Download Accuracy of Quadrat Sampling in Ecological Reproduction Studies: A Case of Dalkena Area and more Schemes and Mind Maps Social ecology in PDF only on Docsity!
ACCURACY OF^ QUADRAT SAMPLING^ IN^ STUDYING^ FOREST
REPRODUCTION ON CUT-OVER AREAS
I. T. HAIG Northern Rocky Mountain Forest Experiment Station, U. S. Forest Service
PURPOSE
The quadrat method, first introduced into ecological studies by Pound and Clements in i898, has been adopted by both foresters and ecologists as one of the most accurate means of studying the occurrence, distribution, and development of vegetation (Clements, '05; Weaver, 'i8). This method is unquestionably more precise than the descriptive method which it super- seded. (^) Nevertheless, the results obtained are subject to certain types of experimental error in common with all methods of sampling in which it is necessary to picture conditions over large areas from representative frac- tions. Despite its long and wide-spread use, few investigators have seen fit to check the accuracy of the quadrat system, and our knowledge on this subject, particularly in the field of^ silvical investigations, is extremely lim- ited. It is the purpose of this paper to discuss the accuracy of quadrat sampling as applied to the study of forest reproduction on cut-over areas, and to describe methods by which this accuracy can be measured.
THE DATA USED The data here presented were gathered in a study of old cuttings in the western white pine forests of northern Idaho. In this study, reproduction counts were obtained on mil-acre quadrats (squares 6.6 feet on a side) laid out at intervals along parallel strips running at right angles to the main topographic features. The usual procedure was 'to lay out quadrats at one half to one-chain intervals (that is at 33 to 66 feet intervals) and to space adjacent strips 2.5, 5 or i0 chains apart. Two important averages obtained from these data were: (i) the percentage of area stocked, a value commonly termed the frequency index or F. I. (Gleason, '20), and (2) the average number of seedlings per acre. These figures are essential for weighing the merits of various methods of cutting in regard to obtaining satisfactory re- production. They are based, in most reproduction studies, on an extremely small sample; in this, instance on a sample of from .i to .8 per cent of the total area; and it is obviously desirable to check their accuracy.
1 Presented at the Annual Meeting of the Northwest Scientific Association, (^) Spokane, Wash., December 28, i928.
374
This file was created by scanning the printed publication. Errors identified by the software have been corrected; however, some errors may remain.
October, I929 ACCURACY OF QUADRAT SAMPLING 375
FREQUENCY INDEX IN OCCURRENCE OF (^) SEEDLINGS Accordingly, frequency index (percentage of area stocked) was checked in the following manner. In addition to tallying reproduction on quadrats at half-chain or one-chain intervals along parallel strips, the usual method of study designated in this paper as the quadrat-at-interval method, a tally was made on strips of contiguous quadrats distributed over the same area in a similar manner. This tally simply noted the presence or absence of reproduction by individual quadrats, and furnished data from which to compute F. I.'s based on from five to twenty times the usual sample as ob- tained by the quadrat-at-interval method. These (^) frequency indices, there- fore, act as a standard against which to check the accuracy of the F. I.'s obtained by the quadrat-at-interval method, and a comparison of this sort for six cut-over areas is shown in Table I.
TABLE I. Check of frequency indices
Cut-over Area Number Method of I^2 3 4 5 Sampling F.^ I. Basis^ F.^ I. Basis^ F.^ I. Basis^ F.^ I. Basis^ F. I. Basis^ F.^ I. Basis
_% Quads^ %^ Quads^ %^ Quads^ %^ Quads^ /% Quads^ %^ Quads Quadrats con- tiguOus...... (^90 468 92 782 84) 9I3 92 479 69 759 86 507 Quadrats at in- tervals.^ 88..... 7I 100 45 96 96 98 I09 77 I23 (^100 )
The reasonably close agreement of these figures indicates that the quad- rat-at-interval method gives sufficiently accurate results in spite of the meagerness of the sample. The significant differences, however, show the need for caution in deciding the relative merits of two cut-over areas on the basis of differences in frequency index. Such differences might be attributed, in part or in whole, to accidents of sampling as well as to dissimilarity in conditions on the cut-over areas or in methods of cutting.
AVERAGE (^) NUMBER OF SEEDLINGS PER ACRE
Checking the values for average number of seedlings per acre proved a much more difficult task. Statisticians (Student, 'o8) offer a very definite method of checking the accuracy of averages through the use of the prob- able error concept. But it is generally assumed that this statistical measure is directly applicable only to normal or nearly normal frequency distribu- tions, illustrated in figure I, A, in which the individual values obtained in sampling are distributed around the average value in an orderly progressive fashion, though it is also held probable that the deviation from normality
October, I929 ACCURACY OF QUADRAT SAMPLING 377
must be extreme to lead to serious error. In this study the frequency dis- tributions are of an extreme type, being very much skewed in a plus direc- tion. This type of distribution, based on the data for one of the cut-over areas studied, is shown in figure I, B. (^) Ordinary statistical methods for measuring the accuracy of averages cannot be applied accurately to this type of distribution, and it was necessary to convert these distributions to normal or more nearly normal distributions before such checks could be made. METHOD OF TESTING DISTRIBUTIONS A review of available statistical literature failed to suggest any method of testing the accuracy of (^) averages obtained from such J-shaped distribu- tions, or of converting such distributions to more normal (^) proportions. Quartile deviation is the measure of dispersion often recommended for use with skewed distributions (Mills, '24), but this obviously is not applicable to the arithmetical average, the item of major interest in this particular study. It was necessary, therefore, to develop methods of treatment appli- cable to this extremely skewed type of distribution. Close study of the frequency curve in figure I, B suggested certain steps that finally led to successful treatment. Although it was apparent that the definite skewness of this curve was the result of a combination of causes, the very definite J- shaped character of the curve was due chiefly to the fact that the average number of seedlings per quadrat was so small as to place the peak of the frequency curve very close to the zero point. The solution seemed the use of a larger quadrat containing on the average a greater number of seedlings. This would push the peak of the frequency curve in a plus direction and eliminate in large measure the J-shaped character of the distribution. There are numerous objections to the use of any quadrat other than the mil-acre as a field unit of measure, but none whatever to the use of a larger unit in the office analysis. Consequently, mil-acre quadrats were selected at ran- dom, arranged in groups of four or eight quadrats and the total sum of seedlings for each of these groups used as the number of seedlings per new unit measure. This grouping had the same effect as if (^) the field quadrats had been four or eight times as large and only one-fourth or one-eighth as many had been installed. Note that groups of 4's or 8's were used simply as a matter of convenience. The number of mil-acres incorporated in each group might have been any number desired, the sole criterion being that the number in each unit must be large enough to move the average number of seedlings per quadrat, and hence the peak of the frequency curve, a rea- sonable distance away from the zero ordinate. This method of grouping eliminates the J-shaped character of the dis- tribution, but it also reduces the number of quadrats on which the frequency curve is based and tends to produce weak and erratic distributions. To (^2) This method of grouping was originally suggested (^) to the author by Mr. Donald Bruce.
378 I. T. HAIG Ecology, Vol. X, No.
overcome this objectional feature, this procedure of grouping was repeated any desired number of times, the entire set of original mil-acre values being used in each series, but the individual units in each series being formed by new combinations of different mil-acre values. In this way a new set of values was built up, the number of quadrats in this array being steadily in- creased by additional combinations of the original data until these values produced a reasonably smooth frequency curve. This type of distribution is shown in figure I, C. Notice that this curve is still skewed. It was found impossible to entirely eliminate this factor, but the distribution is now near enough normal to permit the application of ordinary statistical checks with reasonable safety. The use of the same basic data over and over in building up this new series of quadrats raised some question as to the accuracy of this process and its consequent effect on the probable error. Accordingly, a check was made of the effect of this process. The values in Column 2, Table II, show very definitely that as the number of combinations increases, that is as the same basic data is used over and over an increasing number of times to form values for new quadrats, the probable error appears to decrease with each additional repetition, changing from ii.i per cent (the best available figure for probable error based on an average of the probable errors of six individual groupings by 4's or 8's) to 3.5 per cent on a basis of ten such combinations. TABLE II. Check of probable error (^) comfputations Probable Error Per Cent Number of Combinations Uncorrected Value Corrected Value (I) (^) (2) (3) I...................... II. (^) IIJ (^3) ...... ......... (^) .... 6.5 (^) II. (^5) .5.I (^) ...................... (^) II. 7 .*.^ 4.3 I I. IO10......... 3.5 II.I ? Most reliable figure; average of six probable errors each based on (^) one combina- tion by 4'S or 8's.
This reduction is obviously an inaccuracy introduced by the repeated use of the basic data in various combinations, for ordinarily the probable error for this array would remain in the neighborhood of (^) ii.i per cent unless changes were introduced directly into the basic data. It was reasoned that the decrease in probable error must be directly related to the square root of the number of combinations and that this error could be eliminated by multiplying the probable error of an array by the square root of the num- ber of combinations. This assumption was checked empirically and found correct. The corrected values for probable error listed in Column 3, Table II, were derived by multiplying the uncorrected values in Column 2 by the
380 I. T. HAIG (^) Ecology, Vol. X, No.
probable error) of the actual number of (^) seedlings per acre that would be secured by a one-hundred-per-cent cruise of the cut-over area. This state- ment is based on the usual statistical definition of probable error, namely, that if each cut-over area were resampled, the new samples consisting of the same number of mil-acre quadrats as before, distributed in a similar manner, the new averages would fall within the range formed by the present average plus or minus the probable error in approximately 50 per cent of the cases, and within a range of the present average plus or minus three times the probable error in practically all or approximately 95 per cent of the cases. The practical value of these probable errors is (^) apparent. First, they offer assurance that the method of sampling employed has given averages accurate enough to insure a true picture of reproduction conditions, for all of the probable errors in this group are less than IO per cent. (^) Second, the probable errors show clearly that in weighing the merits of two cut-over areas, differences in values of average number of seedlings per acre of less than ten per cent are probably not significant (these can very easily be due to accidents of sampling alone) and that even differences of as much as (^25) per cent must be classed as significant only with reservation.
LOCATION OF QUADRATS There are two other matters of interest that might be mentioned in con- nection with quadrat sampling of cut-over areas. First, that it is essential to distribute the strips very thoroughly over the area. Errors as high as 6o per cent were incurred by concentrating strips on one portion of the area as opposed to distributing these strips at wider intervals over the entire area. This is in agreement with the conclusions reached by Gleason ('20), who, in studying the location of quadrats in ecological studies, found that the use of one-fourth of the data in contiguous quadrats, that is a concen- tration of the sample, resulted in an error of (^) IO.7; while the use of one- lourth of the data in non-contiguous quadrats gave a corresponding error of only 2.7. The second matter was that within certain common-sense limits, given the same number of mil-acres per unit of cut-over area, the greatest accuracy was obtained by distributing the quadrats along closely spaced strips with larger intervals between quadrats as opposed to widely spaced strips and smaller intervals between quadrats. This statement is based on a study of the effect of distribution of strips on four cut-over areas. The average error introduced by using only half the data with twice the usual interval between quadrats was I7 per cent, as compared to an error of 32 per cent incurred by using the same amount of data with twice the usual distance between strips.
October,i929 ACCURACY OF QUADRAT (^) SAMPLING 38i
SUMMARY The purpose of this paper has been to discuss the accuracy of quadrat sampling as applied to reproduction studies in the western white pine type. and 'to discuss methods by which this accuracy can be measured. In gen- eral, the methods used in this study consisted in counting reproduction on mil-acre (6.6 feet square) quadrats distributed at one-half or (^) one-chain (33 and 66 feet)^ intervals along parallel strips 2.5 to io chains apart. This quadrat-at-interval system, giving a sample of from .I to .8 per cent of total area, was found to give satisfactory values for both frequency index (percentage of area stocked) and average number of seedlings per acre. In checking the values for average number of seedlings per acre, a method is suggested by which the J-shaped frequency distributions, probably common in similar ecological and silvical studies, can be converted into more nearly normal distributions and so strengthened as to permit the application of the probable error concept with a reasonable degree of safety. Indications are given that in sampling by the quadrat-at-interval method it is essential to have the parallel strips well distributed over the area, and that when only a limited sample can be taken the tendency should' be to lengthen the interval between quadrats rather than the distance between strips.
LITERATURE CITED Bruce, Donald. 1924. The rejection of observations. Mimeograph Report. U. S. Forest Service. Clements, F. E. 1905. Research methods in ecology. University Publishing Co., Lincoln, Nebraska. Gleason, H. A. 1920. Some applications of the quadrat method. Contri. N. Y. Bot. (^) Gardens No. 216. Mills, F. C. 1924. Statistical methods. New (^) York, Holt and Co. Pound, R., and Clements, F. E. 1898. A method of deterniining the abundance of secondary species. Minn. Bot. Studies, 2: 19. Student. 1908. The probable error of a mean. Biometrika, 6: 1-25. Weaver, J. E. 1918. The quadrat method of teaching ecology. Plant World, 21: 267-283.
