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

Lake Water Quality in New Zealand

Acknowledgements

Executive summary

1 Introduction

2 Methods

3 Results and discussion

4 Conclusions

References

Appendix 1: Glossary of common limnological terminology used in this report

Appendix 2: Morphological and Catchment Characteristics of the 121 Lakes

Appendix 3: Maps of Water Quality Parameters

Appendix 4: Map of Water Quality Trends (1990-2006) based on Percent Annual Change (PAC) using the Burns et al (2000) Methodology

2 Methods

2.1 Lake database and classification

For this analysis we have used a GIS-based national database of 3,820 New Zealand lakes greater than 1 ha in area, delineated from the NZMS 260 1:50,000 topographic map series. These lakes include over 98% of the total lake area of New Zealand according to the latest Land Information New Zealand (LINZ) database. Water quality data are only available for a small fraction of these lakes, but their ecological character can be classified according to the major limnological variables that drive water quality (eg, mixing regime, morphometry, nutrient inputs), allowing extrapolation to other lakes.

We have used two classification approaches to explore differences in lake condition among different lake types. The traditional method of lake classification in New Zealand and elsewhere is based on mode of geological formation, originally developed by Hutchinson (1957), and adapted for New Zealand by Irwin (1975). This method produces eight lake categories (see Table 1), plus sub-categories and combinations of these.

NIWA has recently developed a lake environmental classification scheme for the Department of Conservation that incorporates these classes, together with other environmental factors that control lake ecosystems (Snelder, 2006), such as climate, catchment type, size and shape, all of which are known to influence water quality and ecosystem structure. These factors are arranged hierarchically from those likely to be important at a national scale (eg, climate) to those likely to be important locally (eg, soil type). The geological emphasis of the Hutchinson and Irwin classifications is captured in factors such as soil type and lake shoreline development. The environmental classes produced from this approach allow condition to be compared between groups of lakes that differ in ecological character at a range of spatial scales. The classification can group lakes according to their natural (pre-European settlement) character or according to modern land-use classes. This method results in seven primary classes of lakes (see Table 2) distinguished at the national scale, with sub-categories of each at more regional scales. Table 2 emphasises that most lakes in New Zealand are small. Note that the Snelder (2006) classification scheme is still provisional and likely to be refined further.

Table 1: Classification of New Zealand lakes* based on mode of geological origin, following Irwin (1975)

* Includes only lakes with at least one axis > 0.5 km (a total of 776 lakes).

Table 2: Classification of New Zealand lakes* based on the lake environmental classification scheme of Snelder (2006)

* Includes all lakes > 1 ha in area (a total of 3,820 lakes).

Differences in depth stratification patterns are central to understanding lake productivity and its responses to human activities. Lakes that are deep enough to undergo full seasonal stratification (monomictic lakes) are almost exclusively in classes 4, 5 and 7. The few large, coastal lakes in class 6 are too shallow and have too great a fetch to stratify seasonally, and most of the small lakes in classes 1 to 3 are too small for seasonal stratification. However, brief intermittent stratification (polymixis) is common in these lakes, sometimes diurnally or sometimes irregularly, depending on solar radiation, wind and air temperature.

2.2 Water quality data and land-use classification

2.2.1 Data collection

We obtained lake water quality data from regional councils, local authorities and research institutes for all lake sites that have been monitored over the last 10 years, and in some cases received data going back as far as 1990. All data were provided with GPS (global positioning system) locations of collection sites, which we used to link data to our lake database. Errors in GPS readings were corrected algorithmically to ensure all sites were correctly located. The water quality data collected are shown in Table 3.

Table 3: Water quality data collected from regional councils, local authorities and research institutes

Nutrient concentrations below detection limits were replaced with values equal to half the detection limit. Measurements that were clearly erroneous were deleted, and in some cases there were errors in units of concentrations that could be corrected by comparison with other data. Data sets varied widely with regard to the inclusion of parameters, and also in the number of data and numbers of missing values for each parameter. As a result, the total number of data points varied widely among parameters.

Sufficient data were available to characterise recent water quality and make statistical comparisons for 121 lakes from 10 regions. This is a much greater number of lakes than previous comparisons, which have usually involved fewer than 20 lakes, and reflects the increase in lake monitoring carried out by councils in recent years. Of the 121 lakes, only 49 had data sets of sufficient duration and number of data points to calculate trends following Burns et al (2000). We have therefore taken two approaches:

• a nationwide analysis of the current condition or state of New Zealand lakes based on the 121 lakes

• trend analysis on data from the 49 lakes.

To carry out a robust trend analysis, our minimum requirements were a three-year data set and at least 15 time points to fit linear regressions. For the nationwide snapshot of condition, we took significant differences in condition between land use and lake morphology identified from the 121 lakes; and then used the GIS database to extrapolate to the 3,820 lakes. Morphological and catchment characteristics of these lakes are shown in Appendix 2.

2.2.2 Trophic Level Index (TLI)

We compared water quality using raw data for the various individual parameters and using the TLI methodology. The TLI method logarithmically transforms TN, TP, Z_SD and chla data to give a trophic level score for each parameter on the same scale. The formulae for these four component indices (TLn, TLp, TLs and TLc) are:

• TLn = -3.61 + 3.01log₁₀ (TN)

• TLp = 0.218 + 2.92log₁₀ (TP)

• TLs = 5.10 + 2.27log₁₀ (1/Z_SD – 1/40)

• TLc = 2.22 + 2.54log₁₀ (Chla).

The method then averages the four component scores into an index of trophic condition, the TLI. Trophic state classes are defined based on the TLI range (see Table 4).

Table 4: Definition of seven trophic states based on range in average Trophic Level Index (TLI), and corresponding nutrient enrichment descriptions*

* As described by Burns et al (2000).

The raw TN, TP, Z_SD and chla data were highly skewed, but the log-transformation ensured that TLn, TLp, TLs and TLc were close to normal distributions, enabling statistical comparisons. Figure 1 compares histograms of the raw data and transformed data for TN as an example, with the raw data being highly skewed (many low values), compared with the normally distributed TLn data. Figure 1 also shows the normal distributions of all four TLI components and the aggregate TLI.

The average TLI and its components were used in parametric statistical tests. Untransformed values are used at all other times.

2.2.3 Land cover/use

Land cover was determined from the latest New Zealand land-cover database (LCDB2, 2001–2002). The LCDB2 classes were grouped to form five land categories that covered almost all land in lake catchments throughout the database:

• alpine

• native forest and scrub (hereafter termed native)

• pasture

• exotic forest

• urban.

We then took two approaches to looking at relationships of catchment type to water quality.

First, we took a broad overview by associating lakes with the predominant land use in their catchment. The rule applied to LCDB2 was that the predominant land use was the largest single land-cover class in the catchment. This approach has limitations. For example, dominant land use may not be the most important factor affecting water quality in lakes where point sources are more important than diffuse sources. An example of this is Lake Rotoiti (Bay of Plenty), where the input from the Ohau Channel (draining Lake Rotorua) has a large effect on trophic state, even though native is the largest single land cover for this lake’s catchment (unless the Lake Rotorua catchment is included as part of the Rotoiti catchment). Lakes Horowhenua and Rotorua (Bay of Plenty) are other examples of lakes with a history of significant point source inputs. However, relatively few New Zealand lakes have large point source inputs, and these outliers were easily identified and are discussed separately. The trophic state of the large majority of New Zealand lakes is driven by diffuse catchment inputs that can be explained from land-use patterns.

Another limitation of simple dominance rules is that some land uses may have a proportionately greater influence on water quality than others. For example, urban land use often has very dramatic effects on water quality even if only present in small amounts. This may also differ in different soil types: a given land use may contribute greater nutrient run-off in porous soils such as sand and pumice than in other less porous soils. Also, land-cover classes such as ‘pasture’ will include agricultural land with a range of livestock species, stocking densities and management practices. In many regions councils have specific land-use information to allow closer examination of land-use practices and water quality. However, we found that the dominant land use identified from LCDB2 data was adequate to provide a useful picture of broad land-use effects.

Second, we examined land-use relationships by plotting TLI values and individual water quality parameters against the proportion of the catchment in different land-use types, to complement the analysis based on simple dominance. This approach allowed more detailed examination of interactions between land use, lake morphology and climate.

2.3 LakeSPI

All LakeSPI data collected by NIWA were incorporated into the database and linked to the appropriate lake ID and water quality set. LakeSPI is an aggregate score based on three parameters:

• plant colonisation depth (defined as depth where plant cover is < 10%)

• native condition index (higher score for greater species diversity)

• invasive condition index (lower score for more species and greater invasion).

There were 39 lakes for which we had both TLI and LakeSPI data, all in Bay of Plenty, Waikato and Northland.

2.4 Data analysis

Of the parameters provided in the data sets, eight water quality parameters (TN, TP, Z_SD, chla, NO₃-N, NH₄-N, conductivity and pH) were available in sufficient detail to allow comparisons between regions and land-cover types, and were used in the state analysis. TLI values were calculated from the four TLI variables (TN, TP, Z_SD, chla) according to the TLI methodology (Burns et al, 2000), but these variables were also examined individually because their individual responses can differ with different land-use impacts. The three LakeSPI parameters and their aggregate LakeSPI score were treated similarly. Current values for water quality parameters used in the state analysis were calculated as de-seasonalised averages of 2004–2006 data.

Seventy-five lakes had only three of the four TLI parameters (TN, TP and chla). However, we found that the average TLI in those lakes for which we had all four parameters correlated moderately well with the average TLI based on TN, TP and chla alone. Accordingly, we used the three-parameter average TLI when Z_SD was not available. Burns et al (2000) also provide evidence that a three-parameter TLI is adequate for lake characterisation in many cases.

Our statistical analysis of the factors driving lake water quality was based on the classification and regression tree method (De’ath and Fabricius, 2000). This is a relatively simple regression method that can deal very robustly with non-linear relationships and data sets with many missing values. Briefly, tree analysis carries out iterative calculations and generates regression trees that identify the independent factors (eg, lake size, shape and location, catchment land use) that are most important in driving the dependent variables (ie, water quality parameters). It then presents them visually as a tree diagram, dividing lakes into groups differing in water quality relating to the important variables. For this analysis we generated regression trees using TLI indices for TN, TP, chla (121 lakes), and alternative trees using TLI indices for all four parameters where these were available (75 lakes). The tree accepted is the smallest one within one standard error.

For trend analysis we took two approaches. For the 49 lakes that have monitoring data long enough for accurate trend analysis, we applied the percent annual change (PAC) methodology as described by Burns et al (2000). Data for these lakes are from post-1990, and include continuous 1990–2006 records in some regions, 1995–2006 records in some regions, and sporadic monitoring in the 1990–2006 period in others. It should be stressed that Burns et al (2000) recommended three-year running averages as a time scale suitable for validly identifying trends. Spurious significant trends, or significant trends related to short-term natural changes rather than long-term pressures, can often arise when shorter records are used, especially in oligotrophic lakes, where a small change in water quality can give a high PAC value.

The Canterbury high-country lakes in the data set are a good example. Their monitoring only began recently (December 2004), and shows decreases in TN, TP and chla over that time. These apparent improvements in what were already near-pristine lakes with little human pressure appear to be related to recent drought conditions in Canterbury and reduced inflows, which decreased nutrient inputs. Such lakes demonstrate that fluctuations in water quality do occur independently of human effects, and emphasise the need to use appropriate long-term data sets for identifying human impacts.

To provide a view of longer-term changes, we also made simple comparisons of the current state against two time points at which the state had previously been described for a range of lakes. White (1983) used 27 lakes ranging from oligotrophic to eutrophic condition in his classic study of the nature of nutrient limitation in New Zealand lakes. We were able to access recent data for 16 of these, and we plotted 1983 values for TN, TP and chla from these lakes against 2004–06 averages. Early Z_SD values were too variable to make comparisons. Burns and Rutherford (1998) monitored 26 lakes over 1992–1996 in developing the New Zealand lake monitoring programme, of which 18 had recent monitoring data for comparison. This exercise allowed a broad-scale comparison of current 2006 lake status against two time points (1983 and 1996), albeit with a different set of lakes in each comparison.

[ Previous|Next