7.1 Statistical designs
The choice of an appropriate test to reliably detect environmental disturbance due to human activities is often difficult if the magnitude of the impact is small relative to natural background parameter variability. As demonstrated for stream environments, the temperature regime varies longitudinally with stream order, width and elevation and can often be modified by land management activities that influence channel width, riparian canopy cover, pool volume, run-off timing and instream flow (Rhodes et al. 1994 cited in McCullough 1999). Moreover, stream temperature varies daily and seasonally and can be further modified by small-scale (wildfires, floods) and large-scale (global warming) disturbances. Therefore, the challenge exists to select a sampling design that can detect unusual patterns of change in a very interactive and variable measurement (Underwood 1994).
Paired studies represent the most robust means of comparing the effects of disturbance. The use of paired lakes or paired watersheds is most desirable among any sampling design, but it is often difficult to establish suitable pairs given the ubiquity of land development. As an alternative, sampling designs have evolved to contrast differences in parameters between disturbed and undisturbed sites within the same basin. In its simplest form in stream environments, environmental monitoring to detect change from a given disturbance has been facilitated by comparison of a measurement upstream and downstream from a known point source. An early environmental sampling design employed measurements before and after the initiation of a disturbance at control and impact locations; the design was originally referred to as BACI (Before-After-Control-Impact; Green 1979). The design has been criticized for its failure to account for measurement differences before and after the disturbance and directly related to an impact that could otherwise be explained by natural variation in space and time (i.e., the "pseudoreplication" argument advanced by Hurlbert 1984). The problem is quickly overcome with suitable replication of control and impact locations sampled over time, but may be restrictive due to the number of samples required (Stewart-Oaten et al. 1986; Underwood 1991, 1992 and 1994; Osenberg et al. 1994). The application of this design to rivers, lakes and streams is highly adaptable to the measurement of a variety of disturbances from human activity, but exceedingly more difficult in marine environments where replication of representative habitats or site conditions are required.
The procedure is relatively straightforward where a single putative impact occurs within a basin and becomes progressively more difficult where cumulative effects result from a variety of point and non-point sources, each with its own level of impact. In the simpler cases, application of a BACI design with paired sampling (Osenberg et al. 1994) or an asymmetrical design (Underwood 1994) is recommended for statistical comparisons. In the latter instance, additional control locations are incorporated into the design along with a single impact location to account for variability in nature and reliably detect impacts at either spatial or temporal scales. As with all statistical designs, the rules of representativeness and randomness must apply, and therefore in the case of single point source assessment, control and impact sites should be restricted to the same stream reach and located in similar habitat units. In the more difficult examples where multiple non-point sources occur, it may be more instructive to monitor temperature change at strategic points throughout the basin to determine where temperature maxima may limit the distribution and abundance of biota or impose restrictions on other beneficial uses. In this instance, strategic monitoring locations would likely correspond to areas of increasing stream order from headwaters to confluence. Locations would necessarily correspond to stream reaches above and below land-based developments within the basin. This approach would be highly consistent with ecosystem-level monitoring (i.e., holistic approach) upon which informed management decisions could be based.
7.2 Sampling protocols
The issue of sampling protocols for temperature collection seems relatively clear. In light of today's technologies, there is little question that continuous temperature records should be maintained. Continuous data loggers not only provide seasonality to temperature collections, but also provide the added advantage of identifying the frequency and duration of temperature extremes. In consideration of their cost relative to the quality and quantity of information provided, continuous records should become the standard with individual measurements collected at hourly intervals. The placement of temperature recorders within the thalweg and at near-shore margins would provide a more reasonable estimate of average temperature within the cross-section at each site. The three stations at each site would also serve as replicates for statistical analysis. The combined procedure would greatly assist calculations of mean weekly maximum temperatures (MWMT) for future enforcement of the recommended guidelines. The incorporation of a temperature recurrence interval (Bartholow 1989 cited in McCullough 1999) during temperature analysis would be beneficial in estimating the probability of exceedence of fixed temperature thresholds throughout the year that coincide with specific life history activities of individual fish species. Bartholow (1989) also recommended calculation of the cumulative probability of a sequence of events within a specified duration. Application of this information would suggest periods where fish may be at risk if the cumulative effects of consecutive days of sub-lethal temperatures impart negative biological responses (e.g., increased susceptibility to disease or poor growth). The output of this approach would certainly aid the decision-making process relative to adjudication's on future land-based developments.