eLearning

among plots can be inferred to represent changes
that would happen within single sites over time.
The strongest inferences come from individual sites,
where repeated sampling over time tests for
changes in forest N content.
In all three approaches to studying N accretion,
replication is critical to the level of confidence war-
ranted by the findings. We emphasize two aspects
of replication: the number of soil pits sampled
within each plot or site, and the number of plots or
sites examined. For example, in the classic chrono-
sequence of primary succession after glacier retreat
in Glacier Bay, Alaska, USA Crocker and Major
(1955) chose a sequence of eight sites that spanned
180 years of time. Each age class of system was
represented by one to three soil pits (the authors do
not state if the pits are in the same site or not), and
each soil pit was chosen to avoid rocks (which
comprise approximately 75% of the volume of the
soil on average). The overall trend in N content of
the soil with time showed a quadratic pattern (r
2
5
0.9, P , 0.01), increasing by approximately 20–25
kgNha-1
y-1
up to age 100, and then declining at a
rate of 7.5 kg N ha-1
y-1
. Chapin and others (1994)
did a more thorough study, with five randomly
selected sites for each of four age classes. They
found much lower rates of N accretion (2–8 kg N
ha-1
y-1
), and no mysterious N loss in older age
classes. The analysis by Chapin and others (1994)
treated each time period as a unique treatment in
an analysis of variance rather than addressing the
trend over time as levels of a single treatment. The
old spruce stage had 1000 kg N/ha by Crocker and
Major’s estimate that avoided rocky microsites,
which is approximately double the estimate found
by random sampling by Chapin and others (1994).
In the rating system we use in this article, the N
accretion estimate of Crocker and Major (1955)
warrants moderate confidence because of the lack
of clear description of sampling design, and the
biasing of the sampling away from the modal sites.
The estimate from Chapin and others (1994) war-
rants high confidence as a result of strong replica-
tion and unbiased sampling.
Repeated sampling within single sites also entails
critical issues of quality assurance in analysis of N
concentrations in soils. Sampling methods are chal-
lenging to reproduce, especially in sampling depth
(particularly the O/A interface) and correction for
changes in bulk density. If bulk density decreases by
10% between sampling intervals, then a 0–10-cm
soil sample would include only 90% of the soil that
would have been sampled at the same depth in the
earlier period. Analytical methods may differ be-
tween sampling periods, and even when methods