Linking Human Disturbance To Biological Change (continued)

Metric testing included safeguards against circular reasoning

One risk associated with using correlative data to demonstrate connections between human disturbance and biological degradation is that the observed correlation may be due to spurious correlation with another underlying cause, such as elevation or watershed size that drives both biology and patterns of human settlement. Or, the observed correlation may be due to some additional factor that was not considered. When developing biomonitoring tools, the goal is to select biological indicators that vary only as a function of human disturbance and are immune to variability associated with natural physical or geographic features. Unfortunately, humans are biota, too, and their land use patterns tend to follow landscape features, thus confounding human activities with physical features. The challenge for the MAIA project was to demonstrate that human disturbance was the most likely agent of biological change.

A variety of safeguards helped reduce the probability of drawing unsubstantiated conclusions from the MAIA data analysis. Five approaches were used to isolate the relationship between human disturbance and biological change from other confounding influences.

First, site selection was randomized across a large geographic area to ensure that the sample was representative of the entire population of possible sites. Unbiased selection of sites provides some protection against confusing the effects of human disturbance with other, natural features (Stewart-Oaten et al., 1986; 1992).
Second, measures of disturbance were selected independently of the biological metrics. Bryce et al. (1999) present an integrated definition of human disturbance for the Mid-Atlantic region derived without consideration of biological indicators.
Third, metrics were tested in multiple years, or part of the data set was reserved to test the final indexes as an independent test (McCormick et al., 2001; Fore, 2002b; Klemm et al., 2003). In this way the observed relationships were demonstrated to be consistent across years.
Fourth, all metrics were tested for correlation with multiple gradients of human disturbance rather than for their simple ability to distinguish between one set of minimally disturbed, or reference, sites and severely degraded, or impaired, sites. Thus, the metrics selected were consistently associated with different aspects of human disturbance.
Fifth, potential confounding factors such as watershed area or elevation that could underlie patterns of both human influence and biological condition were explicitly tested. Where necessary these effects were removed, for example, when fish taxa richness metrics correlated with watershed area. In this way, metrics were selected for their association with disturbance rather than other natural features.

Patterns of human disturbance were complex

The MAIA pilot had the luxury (and the curse!) of data for practically any variable that has ever been recorded to evaluate water resources. Dozens of variables related to water chemistry, metals, nutrients, fish tissue contaminants, habitat, channel morphology, geographic features, human census data, satellite land cover and use, and specific point sources were included in the data set. Hundreds more were derived from the data collected. The hope was that such a complete record of human activity would provide a clear picture of human influence and disturbance within a watershed. The reality was that the different measures of disturbance tended to tell their own story; that is, different measures were associated with specific types of human activity. Therefore, disturbance measures were not necessarily correlated with each other because not all activities were present in every watershed. As a consequence, one of the primary challenges for the MAIA pilot study was to determine which variables most accurately characterized human influence.

During the process of developing biological indicators, much discussion surrounded the choice of appropriate measures of site condition for metric testing. A primary lesson learned in the Mid-Atlantic was that no simple method existed to quantify human influence in such a complex landscape with such a long and varied history of human activity (Herlihy et al., 1998; Bryce et al., 1999, McCormick et al., 2001). A missing piece from this project was a comprehensive study linking the types of human activities (e.g., mining or agriculture) with their specific stressors (e.g., SO₄ or nutrients). Given the tendency to find multiple types of disturbance in each watershed, a clear picture may not have been possible. Nonetheless, a better understanding of which of the many measures of disturbance tended to vary together along with a better understanding of which measures were related to natural geographic or landscape features would have helped clarify metric response to disturbance. For example, diatom metrics were correlated with elevation but so was disturbance because towns and farms tended to be found at lower elevations (Fore, 2002b).

Examination of a correlation matrix of all the variables related to site condition (too large to show here) revealed patterns of correlation among related variables. Sets of variables could be identified that seemed to measure similar or underlying processes. Variables related to water chemistry, nutrients, and water quality tended to be significantly correlated with each other. Other sets of variables related to channel structure, fish cover, or the condition and extent of riparian (streamside) vegetation showed similar patterns of higher correlation among related sets of variables. In contrast, some groups of variables showed little correlation across groups-for example, measures of riparian vegetation did not tend to correlate with measures of water chemistry.

Integrated measures of disturbance were better predictors of index values

In general, specific stressors tended to be more highly correlated with integrative measures of human disturbance than they were with similar measures that measured only a single aspect of disturbance. For example, turbidity, percentage of sand and fine, pebble size, riparian vegetation condition and riparian disturbance were correlated with one or two of each other, but all five were correlated with a habitat index developed to integrate measures of site condition at the reach scale (Table 2). A similar pattern was observed for water chemistry measures such as total N, total P, NH₄, and SO₄ that showed fewer significant correlations with each other than with integrative measures that summarized human influence at the watershed scale. Bryce et al.'s (1999) condition classes were derived from an analysis of patterns of human land use within the watershed and the observed riparian condition at 102 sites. All the listed measures correlated with Bryce et al.'s index of disturbance. The percentage of disturbed land was the sum of land in the upstream watershed used for agriculture, urbanization or mining; all but one of the uni-dimensional measures correlated with this measure.

Similarly for biological indicators, multimetric indexes for all three assemblages showed a higher correlation with integrative measures of disturbance than with specific stressors (see Table 1 on page 18). One chemical measure, chloride, was a strong indicator of general disturbance and also highly correlated with all three biological indexes (Herlihy et al., 1998).

Thus, measures of disturbance that integrated measures of site condition over multiple spatial scales tended to better capture the cumulative effects of human influence. This result supports the idea that much of the scatter observed in plots of biological measures against human disturbance gradients is in fact associated with the x-axis: one-dimensional measures of disturbance simply fail to capture the cumulative influence of human activities on the biota (Karr et al., 2000).

Table 2. Spearman's correlation matrix for measures of human disturbance that were used to test biological metrics for the MAIA study; only correlation coefficients > 0.3 (or < -0.3) are shown. (See Table 1 for description of variables.)

Measure	N	P	NH₄	ANC	SO₄	Turb	%S_F	PbSz	RVeg	RDist	RBP	CL	Bryce	%Dist
N	—	0.50	0.34	0.39								0.51	0.38	0.67
P	0.50	—	0.36	0.31		0.57	0.44	-0.34				0.35	0.30	0.50
NH₄	0.34	0.36	—									0.41	0.36	0.32
ANC	0.39	0.3		—	0.35					0.33		0.45	0.47	0.54
SO₄				0.35	—							0.45	0.39
Turb		0.57				—	0.47	-0.38			-0.32		0.32	0.33
%S_F		0.44				0.47	—	-0.75			-0.39		0.55	0.49
PbSz		-0.34				-0.38	-0.75	—			0.31		-0.36	-0.34
RVeg									—	-0.65	0.60		-0.60
RDist				0.33					-0.65	—	-0.35		0.45	0.30
RBP						-0.32	-0.39	0.31	0.60	-0.35	—		-0.77	-0.41
CL	0.51	0.35	0.41	0.45	0.45							—	0.58	0.68
Bryce	0.38	0.30	0.36	0.47	0.39	0.32	0.55	-0.36	-0.60	0.45	-0.77	0.58	—	0.47
%Dist	0.67	0.50	0.32	0.54		0.33	0.49	-0.34		0.30	-0.41	0.68	0.47	—
Total	6	9	5	7	3	6	6	6	3	5	7	7	13	11

<< previous page
next page >>

Biological Indicators | Aquatic Biodiversity | Statistical Primer

Linking Human Disturbance To Biological Change (continued)

Metric testing included safeguards against circular reasoning

Patterns of human disturbance were complex

Integrated measures of disturbance were better predictors of index values

Local Navigation