated for the period 1971-80 for each station(Table 2). Each station was placed into one of two ecoregion groups as described below.The l2 variables (monthly averages of the runoff) for each station, along with theecoregion group of the station, constituted the data for analysis.

The ecoregion system is organized into a four-level hierarchy, each level representing

R. G. Bailey 243

a subdivision of the area encompassed by the preceding hierarchical level. These levels,from upper to lower, are ecodomain, ecodivision, ecoprovince and ecosection. For thepurposes of this analysis, the ecodomain level, which differentiates regions primarilyaccording to broad climatic similarity, was chosen because of the small sample size. Witha sample of 53 stations and 12 variables, only two groups could be properly analysed(Lachenbruch, 1975). Excluding the southern tip of Florida, the conterminous UnitedStates is partitioned into two ecoregions at the ecodomain level-dry and humidtemperate. The two ecodomains of the conterminous United States are shown in Figure1, which also shows the location of the hydrologic bench-mark stations. Fifteen stationscomprised the dry group; 38 stations comprised the humId temperate group.

Discriminant analysis, a technique for analysing a priori grouped data, was used totest whether the two ecoregions were different on the basis of hydrologic productiyity.Discriminating variables (those characteristics on which groups are hypothesized todiffer) are linearly combined and weighted. The analysis tests how successful thediscriminating variables are in significantly distinguishing between groups. Once asignificant (P < 0'05) set of discriminating variables has been found, the original set ofcases (stations) is classified as to group membership, given those variables and weights.This classification is done by calculating a discriminant score for each case and thencomparing it to the mean (average) score for each of the two groups. By comparingpredicted versus actual group membership, one can measure the success of groupdiscrimination.

The two ecoregions were analysed via the computer program BMDP designed formultivariate discriminant analysis (Dixon, 1981). All of the discriminating variablesexcept August were significant and were used in the analysis.

4. Results and discussion

Calculations of the discriminant scores for the 15 dry and 38 humid temperate stationsgave values ranging from -3.06 to + 2.99 {Table 2). In general, the lower values of thediscriminant score represent dry stations, and the higher values represent humidtemperate stations. The frequencies of stations for various levels of discriminant scoresare shown in Table 3. Although there is overlapping of discriminant scores for dry andhumid temperate groups, this table clearly shows the clustering within the two groups.

The mean value of the discriminant scores for the two groups was:

-1.29

0-51

-0.39.

dryhumid temperate

midpoint value

In theory, when the discriminant score is less than the midpoint value -0.39, thestation wi!l be classified as dry, and when the score is greater than -0.39 the station willbe classified as humid temperate.

The classification results are shown in Table 4. The numbers of stations that werecorrectly classified are given as diagonal elements of the matrix and the numbers ofincorrectly classified stations appear as off-diagonal elements. Seven of 53 stations weremisclassified. The misclassified stations include 23, 27, 32, 36, 46, 50, and 53 (Figure 2).All but one (Station No. 32) of these are in the humid temperate ecodomain. The overall

+--

000

..!.I-~-.Ifc=,~~'"~~E.6uc=Q

).cu'500-0..."0:>

..c:,...::,:E"S~c=cIt::0c=~...:>

.:sc=0E~~[)><~~..J=-<E

--

.dd0~

\Qr-~

0\~

V)-

V)N

66 00\0\ ..:to

NN

NV

.x, 6

0\\0r-r-N

6- 00O

\D0,,6

r'"\V0\-6':"- ,,",\0"""""'.;,.~

'rI"'i"00,,",

":'00

0\0\\O

N600

0000M

OO

6N

Nt--

66 -ir\N

~66 000\00

":'6

1-11")~

-6":'

0\-N

\O66 N

r-N

M66 r-o--N66 .q-1/"j-('I

66 ""1"0\-""1"

66 M~

N'"

6":'

-N0000~

'"

"t"0\-00~

oo

\D

000

.,;.,~

r-oo<

":-I~-- ~

O\O

N6':'

\O~

0\,..,~

..:.I

I

-N""~

I£\

t-t--O

\f'\\r)ooN

\0"'-t-'ot"ot'66",

\OO

\"i-"i-O

v)':":'.;,.

~N

OO

~~

OM

N\O

Ovt-

\ON

NN

N.;.,

0\000N

,

~~

~

0\-00r-oo':t"A

.:..-o

~",,"M

~7'9

--or)

-oot-V

).,...V)

":'6,,",

-0.,.,""",,,"I-.;.,6N

~r--N

~V

)O\

N6N

I--IrII--O

NN

N6N

Nt-a,

t-f"Ia,ooNI

'"N6 0M6 It')0-6 0-~- 00N or)00N -r:" 00-N I-- ~'r\6 -~6 f-.N6 -~6

\01---000\0""""""-('1

o"'~

~O

MM

""~6666N

ON

I-O\O

0~t'"\/X

)~6..:.66N

O\C

-~N

O--.t\C

\C6N

6':"N

oq-oq-oq-V)oq-

oov)ooo\C

:>N

C:>

N":'

.,.,-.tot-t-<

;>":'t-;-":'~

0("'10("'1-

\0\0\0\00-o-.to--6.,;.,.:..~

~

\r')\Q--\r')

00[--,0006N

':":'OO

NO

ON

O\

OM

O\O

N6..:.6N

OO

--oNO

\r-O

-o-or--o6666N

Olr)"ot"ot"ot

ot--\O"ot

6666':"

OM

lrla--O

Irl-IrIa-66666

O--N

N0"'-1"--0066':'66

-NO

\\DV

MN

O-N

":"6":"6":"I

--N

C'"I-:t."

NN

NN

N

-N-r-

oooo~O

'\

6666

O~

NN

-0\'"1"1-":'66":'

VO

\('-.1r\~

~~

<?

ooom

-0...,--...,-,j"-V

N666.ot

'o:t"~\C

\CN

NM

-666

!"'-"'MO

O!"'-t'lt'l0\66':'~

O\vO

\.,..,V

OI-~

';"':'6~

"""-'""'00\0'"M

N6M

C\C

\-C\

0 II"} 0\

00';"';"66

-N""'\O

0,)000\0666":'

"S"ooo"S

"N

V)-.t"O

666':'"

C'"I~

r---~~

~"rt-;-

000-

OO

t"\"ot:"ot:--r-0\

':"OO

NI

~r-

NN

...,...,a-N':"6

-MC\N

';'6

('"INN

N

':"6

c.-.NO

N':'6 0-t-N66 --.I-O

N":"6

t-"f"-N";"6

1-","I-N

6

ON

Ir.N006- C

ON

CO

N

66- rf")NN

N';'6

-NV1N6

01:1-0!""'--6':"I

"'II-V)

-"'IV)

66':"

NO

OO

\N

t'"\\O66';"

VV

NV

V

) \0

66":'

0-"'0-~

~":I

0\0\...,9'f'~---

a-NV

\N

-O6":"':'

or.-.t-.t-r-o-666

...,\O~

-V\~

666

~o~

O\O

N6":"6I

I I

Na-a-

~O

O666

-.r-.rOO

":'~9

oo~

cx>~

va-ocx>

-~~

o6 o..:t

",,6

-o-Ir)Ir)O...,-r-r-O66~

..:..6

""-"'00("--100-."."'-66N

",,6

-or-ooV)

-~oV

)oo":'6-o~

6

NIr)C

O~

1r)0\\00,,",\0

66-b..:t6

001O00,,",!-

-.tv\V\O

\OO

N";"N

O

\OV

IC-M

V)

Oa-V

-O6~

NN

6

\DO

OO

O\-

O,,",t')\D

O66':":'6

N-O

O':t"

0-1£)1£)\00666":'6

~

00 t-

\0 N

,"-"'-'0"-'066':":'6

Ir)o-r--ooo-M

ON

o-66':":'6

O\O

O\N

N\O

N':t:':t:O

66~';"6

-:t"OO

\-:t"O-:t"N

O\O

\O66A

.:.6

N",",~

'""'""""""""""'

Ot--\O

NO

Ir1-V6A

6,,;.,

0-"'00O

N",",

006N

6':'

0'"""""'000,"""",,,-O

NO

N

,,",0\\00\-0\00-6N

61i1

t-~\O

t-~

\O(x)N

6':'6""

O~

M~

-V)~

O6":"":"';-'

MM

M"'1"

O~

Ic-\O666":"

-or-oo9~

9':'10000

00\'o:tt-010-("'16666

\0 !-oo

0\ 8'

...,...,...,...,~

---r"1O

.qoN'"

666":'

ot-VIl')

Ot-N

-666N

o-~~

OI--~

\O6666

-0000"-'-"-'0":'66":'

NO

r ,...,

r ,-~-

ON

':"':" 0

O\r'"\!"-V

lMO

IOV

lVI-

':'r,.':":'6

N-r-vt'")

0-0-0-":"::"'::"':'6

NN

r'lo-No-O

o--V-

ON

':"':" 0

-'Ot'Ir\N

\O'O

t'Ir\Nr

..:.-itN..:.6

~O

\~~

NN

O\!"'-V

lNN

,,;,,":'..:.6

\l"\N\I"\'V

'VN

\1"\

\l"\N

..oA":'..:.6

00\""00000""""'"

ooN';"';"6

..,0\..,"'-0-oO

\O\v-

~66":'6

NO

r-\O~

oo~r--.t-

666':"6

CO

M\f")M

~~

OC

O~

-

6":'6":'6

-N

~",

~~

~~

~

"'\CO

Ol-'"

","","1-.,.,-6':'6':'6

Q\Q

\v.,."",N

---N":"':'66":'I

I I

00\~00

Or---O

N6":"",,"66

ovooo("'\0"""'0("'\6",,';"66

Oc.-lc.-lO

'OO

'OO

O~

6N~

66

..,NI-O

I-""rI"'O

'"6oi)-b66

00-0...,0-00""...,0",6';""'r66

""°0000000-006A

6~

~0-01l")

011")\0011")

6..:..i-I66

0("\""'001£)0-0,0,,",6':'N

66

ICIr-O

OO

\O~

~~

~Ir)

ON

~O

O\

ON

r--N6':";:"66

ON

O-~

O~

O

N6N

N66

Oo\\o-N

OO

OO

-N66N

66

OO

OO

Nin

OO

~lnN

6~~

66

--OO

't"t'N

a-'t'~a-

":"":"666I

I I

-r.-cor--ot"

r--o6":"6';;"':"

.qo \0 0

\0\0\00\

6':" 6V

1 6

...

...,0\...,0\0""""N

-'"6~

6.,.,6N

'tc-OO

OO

Or'"\o-I-'tc

6N6~

6

"'O\\O

r-~M

r---M6';'6~

6

~o~

\O\O

r ,~

6A6A

6

O\o\O

~"'"

""0\""""0\

.:..:.6o,N

OO

'.s"Ir")'.s"--0\\00\0\

N":";"~

~

00r"1r"1-.1-0000-.l-r"10.:...:.. 0\ r=

-':"

\Ct-V

)ot-V

) C

X>

CX

>N

V

)

66N~

6

-oCO

Q\-o-o

r"1CO

r-~-o

666~6

-NO

\MO

OM

N"'M

O6":'6",":'

-N""""'"

""""""'"

N",,"\O

MM

""" 0"""

NII"I

66":"":"6I

I

00 V

)V

)f')66 V

) V

)V

)M

66

r--V

f")66 C

XlIO

""('166 -0-V

rrl66 O

\N~

~66 M

t-c»-N

"","

~;

.r,~

0000oor--N

"'"

--:t-:tr-':'6\Q

t-V

) V

)

~""

r-~

66 N~

O\~

66 N\O

\00':"..;.,I

I

>.

., .>

>

.~

~cn

II .,;

~

N-U

;t)

">,

.-"-00,...

cd0

cd ~0'4;

cd"'"'~0

e.-.p.~

~"""

::>.-

~..e"O., ~

~t) """',...~

II

0

~-Z

+-+

+.00

246 Testing an ecosystem regionalization

TABLE 3. Frequencies of humid temperate and dry stations for various levels ofdiscriminant scores

TotalClass limits fordiscriminant score-~

-3.15 to -2.80-2.79 to -2.10-2.09to-l.40-1.39 to -0.70-0.69 to.O.OO

0.01 to 0-700-71 to 1.401.41 to 2-102.11 to 2.802.81 to 3-50Total

2 2

2147

12104

5412104

91

238

25315

TABLE 4. Classification results for ecodomains based on a lineardiscriminant function utilizing average monthly runoff data

Predicted group membership

Percentage of grouped stations correctly classified: 86.79%

percentage of grouped stations correctly classified was 87%.The high percentage of correctly classified stations indicates a high degree of

discrimination between the two ecodomains. Thus, the regional ecosystems tested in thisstudy exhibit a high degree of ability to circumscribe stations with similar hydrologicproductivity. Although classification results can theoretically achieve a level of 100%,inherent natural variation or inadequacy of the variables is such that this level canprobably never be reached. With this in mind, 87% is good discriminating power,considering the smallness of the sample. Lachenbruch and Mickey (1968) indicated thatusing a classification function to classify the same cases that were used to compute it isbiased in favor of better discrimination. Even allowing for this bias, the discriminationbetween groups is impressive.

The misclassified stations provide a clue to the validity of the map units. Forexample, most of the misclassified stations are relatively near the dry fhumid temperateboundary. This result can be interpreted to mean that the cores of the regional units arevalid but the boundaries, in terms of hydrologic productivity, may need someadjustment. This interpretation would not be possible if the misclassified stations hadbeen scattered throughout the groups.

The misclassified stations also serve to confirm the existence of finer subdivisions.For example, Bailey's (1976) map shows the dry western side of the humid temperate

R. G. Bailey 247

----1

("'"\) ", rd

~. ..46

..

-).....

...50.Group I

.Group 20 500 milesI ,I0 SOOkm

Figure 2. Predicted group membership from discriminant analysis, hydrologic bench-mark stations. ,eastern boundary of the subhumid prairie.

ecodomain in central USA distinguished as subhumid prairie (Figure 2). Four of theseven misclassified stations (Nos 27, 36, 46, and 50) are located in this region.

Discriminant analysis seems to be a particularly appropriate method for testing thevalidity of land units at all scales of interest. In this study, the method was applied to thebroadest level of ecological generalization. The next level of testing suggested asworthwhile is at more detailed l~vels of the hierarchy than was possible with the lowresolution data available for this study. Through multi-level application of this method,a complete analysis of the ecoregion map may be done. Other measures of productivity,such as net primary productivity, need to be used for testing.

The method differs from previous use of multivariate approaches with land units(Radloff and Betters, 1978; Laut and Paine, 1982; Briggs and France, 1983) which usecluster analysis of grid units to provide the initial map units. The approach taken in thisstudy uses multivariate discriminant analysis to test and validate map units initiallyrecognized and delineated by theoretical considerations.

The author wishes to thank G. L. Thompson and G. E. Brink for help with the retrieval of theWATSTORE data, and R. M. King for help with the statistical analysis. J. M. Omernik and R. J.Olson reviewed the manuscript.

ReferencesBailey, R. G. (1976). Ecoregions of the United States. Miscellaneous Publication 1391. Washington, D.C.:

USDA.Bailey, R. G. (1983). Delineation of ecosystem regions. Environ. Mgmt 7,365-373.Briggs, D. J. and France, J. (1983). Classifying landscapes and habitats for regional environmental planning. J.

environ. Mgmt 17, 249-261.Casetti, E. (1964). Multiple discriminant functions. Technical Report No. II, Computer Applications in the

Earth Sciences Project. Evanston, Illinois: Northwestern University, Department of Geography.Cobb, E. D. and Biesecker, J. E. (1971). The national hydrologic bench-mark network. Circular 460-D.

Washington, D.C.: U.S. Geological Survey.Crowley, J. M. (1967). Biogeography. Can. Geogr. 11,312-326.

248 Testing an ecosystem regionalization

Davis, L. S. (1980). Strategy for building a location-specific, multipurpose information system for wildlandmanagement. J. For. 78, 402-408.

Dixon, W. J. (ed.) (1981). BMDP Statistical Software 1981. Berkeley: University of California Press.Hutchison, N. E. (1975). W ATSTORE, National Water Data Storage and Retrieval System. Open File Report

75-426. Washington, D.C.: U.S. Geological Survey.Inkley, D. B. and Anderson, S. H. (1982). Wildlife communities and land classification systems. In

Transactions 47th North American Wildlife and Natural Resource Conference (K. Sabol, ed.), pp. 73-81.Washington, D.C.: Wildlife Management Institute.

Lachenbruch, P. A. (1975). Discriminant Analysis. New York: Hafner Press.Lachenbruch, P. A. and Mickey, M. R. (1968). Estimation of error rates in discriminant analysis.Technometrics 10, 1-10. .

Laut, P. and Paine, T. A. (1982). A step towards an objective procedure for land classification and mapping.Appl. Geogr. 2, 109-126.

Olson, R. J., Kumar, K. D. and Burgess, R. L. (1982). Analysis of ecoregions utilizing the geoecology database. In Proceedings In-Place Resource Inventories (T. Brann, ed.), pp. 149-156,9-14 August 1981, Orono,Maine. Washington, D.C.: Society of American Foresters.

Omernik, J. M., Shirazi, M. A. and Hughes, R. M. (1982). A synoptic approach for regionalizing aquaticecosystems. In Proceedings In-Place Resource Inventories (T. Brann, ed.), pp. 199-218,9-14 August 1981,Orono, Maine. Washington, D.C.: Society of American Foresters.

Omi, P. N., Wensel, L. C. and Murphy, J. L. (1979). An application of multivariate statistics to land-useplanning: classifying land units into homogeneous zones. Forest Sci. 25, 399-414.

Pavlik, H. F. and Hole, F. D. (1977). Soilscape analysis of slightly contrasting terrain in southwesternWisconsin. Soil Sci. Soc. Am. J. 41, 407-413.

Radloff, D. L. and Betters, D. R. (1978). Multivariate analysis of physical site data for wildland classification.Forest Sci. 24, 2-10.

Rowe, J. S. (1962). Soil, site and land classification. Forestry Chron. 38, 420-432.Rowe, J. S. and Sheard, J. W.(1981). Ecological land classification: a survey approach. Environ. Mgmt 5,

451-464.Sharpe, D. M. (1975). Methods of assessing the primary production of regions. In Primary Productivity of the

Biosphere (H. Lieth and R. Whittaker, eds), pp. 147-166. New York: Springer-Verlag.Steiner, D. (1965). A multivariate statistical approach to climatic regionalization and classification. Tijdschr.

K. ned. aardrijksk. Genoot. 82, 329-347.Zobler, L. (1957). Statistical testing of regional boundaries. Ann. Assoc. Am. Geogr. 47, 83-95.