4. Discussion
4.1. Patterns in state and trends in stream and river water quality
4.1.1. Patterns in water quality state
Climate level. The comparisons of water quality state at the climate level indicate that the dry classes (CD and WD) and the wet classes (CW, WW, CX and WX) form two fairly distinct groups. The dry climate classes had higher median NOX, NH4 and E. coli concentrations than the wet classes, and median nutrient and E. coli concentrations and clarity in the dry climate classes did not meet guideline values. In contrast, the climate classes with the greatest annual precipitation, CX and WX, generally had the highest water quality and met the guidelines (Figure 3). E. coli concentrations exceeded the guideline in the WX class, but this class was represented by only two sites, both in pastoral land cover classes. A general pattern of improved water quality with increasing rainfall was reported in the previous nation-wide water quality study using the REC framework (Snelder & Scarsbrook 2002). In that study, nation-wide patterns were based on NRWQN data from 70 sites on large rivers. The current study confirms this pattern, and the rarity of differences between stream orders within climate classes (Results, Section 3.4) suggests that the pattern holds for streams of all sizes.
Two general explanations may be posed regarding differences in water quality among climate classes. The first is that physical conditions and processes that vary at the scale of climate classes (103 - 105 km2) directly control water quality. REC climate classes delineate broad hydrological and thermal patterns, and water quality differences among classes may reflect different regimes, or differences in temperature-regulated nutrient cycling.
The second explanation is that difference among climate classes are due to differences in the predominant source-of-flow, geology and land cover categories within each climate class, and not to climatic conditions per se. It is very likely that predominant land-cover categories strongly influence apparent climate-level patterns. As noted previously (Results, Section 3.1.1), most sites in the CD and WD climate classes are in the pastoral and urban land-cover classes. These land-cover classes had consistently poor water quality, and the inclusion of a small number of sites from the natural and exotic forest classes may have little influence when all sites in the dry climate classes are aggregated. Further, the CX class, with the highest water quality of the five climate classes, is composed primarily of sites in the natural land-cover class. In our dataset, most sites in the CX class were on upper reaches of rivers in national parks, where watershed disturbances are minimal. Nation-wide, CX stream segments occur in high-elevation catchments (e.g., Southern Alps, Mount Taranaki, Raukumara and Tararua Ranges) and in Fiordland, all areas with minimal development.
It is likely that water quality is affected by both large-scaled processes associated with high REC levels, and fine-scaled processes associated with land-cover. Land-cover effects are propagated up to higher REC levels when large areas are dominated by single land-cover classes.
Source-of-flow level. Comparisons among source-of-flow classes clearly indicate that low-elevation river segments have poor water quality compared to hill, mountain and lake classes. Within the low-elevation classes, those in dry climate classes (CD/L and WD/L) generally have lower water quality than those in wet and extremely wet climate classes. As is the case for climate classes, differences in water quality among source-of-flow classes may be due either to processes that vary at the source-of-flow scale (102 - 103 km2), or to the effects of dominant land cover types or geologic formations within each source-of-flow class. Processes operating at the source-of-flow scale that may affect water quality include seasonal runoff, sediment yield, and hillslope and bank erosion. These processes affect water quality via sediment, organic matter and nutrient input from catchments, and channel sediment reworking and suspension. In the present study, low elevation classes had poorer water quality than mountain, hill and lake classes. However, most low-elevation classes were dominated by pastoral land-cover classes. As with climate classes, it is likely that differences in water quality among source-of-flow classes are due in part to fine-scaled factors associated with land cover categories. Specifically, the effects of pastoral land-cover classes at low elevations result in worse water quality than do the predominantly natural land-cover classes at high elevations. A study of the Pomahaka River system in Otago by Harding et al. (1999) supports this proposition. Land cover in high-elevation river segments (in the REC mountain and hill source-of-flow classes) was native tussock and lightly-grazed pasture, while land cover low-elevation segments (REC low-elevation source-of-flow class) were heavily-stock pasture, cropland, and dairy, pig and deer farms. NOX, and DRP concentrations were higher at the low-elevation sites, and clarity significantly lower, compared with the mountain and hill sites.
In five of the six low-elevation source-of-flow classes in the dataset, at least 60% of the sites were classified as pastoral. The sixth low elevation class, CX/L, was composed of sites in natural (55%), pastoral (33%), exotic forest (10%) and urban (3%) classes. Median nutrient and E. coli concentrations were lower in the CX/L class than in the other five source-of-flow classes. The contrast between the CX/L class and the five agriculture-dominated low-elevation classes lends further support to the idea that water quality in high-level REC classes is strongly affected by smaller-scaled land-cover patterns. It also highlights the relatively degraded state of many low elevation rivers in New Zealand (see Land-cover level, below).
Examinations of direct relationships between source-of-flow classes and water quality, without the confounding effects of changing land-cover classes, could be made within catchments where a single land-cover class extends over a wide elevation gradient. Such catchments are rather rare in New Zealand, but do occur in forest and national parks where indigenous forest extends from mountains to lowlands, e.g., Fiordland and Mount Aspiring.
Land-cover level. Overall water quality state in land-cover classes can be ranked as follows: urban streams are clearly in the worst condition, followed by pastoral, exotic forest and natural streams. Very few urban land-cover classes in the dataset met any of the water quality guidelines. In contrast, most natural land-cover classes were within guideline values, except the NH4 guideline (see below). In particular, natural classes had uniformly low NOX concentrations and high clarity. The generally poor water quality in pastoral classes, both nation-wide, and within individual regions, corroborates recent studies conducted at smaller scales (eg., Donnison & Ross 1999, Quinn & Stroud 2002). Regional councils also report widespread water quality problems in streams and rivers draining agricultural areas (e.g., Environment Southland 2000, ORC 2001, Meredith & Hayward 2002).
Agricultural land use and its effects on water quality are issues of growing concern in New Zealand (e.g., MfE 1997, Vant 2001, Meredith & Hayward 2002, Hamill & McBride in press). Pastoral land-cover classes clearly predominate in low elevation areas (Results, Section 3.1.2). This pattern is not an artefact of monitoring site selection by regional councils, as 73% of the low elevation river length in New Zealand is in the pastoral REC class. Agricultural land dominates low elevation catchments in most regions, and large areas of undeveloped low-elevation land are now limited to the Coromandel Peninsula, Wanganui National Park and environs, and coastal margins of the East Cape, eastern Northland, Westland, Marlborough, southern Fiordland and Stewart Island. Most of the pristine or near-pristine low elevation streams that may be useful as references for assessing or restoring degraded streams occur in these areas.
Examination of the plots showing land-cover classes that exceeded water quality guidelines (Appendix 2) indicates that the NH4 is the parameter most often exceeded. Some land-cover classes that met the guidelines for the other four parameters exceeded the NH4 guideline (e.g., CD/H/HS/N and CX/M/HS/N). The ANZECC guidelines for inorganic nutrients such as NH4 are primarily intended to protect river ecosystems from excessive macrophyte and periphyton growth (the NH4 guideline should not be confused with total or unionized ammonia guidelines, which are intended to prevent ammonia toxicity; Alabaster & Lloyd 1982). However, the ANZECC guideline for NH4 (0.015 g/m3) is less than 5% of the NOX guideline (0.31 g/m3). Plants and algae must reduce inorganic nitrogen ions before they can be assimilated. As NH4 is more reduced than NOX, NH4 may be taken up preferentially (Lohman & Priscu 1992), but this is rarely observed in the field. NOX is generally present at much higher concentrations than NH4 in New Zealand streams (Close & Davies-Colley 1990, and the present study), and it is unlikely that preferential NH4 uptake warrants a 20-fold difference in guideline concentrations. We suggest that land-cover classes with median NH4 concentration moderately higher than the ANZECC guideline and otherwise high water quality are not necessarily degraded. Further, we suggest that a single guideline for dissolved inorganic nitrogen (NOX + NH4), as used by Snelder & Scarsbrook (2002) would be more useful in water quality assessments than separate NOX and NH4 guidelines.
4.1.2. Patterns in water quality trends
Snelder & Scarsbrook (2002) reported trends for flow and water quality variables at the national scale for the period July 1995 - June 2001, using data from the 70 NRWQN sites. Despite limited replication, they detected significant negative trends in flow and in flow-adjusted NH4 concentrations, and a significant positive trend in flow-adjusted DRP concentrations. Results of the current analysis indicate that some national-scale trends were consistent across the two datasets, and persisted through 2002. Negative trends still existed in 2002 for flow and NH4. The current analysis also revealed a significant positive trend in flow-adjusted temperature; in the previous study, a trend of similar magnitude and direction was not significant. Major differences between the two studies relate to flow-adjusted trends in DRP and NOX. During the 1996-2002 period, there was a significant negative trend in NOX, and no apparent trend in DRP (i.e., median slope = zero); the previous study reported a positive, but not significant, trend in NOX, and a significant positive trend in DRP. A comparison of the two studies suggests that trends in flow, temperature and ammonia levels may be general trends across New Zealand, whereas trends in NOX and DRP, may be more spatially variable.
The levels of spatial variability in flow and water quality trends can be examined by comparing trends at different REC levels. Trends in flow, temperature, and clarity observed at the source-of-flow level were also observed at the land-cover level. Trends in NH4 concentrations varied between land-cover categories; there was a negative trend in NH4 in pastoral classes, and no NH4 trend in natural classes. The negative trend in NH4 in pastoral classes may be due to improved waste management practices, but other possible causes must be considered. For example, there were differences in the strength of temperature trends across natural and pastoral classes, particularly when comparisons were made at the land-cover level. Trends in NH4 could be related to a changing temperature regime, and its effects on in-stream nitrogen cycling, rather than to changes in NH4 input.
Comparisons among jurisdictional regions at the source-of-flow level revealed some geographic variation in trends within REC classes. For example, there were positive trends in clarity and DRP concentrations, and a negative trend in flow in the CW/H class in Waikato, but not in Southland (Results, Section 3.3.3), and a significant positive trend in temperature in the CD/L class in Southland, but not in Canterbury. A possible explanation for these differences is that the REC system does not account for all climate differences that affect temporal variability in water quality. Scarsbrook et al. (in review) reported that temporal changes in several water quality parameters in New Zealand rivers are related to the strength of the El Niño-Southern Oscillation climate process (ENSO). ENSO effects vary geographically in New Zealand, primarily between the east and west sides of the main islands (Salinger & Mullan 1999). Spatial variation at the scale of ENSO is not specifically considered in the REC system. Therefore, regional differences in water quality trends that develop as a result of ENSO (or other climate anomalies that differ in strength among regions) are likely to be sources of increased variability within REC classes.
In this report we have reported all trends that were statistically significant. Not all significant trends are environmentally consequential, however. At the national scale, the magnitude of most trends was much less than 1% of the parameter median per year. For example, the nation-wide negative trend in NH4 (-0.03), while statistically significant, equates to a 0.2% change in median NH4 concentration from 1996 to 2002. It is difficult to predict the environmental consequences of such trends when we use aggregated data and report median slope values for continuous data that range from negative to positive. Very large trends (i.e., the largest 5% of trends among all sites) were on the order of 5 to 10% of the parameter median. Such large trends are sometimes hidden when sites are aggregated. For example, there was a very large positive trend in NH4 concentration at the Seaward Downs monitoring site on the Mataura River, Southland (classified CD/L/HS/P). At this site, the median NH4 concentration increased by approximately 11% from 1996 to 2002. Despite this site-specific trend, there was not a significant trend in NH4 in the CD/L/HS/P class at the national scale. Identification of rivers at which water quality degradation is progressing rapidly may need to be made by a site-by-site examination.
Most of the patterns observed in water quality state and in temporal trends were dissimilar. For example, consistent differences between natural and pastoral land cover classes in median water quality parameters were not mirrored by differences in trends between these classes. In many cases, large differences were detected in water quality state, while trends were undetectable, or very small in magnitude. There are at least two possible reasons for the apparent uncoupling of state and trends. First, differences in state between classes could have developed at any time before the period used in trend analysis, and the current differences in state represent stable conditions in the classes being examined. Second, the mechanisms that cause some trends may not be the mechanisms that cause differences in state. For example, temperature-controlled changes nitrogen cycling may cause the decreasing trends in NOX and NH4 concentrations in natural and pastoral classes (Results, Section 3.3), while state differences in NOX and NH4 concentrations between natural and pastoral classes are caused primarily by land-use differences.
4.2. Stream order as a source of water quality variability
Stream order is often used as a surrogate for stream size, either to ensure that studies of multiple sites are conducted at similar-sized streams, or to ensure that a range of stream sizes is sampled within a catchment. The particular attributes of stream size that are expected to change with order include depth, velocity, discharge, and proportion of surface shaded by riparian canopy. However, these attributes are not always closely related to stream order: low order streams may be deep, sunlit and swift, and high order streams may be shallow, shaded and slow. Direct measurements of stream dimensions, such as median flow (e.g., Chetelat & Pick 2001) are likely to be better determinants of water quality than stream order.
Most REC classes in the dataset were represented by one or two stream orders; only 14 classes included a sequence of stream orders (e.g., orders 2 through 8). There were no indications in these cases that water quality parameters changed systematically with stream order. The lack of systematic effects of stream order, and the rarity of effects at low REC levels suggests that the REC land-cover level accounts for most fine-scaled systematic variation in water quality. Because stream order is rarely a major source of variability within REC classes, monitoring sites can probably be assigned to classes without regard for order. As a caveat, however, the statistical power of comparisons among stream orders at the land-cover level was limited by low and unequal sample sizes. Analysis using a balanced dataset (i.e., equal replication for each order in each land cover class) would help to confirm or reject the conclusion that stream order has little or no systematic effect on water quality.
With respect to temporal trends, the variability associated with stream order when sites were aggregated at the source-of-flow level was generally small. There were two cases in which differences in trends among stream orders were significant, trends in DRP in the CW/H class, and trends in clarity in the WW/L class (Results, Section 3.4.2). In both cases, there appeared to be systematic changes in trends across stream orders (Fig. 8). In particular, positive trends in clarity in the WW/L class tended to increase in magnitude with increasing stream order. This could indicate an improvement in clarity in larger rivers relative to small streams in the WW/L class, or a deterioration of water clarity in smaller streams superimposed on a climate-driven nationwide improvement in water clarity. We are unable to provide support for either hypothesis at this stage.
4.3. Limitations associated with replication and distribution of monitoring sites
The statistical power of some comparisons made in this study was limited by low and/or uneven replication and high variability within REC classes. Differences among REC classes, stream orders, and regions were assessed using a significance level (α < 0.05) that is conservative given the limitations of the dataset. We did so to ensure that few of the differences we report could have been due to chance alone, rather than to true differences among the groups being compared (i.e., we minimized the probability of type 1 errors). As a consequence, it is likely that some important differences that do exist were not detected (type 2 errors). This situation was particularly apparent in the comparisons of stream orders, where some mean parameter values differed by 2-3 times, but differences in ranks were not detectable in non-parametric tests.
The scope of the study was limited due to the absence of many stream classes from the state and trend datasets. This problem is clearly illustrated by the low number of comparisons we could make among fully-classified REC classes at the land-cover level (10 comparisons of state, Table 5; 4 comparisons of trends, Table 18). Of the 265 REC classes that occur in New Zealand (when classified using the approach in this study), only 88 were included in the state dataset, and 62 of these were represented by fewer than 6 sites. Of the 75 classes in the trends dataset, 53 were represented by fewer than 6 sites. Several REC classes that were not included in comparisons due to the absence or scarcity of data are important in New Zealand in terms of total stream length (e.g., all CD/M and WD/Lk classes). In other cases, comparative assessments of common REC classes could not made because there were too few sites in corresponding classes. For example, the CD/L/HS/P is a major class of low elevation streams in agricultural regions of both islands, but there were insufficient sites representing other land cover classes within the CD/L/HS class to make comparisons, so this important group could not be assessed at the land-cover level.
In addition to limited replication and spatial scope, bias in the selection of monitoring sites is an issue of concern for assessments using the REC. Bias is used here to refer to over- or under-representation of classes in monitoring programmes or in water quality datasets, relative to their abundance in nature. When a group of sites does not accurately represent the true distribution of environmental characteristics of a given class, inferences drawn from the sites to the entire class may be biased. In the REC framework, this effect can propagate from small spatial scales (e.g., land-cover classes) to larger scales (e.g., climate classes). Two examples serve to illustrate the potential for bias, and its effects.
The first example concerns the CD climate class. Water quality in the CD class was generally poor. Median nutrient and E. coli concentrations were higher than the recommended guidelines, and significantly higher than in the other cool climate classes, and clarity was below the recommended guidelines, and significantly lower than in the other cool climate classes. In the dataset, the CD climate class was dominated by sites belonging to the low elevation source-of-flow class and the pastoral land-cover class (source of flow class: 68% low elevation, 31% hill, < 1% lake; land-cover class: 94% pastoral, 3% urban, 3% natural). Across New Zealand, river kilometres in the CD climate class are distributed into source-of-flow classes as follows: 55% low elevation, 39% hill, 5% mountain, 1% lake. CD sites are distributed into land-cover classes as follows: 72% pastoral, 24% natural, 3% exotic forest, and 1% urban. A comparison of proportions of monitoring sites with proportions of total river length indicates that the low elevation and pastoral classes were over-represented in the dataset, and the hill, mountain and natural classes were under-represented. Inaccurate representation of the climate class, together with the strong effects of land-cover classes, suggests that average water quality at the monitoring sites gives a biased estimate of conditions in CD streams and rivers in general.
The second example concerns the WD climate class, which was also characterized by very high nutrient and E. coli concentrations and low clarity. For each parameter, at least 75% of the sites in the WD class did not meet the recommended guidelines. The WD sites in the dataset all belonged to the low elevation source-of-flow class, and to the pastoral and urban land-cover classes. Across New Zealand, river kilometres in the WD climate class are distributed into source-of-flow classes as follows: 99% low elevation, 0.6% lake, <0.1% hill. WD river kilometres are distributed into land-cover classes as follows: 89% pastoral, 4% urban, 4% exotic forest, and 2% natural. In this case, it appears that the monitoring sites in the dataset accurately represented the dominance of low elevation, pastoral streams in the WD climate class, and the conclusion that water quality is generally poor across the climate class is warranted.
4.4. Recommendations for establishing monitoring sites
Each of the limitations described above could be alleviated, at least partially, by increasing the numbers of sites that regional councils and unitary authorities monitor, or by changing the locations of some existing sites, or both. The risk of bias due to non-representative site selection can be minimized by choosing sites randomly from all possible sites in a given class. However, difficulty in accessing many sites, and over-riding needs to monitor particular sites can make random site selection impractical. As an alternative, sites may be deliberately chosen to accurately represent the composition and abundance of REC classes in a region.
Assessment of the environmental state of rivers in a district or region is only one reason that councils invest in water quality monitoring. Many rivers are monitored at specific sites to assess human health risks (e.g., bathing beach monitoring), to address specific research or management problems (e.g., non-point inputs related to selected land uses, or baseline data collection prior to a development), or to investigate acute short-term problems (e.g., effluent discharges, spills, fish kills). It is not realistic to expect resources allocated to these programmes will be reallocated to state-of-environment programmes in order to increase site numbers.
We recommend selecting sites for state-of-environment monitoring (or reviewing current site arrangements) in four steps. First, the REC should be used to determine the classes of rivers that occur in the district or region, and the abundance of each class (based on river length, surface area, or other appropriate measures). Second, the sites monitored in current state-of-environment (SOE), drinking water, and bathing beach programmes should be examined as a whole. Some long-term monitoring sites used for drinking water and bathing beach programmes might serve as SOE sites, although the parameters measured may have to be modified. We note that several district councils have SOE and bathing beach monitoring programmes underway that use an overlapping suite of parameters, but have no sites in common; using some sites for both programmes could increase efficiency. Third, the REC should be used to identify classes that are currently over-represented by monitoring sites, and some of these sites should be relocated into unrepresented or under-represented classes. Fourth, new monitoring sites should be established as necessary in remaining unrepresented or under-represented classes.
Water quality monitoring sites are most valuable when they have generated long-term data of consistent quality. Several district councils provided us with water quality data from sites that were monitored for short periods, or intermittently with long gaps. Such data is of little use for assessing current state and recent trends. In view of the importance of long-term contemporaneous data, much deliberation is needed before relocating sites and ending the record at former sites. Only sites that are clearly redundant should be relocated.
4.5. Increasing standardization in water quality monitoring
Variability among councils in water quality monitoring falls into two general classes, varied methods for field measurements and sample collection, and varied laboratory procedures and instrumentation (see Appendix 1). Significant differences in analytical results among New Zealand water quality laboratories have been reported, some due to differences in methods and instruments (Stansfield 2001). However, laboratory equipment for nutrient and microbiological analysis is costly, and it is not reasonable to expect councils to revamp laboratories in order to increase standardization. Further, councils that contract commercial laboratories for water sample analysis have limited control over methods or instruments [We recommend that all water quality laboratories participate in accreditation programmes and proficiency testing programmes (e.g., interlaboratory comparisons).] . In lieu of major changes in laboratory procedures, we recommend assessing, and if necessary, improving the standardization of field measurements and sample collection. For unpolluted freshwater, differences between the laboratory procedures currently in use will often result in small differences in concentrations and in detection limits. In contrast, differences in field procedures can have very large effects on data, and these differences can be greatly reduced at little expense.
There are no universally-accepted, optimum procedures for field measurements and water sampling, and differences among priorities (e.g., analytical accuracy, rapid fieldwork, budget constraints) are likely to affect the choice of procedures. Below we suggest procedures that are already widely used, and that should result in reasonable accuracy and efficient fieldwork. The suggestions are made with a typical field crew in mind: one or two people, travelling between sites by car or truck, and returning samples to the laboratory several hours or days after collection. These suggestions are the opinions of the authors, and intended primarily as points for discussions with regional council staff. A working group focused on water quality data collection, and composed of council field personnel, analysts, and water quality scientists is strongly recommended.
4.5.1. Sampling locations
To some degree, the point at each monitoring site where samples are collected and field measurements are made can be standardized. Points should be selected with the goal of collecting samples and making field measurements safely under varied flow and weather conditions, and with minimal contamination. The point should be identified with a global positioning system so that all subsequent collections are made at the same location. All points should meet these guidelines: straight reaches with uniform flow; far enough below confluences of streams or point sources of contamination to avoid sampling a cross section where flows are poorly mixed; upstream from bridges or other structures, to avoid contamination from the structure or roadbed.
4.5.2. Field measurements
Several councils use hand-held meters to make conductivity, dissolved oxygen and pH measurements in the field. However, field measurements of oxygen and pH can be problematic. Hand-held meters must be kept calibrated, which is time-consuming. Contamination of solutions and breakage and loss of parts are frequent problems. Non-standard conditions (e.g., stirring to prevent oxygen depletion) can affect measurements. These parameters may be better measured in the laboratory, where instruments can more easily be calibrated, constant conditions can be maintained, and surroundings are clean. Laboratory measurements of conductivity, pH, and dissolved oxygen routinely achieve accuracies of 1 µS/cm, 0.05 pH units, and 0.1 mg O2/L. Iodometric (Winkler) tests for dissolved oxygen samples fixed in the field are accurate and inexpensive. Modifications of the iodometric test for sites with suspected high nitrite, ferrous iron or suspended solids concentrations are given in APHA (1998). Laboratory-grade conductivity and pH meters are relatively inexpensive.
Turbidity and/or clarity are measured in the field by most councils. As with other hand-held meters, field turbidometers are less reliable than laboratory turbidometers, and are likely to be less accurate. If turbidity is routinely measured, we recommend doing so in the laboratory. Turbidity, suspended solids concentrations and water clarity are often considered to be equivalent measurements of optical water quality, but they are not. Clarity is a measure of is a measure of light beam and diffuse light attenuation. Turbidity and suspended solids concentration may be only weakly related to light attenuation because neither accounts for dissolved matter or particle composition and size distribution. Turbidity measurements also have problems associated with a lack of standard methodology. We recommend the following: for assessing optical water quality, clarity should be measured. For assessing non-optical effects of particulate matter, suspended solids concentrations should be measured. Clarity must be measured in the field. We recommend the black disk procedure of Davies-Colley (1988). In very small, sinuous, or deep channels, or in slow water were stirring by the field crew affects clarity, in situ measurements may be impossible. In these cases, offsite measurements with black disks or clarity tubes may be needed. Comparisons of in situ and offsite measurements made with black disks indicate that the two methods give near-identical results for clarity levels from 0.1 to 1.5 m (Davies-Colley & Smith 1992). Comparisons of in situ black disk and clarity tube measurements suggest that these methods give comparable results for clarity levels less than 1 m (Kilroy & Biggs 2002).
4.5.3. Water sample collection
The remaining water quality parameters measured by most district and regional councils require the collection of water samples for laboratory analysis. We recommend that council field teams follow the standardized procedures developed by the U. S. Geological Survey for river sampling (USGS 1997). Two issues in water sampling warrant further consideration, field-filtering and sample preservation and storage.
The need to sample water with mid- to high particulate concentrations is certain to arise, but samples from these waters may undergo changes in dissolved nutrients due to chemical and biological transformations. Such changes can be reduced by filtering samples in the field. Immediate freezing of samples with low particulate concentrations might eliminate the need for filtering (Dore et al. 1996), but few field crews have immediate access to freezers, so we recommend field-filtering all samples as a standard procedure. We note that several regional councils choose not to field-filter samples, in order to minimize contamination. A working group of council field personnel and water quality scientists should meet to examine this issue.
We recommend that all samples be transported on ice, and that nutrient samples are frozen as soon as possible, and kept frozen until analysis. Frozen storage has been reported to cause small reductions in DRP concentrations, as orthophosphate may form an precipitate with calcite or other compounds upon freezing, but DRP loss upon freezing is more severe in unfiltered samples than in filtered samples (Klingaman & Nelson 1976). Silica can form insoluble polysilicates upon freezing; programmes that include silica as a water quality parameter should be aware of this effect. Storage without freezing may result in changes in nutrient concentrations due to microbial metabolism, and these effects may be greater in magnitude that those of freezing. Alternatives to freezing include additions of biocides and strong acids. The most commonly used biocide, HgCl2 is hazardous to use, creates serious disposal problems, and does not stabilize all dissolved nutrients to the same degree. Acidifying and neutralizing increases the chances of contamination. Care should be taken to prevent repeated thawing and refreezing, as microbes that passed through the filter will fracture and add soluble contents to the sample upon each freeze-thaw cycle. Once frozen, filtered nutrient samples can be stored for long periods with little or no alteration (Lambert et al. 1992, Avanzino & Kennedy 1993).
4.6. Conclusions
This study has shown that there are substantial differences in water quality state among climate, source-of-flow and land cover classes. Water quality state in dry climate classes was generally lower than in wet and extremely wet classes. The predominant land-cover classes within the dry climate classes are pastoral. At the source-of-flow level, low elevation classes generally had lower water quality than mountain, hill and lake classes. An exception was the CX/L class, which had moderate water quality, and is dominated by indigenous forest land-cover classes. The other low-elevation classes are dominated by pastoral land-cover classes. These results lead to the conclusion that water quality state at high REC levels (i.e., large spatial scales) is affected both by large-scaled processes, such as temperature and runoff regimes, and fine-scaled processes associated with land cover. Effects of land cover are particularly evident at large scales when single land-cover classes comprise large proportions of higher-level classes.
The water quality state of land-cover classes was assessed by comparison with recommended guidelines, and by way of three inter-class comparisons: fully-classified classes, land-cover classes within climate classes, and land-cover classes across climate classes. The results of these comparisons consistently showed that pastoral classes have lower water quality than natural classes. Results of a nation-wide comparison of the four major land-cover classes indicated that urban classes have the lowest water quality, followed by pastoral, exotic forest and natural classes. Few urban land-cover classes in the dataset met any of the water quality guidelines. In contrast, most natural land-cover classes were within guideline values, except the NH4 guideline. Generally poor water quality in pastoral classes was evident within climate classes, nation-wide, and within individual regions.
Patterns of water quality state in REC classes within Canterbury, Southland and Waikato Regions conformed to nationwide patterns: median parameter values for low elevation source-of-flow classes generally failed to meet recommended guidelines, while those for mountain and hill classes were equal to or near guideline values, and pastoral land-cover classes generally exceeded guideline levels, while natural classes were at or below guideline levels. Results of inter-region comparisons varied among parameters and among REC classes. Therefore, we could not conclude that water quality in any one of the regions was generally higher or lower than in the other two.
At the nationwide level, and within several REC source-of-flow and land-cover levels, negative temporal trends in flow and in flow-adjusted inorganic nitrogen, and positive trends in flow-adjusted temperature and clarity were detected. The negative trends in flow and in NH4 may be long-term features, as they were also detected in a previous analysis covering the period 1995 - June 2001. The negative trend in NH4 appears to be in effect in pastoral rivers, but not in natural rivers, suggesting that the trend is a direct result of improved waste management. However it may be due instead to the indirect effects of increasing water temperatures, which may increase nitrification rates (Hamill & McBride in press)
Significant differences in trends among stream orders were rare at all REC levels. This finding suggests that REC classes can be used to classifying rivers and monitor water quality without regard for order. However, the issue has not been completely resolved, because low replication of stream orders at low REC levels reduced the statistical power of the comparisons. A dataset with higher and equal replication across stream orders should be examined before conclusions are drawn about stream-order effects within land-cover classes.
In general, water quality trends for the 1996 - 2002 period were quite small. Most of the significant trends indicated a change of less than 1% per year, and the average magnitude of all trends was much less than 1%. This pattern was observed at all scales examined (nation-wide, within REC classes, within the Canterbury, Southland and Waikato Regions, and within stream orders). The general pattern of small to non-existent water quality trends (i.e., slow or no recent changes in water quality state) contrasts with the large differences between REC classes that exist for water quality state. For example, nation-wide trends in DRP and E. coli concentrations within the natural and pastoral classes were not statistically significant, and the difference between those trends was also not statistically significant (Table 15). However, the state analyses indicated that the average DRP concentration in the pastoral class (0.29 g/m3) was nearly 3 times that of the natural class (0.10 g/m3). The average E. coli concentration in the pastoral class (186.2/100 ml) was nearly 5 times that of the natural class (851.8/100 ml). As noted previously, differences among classes in water quality state may have developed prior to the 1996-2002 period, and current differences in state may represent stable conditions. It appears that recent changes in land use practices, climate conditions or other sources of natural variability have had relatively small effects on water quality.
The REC framework partitions variability in water quality data into a large number of classes at multiple levels. Each level has a characteristic spatial scale, and each is associated with a suite of physical processes that influence water quality. These features make the REC a powerful tool in the sense that a large proportion of variability is accounted for (and relatively little attributed to "random effects"), and possible mechanisms to explain variability are offered at each level (e.g., temperature regime at the climate level, seasonal runoff patterns at the source-of-flow level). To reach its greatest potential as an analytical tool, however, the REC framework needs to be coupled with data from an appropriately-designed sampling programme. To date, most analyses that have used the REC framework, including the present study, have used data from programmes that were not designed for REC analysis (Snedler & Scarsbrook 2002, Snelder et al. in review). As a result, these analyses have been affected by replication and representativeness problems. The water quality programmes currently conducted by regional councils and unitary authorities were established before the REC was developed, and none were designed for analysis by a multiple-scaled system like the REC. Instead, site selection in most council programmes is based on the need to monitor sites where land-use pressures are greatest, to pair such sites with reference sites, and to distribute sites evenly across the region (Wilson 2001). The array of monitoring sites that suits the needs of a regional council is not necessarily well-suited for inter-regional or nation-wide assessments of environmental classes. A compromise may involve moderately increasing the degree to which council monitoring sites match the REC framework. We suggest several steps to help make such changes.
Datasets collected with the intent of analysis using REC could have two general forms, depending on the issues to be addressed. If the most important issues concern differences between river classes, balanced dataset are best. If the most important issues concern water quality state within a given area, a representative dataset is appropriate. As an example of the first case, it may be of interest to know how the pastoral and natural land-cover classes compare in the CD/L/VA class. Monitoring an equal number of randomly-selected sites from the CD/L/VA/N and CD/L/VA/P classes is the best approach this case. An example of the second case is an assessment of the water quality state of an entire district. The best approach in this case is to classify the streams in the district at the REC land-cover level (or finer), weight the number of sites selected from each REC class by the abundance of the class, then examine the average or median condition of all sites selected. REC classes should be at the land-cover level or finer in this case, because results of the present study suggest that land-cover levels have a strong influence on higher REC levels.
To improve the comparability of water quality data from multiple monitoring programmes, we recommend increasing the level of standardisation for sampling site locations, sample collection and storage, and field measurements. The first step should be the formation of a working group to discuss water quality monitoring procedures for river monitoring programmes.
Results of this study have shown that rivers with poor water quality are numerous and widespread in New Zealand. Every region in the nation has pastoral and urban streams that are currently degraded, or at risk of degradation, as indicated by membership in REC classes that have poor median water quality. Poor water quality (defined here as inorganic nutrient, E. coli and clarity levels that fail to meet recommended guidelines) poses a threat to ecological, recreational and aesthetic values. High inorganic nutrient concentrations may enhance the growth of periphyton, macrophytes, and some bacteria. When periphyton and macrophyte growth rates exceed rates of removal by algae and plant consumers and by scouring flows, living and detrital biomass accumulates and can reduce habitat quality and further reduce water quality. The effects of nutrient enrichment are likely to be most severe in low-elevation streams within developed catchments where low gradients, absence of riparian canopies and changes in invertebrate assemblages can result in heating, reduced aeration, and reduced control of periphyton by consumers. Inorganic nitrogen and phosphorus concentrations that stimulate periphyton growth in New Zealand rivers may be as low as 0.02 g N/m3 and 0.005 g P/m3 (Biggs 2000). Reductions in clarity can adversely affect productivity in macrophytes, bryophytes and algae (Davies-Colley et al. 1992), induce physiological stress in fish and invertebrates (Ryan 1991), and alter fish and invertebrate distributions and densities (Quinn et al. 1992, Rowe et al. 2000). E. coli is an indicator of the presence of pathogenic bacteria. High E. coli concentrations indicate high health risks for drinking untreated stream water, and for aquatic recreation and food gathering. Lowland pastoral and urban streams are the most vulnerable classes to microbiological contamination, due to high livestock densities and high input of contaminated water from outfalls, farm drains, and other point and non-point effluent sources. In view of the cultural and economic importance of aquatic recreation and food gathering in New Zealand, it is imperative that high standards of microbiological water quality are maintained.
