Water Quality Monitoring: From River To Reef Glenn De'ath

Transcription

1 Water Quality Monitoring: From River To Reef Glenn De'ath p.

2 Table of Contents :Overview... :Summary...5 A: Synthesis...5 B: Chapter Summaries...7 3:Recommendations... :River Water Quality Monitoring...5 Summary...5 Data...6 Statistical Analysis...6 Exploratory Analyses...7 Trend Analyses...7 Sampling Considerations...8 List of Tables...9 List of Figures...9 Tables and Figures... Exploratory Data Analyses... Tully...6 Trend Analyses :Chlorophyll Monitoring... Summary... Data... Statistical Analysis... Results... Sampling...3 Tables and Figures...3 6:Lagoon Water Quality Monitoring...5 Summary...5 Data...5 Statistical Analysis...5 Results...55 Tables and Figures :Seagrass Monitoring...65 Summary...65 Data...65 Statistical Analysis...65 Results...66 List of Tables...67 List of Figures...67 Tables and Figures...68 p.

3 8:Benthos Monitoring...8 Summary...8 Data...8 Statistical Analysis...8 Results...8 Video coral cover...8 Manta tow coral cover...83 Coral demography data...83 List of Tables and Figures...83 Tables and Figures :Coral Recruitment Monitoring...96 Summary...96 Data...96 Statistical Analysis...96 Results...96 List of Tables...96 List of Figures...96 Tables and Figures...97 :Statistics, Causality and Ecological Change... Introduction... Statistical issues in the detection of change... Attribution of causality... References... :Estimating Monitoring Performance... River water quality... Lagoon water quality... Long-term chlorophyll... Reef benthos video assessments... Methods... Tables of Precision...6 () Summary All programs...6 (a) Water Quality Tully River Data...6 (b) Water Quality Burdekin River Data...7 (3) Water Quality Lagoon Data...7 () Long-term Chlorophyll Surveys...8 (5) Benthos Data Video Surveys of Inshore Reefs...8 p.3

4 : Overview This document has been prepared for the Reef Water Quality Protection Plan Monitoring (RWQPPM). It is assumed the reader is familiar with the background materials and the context; they are not repeated here. The document has several objectives, namely:. To present exploratory and inferential analyses of a number of data sets foundational to the RWQPPM. The exploratory analyses reveal graphically many characteristics of these data. The inferential analyses show the potential to estimate temporal change, and some of the statistical issues related to that process.. To inform about methodological and statistical issues of monitoring. 3. To present estimates of the expected performance of the established monitoring programs.. To make recommendations about aspects of the RWQPPM. These relate to the objectives of the program, to methodological, statistical and data issues and to allocation of resources. The structure of the document is as follows.. Chapter comprises: (A) brief synthesis, and (B) summaries of key issues relating to monitoring and data management, and summaries of the data analysis chapters.. Chapter 3 is a series of recommendations about the RWQPPM. These concern objectives of the program, 3. Chapters 9 are analyses and discussions, based on long term data sets from established monitoring programs. The results are discussed in relation to the RWQPPM. These data sets include water quality in the Tully and Burdekin rivers, lagoon water quality, chlorophyll concentrations in the GBR lagoon, inter-tidal seagrass, coral recruitment and reef benthos cover data.. Chapter is a detailed and technical discussion of the philosophical and statistical issues associated with the monitoring of complex environmental and ecological systems. 5. Chapter presents estimates of the expected performance of the established monitoring programs with regard to detecting long-term changes. This includes estimates of precision for annual means of responses (e.g. water quality parameters), differences in means, and linear trends. Disclaimer The views expressed in this document are those of the author. They do not necessarily coincide with those of the Australian Institute of Marine Science, Reef CRC or the Great Barrier Reef Marine Park Authority. Acknowledgements Many thanks to Miles Furnas and David Haynes for many useful comments and editorial contributions which have lead to the major improvements of this document. :Overview p.

5 : Summary A: Synthesis The majority of this report deals with the analysis of variability in four long-term water quality data sets from two Queensland rivers and the Great Barrier Reef lagoon. Other major data sets are intertidal seagrass cover and composition and benthos cover of inshore reefs based on video transects. : Water Quality Monitoring The water quality data are highly significant since:. They are from monitoring programs that are likely to form the core of the Reef Water Quality Protection Plan Monitoring (RWQPPM).. The principal objectives of the RWQPPM are to determine baselines and long-term change in the water quality of the rivers and the lagoon. The primary finding of this report concerns the ability of the water quality monitoring programs (river sediment and nutrients, lagoon nutrients and chlorophyll) to provide sufficiently precise estimates of change over periods of ~ years. For river nutrient data, we can reliably detect an increase in concentrations of water quality parameters of ~5% over a -year period. This result is the median estimate of all water quality parameters, and assumes change is linear over time. The median detectable increase rises to ~75% if change is non-linear. For lagoon water quality data, the median detectable increases are ~% (linear) and ~5% (non-linear), and for chlorophyll are ~3% (linear) and 8% (non-linear). These estimates of precision of baselines and long-term change need to be carefully considered in relation to the RWQPPM objectives. In particular, are we likely to meet objectives given the expected levels of precision? Statistical analysis of river water quality reveals the crucial role of river flow rates in determining observed concentrations of most nutrients and suspended sediment. Thus, in order to assess longterm changes in water quality, we need to adjust concentrations for the temporal variation of river flows. Since the variation of river flows is large compared to expected changes due to land management practices, estimates of long-term change in concentrations are relatively imprecise. The statistical analyses also reveal unexplained systematic variation that suggests unobserved factors in the rivers or catchments may be affecting concentrations. Three possible ways to improve precision are:. Increase the frequency of sampling occasions. This can be accompanied by reduced analytical sub-sampling. To improve the precision of estimation by a factor a, an increase in sampling of ~- fold would be required.. Improve the current level of precision by additional statistical data analysis. This needs to be based on improved knowledge of the processes governing the variability of concentrations in the rivers and the lagoon. The analyses of other data, e.g. time series data from intensely sampled rivers may help us better understand the processes. 3. New automated technologies have the capacity to deliver large volumes of data with high temporal resolution. This is likely, but not certain to provide better estimates of concentrations of some sampled parameters. :Summary p.5

6 : Seagrass Monitoring Seagrass monitoring has collected data from at 63 sites in 6 geographical regions along the Queensland coast. It is a multi-purpose program, providing monitoring information to detect change in seagrass communities, and also enables to community participation and education. The seagrass monitoring is providing valuable information about temporal trends and spatial differences, with changes in seagrass cover occuring at various spatial and temporal scales. The monitoring showed recovery in the Hervey Bay Region, from ~3% cover in increasing to ~7% cover in. The value of the monitoring would be greatly enhanced by adding more widely spread regions to the surveys. 3: Benthos Monitoring Analysis of coral cover based on video surveys and manta tow of inshore reefs showed significant declines in the central section of the Great Barrier Reef (Cairns to Townsville). Typical declines were ~5 - % cover over ~ - 5 year periods. The spatial and temporal scales of the effects of changes in water quality on reef benthos are largely unknown. Careful consideration needs to be given to issues of scale and to the selection of reefs and sites to be monitored. Locating reefs along known water quality gradients could increase the ability of the program to relate changes in water quality to changes in reef benthos. :Summary p.6

7 B: Chapter Summaries The following summaries are direct extracts from subsequent chapters.. Environmental and Ecological Monitoring Issues of contemporary environmental and ecological monitoring are discussed. The central points are:. Environmental and ecological systems are constantly changing everywhere. A primary objective of monitoring programs is to assess how much and how systems are changing, not to assess if they are changing.. Objectives of monitoring programs are often poorly defined. This leads to difficulties in evaluation and adaptation. 3. Monitoring programs should be adaptive to new knowledge and methodologies.. Concurrent research and data analysis need to be considered as essential components of monitoring programs. 5. At present, methods development, experimental work to resolve causes for change and variability, and scientific and analytical support are under-resourced in comparison with data collection.. Data sources Data management and access continues to be a major issue. The major elements are:. Knowledge of relevant data. We need to acquire extensive meta-data covering all aspects of the monitoring programs. This should include not only data sets of agencies involved in the RWQPPM, but also relevant data sets from other agencies.. Access to data. Data from RWQPPM agencies should be openly available. Other data can often be obtained through license or collaboration. 3. Data need to be cleaned, documented and safely warehoused. There are obligations to deliver information on-line for reporting purposes. However, the need to provide meta-data and enable access to data for analysis and research are of greater importance to ensure effective performance of monitoring programs, and require fewer resources. 3. River Water Quality Monitoring The river water quality monitoring program is primarily based on nutrient and fine sediment sampling in the Tully and Burdekin rivers over the period 987. The water quality parameters examined in this document are: nitrite (NO ), nitrate (NO 3), dissolved inorganic nitrogen (DIN = NO + NO 3 + NH ), total dissolved nitrogen (TDN), dissolved organic nitrogen (DON = TDN-DIN), particulate nitrogen (PN), dissolved inorganic phosphorous (PO ), total dissolved phosphorous (TDP), dissolved organic phosphorus (DOP = TDP - PO ), particulate phosphorous (PP), silicate (Si) and suspended solids (SS).. Changes in concentrations of water quality parameters, in both the Tully and Burdekin rivers, are most strongly determined by river flow (discharge) rates. Concentrations of most nutrients increase linearly with flow, but some changes are non-linear. The non-linear effects typically occur with :Summary p.7

8 dissolved nutrients, where dilution may be occurring at high flow rates. Since the range of flow varies up to 3-fold for the Tully and -fold for the Burdekin, variation in flow rates results in large changes (5 fold) in concentrations of some water quality parameters (particulate nitrogen and particulate phosphorous, and suspended solids) due to flow alone. Concentrations of PN and PP are strongly related to the concentration of suspended sediment. Clearly changes in water quality parameters have to be very accurately adjusted for river flow rates in order to detect typical temporal changes of less than -fold.. The adjustment of concentrations of water quality parameters for varying flow rates is made more difficult by the inter-annual variability of mean river flow rates (discharge). For example, the mean flow rate of the Tully river more than doubled from 99 to, and the Burdekin river also experienced strong inter-annual variation. 3. In order to accurately adjust concentrations for river flows, water quality sampling should take place at sites as close as possible to points where flow rates are measured.. Linear trends in concentrations over time are typically weak and were only detected in the Tully river data. Particulate nitrogen and phosphorous increase at. and 5.9% per year respectively. Since flow rates have increased >3-fold over the sampling period, these estimates are adjusted to average flow rates (see ). Non-linear systematic changes that occur for several parameters in both the Tully and Burdekin rivers have no obvious explanation given the data used in these analyses. 5. The strongest inter-relationships of variables are between particulate nitrogen (PN) and particulate phosphorous (PP), suspended solids (SS) and river flows. These variables also co-vary with time and this could possibly be exploited to better estimate change over time; e.g. by seeing how ratios of these parameters vary. There is potential to relate such ratios of parameters to changes to in land management. 6. The patterns of change in the other water quality parameters (i.e. excluding PN, PP and SS) is weak. If the prime objective is estimate change, the value of further collection of data for some of these parameters is questionable given the current methodologies. 7. The analyses presented here identify some statistical issues and problems. Estimated trends need to be adjusted for certain covariates e.g. seasonal and river flow effects. Correlation between the effects we wish to estimate (e.g. long-term trends) and effects we wish to control for (e.g. flow rates), and correlation between successive samples over time, can lead to inaccurate estimates of the former. There is systematic variation in concentrations of water quality parameters that is unrelated to data used in these analyses. Estimated effects (such as trends over time) are often non-linear and the sampling is highly unbalanced. Analysis of these data is thus difficult and simple statistical solutions are almost certain to produce simplistic findings. Given the statistical complexities, the findings should (as always!) be treated with caution. 8. Additional sampling of rivers should take place during the winter dry season to regularize the data set. In addition to sampling high flow events (as is the current practice), each river should be sampled at least once each month if possible. 9. Sub-sampling, in terms of duplicates and analytical replicates, could be reduced, and pooledsampling could be used. For example, suppose the current practice is to take water samples, split each into, and analyse in the laboratory the resulting sub-samples. There would be minimal loss of information (precision of the estimate values obtained from the sample) if the water samples were pooled and a single laboratory analysis was used. QA & QC issues can be catered for using occasional full sampling; perhaps in every samples. This should result in a >7% reduction in processing costs..the unexplained apparent systematic changes over time are a challenge to our knowledge that needs to be overcome if we are to successfully monitor and understand long-term change in river water quality. What is driving such changes? There are many possibilities. Spatial and temporal variation in rainfall patterns may result in variable contributions from sub-catchments to river mouth concentrations. Sample collection and/or laboratory analysis may vary over time. Environmental covariates which affect the concentrations may be unknown, uncollected, unavailable or simply :Summary p.8

9 unused..the cost-information ratio of water quality data based on collected water samples and chemical analysis is high. Alternative technologies such as in-situ continuous recording are attractive since they can deliver far lower cost-information ratios. The development and deployment of these technologies should be a high priority.. Lagoon Water Quality Monitoring The data used for this analysis were collected by AIMS oceanographers in coastal waters off Cairns over the period November 989 March. Hydrographic, water quality and biological sampling has been carried out at stations in coastal and GBR lagoon waters between Cape Tribulation and Cape Grafton.. There were significant increases in nitrite (NO ), total dissolved nitrogen (TDN), total dissolved phosphorous (TDP) and suspended solids (SS). Nitrite and suspended solids concentrations increased linearly by ~%, The total dissolved nitogen and phosphorous concentrations increased in the later years of surveys after a relatively constant period. Increases were ~5% and ~% respectively. Chlorophyll and nitrate (NO 3) concentrations showed significant non-linear decreases of 3% and 5% respectively. 3. Dissolved inorganic phosphorous (PO ) concentrations decreased in early years but then increased in later years.. The variation of water quality parameters over time was, in most cases, consistent across locations. However, for some parameters, most notably particulate nitrogen (PN) and phosphorous (PP), suspended solids (SS), chlorophyll and phaeophytin, there was large variation between locations. 5. There was little systematic seasonal variation. 6. It would be interesting to return to the non-depth averaged data, to include covariates such as the weather and sea state conditions, and re-analyse the data. Such an approach may result in more precise estimates of temporal change. 5. Chlorophyll Monitoring Phytoplankton chlorophyll a has been monitored monthly since 99 at a large number of sites (stations) in the GBR lagoon. Over time, 86 sites have been visited on this program - ca. have been sampled on an ongoing basis. The stations are located within 8 latitudinal groups of stations (transects) distributed across the shelf from 3 o S to 3 o S. A primary objective of the chlorophyll monitoring program is to estimate long term change in the inshore environment of the GBR resulting from the increasing loads of nutrients being exported from the river catchments adjoining the GBR.. Strong differences in chlorophyll exist across the shelf. Mean concentrations in inshore stations along most transects are greater than offshore except in the far north (3 o 5 o S). In northern transects mean chlorophyll concentrations are low (.5 µg/l) both inshore and offshore. Mean concentrations in offshore stations in the rest of the GBR are similar (.5. µg/l) except in the Capricorn region in the far south where offshore mean chlorophyll is.68 µg/l. In contrast to northern transects inshore mean chlorophyll concentrations from Port Douglas south fall in the range.5.65 µg/l.. Strong seasonal effects are evident in chlorophyll with mean summer/wet season (December April) values ~5% greater than those in winter/dry season (May November). 3. Significant patterns of spatial differences and temporal change were observed in the data over the ten-year period in each of the cross-shelf transects. Long term change was smaller in magnitude than spatial or seasonal differences; typical change being of ~ % over several years. :Summary p.9

10 . There are strong differences between sites and similarities between site profiles over time. In the context of estimating change it is thus extremely important to maintain sampling at the same sites, otherwise spatial and temporal change will be confounded. 5. Phaeophytin patterns are very similar to those of chlorophyll. Levels of phaeophytin are consistently 5% of chlorophyll; with one exception, namely that phaeophytin levels consistently lag those of chlorophyll by one month. 6. The current practice of sampling involves duplicate field samples each analysed twice in the laboratory. An analysis of variance components shows potential to reduce replication with little loss of precision. 7. Size-fraction sampling would enhance the value of chlorophyll and phaeophytin data. For example, larger particles are known to be of food source for the crown-of-thorns starfish (Acanthaster planci). 6. Seagrass Monitoring The Seagrass Watch Monitoring Project has collected data from at 63 sites in 9 locations in 6 geographical regions along the Queensland coast. It is a multi-purpose program, and in addition to providing monitoring information designed to detect change in seagrass communities, it also has objectives that relate to community participation and education.. The seagrass monitoring is providing valuable information about temporal trends and spatial differences.. Changes in seagrass cover occur at all spatial scales (region and site within region) and temporal scales (years), and are clearly identified. 3. The greatest change in cover at the regional scale occurs in Hervey Bay, with levels of ~3% cover in increasing to ~7% cover in. Of the 6 Hervey Bay sites, 8 recovered to some degree.. Changes in the relative composition of Halodule uninervis, Halophila ovalis, Zostera capricorni, epiphytes and algae (the more common taxa) were weaker than for seagrass cover. However, moderate spatial differences and temporal changes were observed for these data. 5. Increasing the spatial coverage of sites by monitoring in additional regions would greatly enhance the value of these surveys. 7. Benthos Monitoring The AIMS Long-Term Monitoring Program has been monitoring reefs of the GBR since ~986. The benthos data analysed in this report comprise video and manta tow coral cover estimates recoded over years, and one-off demography data which focuses on small coral recruits.. Analysis of coral cover based on video surveys showed significant declines in the Cairns and Townsville sectors of 5 - %, but no evidence of change in the Cooktown-Lizard or Whitsundays sectors.. Manta tow coral cover for 8 inshore reefs shows a consistent pattern of either decline or failure to recover over the period Partitioning of variance of coral cover based on video surveys showed that the choice of analysis of statistical model can be crucial. Such choices depend on (among other things) the spatial unit (region, reef, site within reef) that we use for determining change. These questions of inference space are crucial and can be clarified by consideration of reefs selection and sampling. :Summary p.

11 . The use of reduced temporal sampling should be considered with regard to the program objectives. Biennial sampling could replace the current annual sampling with minimal loss of precision of estimated change or trends. 5. For the coral demography data, the distributions of small recruits (<5cm) showed little systematic change across sectors or reefs. 8. Coral Recruitment Monitoring Coral recruits were recorded over 7 years on 3 reefs of the GBR. The numbers of recruits of coral spat per tile varied greatly with typical standard deviations greater than the mean. Coral recruitment on John Brewer Reef declined over the 7 years. This reduction occurred mainly in the last years. For Lizard Island and Yonge Reef, coral recruitment was relatively stable. A large recruitment event occurred at a shallow site on Yonge Reef in Statistics, Causality and Ecological Change Determination of change and causality in ecosystems is difficult, both philosophically and practically, and these difficulties increase with the scale and complexity of ecosystems. This chapter addresses philosophical and statistical issues of monitoring change.. Estimating Monitoring Performance The expected performance of major programs the RWQPP monitoring was estimated for () river water quality, () lagoon water quality, (3) long-term chlorophyll, and () inshore reef benthos video assessments. For these programs, sufficient data has been gathered over several years and a diversity of conditions to enable reliable estimation of monitoring performance. For each of the programs, measures of precision are provided. These are () the confidence interval (CI) of the mean for any single year, () the CI for the difference between any two years, (3) the CI for the linear trend over a year period, and () the worst and best expected detectable difference over a year period. Expressed as percentage change, average confidence interval for year trends across the parameters of the programs were () river water quality (%, +6%), () lagoon water quality (-39%, +65%) (3) chlorophyll monitoring (-%, +7%), and () reef benthos (%, +6%). :Summary p.

12 3: Recommendations The recommendations listed below are based on the proposed Reef Water Quality Protection Plan Monitoring (RWQPPM) EOI, knowledge of design and analysis of monitoring programs and analyses of water quality data and benthos included in this report. Recommendation The objectives should be clarified and made commensurate with the -year time-scale of the contract. The RWQPPM is currently a -year program, though it is likely to extend well beyond that period. The current objectives of the RWQPPM are (from the EOI):. To determine long-term (decadal) trends for water quality.. To produce annual time series... as a basis for detecting changes related to water quality for biological monitoring. These objectives are questionable since:. The contract is for a -year program, and the objectives should relate to that time period, not to decades.. Detecting change implies the question has it changed, or not?, yet we know environmental and ecological systems are constantly changing everywhere. The objective should be to assess how much and how systems are changing, not to assess if they are changing. 3. The issue of causality has been discussed, but is not clearly addressed or resolved in the objectives. For example does the phrase detecting changes related to water quality imply that the changes are accepted as being related to water quality, or is there a requirement to demonstrate this. If causal attribution is to be objective, then this has major ramifications for the design and implementation of the program. Recommendation The two-year program should be used to evaluate both established monitoring techniques and also more novel untested methods. A major focus of the -year program should be the process of determining the effectiveness of the proposed strategies with the objective of putting in place a monitoring program which will effectively determine change over ~ years. That longer term program will, of course, be subject to adaptation in light of new knowledge and technologies. Some monitoring methods proposed for the RWQPPM are well-established, e.g. water quality analysis of river and lagoon data, lagoon chlorophyll monitoring, and video monitoring of reef benthos. For these methods we have access to sufficient data to assess their effectiveness. Other proposed methods are more recent and novel. Their performance cannot be assessed at this time, though they have strong support from scientists. They include the use of passive sampling of pollutants, surveys of coral recruits and the use biological indicators. Newer technologies, e.g. data loggers and remote sensing, have the potential to deliver large volumes of data in a cost-effective manner, and should be considered as alternative and/or complementary to more traditional methods. 3:Recommendations p.

13 We should also be aware of other methods such as the use of coral cores as sentinel monitors. Historical information is vital to long term changes in water quality. Where the distant past is unknown, we need to find techniques to uncover it. Recommendation 3 Additional statistical analysis should be undertaken over the -year program to extract the full potential of the water quality data. This work should result in more precise estimation of change in water quality parameters. For example, estimation of concentrations in river water quality can be improved by adjusting for flow rates, seasonality, total annual discharge of rivers (which varies on scales of several years), and for correlations between consecutive samples. Additional analyses of other water quality data sets, e.g. data from intensely monitored rivers may shed further light on the processes involved, and lead to better estimation. This work requires close collaboration between statistical and water quality experts. To turn data into knowledge requires resources, and a full-time statistician should be appointed to the program. This position would help balance of program resources between data collection on the one hand, and analysis and interpretation on the other. It is also essential in order to meet the requirement to publish results from the monitoring programs in quality journals. Recommendation An objective of the RWQPPM should be to establish baseline levels for specified water quality and benthos parameters from the -years data collection. The -year time scale is suitable for the establishment of baseline levels for the parameters measured by each of the methods. These baseline data will be useful in the future even if the programs that generated them are not continued in the short term. The data will also be essential in determining the temporal frequency and spatial extent of future monitoring for the individual programs. Recommendation 5 All relevant data sets need to be discovered, and if possible, obtained. Data need to cleaned, cataloged, warehoused, maintained and made available to researchers. Data management and access continues to be a major issue. The major elements are:. Knowledge of relevant data. We need to acquire extensive meta-data covering all aspects of the monitoring programs. This should include not only internal data sets (AIMS), but all relevant external data.. Access to data. Internal data should be openly available. External data can often be obtained through license or collaboration. 3. Data need to be cleaned, documented and safely warehoused. There are obligations to deliver information on-line for reporting purposes. However, the need to provide meta-data and enable access to data for analysis and research are of greater importance to ensure effective performance of monitoring programs, and require fewer resources. 3:Recommendations p.3

14 Recommendation 6 Collaboration with NRM board research and monitoring of river flows and nutrients. It can be argued that mouth of river water quality monitoring is the most essential component of the proposed program. This is particularly the case if an objective is to relate long-term changes in land management to the quality of water at river mouths, and in turn to relate river water quality to lagoon water quality and reef benthos. To better understand the complexities of water quality concentrations in rivers we need data and knowledge from the catchments. This is the especially the case when different sub-catchments are delivering waters of different composition. Collaborative links and common protocols should be established. Recommendation 7 Strategies for more efficient sampling in the rives, lagoon and for chlorophyll concentrations. Sub-sampling of water in rivers and the lagoon can be reduced with minimal loss of precision, maintained quality control and assurance, and substantial cost savings. Duplicates and analytical replicates could be dropped, and pooled-sampling with a single laboratory analysis used. QA & QC issues can be catered for by use of occasional full sampling using duplicates and analytical replicates. One full sample for every pooled samples should be sufficient to maintain quality. This could result in a >7% reduction in laboratory costs. These suggestions are based largely on data and QA & QC issues. A cost benefit analysis should be done to more precisely estimate the savings. Recommendation 8 The spatial and temporal sampling for reef benthos based on video transects should be reviewed. The use of reduced temporal sampling should be considered with regard to the program objectives. Analyses of current data show that biennial sampling could replace the current annual sampling with minimal loss of precision of estimated change or trends. The loss of precision of estimates (e.g. SEs and CIs) will be less than 5%, and may be was low as -5%. Biennial sampling would reduce field costs by 5%, or release the resources for equivalent sampling elsewhere or at other times, and the cost information ratio could be reduced to by - 8%. 3:Recommendations p.

15 : River Water Quality Monitoring Summary The analysis of trends and variation in river water quality is primarily based on data derived from samples of the Tully and Burdekin rivers over the period 987. The water quality parameters examined in this document are dissolved inorganic phosphorous (dip), dissolved inorganic nitrogen (din), nitrite (no), dissolved organic phosphorous (dop), dissolved organic nitrogen (don), nitrate (no3), total dissolved phosphorous (tdp), total dissolved nitrogen (tdn), silicate (si), particulate phosphorous (pp), particulate nitrogen (pn), and suspended solids (ss). The analyses presented here are both descriptive and inferential. The former present the data to the reader for self-assessment of patterns and information in the data. The latter address: () the issue of long-term change of water quality parameters, () the influence of flow rates and seasonal variation on temporal change, and (3) sampling issues. Findings are as follows:. Changes in concentrations of water quality parameters are most strongly determined by river flow rates for both the Tully and Burdekin rivers. Concentrations of most nutrients increase linearly with flow, but some effects (changes) are non-linear. The non-linear effects typically occur with dissolved nutrients. Dilution of some dissolved nutrient species may be occurring at high flow rates. Since the range of flow varies up to 3-fold for the Tully and -fold for the Burdekin, variation in flow rates results in large changes (5 fold) in concentrations of some water quality parameters (particulate nitrogen and particulate phosphorous, and suspended solids). Clearly these changes have to be very accurately adjusted for flows in order to detect typical temporal changes of less than -fold which have to be standardised to a common flow.. The adjustment of concentrations of water quality parameters for varying flow rates is made more difficult by the inter-annual variability of mean river flow rates. The mean flow rate of the Tully more than doubled from 99 to, and large changes in flow rates also occurred in the Burdekin River. 3. In order to accurately adjust concentrations for river flows, we need to measure water quality parameters as close as possible to points where flow rates are measured.. Linear trends over time are typically weak and are only present in the Tully river data. Particulate nitrogen and phosphorous increase at. and 5.9% per year respectively. Since flow rates have increased >3-fold over the sampling period, these estimates are adjusted to average flow rates (see ). Non-linear change that occurs for several parameters for both the Tully and Burdekin rivers has no obvious explanation. 5. The strongest inter-relationships of variables are between particulate nitrogen and particulate phosphorous, suspended solids and river flows. These variables also co-vary with time and this could possibly be exploited to better estimate change over time; e.g. by seeing how ratios of these parameters vary. There is potential to relate such ratios of parameters to changes in land management. 6. The patterns of change in the dissolved water quality parameters (i.e. excluding particulate nitrogen and phosphorous and suspended solids) are weak. Thus the value of further collection of data for these parameters using the current methodologies is questionable if the prime objective is to estimate change. 7. The analyses presented here identfiy statistical issues and problems. Estimated trends need to be adjusted for certain covariates e.g. seasonal and river flow effects. Correlation between the effects we wish to estimate (e.g. long-term trends) and effects we wish to control for (e.g. flow rates), and correlation between successive samples over time, can lead to inaccurate estimates of the former. Estimated effects (such as trends over time) are often non-linear and the sampling is highly :River Water Quality Monitoring p.5

16 unbalanced. Analysis of these data is thus difficult and simple statistical solutions are almost certain to produce simplistic findings. Given the statistical complexities, the findings should (as always!) be treated with caution. 8. Sampling of rivers needs to be more regularised. In addition to sampling high flow events (as is the current practice), each river should be sampled each month if river flows are adequate. 9. Sub-sampling, in terms of duplicates and analytical replicates, could be reduced, and pooledsampling could be used. For example, suppose the current practice is to take water samples, split each into, and analyse in the laboratory the resulting sub-samples. There would be minimal loss of information (precision of the estimate values obtained from the sample) if the water samples were pooled and a single laboratory analysis was used. QA & QC issues can be catered for using occasional full sampling; perhaps in every samples. This should result in a >7% reduction in processing costs..the unexplained apparent systematic temporal changes in some water quality parameters are a challenge to our knowledge that needs to be overcome if we are to successfully monitor and understand long-term change in river water quality. What is driving such changes? There are many possibilities. Spatial and temporal variation in rainfall patterns may result in variable contributions from sub-catchments to river mouth concentrations. Sample collection and/or laboratory analysis may vary over time. Environmental covariates which affect the concentrations may be unknown, uncollected, unavailable or simply unused..the cost-information ratio of water quality data based on collected water samples and chemical analysis is high. Alternative technologies such as in-situ continuous recording are attractive since they can deliver far lower cost-information ratios. The development and deployment of these technologies should be a high priority. Data The data sets used for this analysis were collected by the Australian Institute of Marine Science (AIMS) biological oceanography group. They comprise sampling for nutrients and suspended sediments in the Tully (987 ) and Burdekin (988 ) rivers. The statistical analyses are based on surface water samples collected by bucket from the highway bridges crossing the Tully (wet catchment) and Burdekin (dry catchment) rivers at Euramo and Ayr, respectively. Most samples were collected during the summer wet season, and within each individual wet season, concentrated on flow events of varying magnitude. Water flows in both rivers were sufficient to keep the water column well mixed with regard to dissolved nutrients and particulate matter associated with very fine sediment particles. Daily river discharge rates on the sampling occasions were obtained from The Queensland Department of Natural Resources and Mines (Watershed: Statistical Analysis The data analyses are presented in two sections: () Exploratory Data Analyses contains exploratory and descriptive analyses and () Trend Analyses presents analyses of long-term trends including options for defining long-term trends statistically and illustrating the importance of adjusting for covariates such as seasonal effects and river flow. The details of the latter are presented in the section Trend Analyses. All concentraions and flow rates are log transformed (base ), since under this transformation the data are approximately normally distributed. If data are not transformed, patterns in the data are largely obscured by a few extreme large values. This is clearly illustrated in the time series plots (Figure ) of the river flow data. Base has been used for the log transformation, rather than natural logarithms (base e) or base, for ease of interpretation. For base, a one unit change corresponds to a doubling or halving. The R statistical software was used for all analyses and graphics. Packages nlme, mgcv, and lattice were extensively used. Reference: R Development Core Team (). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. ISBN , URL :River Water Quality Monitoring p.6

17 Exploratory Analyses Exploratory analyses of the data (Figures -) show strong patterns of change over years, seasons (months), and with river flow. The patterns across the response variables (Table ) vary greatly. Some patterns of change, differences between the Tully and Burdekin rivers, and associations between variables are in agreement with current views of water quality processes however others are unexplained. Examples of the former are associations between particulate nutrients, suspended solids and river flows. Examples of the latter are highly non-linear systematic changes across years; particularly for the Burdekin river. The legends of each figure document some prime features of the data, and these figures should be carefully inspected prior to reading the trend analyses. Trend Analyses Methods Variation of the water quality parameter data with years, months and flow were first explored graphically (Figures -). The responses were all log (base ) transformed since this resulted in approximately normally distributed residuals of reasonably constant variance from exploratory linear models. There was strong evidence of non-linear dependencies of the responses on decimal years (e.g = 3--99, not 99, 99 etc), river flow (which was also log-transformed [base ]), and also dependencies on months. Additional analysis indicated the presence of temporal correlation which was approximated by a first order autoregressive process. Thus the statistical model used for each analysis was a generalized additive model. The non-regular sampling made the temporal correlation more difficult to model and data were averaged over months within years for both the Tully and the Burdekin. This reduced the number of cases from 39 to for the Tully and from 6 to 58 for the Burdekin. The relatively greater reduction in cases for the Burdekin was due to less regular sampling and more intense clusters of sampling in high flow periods. Additionally there were substantial numbers of haphazardly missing cases; particularly for the Burdekin data. Cases were weighted by the numbers of samples in each month. The response was the log (base ) transformed parameter (e.g. dip), and the explanatory variables were smooth trends in decimal years, river flow (log (base )), and months as a factor. The smoothness of the temporal and flow terms was selected by cross-validation, and a first-order autocorrelation term was also included. Examination of the distribution of residuals and serial correlations suggested that the chosen statistical model was adequate with one exception; estimation was unreliable for total dissolved phosphorous in the Burdekin, most likely due to the high autocorrelation. Alternative models were assessed based on untransformed responses but using the same predictors and autocorrelated errors. These used log-linear models rather than linear models, but the results are not presented here. Log-linear models fit the arithmetic mean rather than the geometric mean, as is the case with the log transformed response and linear model. Arguments can be made for both approaches. Results The numerical results in the form of estimates of smoothness for temporal trends and flow effects, estimates of the autocorrelation and tests of significance are shown in Table. Tests for month effects for the Tully river data are omitted. The effects for trends and flow effects are shown in Figures & (Tully) and & 5 (Burdekin) and should be closely examined. These effects vary considerably across the water quality parameters in both strength and form. For the Tully (Figure ), there are moderately strong linear increases in particulate nitrogen and particulate phosphorous. There are also strong non-linear changes in dissolved organic nitrogen, total dissolved nitrogen, total dissolved phosphorous and silicate. For the Burdekin (Figure ), there are weak increases in dissolved organic phosphorous and silicate. There are also strong non-linear changes in dissolved organic nitrogen, total dissolved nitrogen, total dissolved phosphorous and dissolved inorganic phosphorous. There are strong effects of flow for the Tully river data (Table & Figure ), with significant effects for 9 of the water quality parameters; the 3 non-significant effects are for nitrite, dissolved organic phosphorous and dissolved organic nitrogen. All of the effects appear plausible with some dilution apparent at high flow rates. For the Burdekin (Table & Figure ) river data there also are significant :River Water Quality Monitoring p.7

18 flow effects for 9 of the water quality parameters; dissolved organic phosphorous, dissolved organic nitrogen, and silicate were non-significant. For the Tully river data there are significant correlations between samples over time for 9 of the water quality parameters; of the 3 non-significant correlations one is also marginal. For the Burdekin river data there are significant correlations between samples over time for 9 of the water quality parameters; of the 3 non-significant correlations one is also marginal. Also shown are the trend effects when autocorrelation is not included in the models (Figures & 7), and apparent structure which is simply an artifact of ignoring (or missing) autocorrelation. Neglecting river flow in the estimation of trends can also be deceiving. In many cases, there are positive dependencies between flow and the responses, Since flow rates are higher in later years of surveys for both the Tully and Burdekin rivers, this can lead to considerable over-estimation of increasing trends in responses over time. Examples are particulate nitrogen and phosphorous rates of increase which would be over twice as high if unadjusted for flow rates. The results of the analyses for the Tully river show clear trends in particulate phosphorous and particulate nitrogen, and flow is a strong determinant of concentrations for most of the water quality parameters. For the Burdekin, there are no clear trends over time, though flow is again a strong determinant of concentrations for most of the water quality parameters. The smaller data set (when averaged by months) and the less stable estimation are problematic. Sampling Considerations Sampling frequency and balance over time There are clear differences in the patterns of sampling over time for the Tully and Burdekin (Figure 3). The range of monthly frequencies was -9 samples for the Tully, and -59 for the Burdekin. Also for the Burdekin, sampling in the early years was negligible. These two factors, the imbalance over months and the shorter period of sampling, contribute to the difficulty of determining trends for the Burdekin. A more balanced approach for all rivers is advisable, with at least one sample per month, and more frequent sampling in times of high flow. Sub-sampling The variances of duplicate samples within sampling occasions and analytical replicates within subsamples is relatively small compared to the total variation of all data (Table 5). This suggests considerable gains in efficiency could be made by dropping analytical replicates and possibly duplicate samples. There is some doubt about the classification of sampling occasions, duplicates and analytical replicates in the data, and if less questionable data were available, then accurate calculations on the precision of estimation could be made for alternative sampling strategies. :River Water Quality Monitoring p.8

19 List of Tables Table. Water quality variables used for statistical analyses. Table. Analysis of water quality parameters for the Tully and Burdekin rivers over the period 987. Table 3. Estimation of linear change in concentrations of water quality parameters for the Tully river. Table. Estimation of change in concentrations of water quality parameters for the Tully and Burdekin rivers. Table 5. Sampling variance analysis for water quality parameters for the Tully river. List of Figures Figure. Plots of variation in river flow for the Burdekin and Tully rivers over the period 97. Figure. Seasonal and trend analyses of river flow for the Burdekin and Tully rivers over the period 97. Figure 3. Illustration of the sampling intensity. Figure. Temporal variation in water quality parameters for the Tully river over the period 987. Figure 5. Temporal variation in water quality parameters for the Burdekin river over the period 988. Figure 6. Monthly variation in water quality parameters for the Tully river over the period 987. Figure 7. Monthly variation in water quality parameters for the Burdekin river over the period 988. Figure 8. Plots showing the relationships of water quality parameters to river flow for the Tully river. Figure 9. Plots showing the relationships of water quality parameters to river flow for the Burdekin river. Figure. Comparison of water quality parameters for the Burdekin and Tully rivers averaged over the months January, February and March for the period 987. Figure. Plots showing the relationships of water quality parameters to suspended solids for the Burdekin and Tully rivers. Figure. Estimation of temporal trends in water quality parameters for the Tully river over the period 987. Figure 3. Estimation of flow effects on water quality parameters for the Tully river over the period 987. Figure. Naïve estimation of temporal trends in water quality parameters for the Tully river over the period 987. Figure 5. Estimation of temporal trends in water quality parameters for the Burdekin river over the period 988. Figure 6. Estimation of flow effects on water quality parameters for the Burdekin river over the period 988. Figure 7. Naïve estimation of temporal trends in water quality parameters for the Burdekin river over the period 988. :River Water Quality Monitoring p.9