HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via ISI Web of Science (1) Citing Articles via Google Scholar Google Scholar Articles by Lindbladh, M. Articles by Jacobson, G. L. Search for Related Content PubMed Articles by Lindbladh, M. Articles by Jacobson, G. L., Jr Agricola Articles by Lindbladh, M. Articles by Jacobson, G. L.

Morphometric analysis of pollen grains for paleoecological studies: classification of Picea from eastern North America¹

²Institute for Quaternary and Climate Studies and Department of Biological Sciences, University of Maine, Orono, Maine 04469 USA; ³Department of Wildlife Ecology, University of Maine, Orono, Maine 04469 USA

Received for publication December 6, 2001. Accepted for publication April 16, 2002.

	ABSTRACT

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION LITERATURE CITED

 
Little is known about the paleoecological histories of the three spruce species (white spruce, Picea glauca; black spruce, P. mariana; and red spruce P. rubens) in eastern North America, largely because of the difficulty of separating the three species in the pollen record. We describe a novel and effective classification method of distinguishing pollen grains on the basis of quantitative analysis of grain attributes. The method is illustrated by an analysis of a large sample of modern pollen grains (522 grains from 38 collections) of the three Picea species, collected from the region where the three species co-occur today. For each species X we computed a binary regression tree that classified each grain either as X or as not-X; these three determinations for each grain were then combined as Hamming codes in an error/uncertainty detection procedure. The use of Hamming codes to link multiple binary trees for error detection allowed identification and exclusion of problematic specimens, with correspondingly greater classification certainty among the remaining grains. We measured 13 attributes of 419 reference grains of the three species to construct the regression trees and classified 103 other reference grains by testing. Species-specific accuracies among the reliably classified grains were 100, 77, and 76% for P. glauca, P. mariana, and P. rubens, respectively, and 21, 30, and 22% of the grains by species, respectively, were problematic. The method is applicable to any multi-species classification problem for which a large reference sample is available.

Key Words: classification and regression-tree (CART) analysis • modern pollen grains • Picea glauca • Picea mariana • Picea rubens

	INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION LITERATURE CITED

 
Identification of fossil pollen usually involves a combination of analyst experience and taxonomic keys (e.g., Faegri and Iversen, 1964 ; Moore, Webb, and Collins, 1991 ) that requires binary decisions based on the morphology of the grain in question. Such keys, together with modern reference material, usually allow reliable identification of most pollen taxa. The taxonomic precision of the identifications depends on the natural variability of the group involved and on the degree to which the taxonomic variability has been recognized by palynologists. In many cases, only small and subtle differences in pollen morphology exist between closely related taxa, making precise identifications difficult or even impossible, and in other difficult taxa variability is evident only through careful statistical analysis. In some cases, changes in statistical distributions through time can reveal shifts in abundance of taxa (e.g., species within a genus), even though the identification of individual grains is often problematic.

Many difficult groups exhibit interesting ecological variability that would make reliable identification of species valuable. Pinus, Quercus, Rosaceae, Poaceae, and Picea are examples of genera or families that fall into this category. In a few cases, numerical and statistical approaches have been used with notable success. Hansen and Cushing (1973) , for example, gained useful paleoecological insights by applying numerical methods to differentiate pollen grains of five species of Pinus from southwestern North America.

Another interesting challenge involves Picea in northeastern North America. Given the difficulty of separating the three Picea species—Picea glauca, P. mariana, and P. rubens (white, black, and red spruce)—in the pollen record, little is known about their specific histories in eastern North America following deglaciation. Previous attempts to differentiate the pollen of these species were based on relatively limited reference collections and on collections from different parts of North America (Cain, 1948 ; Richard, 1970 ; Birks and Peglar, 1980 ). In this study, we present the results from analysis of a large sample of modern pollen grains from the three northeastern Picea species. We examined 522 pollen grains from 38 different collections and made seven quantitative and six qualitative measurements on each grain. All collections were from Maine and Maritime Canada, the only region where the three species co-occur today. Previous studies demonstrated clearly that a single character would not suffice for identification of the Picea species. Accordingly, Birks and Peglar (1980) , Hansen and Engstrom (1985) , and Brubaker, Graumlich, and Anderson (1987) used discriminant function analysis for separation of P. glauca and P. mariana. Although this method is widely available in modern statistical packages, and although the species assignments can be made with an individual probability estimate, the procedure lacks the transparency of the traditional taxonomic key. In this study, we applied a classification and regression-tree (CART) analysis to the data in order to distinguish pollen grains of all three Picea species. We wished to determine whether a classification method that combines the transparency of the traditional key with an objective and optimized numerical analysis of the data could be used to differentiate the reference grains.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION LITERATURE CITED

 
Pollen
We used pollen available from reference collections housed at the University of Maine Orono, Maine, USA, and the University of Minnesota, Minnesota, USA. In addition, we received pollen from two separate tree nurseries: Sussex Tree Nursery, New Brunswick, and Department of Natural Resources, Nova Scotia, Canada. All collections were from different individual trees from northeastern North America, where ranges of the three species overlap.

The pollen grains were washed in potassium hydroxide. They were treated with glacial acetic acid and then acetolysed with acetic anhydride and sulfuric acid for 2 min at 90°C. Silicon oil was added after the samples were dehydrated with tert-butyl alcohol. We avoided glycerin, as studies have shown that pollen may swell and to some extent change in morphological character when stored in this medium (Andersen, 1960 ). The samples were not stained. Prepared slides were sealed with paraffin. We added small amounts of sand grains (63–74 µm) to the slides in order to avoid flattened pollen grains because of cover-slip pressure (Cushing, 1961 ; Hansen and Engstrom, 1985 ). This was done in all but eight collections equally distributed among the three species. The mean grain size for the non-sand slides was slightly, but not significantly, bigger in P. glauca (Student's t test, P = 0.33) but was nonsignificantly smaller in P. marina (P = 0.97) and P. rubens (P = 0.64).

We examined a total 522 grains from 38 different collections. At least 20 grains were counted from each collection. Only grains that were unbroken, symmetrical, and fully expanded in equatorial view were used. The same person using a Leitz light microscope with 400x magnification and an ocularmeter that was calibrated to a stage micrometer did all the counting and measuring.

All except one of the seven quantitative variables measured by Birks and Peglar (1980) were measured in this study (Table 1, Fig. 1). We also tested all the qualitative variables suggested by other authors (Richard, 1970 ; Birks and Peglar, 1980 ; Hansen and Engstrom, 1985 ) as possibly useful for distinguishing Picea species (Table 2). Following a preliminary analysis, we excluded three of these as unimportant, either because there were no significant differences among the species or because the character was regarded as unclear or difficult to judge. The three excluded were: "thinning exine at the end of cappus," "ratio of the saccus/corpus area," and "saccus shape."

View larger version (14K): [in this window] [in a new window]   Fig. 1. The seven quantitative variables measured in this study. The variable designation (X1, etc.) are the same as in Birks and Peglar (1980)

Of the 13 remaining variables tested in this study, only four quantitative and two qualitative characters were ultimately determined to be useful in distinguishing the species. The four quantitative characters used were "total grain size" (X₁), "corpus breadth" (X₂), "cap thickness" (X₇), and "position of attachment of saccus to corpus" (X₈) (hereafter termed "saccus attachment") (Table 1). The two qualitative characters were the "degree of undulating margin of the corpus cap" (Y₂) (hereafter: "cap undulation") and the "degree of verrucation of the exine in the sinus area of the corpus" (Y₆) (hereafter "exine verrucation") (Table 2).

Measurements of these quantitative characters are straightforward. Of the qualitative characters used, "cap undulation" is relatively easy to detect (Fig. 2). The "exine verrucation" is a somewhat more subtle character, but consult Fig. 2, the drawing in Birks and Peglar (1980) , or photographs in Richard (1970) . (Those wishing to apply the method presented here will find that examination of reference grains helps enormously in recognizing the within- and between-species variability of the two qualitative characters in question.)

View larger version (122K): [in this window] [in a new window]   Fig. 2. Characteristic modern pollen grains of the three Picea species. (A–B) Picea glauca, (C–D) Picea mariana, and (E–F) Picea rubens

Classification and regression-tree (CART) analysis
The CART models recursively partition data sets on the basis of a set of independent variables. Here such a classification analysis has been applied to the morphometrics of the three Picea species (Fig. 3). The CART procedure first tested each morphometric variable to find the best combination of variable and split threshold that separated the entire sample into two groups that were internally as homogenous as possible with respect to species composition. In this case (Fig. 3), "total grain size" was the best such variable, with a size threshold of 88.5 µm. Relatively few grains of or below this size were P. glauca and most of those above this size were P. glauca. Each of these two subsets was then partitioned in turn, producing a substantial separation of P. rubens from P. mariana on the basis of the score for "cap undulation" with P. rubens typically scoring 1 or 2 and P. mariana scoring 3, 4, or 5. In the subset of larger grains, on the other hand, "corpus breadth" was the best discriminator variable, with the broader grains typically being P. glauca. Note how different criteria can apply within the two sibling subsets, allowing the types of contingent rule typical of taxonomic keys to emerge in a way not possible with multiple regression or discriminant function analysis. (There are other ways of achieving this with statistical methods, but they are either not straightforward or they require prior knowledge not needed with the CART procedure.) The process is repeated recursively through descendent nodes so that a decision tree is grown until certain stopping rules are encountered. At this point the resulting tree was pruned back to an optimally fitting version determined by cross-validation. The result is a decision tree that specifies a hierarchically organized suite of correlates of species identity.

View larger version (10K): [in this window] [in a new window]   Fig. 3. Structure of classification tree for the simultaneous classification of all three Picea species. At each interior node the splitting variable and threshold value are shown and at each end node the commonest species and the probability of being that species with the combination of splitting criteria from end node to root are shown

The classification accuracy of the classification tree constructed in this way can be assessed in the usual way, as the proportion of the sample of known identity that are correctly classified by the classification tree. However, this proportion is likely to be inflated because the tree is constructed (via the exhaustive search procedure) so as to maximize the classification accuracy. This bias is controlled via the cross-validation pruning, but this involves a trade-off coefficient that weights classification accuracy against tree complexity (Breiman et al., 1984 ; Clark and Pregibon, 1992 ). Hence, a permissive choice of this coefficient allows optimistic results, akin to the use of alpha levels of 0.10 or 0.15 in admitting terms to a step-up, multiple linear regression. An independent assessment of the efficacy of the pruned tree is therefore best obtained by making predictions from the model against a reserved sample of data (test set) not used in the derivation of the initial CART model (in this study we used a training set of 419 grains and a test set of 103 grains).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION LITERATURE CITED

 
Table 1 summarizes the distribution of the quantitative variables. Picea glauca was markedly larger in all metrics except "cap thickness," while P. mariana and P. rubens were relatively similar and P. mariana had smaller "total grain size" and "cap thickness" and larger "saccus attachment" than P. rubens. Nevertheless, despite these significant differences in the quantitative characters the considerable overlaps preclude differentiation of the three taxa on the basis of any one character alone. The mean sizes for P. glauca and P. mariana for all variables were considerably larger in this study compared to the study of Birks and Peglar (1980) . For example, "total grain size" for P. glauca was 98.9 µm in this study compared to 89.7 µm in the Birks and Peglar study; for P. mariana the measurements were 84.6 and 74.7 µm, respectively. These differences are so large they cannot be explained by the difference in sample size (i.e., number of collections) alone and probably reflect differences in collection locality: all collections in the Birks and Peglar study were from the Midwest, while in this study all were from eastern North America. The collections of P. rubens in both studies were from eastern North America, and the means of each variable were similar in the two studies. (The mean for "total grain size" reported in Birks and Peglar [1980 ] is actually slightly larger [91.2 µm compared to 89.7 µm]).

Only two qualitative variables exhibited consistently large differences among the species (Table 2). Pollen from Picea rubens had more often a "cap undulation" than pollen from P. glauca and P. mariana. On average 66% of the P. rubens pollen had this feature, but significant variation existed among collections (see Appendix, http://ajbsupp.botany.org/v89). The second variable with some differences was the "exine verrucation." It was more commonly a character on P. rubens and P. mariana grains (88 and 78%) than grains of P. glauca (54%). Interestingly, these two characters were also identified as critical discriminators in other studies, "cap undulation" by Birks and Peglar (1980) and "exine verrucation" by Richard (1970) . However, none of the qualitative variables could be used alone to separate the three spruce species without an unacceptable rate of misclassification. See Appendix at http://ajbsupp.botany.org/v89 for the entire data from all collections.

Traditional classification trees
Standard CART analysis across the three species reveals that Picea mariana grains could be segregated from the others by a combination of a small "total grain size" (<88.5 µm) and a large "cap undulation" score (>2.75) (Fig. 3). Picea rubens grains were identifiable either by the combination of a small "total grain size" (<88.5 µm) and a low "cap undulation" score (<2.75) or by a combination of short "corpus breadth" (<69.5 µm) and low "exine verrucation" score (<2.75) if "total grain size" was >88.5 µm. Longer grains were otherwise classified as P. glauca.

The probability of grains in a given end node in Fig. 3 being of the species shown for that end node varied across nodes. From left to right across the end nodes, these probabilities were 0.65 (P. rubens), 0.74 (P. mariana), 0.62 (P. rubens), and 0.66, 0.55, and 0.96 (all P. glauca). Note that the two rightmost end nodes are both P. glauca, so what knowledge the "exine verrucation" score provides is not discrimination between two species but rather greater certainty of the identification as P. glauca when this score is above 2.25. Thus, although the overall misclassification rate of the tree is 30.6%, the rate of misclassification of grains varies across nodes.

Table 3 shows the classification accuracy of the tree for each species in the training set. Species-specific classification accuracies were 77.1, 59.7, and 71.3% for P. glauca, P. mariana, and P. rubens, respectively. However, these figures reflect a classification tree structured to optimize its fate to the training data and therefore are likely to be overoptimistic. Table 3, therefore, also shows the results of applying the classification tree of Fig. 3 to the reserved data set of 103 cases (i.e., grains not used in the training/calibration analysis), yielding figures of 86.2, 59.5, and 73.0%, respectively. Thus, those of the test-set analysis matched closely the accuracy estimates from the training set.

Figure 4a–c presents the binary trees for each species. Picea glauca (Fig. 4a) was discriminated from the other species by the combination of narrow "corpus breadth," large "saccus width at base" and a high score of "exine verrucation," or by a broad "corpus breadth" and medium to high score of "exine verrucation" (>1.75). Classification accuracy for the training set was 84.0%. For P. mariana (Fig. 4b), classification as the species was based on a short "total grain size" (<86.5 µm) and a high score of "undulating exine" (>2.75). All other cases (i.e., those not classed as P. mariana (see previous sentence) were classed as non-P. mariana. Overall P. mariana classification accuracy was 79.7%. Finally, P. rubens (Fig. 4c) was discriminated from the other species by virtue of a low score of "undulating exine" (<2.75) and narrow "corpus breadth" (<64.5 µm), with overall accuracies of classification equal to 78.5%. These classification accuracies are again from the training set and lower estimates are assessed from the test set, namely 79.3, 56.8, and 59.5% for P. glauca, P. mariana, and P. rubens, respectively.

View larger version (11K): [in this window] [in a new window]   Fig. 4. Structure of binary classification trees for (a) Picea glauca, (b) Picea mariana, and (c) Picea rubens. For each end node a "1" indicates that the focal species was the commonest in that node and a "0" indicates that individuals of the two nonfocal species for that tree were commonest there. The P value gives the probability in a given end node for being the focal species. For example, in Fig. (a): 84% of the grains in the training set that had the combination "Corpus breadth > 69.5 µm" and "Exine verrucation > 1.75" were grains of Picea glauca

View larger version (29K): [in this window] [in a new window]   Fig. 5. Distributions of "total grain size" for grains of each of the three Picea species according to whether the binary triplet procedure classified the grains correctly to species or whether the grain proved problematic in all cases for the training data. From top to bottom the distributions are: Picea glauca (unequivocally classified [black]; unidentifiable or equivocal [shaded]), Picea mariana (unequivocally classified [black]; unidentifiable or equivocal [shaded]), and Picea rubens (unequivocally classified [black]; unidentifiable or equivocal [shaded])

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION LITERATURE CITED

The notion of using Hamming codes to combine the results of binary classification trees appears to be an innovation in classification work. The resulting raw classification rates were necessarily lower than those for the binary trees (Table 5), but reliance on the binary trees alone may be risky: the training set binary classification rates were markedly higher than those from the standard multispecies classification tree—84.0 vs. 77.1% for P. glauca, 79.7 vs. 59.7% for P. mariana, and 78.5 vs. 71.3% for P. rubens—but the values obtained from the test set with the binary trees were lower (79.3, 56.8, and 59.5%, respectively). It is likely, therefore, that the binary results for individual species involve significant of overfitting to the training data. When combined as Hamming codes, however, the species-specific binary trees have the potential to reinforce each other in clear-cut cases and to reveal conflicting classifications with intermediate grains. Thus, a major advantage of the Hamming codes is to reveal those cases for which classification is uncertain. That it is the difficulty of classification rather than a deficiency in the classifying algorithm is indicated by the finding here that the problematic cases in our training set were of intermediate morphology, often too large in a given dimension to be P. mariana or P. rubens but at the same time smaller than typical P. glauca grains (Fig. 5). In operational terms, this ability of the Hamming codes to identify problematic grains that would otherwise reduce the overall classification accuracy is of special value in palynological studies. (It is, of course, possible in a discriminant function analysis to plot the location of individual grains within the n-dimensional space of the n discriminant functions and to determine which grains are peripheral to the species-specific clusters identified. This approach is less direct, and more complex, than the generation of an unequivocal Hamming code). Intuitively, one would expect that conclusions based on relatively reliable estimates of species composition from a subset of available grains are scientifically superior to less reliable estimates made from the complete sample. Furthermore, any spatial/temporal variation in the proportion of unclassifiable grains in the samples for different locations or times may itself be informative. It is worth noting that the Hamming procedure may be of relatively greater value as the number of species involved in the study increases. In principle, Hamming codes can be created for samples with four, five, six, etc. species, yielding quadruplet, quintuplet, and sextuplets instead of triplets (with correspondingly more values, e.g., 2⁶ = 64 values for a six-species mix). One would intuitively expect discriminant function analysis to be relatively more difficult as the number of species in the mixture increases, so that Hamming codes identifying problematic cases would be increasingly valuable. On the other hand, the possibilities for equivocal codes increases in a binomial distribution, and one could envisage a relatively greater frequency of equivocal rather than unclassified grains as species richness increases. However, this is not a problem for analysis of grains in less diverse samples.

Conclusions
The large number of pollen (522) from many collections (38) used in this study was the prerequisite for the separation of the three closely related Picea species. The large between- and within-species similarities discovered in the study emphasize the need for a large collection of reference grains. Furthermore, the apparently large differences in the quantitative characters among the Picea pollen from different regions (i.e., eastern North America and Midwest) suggest that applying the comparative data from one region to another might not work. It may be useful for the analysis to be repeated for each region of interest. Classification tree analysis appears to provide a robust and transparent procedure, and it is at least as powerful as traditional discriminant function analysis, while retaining the conceptual simplicity of traditional taxonomic keys. Use of Hamming codes to link multiple binary trees allows identification and exclusion of problematic specimens and correspondingly greater classification certainty among the remaining grains. This study shows that there is a large proportion of problematic specimens in a modern reference collection of Picea pollen, and thus it is likely that a number of fossil grains will also fall into this category. Hence, the new method allows more confidence than is possible with previous methods about the identity of unequivocally classified grains.

The method has successfully been applied to fossil Picea grains from eastern North America (Lindbladh, Jacobson, and Schauffler, in press ). It is not restricted in applicability to Picea but can readily be applied to any other taxon for which a large reference collection is available. Extensive experience with CART indicates that an initial sample of several hundred cases is needed before adequate discrimination is achieved. With smaller samples, the algorithm fails gracefully rather than proceeding to misidentify cases.

	FOOTNOTES

 
¹ The authors thank the following for their kind and constructive help with this project: Ed Cushing (Minnesota), Hartmund Kunze (JD Irving Ltd. New Brunswick) and Howard Frame (Nova Scotia Dept. of Natural Resources) for providing modern pollen grains of Picea; Isabelle Larocque (Sweden/Switzerland) for translation of some French text; to Ann Dieffenbacher-Krall (Maine) for considerable assistance in the laboratory and for helpful discussions; PM Ekö (Sweden) for his exceptional knowledge in pollen-related computer programming and statistical analysis; and Molly Schauffler and Shawn Fraver (Maine) for helpful discussions. The study was supported in part by grants to M. Lindbladh from The Swedish Foundation for International Cooperation in Research and Higher Education (STINT) and from the Fulbright Commission; by a grant to A. White, G. L. Jacobson Jr., and W. Krohn from The Nature Conservancy (U.S.); and by the University of Maine Institute for Quaternary and Climate Studies.

⁴ Author for reprint requests, current address: Southern Swedish Forest Research Centre, Swedish University of Agricultural Science (SLU), P.O. Box 49, S-230 53 Alnarp, Sweden (matts.Lindbladh{at}ess.slu.se) .

	LITERATURE CITED

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION LITERATURE CITED

 
Andersen T. 1960 Silicone oil as a mounting medium for pollen grains. Danmarks Geologisk Undersögelse 4 4(1): 1-24

Birks H. J. B. S. M. Peglar 1980 Identification of Picea pollen of Late Quaternary age in eastern North America: a numerical approach. Canadian Journal of Botany 58: 2042-2058

Breiman L. J. H. Friedman R. A. Olshen C. J. Stone 1984 Classification and regression trees. Wadsworth, Belmont, California, USA

Brubaker L. B. L. J. Graumlich P. M. Anderson 1987 An evaluation of statistical techniques for discriminating Picea glauca from Picea mariana pollen in Northern Alaska, USA. Canadian Journal of Botany 65: 899-906

Cain S. A. 1948 Palynological studies at Sodon lake. I. Size frequency study of fossil spruce pollen. Science 108: 115-117[Free Full Text]

Clark L. A. D. Pregibon 1992 Tree-based models. In J. M. Chambers and T. J. Hastie [eds.], Statistical models in S, 377–419. Wadsworth & Brooks/Cole Advanced Books & Software, Pacific Grove, California, USA

Cushing E. J. 1961 Size increase in pollen grains mounted in thin slides. Pollen et Spores 3: 265-274

Faegri K. J. Iversen 1964 Textbook of pollen analysis, 2nd ed. Blackwell Scientific, Oxford, UK

Fisher R. G. 1936 The use of multiple measurements in taxonomic problems. Annals of Eugenics 7: 179-188[ISI]

Hamming R. W. 1950 Error detecting and error correcting codes. Bell System Technical Journal 29: 147-160

Hansen B. S. E. J. Cushing 1973 Identification of pine pollen of Late Quaternary age from the Chuska Mountains, New Mexico. Geological Society of America Bulletin 84: 1181-1200[Abstract]

Hansen B. C. S. D. R. Engstrom 1985 A comparison of numerical and qualitative methods of separating pollen of black and white spruce. Canadian Journal of Botany 63: 2159-2163

Jackson S. T. C. Weng 1999 Late Quaternary extinction of a tree species in eastern North America. Proceedings of the National Academy of Science 96: 13847-13852[Abstract/Free Full Text]

Lindbladh M. G. L. Jacobson Jr. M. Schauffler In press The postglacial history of three Picea species in New England, USA. Quarternary Research

Moore P. D. J. A. Webb M. E. Collins 1991 Pollen analysis, 2nd ed. Blackwell Scientific, Oxford, UK

Richard P. 1970 Atlas pollinique des arbres et de quelques arbustes indigenes du Quebec. I Introduction generale. II Gymnospermes. Naturalist Canadian 97: 1-34

This Article Abstract Full Text (PDF) Submit a response Alert me when this article is cited Alert me when eLetters are posted Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Similar articles in ISI Web of Science Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via ISI Web of Science (1) Citing Articles via Google Scholar Google Scholar Articles by Lindbladh, M. Articles by Jacobson, G. L. Search for Related Content PubMed Articles by Lindbladh, M. Articles by Jacobson, G. L., Jr Agricola Articles by Lindbladh, M. Articles by Jacobson, G. L.