• About this site|
• Site index|
• Contact us|
• Jobs

Spatial patterns in state and trends of environmental quality in New Zealand rivers: an analysis for State of Environment reporting

Executive Summary

1. Introduction

2. Methods

3. Results of analysis of physico-chemical and biological state

4. Results of analysis of trends in water quality and biological variables

5. Discussion

6. Conclusions

7. Acknowledgements

8. References

Appendix 1

Appendix 2: State analysis results summary table

2. Methods

2.1. Sources of environmental data

The analysis used long-term river water quality and biological data from two monitoring networks, the NRWQN and the ESWQN.

2.1.1. National Rivers Water Quality Network

The NRWQN includes 77 sites (Figure 1), distributed throughout the North Island (44 sites) and South Island (33 sites), at which river flow, and 12 other physico-chemical variables have been measured monthly since 1989. NRWQN sites were selected to reflect both baseline conditions (32 upstream sites) and impact conditions (45 downstream sites). Fieldwork is carried out by NIWA's 14 regional hydrometric field teams. All laboratory analyses are performed at NIWA's Hamilton water quality laboratory.

Each site is sampled at approximately the same time of day to minimize variability due to diurnal changes. Flow is measured on each sampling occasion. Dissolved oxygen, temperature, and visual clarity (measured by horizontal black disc visibility; Davies-Colley 1988) are measured in the field. Conductivity, pH, turbidity, biochemical oxygen demand (5-day test; hereafter BOD₅), absorption coefficient at 440 nm (g440), oxidised N (NO₂-N + NO3-N), Total Ammonia (NH₄+-N + NH₃-N), soluble reactive phosphorus (SRP, often referred to as dissolved reactive phosphorus DRP) and total phosphorus are measured in the laboratory. Further details on sampling techniques and field and laboratory measurement are given in Smith et al. (1989), Smith & McBride (1990) and Smith et al. (1996).

Periphyton cover is monitored monthly at 68 sites by wading as far into each river as is safe (the whole width where possible) and using an underwater viewer to observe periphyton cover in 10, equally spaced, areas of riverbed. The percentage cover of filamentous algae and thick mats (>3 mm thick) is assessed visually.

Benthic macroinvertebrates were sampled annually over 10 years (1989-98) at 66 NRWQN sites, between January and April. Most sites were sampled at the same time each year. Sampling occurred under baseflow conditions (Q < Qmedian) and not less than four weeks after bed-scouring floods. To reduce small-scale spatial variability, samples were collected from a range of depths (0.2 - 0.4 m), velocities (0.6 - 1.0 ms^-1) and substrate conditions (dominated by cobbles and gravels). Seven Surber samples (0.1 m², 250 mm mesh nets) were collected from a site on each occasion and pooled into a single sample. Samples were preserved in 10% Formalin (1990-95) or 70% Isopropyl alcohol (1996-98). Macroinvertebrates in sub-samples were identified to the lowest practical taxonomic level (usually species or genus) following the keys listed in Quinn & Hickey (1990). In preparation for analysis, the invertebrate dataset was first reduced to a taxonomic level where changes in knowledge and variations in experience of sample processors over the 9-year period would not impact on the integrity of the data. In general the reduction was to a genus level, although several groups were reduced to higher taxonomic levels.

2.1.2. Environmental Southland Water Quality Database

Environment Southland currently monitors the water quality of Southland rivers to compare their current state, determine trends over time and identify sources of contaminants (Environment Southland 2000). Comprehensive monitoring with the intent of detecting changes in river water quality began in Southland in 1989 with NIWA's NRWQN. In 1995 water quality monitoring began at a further 26 sites (integrated with but not including the NIWA sites). Monthly tests for faecal indicator bacteria (FIB) at all these ESWQN water quality sites began between July 1999 and July 2000. At the same time, freshwater sites used for recreational bathing were identified for intensive sampling of FIB during summer (the bathing water network). Southland Regional Council began annual monitoring of aquatic macroinvertebrates in December 1994 to assess ecosystem health. The identification and analysis of algae samples were added to the programme in January 1999. A map of the location of ESWQN sites is shown in Figure 5 and full details are appended in Appendix 1.

2.2. River Environment Classification

2.2.1. Spatial framework

The REC is a spatial framework that has been developed for similar application to frameworks that have been applied in other countries (e.g. Ecoregions). The following section provides a brief explanation of how REC classes are defined and mapped.

2.2.2. Features of the REC system

The REC was developed using similar principles to other forms of regionalisation, in particular the 'Ecoregion' approach (e.g. Omernik, 1987). Regionalisations use landscape scale factors to delineate regions of distinctive ecological character. For example, Ecoregions are delineated by a combination of factors that are assumed to cause or show aspects of ecological character.

The REC uses controlling factor principles (sensu Bailey 1995) and includes three key principles. First, the REC introduces the idea that ecological patterns are dependent on a range of factors, and associated landscape scale processes (e.g. Wiens 1989; Levin 1992). The REC systematically accounts for ecological variation, at a range of scales, by stratification of river environments at multiple scales using a consistently applied set of factors. This differs from the Ecoregion approach, which uses several inconsistently applied factors to define individual ecoregions at a single characteristic scale. REC arranges the factors hierarchically so that each level of the classification is defined by a component of the environment, the 'controlling factor', that is the 'cause' of ecological variation at that level.

The second key principle is the 'network' approach to classifying rivers used by REC. Previous work in New Zealand's highly heterogeneous environments (e.g. Biggs et al. 1990) showed that it is difficult to define meaningful ecoregions, particularly for water quality variables (Harding, et al., 1997). This and other work showed that spatial variations in ecological characteristics of rivers are essentially responses to fluvial (i.e. hydrological and hydraulic) processes. These processes are fluvial, and thus, factors must be considered with respect to the catchment that generates flows of water and water borne constituents. Therefore, the REC treats rivers as networks of catchments connected by 'sections' of the river network. Classification is carried out for individual sections of the network and is based on four factors that characterise the upstream catchment (Climate, Topography, Geology and Land Cover) and two factors that characterise the section itself (Network Position and Valley Landform).

When mapped REC has the form of a 'linear mosaic' of poly lines rather than polygons such as for 'Ecoregions'. Classes change in the downstream direction as the characteristics of the upstream catchments change. Tributary streams may therefore have very different classifications to the mainstems they meet, and tributaries may collectively change the classification of the mainstem. This produces longitudinal spatial patterns (Figure 1) that are typical of patterns of many properties of river ecosystems (e.g., Vannote et al. 1980).

The third key principle concerns the structure of the classification and the assignment of sections of the river network to a class. REC assigns individual river sections to a class independently and objectively according to criteria resulting in a geographically independent (sensu Detenbeck et al. 2000) framework where classes may show wide geographic dispersion (see Figure 2).

Figures 3 and 4 show the mainstem rivers of the North and South Islands classified at the Source of Flow level. These classification levels break up New Zealand's rivers into a small number of groups. These groups have been used in this analysis as the basis for characterisation of environmental state and trends in New Zealand's rivers.

2.2.3. REC classes

The REC characterises river environments at six hierarchical levels, the names of which reflect the controlling factors that define each level; Climate, Source of Flow, Geology, Land Cover, Network Position and Valley Landform (Table 1). Each level of the hierarchy subdivides the preceding level using categories of the defining factor.

Each REC class is denoted by the concatenation of each controlling factor and has the form: Climate/Source of Flow/Geology/Land Cover/Network Position/Valley Landform.

Categories at each classification level and their shortened notation are shown on Table 1. The number of potential classes at any level is equal to the number of categories at that level multiplied by the number of classes at the preceding level. Thus the Source of Flow level has 30 potential classes (6 climate classes x 5 Source of Flow classes). Not all of these potential classes occur however, for example there are no Warm Humid Mountain classes. Classification at any particular level is denoted by truncating at that level. For example, a class at the Geology level may be Cool Dry/Hill/Hard Sedimentary, denoted CD/H/HS.

The process of characterization and labelling of the REC is distinct from the process of mapping. Mapping involves assignment, which is the process of choosing or recognizing the class to which each section of the network belongs to (Klijn, 1994). The REC assigns each section in the network to a class based on 'mapping characteristics' (sensu Klijn, 1994) and criteria (sensu Bailey, 1995). The criteria used for assignment are included in Table 1 and are discussed more fully in Snelder and Biggs (2002). The mapping characteristics are based on existing maps of environmental attributes (e.g. geology, land cover, topography) or spatial coverage of interpolated climatic data. The REC has been used to classify and map all the rivers of New Zealand at a 1:50,000 scale and is stored as a Geographic Information System (GIS).

Table 1: REC classification levels, categories and their notation, mapping characteristics and class assignment criteria

View REC classification levels, categories and their notation, mapping characteristics and class assignment criteria (large table)

2.3. Using REC as a spatial framework for data analysis

The REC defines a hierarchy of classes, within which ecological similarity (e.g. water quality or biological communities) varies from general to specific, as the classification level is decreased. The highest REC classification level (Climate) groups rivers that are very generally similar. The 4th (Land Cover) level of REC will group rivers that share more specific similarities. Thus, when we look at a water quality variable for a group of sites, the variation within a group that share a similar Climate class is quite large, as the sites are only 'generally' similar. As the classification level is reduced, variation within a class decreases because the number of shared controlling factors increases. These similarities are linked to spatial scales; rivers may be generally similar over large areas but specific characteristics remain similar over only small scales.

In this study, the water quality and biological state and trends of each REC class are described by aggregating data from sites on different rivers that share the same class. We take he state or trend for the class as whole to be the median value for the variable (i.e. state or trend). Our ability to describe river classes at a given level of classification is dependent on the number of sites (or 'replicates') assigned to that class. Replication is needed to ensure that a particular site, which may be atypical of the class, does not introduce bias.

Replication was a limitation for some of the analyses reported here. In general, the NRWQN data allowed us to characterise rivers at two spatial scales defined by the Climate and Source of Flow levels of the REC, but replication within classes was not uniform. For example, at the Climate level, the WX class was represented by only one site. Replication decreased when sites were further subdivided by the Source of Flow level. Several Source of Flow classes (CX/L, CX/Lk, WW/Lk, WX/L) are represented by a single site. This reduced the statistical power of the analysis and increased the risk of bias. A case in point is WW/Lk class. This class was represented by a single site, the Tarawera at Awakaponga. This site is heavily impacted by an upstream industrial discharge and cannot be assumed to be representative of the class.

By supplementing the NRWQN sites in Southland with ESWQN data, we have increased replication within the region. This allowed us to have replication of sites classified at the REC Land Cover level. However, the Land Cover level has a large number of classes and replication again became a limitation. Some Land Cover classes in Southland are not represented at all or only represented by one or two sites.

2.4. Classification of monitoring sites

All of the NRWQN and ESWQN monitoring sites were classified by assigning the REC class of the section of river on which they are located. This process is carried out using GIS maps of the REC of the location of all sites.

Instream processes are known to change the characteristics of water quality moving down the river network (i.e. from headwaters to the mainstem). Thus sites that share similar catchment characteristics and, therefore, have the same REC class may have dissimilar water chemistry. We therefore examined the Network Position classification of the 77 NRWQN sites and the 49 ESWQN sites and found that the majority are located on main stem rivers (i.e. Network Position classification of HO). In order to retain a consistent set of river sites we removed all sites from both datasets with a Network Position classification of LO and MO, leaving a set of 70 NRWQN and 30 ESWQN mainstem sites.

2.4.1. NRWQN sites

Table 2 details the number of NRWQN sites in each class at both the Climate and Source of Flow REC levels. At the Source of Flow level, thirteen REC classes are represented by NRWQN sites. The REC has been used to map the total river length of each Source of Flow class represented by a site in the NRWQN (Figures 3 and 4). A list of NRWQN sites sorted by class is in Appendix 1.

2.4.2. Environment Southland sites

Because few pre-1995 data are available, we restricted the ESWQN data to the six year period from 1995 to 2001. The resulting dataset could be compared to the NRWQN data from the same period. A number of the mainstem sites had less than 6 years of data. However, omitting these data would have greatly reduced the size of the dataset. We retained these sites for the state analysis, but did not include them in the trend analysis. We set a limit of at least one year of data to avoid bias due to seasonal variation. We also removed three sites that were specifically for monitoring point source discharges. The final dataset had 30 sites for the state analysis and 21 sites for the trend analysis.

All ESWQN sites were classified to the Land Cover level. This level provides a smaller spatial scale for analysis than the Source of Flow level used for the NRWQN dataset. Table 3 details the number of sites in each class. Figure 5 shows the location of each site (see Appended Table 1.2 for site keys) and the mainstem rivers in Southland classified at the Source of Flow level.

To further increase the replication of the ESWQN at the Land Cover level of the REC, we combined some classes. First we aggregated all sites with indigenous forest (IF) and tussock (T) land cover categories to a 'Baseline' (B) class. We also converted all streams with a Alluvial (Al) geology category to hard sedimentary (HS) because alluvium in Southland is predominantly composed of hard sedimentary material. We also merged the one Mountain (M) source of flow site with the Hill (H) class. The simplified classification and sites that belong to each class are shown on Table 4 and detailed in appended Table 1.3. The extent of each of the classes is shown on Figure 6.

2.5. Analysis of State Using Water Quality and Biological Variables

Environmental 'state' is the physico-chemical or biological condition of a river, or group of rivers. All of the variables measured by the NRWQN and ESWQN describe an aspect of the state of a site at a particular time. State may be variable in time and reflects both natural variation as well as human-induced effects. To characterise the general or long-term state at a site, we calculated measures of the central tendency (means and medians), and range (minimum, maximum and percentiles) of each variable for sampling occasions within a specified time period. Bar and whisker plots of the aggregated data were used to characterise the median state and the variation among sites in the class.

2.5.1. Water quality variables

We focused our analysis on six water quality variables; clarity, BOD₅, SIN, SRP, total ammoniacal nitrogen (referred to hereafter as Total Ammonia) and the microbiological indicator Escherichia coli (E. coli). The ESWQN dataset does not include BOD₅ and the NRWQN does not include E. coli.

Clarity is a measure of the level of suspended and dissolved material in the water column, and is measured by the horizontal distance that a standard black disc can be seen through the water column. Clarity is an important determinant of ecological functioning and a minimum clarity guideline is applied for waters with contact recreation use.

Nitrogen and Phosphorus are required for algal growth. Total Nitrogen (TN) and Total Phosphorus (TP) include particulate nitrogen and phosphorus, which may not be immediately available to primary producers. Consequently, these variables may be poor indicators of water quality conditions, and were not the focus of our analysis. However, Soluble Inorganic Nitrogen (SIN) and Soluble Reactive Phosphorus (SRP) were used for the state analysis because these variables are more closely linked to the 'response variable' of interest being algal (periphyton) biomass. SIN is the sum of nitrate (NO₃) and Total Ammonia. For the trend analysis we report on NO₃ and NH₄ separately so that their contributions to trends in SIN can be distinguished.

BOD₅ is a measure of the organic enrichment of the water and reflects point and non-point discharges of organic material. Prolonged periods of elevated levels of dissolved organic matter promote heterotrophic growths (e.g. sewage fungus), which has ecological and aesthetic impacts.

Unionised ammonia (NH₃) is potentially toxic. The concentration of unionised ammonia is related to the concentration of Total Ammonia (NH₄+ + NH₃), pH and temperature.

Concentrations of E. coli are an indicator of risk of pathogenic bacteria spread by faecal contamination from humans and animals. Faecal contamination may affect human health through contact recreation and stock health may be affected by drinking contaminated water.

2.5.2. Guidelines for Water Quality

Water quality guidelines reflect particular value judgments about the use or value of the environment to which they are applied. Guidelines are, therefore, not seen as absolutes, but benchmarks that may indicate a risk of adverse effects. We have applied published guidelines from a variety of sources (see Table 5) to the six water quality variables.

For five of the variables (Clarity, BOD₅, SIN, SRP, E. coli), the mean of measured monthly value was compared to guidelines. In order to report on the state of rivers in a class, we aggregate the means for each site in the class, and compare the median value of this statistic to the guideline. Thus, we show the state of REC classes with respect to Clarity, BOD₅, SIN, SRP, E. coli on box and whisker plots by comparing the median of the means (i.e. the small square in the center of each box) with the guideline, which is superimposed onto the box and whisker plots (a red horizontal line).

The toxicity of Total Ammonia depends on its ionic state, which in turn depends on pH. We have assumed a pH of 7.7 and applied ANZECC standards to derive guidelines that are then applied across all classes. Ammonia toxicity will adversely affect 'state' even if it occurs infrequently. Excursions above the guidelines on individual monthly sampling occasions are an exceedance rather than the guideline being exceeded by the mean monthly concentration. Therefore, we compared the guidelines for Total Ammonia with the median of 99th percentile of concentration for each site, rather than the median of means as for other variables.

We note that Dissolved Oxygen concentration (DO), Dissolved Oxygen saturation (DO%), pH and Temperature are measured by the NRWQN and ESWQN. DO and Temperature are confirmed freshwater indicators (MFE 1998). However, these variables are hard to interpret meaningfully because they show wide daily fluctuations that require a consistent sampling time to be useful as an indicator. We therefore discarded them from the analysis.

2.5.3. Biological variables

For the NRWQN we focused our analysis on two biological variables; periphyton cover and the macroinvertebrate community index (MCI). Periphyton is algae attached to the riverbed and is an indicator of nutrient enrichment. High periphyton cover may indicate high levels of nutrients and has biological, recreational, as well as visual effects. The analysis first calculated the mean of up to 10 quadrats along transects perpendicular to the riverbank to find the percentage of filamentous algae and mat cover for each month. The annual maximum of monthly periphyton cover was then found for each site. The average of maximum annual cover for each site was then found.

Macroinvertebrates (insects, worms and snails) are commonly used as indicators of water quality. In New Zealand, the MCI was developed as a biological index of organic pollution (Stark 1993). The MCI gives individual taxa a score relative to their assumed tolerance to organic pollution. The MCI is similar to other indices around the world and was modeled on the Average Score Per Taxon (ASPT) (Armitage et al. 1983). In general, high values of the MCI are associated an invertebrate community made up of organisms that are sensitive to poor water quality.

2.5.4. Guidelines for biological variables

Guidelines for acceptable periphyton cover to protect aesthetic and recreational values of rivers are annual maximum less than 60% cover of the visible streambed by mats and less than 30% cover by filamentous algae (Biggs 2000). We compared these guidelines to the median of a series of annual maxima for the period 1995 to 2001. In determine the state of rivers in a class, we aggregate the medians of annual maxima by REC class and compare the guideline to the median.

Guidelines for the MCI are given in Stark (1998). An MCI score greater than 120 indicates unpolluted conditions, scores between 120 and 100 indicates probable mild pollution, MCI scores less than 100 indicates probable moderate pollution and scores less than 80 indicate probable severe pollution. In order to determine the state of rivers in a class, we found the mean of annual MCI scores between 1995 and 2001 for each site. We then aggregated these means by REC class and compared the median MCI score for each class to the guidelines provided by Stark (1998).

2.6. Analysis of temporal trends in water quality and biological variables

The ability to detect trends is dependent on the temporal scale of the dataset. Over short time scales, a trend in water quality or biological data may be observed. However, given a long enough time period most trends will tend to disappear (Loftis et al. 1991). In our analysis data availability constrained the time period for trend analysis. NRWQN water quality and periphyton data were available for the period 1989 to 2001 and the ESWQN data for the period 1995 to 2001. In addition, invertebrate data from the NRWQN was available for the period 1989 to 1998.

The time period of observation is important because different time periods may coincide with climatic trends that may in turn contribute to trends in some variables. Climate variability results in varying rainfall (thus varying river flow) and solar radiation (thus varying water temperature). While we routinely adjust water quality variables for flow variability, we know of no cases where water quality data have been adjusted for climatic variability. In examining trends we are mindful that flow adjustment alone may not remove the natural 'noise' associated with climate from long-term datasets.

2.6.1. Simple linear regression method

We used two methods for trend analysis: linear regression and Seasonal Kendall tests. For linear regressions, we calculated the annual median of each variable for each site. The 10th and 90th percentile values were also calculated to describe the range of data, excluding outliers. This data was then aggregated, both for all sites and by REC classes by calculating median values of the percentiles for each aggregated group of sites (for example, the median value of the 10th, 50th and 90th percentile for 5 CD/H sites in the NRWQN). The median percentiles were plotted. These plots show changes in median state through time, and the range in each variable in space. Linear regression was used to assess magnitude and significance of linear models applied to the median state through time. Initial observations of the data suggested that linear models were appropriate in the majority of cases.

2.6.2. Seasonal Kendall methods

Reducing monthly water quality data to annual medians greatly diminishes sample size, and as a result the power to detect trends is diminished. However, monthly water quality data often exhibits seasonal patterns. Linear regressions were not appropriate for analysing data at a monthly time step because the method does not account for strong seasonal components of variability. Instead, the non-parametric Seasonal Kendall trend tests were used to assess the magnitude and significance of trends (Gilbert 1987; Harcum et al. 1992; Helsel & Hirsch 1992). These tests were made using the WQStat II package (Colorado State University 1989).

The seasonal Kendall Sen Slope Estimator (SKSSE) is the median of all possible combinations of slopes within each month. Consider all the January data for n = 5 years of record. All possible slopes between these data are calculated (10 in number). This is then repeated for all other months, giving 120 slopes, so that the slope estimator is the average of the two slopes with ranks 60 and 61.

Seasonal Kendall tests are two-sided tests of the null hypothesis that there is no trend. A strong advantage of this test is that the only assumption is monotonicity meaning the functional form of any trend (e.g., linear, exponential) need not be considered; the test merely contemplates whether within-month/between-year differences tend to be monotonic. The slope estimate does not enter directly into the test procedure, so small inconsistencies can occasionally arise between slopes and p-values (particularly if there are many tied values). The mechanics of the test and the explanation of these inconsistencies are given elsewhere (McBride & Loftis 1994).

The SKSSE for each site was expressed as a percentage of the site median to allow direct comparison of trends between sites. This standardised trend is referred to as the Relative Seasonal Kendall Sen Slope Estimator (RSKSSE). We tested the statistical significance of aggregate RSKSSE using a Binomial test to determine whether the aggregate trend within a class is statistically different to zero. A criterion for statistical significance of p less than 0.05 was used. The binomial test is highly dependent on sample size and cannot be made with less than five sites (i.e. results will never be statistically significant if n is less than five).

We present the slopes of trends for individual sites or for sites aggregated either in total or by REC classes. On the y-axes of graphs, the values given correspond to the slope of the trend as % of the median. The units are per annum. Therefore a value of 1% corresponds to a 10% increase in the median value of that WQ variable over 10 years.

2.6.3. Flow adjustment

Flow is a source of temporal variation in water quality. The two main mechanisms are dilution and wash-off. If material enters a river at constant rate, then concentrations decrease as flow increases. This is often seen with dissolved solids, or downstream of major point sources (Helsel & Hirsch 1992). If material enters a stream as a consequence of wash-off (e.g., from a hillside or from bank erosion), then concentrations tend to increase with flow. This effect is often seen with suspended solids and total phosphorus (Helsel & Hirsch 1992). For many variables and sites both processes are important and variation with flow is not predictable. For example, hysteresis may be present whereby concentrations on the rising and falling limbs of a flood hydrograph differ (e.g., McDiffert 1993; Sokolov & Black 1996). Consequently, concentration-flow plots do not always show such a clear pattern, and flow adjustment is needed.

Accordingly Seasonal Kendall tests have all been performed with flow adjustment. Flow-adjustment was made using log-log transformations, the default procedure in WQStatII. Log-transformed water quality data were plotted against log-discharge. Then linear regressions were used to determine slopes and intercepts. From each observation, the predicted value was subtracted. This produced a flow-adjusted set of observations with a mean of zero. To complete the adjustment, the mean of the water quality series was added to each observation. The Seasonal Kendall test was then applied to the adjusted time series, and the RSKSSE was calculated.

2.7. Representation of rivers by the available monitoring data

The REC enabled us to examine the representativeness of monitoring networks, based on the distribution of sites among classes. One of the pieces of information provided by the REC is the total river length assigned to a given class. Representation of each class, by each monitoring network, was calculated by comparing the proportion of sites in a class to the total number of sites, and then comparing this number to the river length of that class relative to the total river length. Numbers close to one indicate that the ratio of sites in the monitoring network is similar to the ratio of river length in that class to total river length. Numbers greater than 1 indicate 'over-representation' and numbers lower than 1, 'under-representation'. In effect, the analysis indicates whether the network as a whole provides a fair representation of rivers in general.

The results of the analysis of representativeness are shown in Table 6. The NRWQN is reasonably representative of main stem rivers at the national scale. The NRWQN over-represents 'Lake' Source of Flow classes (WW/Lk and CW/Lk) and the Warm Extremely Wet Low Elevation class (WX/L). This suggests that aggregating all the NRWQN data and using it to report on the nation-wide state of rivers will provide a reasonable 'snapshot'.

The analysis of representativeness of the ESWQN is shown on Table 7. The ESWQN is not as representative of Southland's rivers as the NRWQN sites are of New Zealand's rivers. This is not a reflection of the quality of Southland's water quality network, rather it may reflect different criteria used by Environment Southland for site selection. In addition, the process of 'filtering' we used to remove sites from the data we used may have distorted the representativeness of Environment Southland's water quality network.

Table 6: Representation of the high order rivers of New Zealand by the NRWQN by REC class

View representation of the high order rivers of New Zealand by the NRWQN by REC class (large table)

Table 7: Representation of the Southland region's high order rivers by the selected sites from the ESWQN

View representation of the Southland region's high order rivers by the selected sites from the ESWQN (large table)