Hampson, D.I., Crowther, J., Bateman, I.J., Kay, D., Posen, P., Stapleton, C.M., Wyer, M.D., Fezzi, C., Jones, P. and Tzanopoulos, J. (2010). “Predicting Microbial Pollution Concentrations in UK Rivers in Response to Land Use Change”, presented at the International Water Association Water Research Conference, Marriott Lisbon, Portugal, 11-14 April 2010.
The Water Framework Directive has caused a paradigm shift towards the integrated management of recreational water quality through the development of drainage basin wide programmes of measures. This has increased the need for a cost-effective diagnostic tool capable of accurately predicting riverine faecal indicator organism (FIO) concentrations. This presentation outlines the potential applications of models developed to fulfil this need.
Predicting Microbial Pollution Concentrations in UK Rivers in Response to Land Use Change
1. Predicting Microbial Pollution
Concentrations in UK Rivers in Response
to Land Use Change
Danyel Hampsona
, John Crowtherb
, Ian Batemana
, David Kayc
,
Paulette Posena
, Carl Stapletonc
, Mark Wyerc
, Carlo Fezzia
,
Philip Jonesd
and Joseph Tzanopoulose
a
School of Environmental Sciences, University of East Anglia
b
Centre for Research into Environment and Health, University of Wales
C
River Basin Dynamics and Hydrology Research Group, University of Aberystwyth
d
Centre for Agricultural Strategy, University of Reading
e
Centre for Agri-Environment Research, University of Reading
2. Presentation outline
• Introduce the problem and the research imperative
• Describe the faecal indicator organism (FIO) models
underpinning this research
• Examine catchment and subcatchment scale applications
• Indicate land use policy relevance and future research
3. Policy drivers
Microbial pollution remediation central to the Water Framework Directive (WFD)
Bathing Water Directive and Shellfish Water Directive contain
strict microbiological standards
Land use planning guided by research
Increased need for a modelling tool
4. Objective – create models to predict
microbial pollution
• Empirical foundation
• Capture high river discharges
• Transferrable
• Use nationally available data
• Incorporate human and livestock
population variables
5. Data used in the meta-analysis
This research remodels 15 catchment
studies conducted by Centre for
Research into Environment and
Health (CREH) to predict:
• GM E. coli (cfu 100ml-1
)
• GM enterococci (cfu 100ml-1
)
CREH Catchment
study sites
6. Data used in the meta-analysis
Agricultural census data:
Density of twelve livestock types per km2
per hydrological response unit (HRU)
Catchment
Year
Sampled
Agricultural census
year used for
livestock enumeration
Census year used
for population
enumeration
Clacton 1998 1997
2001 used
throughout
River Ribble 2002 2003
Staithes Beck 1995 1995
Lake Windermere 1999 1997
River Leven/Crake 2005 2004
Sandyhills 2004 2004
Brighouse Bay 2004 2004
Troon coastal inputs
2000 2000
Killoch Burn 2004 2004
River Irvine/Garnock 1998 1997
Ettrick Bay 2004 2004
River Nairn 2004 2004
Afon Ogwr 1997 1997
Afon Nyfer 1996 1996
Afon Rheidol/Ystwyth 1999 1997
Humber
Predicted
2004
2004 2001
7. Data used in the meta-analysis
Human census data:
Human density per km2
per HRU
Catchment
Year
Sampled
Agricultural census
year used for
livestock enumeration
Census year used
for population
enumeration
Clacton 1998 1997
2001 used
throughout
River Ribble 2002 2003
Staithes Beck 1995 1995
Lake Windermere 1999 1997
River Leven/Crake 2005 2004
Sandyhills 2004 2004
Brighouse Bay 2004 2004
Troon coastal inputs
2000 2000
Killoch Burn 2004 2004
River Irvine/Garnock 1998 1997
Ettrick Bay 2004 2004
River Nairn 2004 2004
Afon Ogwr 1997 1997
Afon Nyfer 1996 1996
Afon Rheidol/Ystwyth 1999 1997
Humber
Predicted
2004
2004 2001
8. Overview of the FIO models
The FIO models can be written as:
y = a + b1x1 + b2x2 + ….. + bixi + e
Where:
a is the intercept (y at x = 0),
b is the slope (change in y per unit change in x),
and e is the random error term
10. Overview of the FIO models
• High flow concentrations are an order of magnitude higher than base flow
• E. Coli concentrations are an order of magnitude higher than enterococci
** = strong significance at
p<0.01
* = significance at p<0.05
Model type Intercept Primary coefficient Secondary coefficient Tertiary coefficient R2
(Adj)
GM E. Coli (Log10 cfu/100ml)
High-flow + ** + Log10Dairy/km2
** + Log10Human/km2
** + Sheep/km2
** 0.622
Base-Flow + ** + Log10Human/km2
** + Log10Dairy/km2
** - 0.418
GM Enterococci (Log10 cfu/100ml)
High-flow + ** + Log10Dairy/km2
** + Log10Human/km2
** + Sheep/km2
** 0.624
Base-flow + ** + Log10Human/km2
** + Log10Dairy/km2
* - 0.311
11. Overview of the FIO models
• At high flow livestock dominate
• At base flow human sources become dominant
** = strong significance at
p<0.01
* = significance at p<0.05
Model type Intercept Primary coefficient Secondary coefficient Tertiary coefficient R2
(Adj)
GM E. Coli (Log10 cfu/100ml)
High-flow + ** + Log10Dairy/km2
** + Log10Human/km2
** + Sheep/km2
** 0.622
Base-Flow + ** + Log10Human/km2
** + Log10Dairy/km2
** - 0.418
GM Enterococci (Log10 cfu/100ml)
High-flow + ** + Log10Dairy/km2
** + Log10Human/km2
** + Sheep/km2
** 0.624
Base-flow + ** + Log10Human/km2
** + Log10Dairy/km2
* - 0.311
12. Transferability
• Confirmed via out-of-sample testing (Tashman, 2000)
Study catchment tested
Mean error
(log10 cfu 100 ml 1‑
)
Mean absolute error
(log10 cfu 100 ml 1‑
)
Holland Brook 0.4975 0.4975
River Ribble -0.1883 0.2985
River Leven/Crake -0.0513 0.2215
River Irvine/Garnock 0.6227 0.6227
River Nairn -0.2126 0.3059
Afon Ogwr 0.0417 0.2116
Afon Rheidol/Ystwyth 0.4609 0.4912
Mean 0.1672 0.3784
Inter-study transfer errors in high-flow enterococci model (from Crowther et al . 2010)
19. Introducing the scenarios
• Fiscal constraint: taxing fertilizer by £50 per tonne
• Area intervention: designating the Humber RBD as an
Environmentally Sensitive Area (ESA)
• Cost intervention: raising the price of the EU milk lease quota by £20
• Production constraint: reducing dairy cattle stocking rates by 20%
• Input constraint: reducing fertilizer application by 20%
• Demand side constraint: adopting a nutrition driven food policy
• Micro-level land use management: stream bank fencing
Econometric and linear programming land use models used to generate land
management scenario data
(Fezzi and Bateman, 2009; Jones and Tranter, 2008)
24. Farm best management practices
• Stream bank fencing
• 95% reduction in pathogens when 2.13m separate cattle from watercourses
(Larsen et al. 1994)
27. The effectiveness of the remediation
strategies in the Aire subcatchment
Note: Nutrition driven food policy results excluded from this table because, being generated from a different
dataset, comparison is misleading
Remediation
measure
Predicted E. coli
concentration
(cfu 100ml-1
) exiting
dairy region
Percent reduction
from baseline
concentrations
exiting dairy region
Predicted E. coli
concentration
(cfu 100ml-1
) at
subcatchment exit
Percent reduction
from baseline
concentrations at
subcatchment
exit
Baseline
prediction, high
flow, 2004
92776 - 118169 -
Taxing fertilizer by
£50/tonne
91236 1.66 116514 1.4
Increase milk
quota cost by £20
87450 5.74 100420 5.02
Reduce fertilizer
application by 20%
83537 9.96 109161 7.62
ESA designation in
Aire
84435 8.99 108006 8.6
Reduce dairy
stocking by 20%
82035 11.58 104893 11.23
Installation of
stream bank
fencing
38418 58.59 77176 34.69
28. Policy relevance of this research
• Cost-effective diagnostic tool
• Inform integrated catchment management
• Insight into cost-effective remediation strategies
• Can be used to aid source apportionment.
29. Future work
• Impact of climate change
• Non-market benefit valuation of reduced microbial
pollution
• Cost-benefit analysis of remediation strategies
30. Contact Details
Danyel Hampson
Research Office 1.15
Zuckerman Institute for Connective Environmental
Research
School of Environmental Science
University of East Anglia
Norwich, UK
NR4 7TJ
D.Hampson@uea.ac.uk
+44 (0)1603 591545
Skype: danyel.hampson
Editor's Notes
I’m Danyel Hampson, and I’d like to present research, undertaken by economists, geographers, water quality modellers, and agricultural strategists, based at several, UK, institutions. This research has been funded, primarily, by the Economic and social research council, the Catchment hydrology, resources, economics and management project, and the Rural Economy and Land Use Programme. Please excuse the script. Without it, I’ll go off at a tangent, and forget what I need to say. I’m not going to give a highly technical presentation, because my work is primarily social science, designed to inform UK land use policy.
Here’s the structure of the presentation. I’ll introduce the problem, and the research imperative, describe the FIO models, underpinning this research, examine catchment and subcatchment scale applications, and indicate land use policy relevance, and future research.
Contaminated bathing, and shellfish waters, are risks to human health. Microbial pollution remediation, is central to the Water Framework Directive strategy, for water quality improvements. The revised Bathing Water Directive, and Shellfish Water Directive, contain microbiological standards, using enterococci, and E. Coli, as surrogates for infection risk. Land use planning, must be guided by research, to identify the optimal locations, for water framework directive induced, environmental improvements. To achieve compliance with the WFD, significant reductions in diffuse agricultural pollution, and improvements to waste water treatment facilities, are required. The WFD requires drainage basin wide programmes of measures, so there is an increased need, for a modelling tool, to predict riverine FIO concentrations.
We assembled a solid empirical dataset, to be able to model high flow rates, as these are the critical periods of pollution discharge. We wanted to produce transferrable models, to predict pollution concentrations, in areas where there is limited, or no empirical FIO data. We then used nationally available population data, to apply the models, at a range of spatial scales. We believe, that these are the first, transferrable, population based models, produced to model FIOs in the UK.
We remodelled the FIO data, from 15 catchment studies, collected over a ten year period, within a meta-analysis, to predict geometric mean e. coli, and enterococci concentrations.
We populated our model with explanatory variables. Livestock populations, tend to fluctuate over short periods, so we used the data from the agricultural census, which most closely corresponded, to the year, of the catchment studies.
Human population data, was drawn from the 2001 census. This is the approximate, mid point of the study period.
Multiple regression techniques, using a stepwise selection procedure, were used to model the relationships, between geometric mean FIO concentrations at base and high-flow, the dependent variables, y, and the various independent variables, x.
We tried to incorporate many variables, that are known to affect FIO transfer, and survival. We experimented with different soil types, livestock pathogen shedding rates, altitude, slope, rainfall, and soil temperature. But none entered the models, as significant variables, in their own right.
We found three highly significant sources of FIOs in rivers. Dairy cattle, humans and sheep.
We’ve built parsimonious models, with high levels of explained variance, because FIO concentrations depend, on a huge range, of complex determinants.
The performance of these population based models, is consistent with previous research, using land use variables. Predicted high flow concentrations, are an order of magnitude higher, than base flow concentrations, and E. Coli concentrations, are an order of magnitude higher than enterococci. All coefficients are positive, and all but one, are highly significant.
At high flow, dairy is the dominant source of FIOs, because large quantities of manure, and slurry, are flushed from farm hard standings, and fields, into watercourses. Sheep enter the high flow models, allowing us to differentiate, between different livestock types, within our modelling.
At base flow, human sources become dominant. This is possibly because of constant background emissions from wastewater treatment plants. In the enterococci models, human sources play a more dominant role, reflecting the fact that enterococci, is primarily human specific.
A programme of out-of-sample testing, was undertaken on the high flow, enterococci model, to evaluate the extent to which, the models are transferable to other UK catchments. This model, was re-run seven times, with data for one of these seven catchments, omitted in turn. The resulting model, was then used to predict enterococci concentrations, for subcatchments in the omitted catchments, and the mean error, and mean absolute error, were calculated for each study catchment.
Although we found some inter-catchment variability, the strength of our models, lies in the fact that they are based on a FIO database, that has extensive geographical coverage. By combining data, from 15 catchment studies, the effects of temporal, and inter-catchment errors, are reduced. The resulting models can be used, with some confidence, to predict FIO concentrations during the summer bathing season, in unmonitored, UK watercourses.
We transferred to the Humber river basin, in the east of the UK, for two reasons: It’s the site of the ChREAM project, which enabled us to share data, and model similar land use scenarios, and because the Humber basin, which drains 28% of the land surface of England, has a wide range of farming activities, and human population densities.
This section of the presentation, explores the results of our catchment scale modelling.
This map shows the results of the transfer exercise, applying the high flow E. coli model. The distributions of E. coli, and enterococci, predicted by our models are similar, as are low flow distributions, but, these, are obviously at lower concentrations. From this point onwards, I’ll focus on the high flow E. coli results.
The lowest concentrations, are predicted in areas of low livestock density, and low human population density, for example, the Yorkshire Dales, and the North York Moors.
High levels of pollution, are found in areas of high human, and high livestock, density. The highest levels of E. coli, are found in the Upper Trent, due to a combination of high human density, and very high, dairy cattle density.
Both indicators are predicted to greatly exceed EU guidelines, at both flow rates. It’s widely acknowledged, that UK rivers are highly polluted. There are only 11 designated, inland bathing areas, within the UK, none of which, are in Humber.
It’s important to consider, integrated basin wide remediation measures, to reduce FIO discharges, to coastal receiving waters. In addition, we hypothesize, that there may be, a large latent demand for riverine recreation, which cannot be fully realised, because rivers are highly contaminated.
We analysed the differential effects, of seven land use management, and policy instruments, all of which can be used, or have been considered for use, by defra, as a means to reduce livestocking rates. As Dave Kay said yesterday, reducing population, reduces river pollution.
We acknowledge the ESA scheme, has been replaced by the Environmental Stewardship Scheme. The focus here, is on the use of the ESA concept, as an area intervention. Similarly, milk quotas may be phased out by 2013. We use milk quota, purely, as an example of a cost intervention.
The land use models we used, are described within the full conference paper, not here. For two reasons. This is a water conference, and I’m focussing on the water, quality, modelling, and there isn’t time to cover them now, or to cover the scenarios in detail, but here are some interesting points:
The econometric land use model, underpinning the fertilizer tax scenario, predicts that an increase in fertilizer price, encourages conversion to low biomass yield land uses, such as rough grazing. However, increasing fertilizer tax, is predicted to achieve very small reductions in E. Coli concentrations. Averaging 1.9% across Humber, possibly, because dairy farmers use livestock waste, to fertilize their grassland.
Increasing the price of the milk quota is more effective, with mean reductions of E. coli, of 6.4% across Humber. Interestingly, areas of high dairy density in the Aire, Dove, and Upper Trent subcatchments, do not see high reductions, for a measure designed specifically, to put pressure on milk production. The larger producers in these areas, enjoy greater efficiencies of scale, and are more likely to weather the increased transaction costs, associated with a milk quota price rise.
While not strictly a measure designed to reduce diffuse pollution, the adoption, by the UK population, of a healthier diet, would see large reductions in dairy cattle, and sheep populations, as milk, mutton, and lamb consumption, decrease. The largest reductions to E. Coli, are predicted to occur in the upland areas, to the west of Humber. Dairy numbers, and consequent FIO concentrations, are typically maintained, or, are predicted to rise slightly, in lowland areas, in the east, as production is transferred, to more economically viable areas.
I’ll now discuss subcatchment scale modelling, and targeted improvements, via farm best management practices in the Aire subcatchment.
Farmers can minimise pollution, by adopting farm scale measures. Stream bank fencing prevents cattle defecating directly into watercourses, and it’s a potentially cost effective remediation strategy. Larsen found a 95% reduction in pathogens, when a little over 2 metres separated cattle, from watercourses. Funding is already available in the UK, for this measure. Fencing for buffer strips, is an eligible expense, in the Catchment Sensitive Farming Capital Grant Scheme.
The Aire subcatchment contains key features of interest. In the northwest, the catchment has low human population, and low intensity livestocking. The river then passes, through an area of intensive dairying, before flowing through high density urban areas, and exiting the subcatchment in the east.
With stream bank fencing erected, within the intensive dairy regions, we see marked reductions in E. coli concentrations, downstream of the environmental improvements.
The majority of the results of our scenario modelling, show trivial improvements in river water quality, for measures which could potentially result in large reductions in farm income. Some of the more aggressive macro policies, for example, raising the price of milk quota, or designating Aire as an ESA, may result in the concentration of milk production, into large enterprises, exacerbating localised microbial pollution concentrations.
Stream bank fencing greatly exceeds the reductions achieved by the other policy instruments, and, although there is still uncertainty surrounding its effectiveness, stream bank fencing may be, a low-cost, high-yield, remediation strategy.
This research provides a cost-effective diagnostic tool, capable of identifying and predicting microbial pollution sources, and distributions. It can inform integrated catchment management programmes, as required by the water framework directive, and offers insights, into the optimal, cost-effective mix of catchment scale, and farm scale remediation strategies. By incorporating human, and livestock sources, as explanatory variables, these models can be used to help apportion the responsibility, for microbial pollution, between the water industry, and the agricultural sector, and aid real-time prediction of health risk.
Climate change will affect land use. This in turn, will redistribute microbial pollution sources. We have already begun preliminary work, to model the effects of climate change on land use in the UK, and we will model the resulting FIO distributions, accordingly.
Water Framework Directive compliance, need not incur, excessive, or disproportionate, implementation costs. We will conduct research, into the non-market benefits of cleaner rivers, and undertake economic modelling, to produce a cost benefit analysis, of the different remediation strategies introduced here.
I hope you found this presentation useful. Please contact me if you’d like more information. Are there any questions?