Environ Monit Assess (2014) 186:1939–1950 DOI 10.1007/s10661-013-3508-5
Water quality trends in New Zealand rivers: 1989–2009 Deborah J. Ballantine & Robert J. Davies-Colley
Received: 28 June 2013 / Accepted: 23 October 2013 / Published online: 7 November 2013 # Springer Science+Business Media Dordrecht 2013
Abstract Recent assessments of water quality in New Zealand have indicated declining trends, particularly in the 40 % of the country’s area under pasture. The most comprehensive long-term and consistent water quality dataset is the National Rivers Water Quality Network (NRWQN). Since 1989, monthly samples have been collected at 77 NRWQN sites on 35 major river systems that, together, drain about 50 % of New Zealand’s land area. Trend analysis of the NRWQN data shows increasing nutrient concentrations, particularly nitrogen (total nitrogen and nitrate), over 21 years (1989–2009). Total nitrogen and nitrate concentrations were increasing significantly over the first 11 years (1989–2000), but for the more recent 10-year period, only nitrate concentrations continued to increase sharply. Also, the increasing phosphorus trends over the first 11 years (1989–2000) levelled off over the later 10-year period (2000–2009). Conductivity has also increased over the 21 years (1989–2009). Visual clarity has increased over the full time period which may be the positive result of soil D. J. Ballantine (*) : R. J. Davies-Colley National Institute of Water and Atmospheric Research, Hamilton, New Zealand e-mail: [email protected] R. J. Davies-Colley e-mail: [email protected] Present Address: D. J. Ballantine Department of Environmental Sciences, Xi’an Jiaotong– Liverpool University, 111 Ren’ai Road, Dushu Lake Higher Education Town, Suzhou Industrial Park, Suzhou, Jiangsu Province, China
conservation measures and riparian fencing. NRWQN data shows that concentrations of nutrients increase, and visual clarity decreases (i.e. water quality declines), with increasing proportions of pastoral land in catchments. As such, the increasing nutrient trends may reflect increasing intensification of pastoral agriculture. Keywords Water quality . Rivers . New Zealand . Trends . Agriculture . Diffuse pollution
Introduction Freshwater bodies in New Zealand support a healthy array of flora and fauna and are highly regarded internationally for their recreational value. However, protecting the country’s freshwater bodies is a growing challenge. Water quality in New Zealand varies considerably; water quality in urban and pastoral agricultural areas is degraded, and is coming under increasing pressure as land use intensifies, which has negative connotations for aquatic life, drinking water supplies and water-based recreation. The links between water quality and land use are well recognised in New Zealand and elsewhere (e.g. Jones et al. 2001; Galbraith and Burns 2007; Monaghan et al. 2009; Abell et al. 2011; McDowell et al. 2011), and as land use intensifies throughout New Zealand, we can expect increases in diffuse source contaminants. The National Rivers Water Quality Network (NRWQN), operated by the National Institute of Water and Atmospheric Research (NIWA), is New Zealand’s only national-scale, freshwater quality monitoring
1940
network, although it is increasingly augmented by State of Environment monitoring conducted by Regional Councils (Ballantine et al. 2010). The NRWQN aims to provide a picture of water quality across the whole country through sampling large rivers (essentially ‘integrating’ the quality of all the tributary rivers which drain their catchments), while Regional Council State of Environment monitoring includes smaller rivers and sites which reflect regional issues and pressures. The NRWQN consists of 77 sites on 35 rivers that are fairly evenly distributed over both main islands of New Zealand and drain about one half of the nation’s land area (see Smith and McBride (1990) for a description; Fig. 1 shows site locations). Most of the river systems sampled have an upstream and downstream site. Sites are visited monthly as a compromise between detecting seasonal trends and minimising serial correlation (Smith and McBride 1990). NIWA staff do the monthly sampling. Dissolved oxygen (% DO), pH, conductivity, temperature, and visual clarity (measured by the horizontal black disc visibility method (Davies-Colley 1988)) are measured at the field sites. Samples are freighted by overnight courier to the NIWA-Hamilton laboratory (in chilled, insulated, opaque containers) for laboratory assessment of turbidity, coloured dissolved organic matter (CDOM), oxidised nitrogen, (the sum of NO3– and NO2 (NOx-N)), total nitrogen (TN), ammoniacal nitrogen (NH4-N), dissolved reactive phosphorus (DRP) and total phosphorus (TP) and the microbial indicator, Escherichia coli. Additionally, benthic biological variables are also monitored: periphyton cover is assessed visually on monthly visits, and macro-invertebrates are sampled annually (late summer). The benthic biological variables are not considered further here. Variables were chosen to reflect the objectives of the network, and detailed information on the rationale for their inclusion is given in Smith and McBride (1990) and Davies-Colley et al. (2011). All NRWQN sites are located at, or close to, hydrometric stations from which flow is estimated for each sampling visit. The goal of the network remains: “to provide scientifically defensible information on the important physical, chemical and biological characteristics of the nation’s river waters as a basis for advising government agencies on the trends and status of these waters”. The objectives are to “(1) allow detection of significant trends in water quality, and (2) develop a better understanding of the nature of the water resources so as to assist with their management” (Smith and McBride 1990).
Environ Monit Assess (2014) 186:1939–1950
The NRWQN design has remained stable throughout its history, with no changes to monitoring sites, apart from minor shifts in precise sampling locations (less than 1 km), essentially no change in methods of measurement and only one change in variables measured. Five-day biochemical oxygen demand (BOD5), a measure of biodegradable organic material that is most relevant to organic wastewater discharge, was replaced in 2005 by measurement of the faecal microbial indicator, E. coli. This network emulates ISO (ISO 2004, 2005) standards (i.e. design objectives are clearly defined and documented), even though its design predates the standards by many years. The value of stability in the NRWQN is becoming increasingly evident over time, particularly for the task of detecting water quality trends and for ‘anchoring’ temporary special purpose monitoring campaigns and experimental studies. For example, suspended sediment was monitored through 2011–2012 to allow water quality variables to be related to suspended particle concentrations. Further, detail regarding the NRWQN was given in a recent review (Davies-Colley et al. 2011). Examination of trends in long-term river water quality data is an effective means of assessing environmental change over time. It allows us to identify whether water quality is increasing or decreasing, so causal factors may be sought, such as improved wastewater treatment or changed land use. Assessment of temporal trends in national water quality from monitoring networks is rarely published in scientific journals. This may be partly due to the difficulties associated with maintaining long-term, stable (consistent over time) and un-interrupted water quality monitoring, which means that suitable long-term data records are uncommon. Furthermore, many national networks are not driven by hypotheses that can be easily incorporated into scientific research, and scientific publication of network data is not a priority. Finally, the sheer size of many nations, and the range of environmental variation therein, renders the integration of regional data sets for national-scale trend assessment of water quality impractical. Scotland and Japan have reported national trends in surface water quality (Ferrier et al. 2001; Luo et al. 2011), while trends in groundwater quality in the USA have been assessed by Rosen and Lapham (2008). Trends in phosphorus and nitrogen concentrations were assessed from 1975 to 1994 at 250 long-term monitoring sites in the US Geological Survey’s National Stream Quality Accounting Network (NASQAN, Hooper et al. 2001) by Alexander and
Environ Monit Assess (2014) 186:1939–1950
1941
Fig. 1 Locations of 77 NRWQN sampling sites on 35 major river systems in New Zealand. NRWQN rivers are labelled where they enter the sea
Smith (2006), however this stream network has been significantly scaled down and, at present, only 33 sites are monitored. Assessments of water quality trends at the catchment and regional level are common, e.g. Johnson et al. (2009) have assessed water quality trends in the Minnesota River, while Rothwell et al. (2010) have provided a spatial and seasonal assessment of water quality in North East England. Trend assessment
is also useful to evaluate changes in water quality to ensure protection of important natural ecosystems, for example, the Everglades National Park (Hanlon et al. 2010), Lake Okeechobee Watershed (Zhang et al. 2011) and Chesapeake Bay (Murphy et al. 2011). In New Zealand, trends in water quality are currently best examined at the national scale using the NRWQN data, though it probably gives an optimistic view as the
1942
NRWQN comprises mainly larger rivers, and many smaller, more polluted rivers are included. In the past two decades since the NRWQN commenced, several formal analyses of trends in the NRWQN data have been conducted, broadly following the approach of Smith et al. (1996). In this contribution, we aim to describe trends in water quality over the 21 years (1989–2009) since establishment. Trends for two shorter periods within the longer period are also considered: 1989–2000 and 2000–2009. We considered that trends over shorter periods (e.g. 5 years) might not provide useful information on trends in water quality at the national scale, and indeed, analysis of long-term data has shown that windows of 5 years tend to be too noisy for trend analysis and give misleading results (Burt et al. 2008; Howden et al. 2011).
Methods Data Trend analysis was applied to data for TN, TP DRP, NOx-N, % DO, water temperature, conductivity, pH, turbidity, visual clarity and CDOM (NH4-N was not included due to sample contamination issues). Flow adjustment The values of most water quality variables are dependent on flow. Relationships between water quality variables and flow have been discussed by various authors (Hirsch et al. 1982; McDiffett 1993; Smith et al. 1996). Positive correlations occur particularly for diffuse source pollutants such as TP, or suspended solids, which increase strongly with flow because of wash-in (mobilisation from the landscape) during rainstorms or entrainment from channel stores. Negative correlations occur particularly for pollutants derived from point sources, which are diluted as flows increase, together with solutes derived from interactions of water with minerals (typified by conductivity). It is important to note that the flow-adjusted trend does not necessarily represent all the water quality changes that result from human influence and management actions; it only describes those separate from flow. A change in farming practices that reduces surface runoff but increases groundwater recharge or a change in atmospheric deposition may not be captured in the flow-
Environ Monit Assess (2014) 186:1939–1950
adjusted trend. Therefore, while flow-adjusted trends are an indicator of human activities affecting water quality within a catchment, the relative magnitude must be considered in terms of the hydrologic variability (Langland et al. 2006). Constituent concentrations and loads in streams vary through time in response to many factors, including changes in sources, processes that might modify concentration, human actions, and variations in climatology, notably precipitation. Non-flow-adjusted trends allow for the examination of the overall response of the ecosystem to these changing factors, and comparison to changes in the ecosystem downstream (Langland et al. 2006). Calculation of trends for individual variables Before trend analysis, data for each variable at each site was flow adjusted. Flow adjustment was done using LOcally WEighted Scatterplot Smoothing (LOWESS) with a 30 % span. This allowed us to detect the overall flow-trend of all the data for each variable at each site. Every data-point in the record was then adjusted depending on the value of flow as outlined by Smith et al. (1996): adjusted value=raw value–smoothed value+ median value (where the “smoothed value” is that predicted from the flow using LOWESS). Data was not adjusted for serial correlation, as the network was originally designed so that the effects of serial correlation were minimal (Smith and McBride 1990). The trend analysis was applied to the whole 21 year dataset (1989–2009 inclusive). Following Smith et al. (1996), the non-parametric Seasonal Kendall Slope Estimator (SKSE) was used to represent the magnitude and direction of trends in the flow-adjusted data. The Seasonal Kendall test is a non-parametric test that accounts for seasonality by calculating the Mann–Kendall test on each of m ‘seasons’ separately, and then combining the results. The Sen Slope estimator is the median of all possible combinations of slopes within each season. For example, consider all the January data for n=5 years of records with monthly sampling. 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. This slope estimate is minimally affected by outliers and missing data. A robust feature of this method is that it accounts for seasonality by considering only within-month slopes for monthly data, i.e. data
Environ Monit Assess (2014) 186:1939–1950
is considered on a monthly basis, and January data are compared only with January, February with February, etc. The seasonal Kendall test poses the null hypothesis that there is no trend, with the alternative hypothesis being that there is an upward or downward trend (a two-sided test). This test provides the advantage that one does not have to make an assumption regarding the form of any trend that may be present (e.g., linear, exponential); the test merely considers whether between-year differences (for any given ‘season’) tend to be monotonic. Values of the SKSE were normalised by dividing the SKSE value by the raw data median to give the relative SKSE (RSKSE) and converted to a percent value, allowing for direct comparison between sites measured as percent change per year. The RSKSE may be thought of as an index of the relative rate of change. All statistics are reported at the 95 % confidence level (i.e., P0.05); 2. significant increase/decrease—the null hypothesis for the Seasonal Kendall test was rejected (i.e., P
Water quality trends in New Zealand rivers: 1989–2009 Deborah J. Ballantine & Robert J. Davies-Colley
Received: 28 June 2013 / Accepted: 23 October 2013 / Published online: 7 November 2013 # Springer Science+Business Media Dordrecht 2013
Abstract Recent assessments of water quality in New Zealand have indicated declining trends, particularly in the 40 % of the country’s area under pasture. The most comprehensive long-term and consistent water quality dataset is the National Rivers Water Quality Network (NRWQN). Since 1989, monthly samples have been collected at 77 NRWQN sites on 35 major river systems that, together, drain about 50 % of New Zealand’s land area. Trend analysis of the NRWQN data shows increasing nutrient concentrations, particularly nitrogen (total nitrogen and nitrate), over 21 years (1989–2009). Total nitrogen and nitrate concentrations were increasing significantly over the first 11 years (1989–2000), but for the more recent 10-year period, only nitrate concentrations continued to increase sharply. Also, the increasing phosphorus trends over the first 11 years (1989–2000) levelled off over the later 10-year period (2000–2009). Conductivity has also increased over the 21 years (1989–2009). Visual clarity has increased over the full time period which may be the positive result of soil D. J. Ballantine (*) : R. J. Davies-Colley National Institute of Water and Atmospheric Research, Hamilton, New Zealand e-mail: [email protected] R. J. Davies-Colley e-mail: [email protected] Present Address: D. J. Ballantine Department of Environmental Sciences, Xi’an Jiaotong– Liverpool University, 111 Ren’ai Road, Dushu Lake Higher Education Town, Suzhou Industrial Park, Suzhou, Jiangsu Province, China
conservation measures and riparian fencing. NRWQN data shows that concentrations of nutrients increase, and visual clarity decreases (i.e. water quality declines), with increasing proportions of pastoral land in catchments. As such, the increasing nutrient trends may reflect increasing intensification of pastoral agriculture. Keywords Water quality . Rivers . New Zealand . Trends . Agriculture . Diffuse pollution
Introduction Freshwater bodies in New Zealand support a healthy array of flora and fauna and are highly regarded internationally for their recreational value. However, protecting the country’s freshwater bodies is a growing challenge. Water quality in New Zealand varies considerably; water quality in urban and pastoral agricultural areas is degraded, and is coming under increasing pressure as land use intensifies, which has negative connotations for aquatic life, drinking water supplies and water-based recreation. The links between water quality and land use are well recognised in New Zealand and elsewhere (e.g. Jones et al. 2001; Galbraith and Burns 2007; Monaghan et al. 2009; Abell et al. 2011; McDowell et al. 2011), and as land use intensifies throughout New Zealand, we can expect increases in diffuse source contaminants. The National Rivers Water Quality Network (NRWQN), operated by the National Institute of Water and Atmospheric Research (NIWA), is New Zealand’s only national-scale, freshwater quality monitoring
1940
network, although it is increasingly augmented by State of Environment monitoring conducted by Regional Councils (Ballantine et al. 2010). The NRWQN aims to provide a picture of water quality across the whole country through sampling large rivers (essentially ‘integrating’ the quality of all the tributary rivers which drain their catchments), while Regional Council State of Environment monitoring includes smaller rivers and sites which reflect regional issues and pressures. The NRWQN consists of 77 sites on 35 rivers that are fairly evenly distributed over both main islands of New Zealand and drain about one half of the nation’s land area (see Smith and McBride (1990) for a description; Fig. 1 shows site locations). Most of the river systems sampled have an upstream and downstream site. Sites are visited monthly as a compromise between detecting seasonal trends and minimising serial correlation (Smith and McBride 1990). NIWA staff do the monthly sampling. Dissolved oxygen (% DO), pH, conductivity, temperature, and visual clarity (measured by the horizontal black disc visibility method (Davies-Colley 1988)) are measured at the field sites. Samples are freighted by overnight courier to the NIWA-Hamilton laboratory (in chilled, insulated, opaque containers) for laboratory assessment of turbidity, coloured dissolved organic matter (CDOM), oxidised nitrogen, (the sum of NO3– and NO2 (NOx-N)), total nitrogen (TN), ammoniacal nitrogen (NH4-N), dissolved reactive phosphorus (DRP) and total phosphorus (TP) and the microbial indicator, Escherichia coli. Additionally, benthic biological variables are also monitored: periphyton cover is assessed visually on monthly visits, and macro-invertebrates are sampled annually (late summer). The benthic biological variables are not considered further here. Variables were chosen to reflect the objectives of the network, and detailed information on the rationale for their inclusion is given in Smith and McBride (1990) and Davies-Colley et al. (2011). All NRWQN sites are located at, or close to, hydrometric stations from which flow is estimated for each sampling visit. The goal of the network remains: “to provide scientifically defensible information on the important physical, chemical and biological characteristics of the nation’s river waters as a basis for advising government agencies on the trends and status of these waters”. The objectives are to “(1) allow detection of significant trends in water quality, and (2) develop a better understanding of the nature of the water resources so as to assist with their management” (Smith and McBride 1990).
Environ Monit Assess (2014) 186:1939–1950
The NRWQN design has remained stable throughout its history, with no changes to monitoring sites, apart from minor shifts in precise sampling locations (less than 1 km), essentially no change in methods of measurement and only one change in variables measured. Five-day biochemical oxygen demand (BOD5), a measure of biodegradable organic material that is most relevant to organic wastewater discharge, was replaced in 2005 by measurement of the faecal microbial indicator, E. coli. This network emulates ISO (ISO 2004, 2005) standards (i.e. design objectives are clearly defined and documented), even though its design predates the standards by many years. The value of stability in the NRWQN is becoming increasingly evident over time, particularly for the task of detecting water quality trends and for ‘anchoring’ temporary special purpose monitoring campaigns and experimental studies. For example, suspended sediment was monitored through 2011–2012 to allow water quality variables to be related to suspended particle concentrations. Further, detail regarding the NRWQN was given in a recent review (Davies-Colley et al. 2011). Examination of trends in long-term river water quality data is an effective means of assessing environmental change over time. It allows us to identify whether water quality is increasing or decreasing, so causal factors may be sought, such as improved wastewater treatment or changed land use. Assessment of temporal trends in national water quality from monitoring networks is rarely published in scientific journals. This may be partly due to the difficulties associated with maintaining long-term, stable (consistent over time) and un-interrupted water quality monitoring, which means that suitable long-term data records are uncommon. Furthermore, many national networks are not driven by hypotheses that can be easily incorporated into scientific research, and scientific publication of network data is not a priority. Finally, the sheer size of many nations, and the range of environmental variation therein, renders the integration of regional data sets for national-scale trend assessment of water quality impractical. Scotland and Japan have reported national trends in surface water quality (Ferrier et al. 2001; Luo et al. 2011), while trends in groundwater quality in the USA have been assessed by Rosen and Lapham (2008). Trends in phosphorus and nitrogen concentrations were assessed from 1975 to 1994 at 250 long-term monitoring sites in the US Geological Survey’s National Stream Quality Accounting Network (NASQAN, Hooper et al. 2001) by Alexander and
Environ Monit Assess (2014) 186:1939–1950
1941
Fig. 1 Locations of 77 NRWQN sampling sites on 35 major river systems in New Zealand. NRWQN rivers are labelled where they enter the sea
Smith (2006), however this stream network has been significantly scaled down and, at present, only 33 sites are monitored. Assessments of water quality trends at the catchment and regional level are common, e.g. Johnson et al. (2009) have assessed water quality trends in the Minnesota River, while Rothwell et al. (2010) have provided a spatial and seasonal assessment of water quality in North East England. Trend assessment
is also useful to evaluate changes in water quality to ensure protection of important natural ecosystems, for example, the Everglades National Park (Hanlon et al. 2010), Lake Okeechobee Watershed (Zhang et al. 2011) and Chesapeake Bay (Murphy et al. 2011). In New Zealand, trends in water quality are currently best examined at the national scale using the NRWQN data, though it probably gives an optimistic view as the
1942
NRWQN comprises mainly larger rivers, and many smaller, more polluted rivers are included. In the past two decades since the NRWQN commenced, several formal analyses of trends in the NRWQN data have been conducted, broadly following the approach of Smith et al. (1996). In this contribution, we aim to describe trends in water quality over the 21 years (1989–2009) since establishment. Trends for two shorter periods within the longer period are also considered: 1989–2000 and 2000–2009. We considered that trends over shorter periods (e.g. 5 years) might not provide useful information on trends in water quality at the national scale, and indeed, analysis of long-term data has shown that windows of 5 years tend to be too noisy for trend analysis and give misleading results (Burt et al. 2008; Howden et al. 2011).
Methods Data Trend analysis was applied to data for TN, TP DRP, NOx-N, % DO, water temperature, conductivity, pH, turbidity, visual clarity and CDOM (NH4-N was not included due to sample contamination issues). Flow adjustment The values of most water quality variables are dependent on flow. Relationships between water quality variables and flow have been discussed by various authors (Hirsch et al. 1982; McDiffett 1993; Smith et al. 1996). Positive correlations occur particularly for diffuse source pollutants such as TP, or suspended solids, which increase strongly with flow because of wash-in (mobilisation from the landscape) during rainstorms or entrainment from channel stores. Negative correlations occur particularly for pollutants derived from point sources, which are diluted as flows increase, together with solutes derived from interactions of water with minerals (typified by conductivity). It is important to note that the flow-adjusted trend does not necessarily represent all the water quality changes that result from human influence and management actions; it only describes those separate from flow. A change in farming practices that reduces surface runoff but increases groundwater recharge or a change in atmospheric deposition may not be captured in the flow-
Environ Monit Assess (2014) 186:1939–1950
adjusted trend. Therefore, while flow-adjusted trends are an indicator of human activities affecting water quality within a catchment, the relative magnitude must be considered in terms of the hydrologic variability (Langland et al. 2006). Constituent concentrations and loads in streams vary through time in response to many factors, including changes in sources, processes that might modify concentration, human actions, and variations in climatology, notably precipitation. Non-flow-adjusted trends allow for the examination of the overall response of the ecosystem to these changing factors, and comparison to changes in the ecosystem downstream (Langland et al. 2006). Calculation of trends for individual variables Before trend analysis, data for each variable at each site was flow adjusted. Flow adjustment was done using LOcally WEighted Scatterplot Smoothing (LOWESS) with a 30 % span. This allowed us to detect the overall flow-trend of all the data for each variable at each site. Every data-point in the record was then adjusted depending on the value of flow as outlined by Smith et al. (1996): adjusted value=raw value–smoothed value+ median value (where the “smoothed value” is that predicted from the flow using LOWESS). Data was not adjusted for serial correlation, as the network was originally designed so that the effects of serial correlation were minimal (Smith and McBride 1990). The trend analysis was applied to the whole 21 year dataset (1989–2009 inclusive). Following Smith et al. (1996), the non-parametric Seasonal Kendall Slope Estimator (SKSE) was used to represent the magnitude and direction of trends in the flow-adjusted data. The Seasonal Kendall test is a non-parametric test that accounts for seasonality by calculating the Mann–Kendall test on each of m ‘seasons’ separately, and then combining the results. The Sen Slope estimator is the median of all possible combinations of slopes within each season. For example, consider all the January data for n=5 years of records with monthly sampling. 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. This slope estimate is minimally affected by outliers and missing data. A robust feature of this method is that it accounts for seasonality by considering only within-month slopes for monthly data, i.e. data
Environ Monit Assess (2014) 186:1939–1950
is considered on a monthly basis, and January data are compared only with January, February with February, etc. The seasonal Kendall test poses the null hypothesis that there is no trend, with the alternative hypothesis being that there is an upward or downward trend (a two-sided test). This test provides the advantage that one does not have to make an assumption regarding the form of any trend that may be present (e.g., linear, exponential); the test merely considers whether between-year differences (for any given ‘season’) tend to be monotonic. Values of the SKSE were normalised by dividing the SKSE value by the raw data median to give the relative SKSE (RSKSE) and converted to a percent value, allowing for direct comparison between sites measured as percent change per year. The RSKSE may be thought of as an index of the relative rate of change. All statistics are reported at the 95 % confidence level (i.e., P0.05); 2. significant increase/decrease—the null hypothesis for the Seasonal Kendall test was rejected (i.e., P
