Methods for Estimating Impacts of
Rainfall on Bathing Water Quality Report on results for Saltcoats,
Irvine, Fleetwood and Paignton
UKLQ07
August 2008
EXECUTIVE SUMMARY
Introduction
Rainfall is acknowledged as having a primary influence in causing
episodes of high faecal coliform concentration in bathing water, a
principal indicator of poor water quality. This pollution occurs
through two key pathways; increasing run-off from agricultural land and
from combined sewer overflows (CSOs) spilling during times of heavy
rainfall. This occurs throughout the UK, but in particular areas that
experience higher rainfall, such as the west coast of Scotland,
northwest England and Northern Ireland.
A revised Bathing Water Directive (2006/7/EC, repealing current
Directive 76/160/EEC), came into force on 24 March 2006. With
it comes new responsibilities from 2012 to provide bathers with
information regarding potential risks to their health,
including advance warnings of pollution incidents where
practicable. As a result, interest is increasing throughout
the UK in the development of techniques to predict poor bathing water
quality to enable information to be provided to the public and
therefore enhancing human health.
As a means of managing bathing beaches that are at risk of failing to
meet the current EU Bathing Water Directive mandatory water quality
standards, and in anticipation of the revised Directive, SEPA piloted a
Scottish Government project to predict bathing water quality based on
prior rainfall. These predictions are then displayed at electronic
messaging signs at selected sites. Six sites in southwest Scotland were
targeted in the 2003 bathing water season and ten from 2004 including
two outwith the southwest area.
SNIFFER on behalf of SEPA and NIEA initiated this rain radar project on
the premise that:
- Existing rain-gauge prediction methods could potentially be
improved by using rainfall radar data as an alternative or in addition
to rain and river gauge data.
- The revised Directive includes tighter faecal coliform
standards, requiring increased accuracy of predictions.
- Using rain radar may be more practical than installing
telemetered rain and river gauges at new sites.
- Using rain radar may be better for obtaining midday updates
and projections than rain gauge networks
The aim of this study is to examine whether radar rainfall data can
improve on rain-gauge methods of prediction of faecal coliform
exceedances in bathing waters based on rain-gauge data. This research
project is not aiming to compare and contrast radar and rain-gauge data
directly, but to examine whether radar-derived thresholds could improve
the prediction of faecal coliform concentration exceedances, which
indicate poor bathing water quality.
Halcrow Group Limited was appointed by SNIFFER on 8 June 2007 to
undertake this study. The original contract was to analyse two sites on
the southwest coast of Scotland, Saltcoats and Irvine. Subsequently,
this contract was extended to analyse two further sites in England,
Fleetwood in the northwest and Paignton-Preston Sands in the southwest.
This report is the culmination of the study and encompasses the
findings of the complete project covering all four sites.
Methodology
Met Office radar data were obtained from three different radars to
cover the four study areas. The radar data were provided in the format
of rainfall intensities at a temporal resolution of 5 minutes and a
spatial resolution, depending on site, of either one or two kilometres
squared. A GIS software utility was developed for the project to
construct the time-series of instantaneous rainfall values for each
event and each catchment. This was used to evaluate the average
rainfall over specified areas at a 5-minute intervals during multiple
events. The end output was then a calculation of average rainfall
across a number of grid cells within the areas specified in a shape
file. Using this function, grid squares with centres located within the
boundaries of a polygon area are included in the calculations.
The radar data were processed to derive rainfall averages over a
variety of potential pollution source areas. In total, 11 pollution
source areas were analysed for the 4 sites.
The rainfall depth data were distributed into different durations and
plotted against faecal coliform count to identify thresholds. The
thresholds of rainfall for a given duration were determined by fitting
a best-fit line through the data. Where the best fit line crosses the
500 cfu/100 ml faecal coliform (FC) line a threshold rainfall depth is
defined. This is referred to as an ‘optimising’
approach, which aims to achieve the maximum number of predictions of
FC>500 and minimise ‘false alarms’ i.e.
incorrect predictions of FC>500. This method differs from the
method employed by SEPA using rain-gauge data, referred to as a
‘precautionary approach’ in which the principal aim
is to correctly predict FC>500 events.
Revisions to the methodology made during the project included:
- Improvements in the ascribed method of deriving the
rainfall threshold from the data
- Inclusion of a short duration to capture peak rainfall
intensity and examine the influence this has on prediction. Hence for
Paignton, where it was suspected that peak intensities might play a
greater role due to the type of rainfall experienced, a 1-hour duration
was incorporated as a ‘sliding’ duration
– i.e. the peak 1-hour intensity in a given time period such
as 24-hours.
Results
In order to best compare the rain-gauge results with the radar results
a like-for-like comparison was made by comparing the data from both
sources using the thresholds determined by the radar data (optimising)
method. An overview of results is shown in the table below:
Site |
Data
Source |
Number of correct
predictions
(rain gauge predictions use radar derived thresholds) |
Percentage
correct (%) |
Saltcoats |
Ashgrove
rain gauge
Saltcoats Rural Radar
Saltcoats Urban Radar
Saltcoats Radar aggregated |
13/23
16/23
15/23
15/23 |
57%
70%
65%
65% |
Irvine |
Rain
gauges (aggregated)
Irvine Urban Radar
Kilmarnock Urban Area
Irvine Rural Area
Irvine Radar aggregated |
17/26
9/26
9/26
12/26
10/26 |
74%
35%
35%
46%
38% |
Fleetwood |
Rain
gauges (aggregated)
Wyre Urban Radar
Wyre Rural Radar
Lune Rural Radar
Fleetwood radar aggregated |
25/40
23/40
27/40
22/40
27/40 |
63%
58%
68%
68%
68% |
Paignton |
Rain
gauges (aggregated)
Torbay Radar
Okham Radar
Paignton radar aggregated |
6/11
9/11
9/11
9/11 |
55%
82%
82%
82% |
We
wish to emphasise that the method employed in this project is
relatively simplistic and purposefully aims to determine whether a
‘black box’ method in which rainfall is the only
input and faecal coliform exceedance the only output is adequate for
poor bathing water quality predictions. The results presented in this
report can be improved by the use of more event data, including for
example river flow, tide and wind and, therefore, opportunities exist
to improve the results further in the future.
An analysis of radar data values against rain-gauge data for the Irvine
urban area in which there is rain-gauge showed that the two measurement
techniques gave similar results but that radar consistently recorded
slightly higher rainfall than the Irvine rain-gauge. This finding is
useful as we can state with confidence that radar is measuring rainfall
quantities well and may be recording some events that the rain-gauge is
missing.
Conclusions
Overall this study concludes that the rain radar performs well,
particularly for smaller catchments in which it is shown to improve on
the SEPA and Environment Agency methods using rain-gauges. The radar
performs slightly less well for larger catchments. This is likely to be
due to factors including peak intensities in radar data being
‘smoothed’ due to spatial averaging and additional
complexities in catchment processes affecting the results rather than
the quality of the radar data.
The fact that radar data has been shown to be at least as good as
rain-gauges in predicting exceedances of faecal coliform concentrations
means that a radar-based system could operate where no rain-gauges
exist and may be preferable for cost and practicality reasons (a cost
benefit analysis is a recommendation of the project). Furthermore, the
ability to use forecast rainfall products from the Met Office Nimrod
system up to 6-hours ahead mean that increased lead time can be
achieved, provided the forecast quantities are reasonable predictions
of actual quantities.
A further benefit of radar is that it can measure localised, convective
rainfall which may occur in individual events or within frontal systems
as ‘embedded convection’. At sites at which
localised, high intensity rainfall is known to result in FC
exceedances, the ability to analyse the peak intensity using radar, and
have data available throughout the day, is an added advantage.
Other key conclusions from the study are as follows:
- Where radar data can excel is during highly localised,
convective rainfall events (for example, thunderstorms), or frontal
events with embedded convective rainfall, which are missed by
rain-gauges unless these are installed in a very dense network. At the
Paignton site there was evidence that radar had measured a localised
event that rain-gauges had missed.
- For the Scottish sites, it appeared that that the rainfall
events were predominantly widespread frontal systems crossing the
western coast of the UK from the south-west or west resulting in
relatively uniform rainfall, with the exception of orographic
enhancement (rainfall totals increasing with ground altitude). For this
reason the radar data did not manage to explain the rain-gauge
method’s ‘misses’ at the Scottish sites.
Similarly, the results at Fleetwood from the like-for-like comparison
do not indicate that radar data have improved in identifying localised
intense rainfall missed by rain-gauges.
- Spatial averaging of radar rainfall data, where the radar
pixels are averaged over polygon pollution-source areas, may result in
‘smoothing’ of localised intense rainfall events
(for example, from convective rainfall). Spatial
‘smoothing’ is likely to be particularly
significant for the larger pollution-source areas of Irvine rural, Wyre
urban and rural and Lune rural. In general, the rain radar time-series
for each of the pollution-source areas shown in Appendix 3, 4, 5 and 6
are fairly similar for each of the pollution-source areas.
- Once the pollution-source area radar time series has been
obtained this is accumulated into rain radar totals for different
durations. This is effectively ‘temporal
smoothing’. This ‘temporal smoothing’
results in a further loss of resolution in the data as the rainfall
peaks shown in the radar time-series (Appendix 3, 4, 5 and 6) are
accumulated into rain radar total depths for different durations. In an
effort to assess the impact of this at the Paignton site, where peak
intensities are likely to be influential in predicting FC levels, peak
1-hour rainfall was selected for analysis, having a sliding duration
over a 24-hour period. In this case, no exceedances resulted from the
hourly maxima that did not result in an exceedance of a longer
duration, though up to 13mm in an hour were measured, indicating that
this duration could be useful with a greater number of events to
analyse.
Recommendations
Recommendations for improving the results and for further study are
made at the end of this report. They are detailed under the following
headings:
- Number of events analysed
- Use of further study areas
- Radar data –
spatial averaging
- Radar data –
fixed or ‘sliding’ duration
- Radar quality
- Forecast rainfall
- ‘Optimising’ versus
‘Precautionary’ approach
- Data variability (Noise):
- Antecedent wetness
conditions;
- Water quality samples
– spatial variation;
- Water quality samples
– temporal resolution;
- Water quality
–
relationship with river flows
- Cost-benefit of
implementing a radar-based operational system
Copies of this report are available from the Foundation, in electronic
format on CDRom at £20.00 + VAT or hard copy at
£50.00, less 20% to FWR members.
N.B.
The report is available for download from the SNIFFER Website