3.6 Appendix
5.2.1 Study design and location
A broad-scale, spatial survey of 22 streams was conducted in April-May 2003 w forestry “status” represented as a single fixed factor with two levels: “unlogged “logged” (Table 5.1). Unlogged streams were those that had been subjected to minimal human disturbance during the last 100 years, while logged streams had subject to clear fell-burn-and-sow (CBS) harvesting and regeneration five
ith ” and been or fewer s a in years prior to sampling. Crossed with logging status was “geology” (2 levels:
“granite” and “dolerite-sedimentary” complex, hereafter “dolerite”). Eight stream located on granite were in the northeast of Tasmania and 14 streams located on dolerite in the south of Tasmania (Figure 5.1). All catchments were less than 50 h size and dominated by tall wet eucalypt forest prior to harvest.
Table 5.1 Study design of the spatial survey of 22 streams.
Forestry status Geology and location
Dolerite southern Tasmania Granite north-eastern Tasmania
Unlogged 6 4
Logged 8 4
5.2.2 Metabolic measures
production (GPP), daily respiration (R24) and net daily metabolism (NDM) were
d as All of the metabolic methods used in this part of the study were described in Chapter 2 and are as follows. The mini chambers were deployed in-situ to measure the benthic metabolism of four in-stream patches (depositional, coarse gravel, cobble or woody debris). Respiration and productivity were measured in mg C m-2 hr-1. Gross primary calculated for each patch, and reach scale values were estimated by multiplying average patch values by the relative proportion of habitats (see section 5.2.3). Reach scale sediment respiration (Reach R, in mg C m-2 hr-1) was determined using the ex-situ slurry method. The cellulose decomposition potential of depositional and coarse gravel sediments was measured using the cotton strip assay, and expresse cotton tensile strength loss (CTSL, in kg). Finally, the autotrophic potential was assessed using the algal bioassay to identify potential limiting nutrients, with the standing stock of algae measured in mg Chl-a m-2.
N
8 km
Figure 5.1 Maps of 22 study stream locations in the northeast and south of Tasmania, labelled by forestry coupe name.
N 8 km N 8 km N 8 km N 8 km N 8 km N 8 km N 8 km
5.2.3 Environmental variables
Catchment area and land use for each stream was calculated using GIS layers in ArcInfoTM 8.2 (ESRI, California, USA) based on 1:10 000 maps supplied by Forestry Tasmania, and ground truthed during the survey. Riparian vegetation at all sites was described using several indices based on presence/absence, diversity and canopy cover, as described in Chapter 2. In-stream habitat distribution was assessed by the random placement of a 250 cm2 sampling grid at 40 sites along each stream that identified the relative proportion of in-stream habitats. Other environmental data recorded at each site included water chemistry: total nitrogen (WTN), total phosphorus (WTP), dissolved reactive phosphorus (DRP), nitrate + nitrite (NOX), ammonia (NH4), dissolved oxygen (DO), pH, temperature, conductivity), sediment chemistry: total organic carbon (TOC), total nitrogen (TN), total phosphorus (TP), and sediment size classes were recorded as described in Chapter 2.
5.2.4 Data analysis
Uneven replication, small sample sizes (i.e. numbers of streams) and the presence of outliers in the data violated the assumptions of conventional factorial ANOVA, thus necessitating the use of alternative, robust methods (Tukey 1977, Staudte and
Sheather 1990, Quinn and Keough 2002). Instead of the arithmetic mean, Rousseeuw and Verboven’s (2002) version of Huber’s M-estimator modified for small sample sizes (their Equation 20) was used to compute a “robust mean”, and it will be referred to as such from now on. Instead of variance, “dispersion” was estimated using
Rousseeuw and Verboven’s (2002) Sn, which, again is a robust estimator modified for
small sample sizes (their Equation 25).
For the metabolic variables examined here, the design was a two-way layout with logging status (two levels: unlogged and logged) crossed with geology (two levels: granite and dolerite). Bootstrapping (Wilcox 2005) was conducted to test for the interaction between these factors and, where the interaction was non-significant, for
ain effect. Where logging status did interact ducted ted in R version 2.3.1 using functions modified from Wilcox (2005) to use Rousseeuw and Verboven’s (2002) robust differences between the levels of each m
significantly with geology, tests for the simple effect of logging status were con within geology. All analyses were conduc
estimators; examination of diagnostics from the bootstrapping followed those
recommended by Davison and Hinkley (1997). No problems emerged from any of the
ble
-2
e, observation time). In this case, however, data are “left-censored”, i.e. readings are
etection limit, but not by how much. As with survival analysis, the empirical cumulative distributions within each level of a factor can be
n tree analysis since this procedure permits exploration of non-linear and interactive effects while imposing very few assumptions
up. analyses.
For cotton strip assay data and mini-chamber data separate analyses were conducted for each patch type, except for wood, where descriptive data only is presented (Ta 5.2). All subsamples within a stream were averaged for each patch type.
Many of the chlorophyll-a measurements were below the detection limit of 1 mg m (Table 5.2), and so methods described by Helsel (2005) and implemented by Lee and Helsel (2005) were used to analyse these data. Essentially, these methods draw on the theory and practice developed for survival analyses, which usually deal with “right- censored” data (e.g. some subjects in study survive beyond the last observation tim so all that is known about them is that they have a survival longer than the last known to be below the d
compared with each other statistically, and this was done using the Peto & Peto test, which is most appropriate for left-censored data (Helsel 2005). A significant result from such a test means that at least one of the empirical cumulative distributions differs from the others. There are no formal tests of interactions in these procedures, but inspection of plots showed no evidence of any interactions. Accordingly, tests were conducted on main effects, where possible.
Relationships between metabolic response variables and environmental predictor variables were examined using regressio
(Venables and Ripley 2002). Splits were based on minimising sum-of-squares, with the minimum proportional reduction in residual variance being 5 %. The stopping size per group was set at 4, being the minimum number of replicates per treatment gro There was insufficient data to cross validate the models, and this was purely an exploratory analysis designed to reveal any potential environmental influences that could confound the patterns presumed to result from logging in a particular geology.
Table 5.2 Number of sites where data was available for analyses of patch level metab
measures. For chlorophyll-a, site numbers are where above detection limit values occurred.
Metabolic m
olic
easure Patch Geology and forestry status
Dolerite unlogged Dolerite logged Granite unlogged Granite logged
Productivity/Respiration Depositional sediment 6 8 4 4
Coarse gravel 4 7 4 4 Cobble 6 4 4 2 Wood 0 4 2 0 CTSL Depositional sediment 5 8 Coarse gravel 2 7 4 4 4 4 Chlorophyll-a Control 0 6 2 3 3 +Nitrogen 0 5 1 2 +Phosphorus 0 7 1
+Nitrogen and phosphorus 0 5 1 3
5.3 Results