Premium
MICRODISTRIBUTION AND OTHER ERROR COMPONENTS OF C 14 PRIMARY PRODUCTION ESTIMATES 1 , 2
Author(s) -
Morrison Cassie R.
Publication year - 1962
Publication title -
limnology and oceanography
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.7
H-Index - 197
eISSN - 1939-5590
pISSN - 0024-3590
DOI - 10.4319/lo.1962.7.2.0121
Subject(s) - partition (number theory) , production (economics) , systematic error , mathematics , statistics , combinatorics , economics , macroeconomics
Four experiments have been carried out to partition the error components of C 1 4 primary production measurements into those associated with errors of technique and those due to real spatial variations in production. In Exp. 1, a well‐stirred, bacteria‐free culture of Skeletonema costatum was sampled to measure the minimum error obtainable by the technique. In Exps. 3 and 4, each sample was divided into two aliquots which were incubated separately to give a measure of error. The minimum coefficient of variation was found to be approximately 10%, both in the Skeletonema cultures and in natural populations. This is greater than would be expected from purely physical errors of technique or from chance variations in the number of organisms. It is suggested that a large component of the error is attributable to random variations in the photosynthetic capacity of the phytoplankton, possibly induced by disturbances in handling the samples. This is supported by the fact that, in Exps. 3 and 4, there was a systematic difference in production between the first and second aliquot, the second usually being the higher. Both Exps. 2 and 3 showed clearly that real variations of production could be detected between samples even within the relatively small spatial scale of sampling. From the autocorrelation, it would appear that any two water samples more than 10 cm apart are likely to show a significant difference in productivity. The frequency distribution of in situ production estimates is at least approximately log‐normal. Hence statistical tests based on the normal distribution can best be made on log‐transformed data.