close

Вход

Забыли?

вход по аккаунту

?

A method for the automated extraction of environmental variables to help the classification of rivers in Britain.

код для вставкиСкачать
AQUATIC CONSERVATION: MARINE AND FRESHWATER ECOSYSTEMS
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
Published online in Wiley InterScience
(www.interscience.wiley.com). DOI: 10.1002/aqc.534
A method for the automated extraction of environmental
variables to help the classification of rivers in Britain
F.H. DAWSON*, D.D. HORNBY and J. HILTON
NERC-CEH Dorset, Winfrith Technology Centre, Dorchester, Dorset DT2 8ZD, UK
ABSTRACT
1. A number of systems for assessing river quality or for classifying rivers use data obtained from
maps as a means of linking a given stretch of river with a reference site. The embodiment of the
reference-state principle within the EC Water Framework Directive means that this trend will
accelerate.
2. Historically, data such as the distance to source of river, altitude at source, altitude at site, and
local slope have been derived from maps by hand, a process which is time consuming and prone to
both random and systematic errors.
3. An automated computer-based extraction procedure has been developed on a GIS system to
improve the speed and repeatability of such data collection. This paper describes the processes
required to produce reliable data, to ensure quality control and to obtain optimal speed of data
extraction.
4. The system has several applications including determination of catchment areas for the EC
Water Framework Directive and delineation of reaches to support the development of physical
quality objectives for rivers in England and Wales.
Copyright # 2002 John Wiley & Sons, Ltd.
KEY WORDS:
GIS; river habitats; map-based data; environmental variables
INTRODUCTION
Several systems for assessing and classifying river quality in Britain use physical data from maps as a basis
for comparison and as a means of linking a given stretch of river with a reference site (Raven et al., 1997;
Wright et al., 2000). The embodiment of the reference-state principle within the EC Water Framework
Directive means that this trend will accelerate. Historically, data such as the distance of river to source,
altitude at source, altitude at site and local slope, have been derived from maps by hand, a process which is
time consuming and prone to both random and systematic errors. An automated extraction procedure has
been developed on a GIS system to improve the speed and repeatability of such data collection. This paper
describes the processes required to ensure quality control and obtain optimal speed of data extraction.
*Correspondence to: Dr. F.H. Dawson, NERC-CEH Dorset, Winfrith Technology Centre, Dorchester, Dorset DT2 8ZD, UK.
E-mail: hugh.dawson@ceh.ac.uk
Copyright # 2002 John Wiley & Sons, Ltd.
Received 23 February 2002
Accepted 4 May 2002
392
F.H. DAWSON ET AL.
The system has been developed to underpin the River Habitat Survey (RHS) method (Raven et al., 1997).
It has also been used to supply data for the River InVertebrate Prediction And Classification System
(RIVPACS) (Hornby et al., 2002), for ecological studies on aquatic plant communities in Britain (Dawson
and Szoszkiewicz, 2000) and for the initial development phase for deriving physical quality objectives
(PQOs) for rivers in England and Wales (Walker et al., 2002). These initiatives, in turn, will help the UK to
fulfil requirements under the EC Water Framework Directive (European Commission, 2000). This
Directive requires that all surface waters, including rivers, are classified into types based on various physical
parameters such as altitude, catchment size, and geology and their hydromorphological state assessed in
terms of deviation from ‘undisturbed conditions’.
There are several advantages of obtaining data automatically from maps. Straightforward variables such
as altitude, position and geology, but also more sophisticated ones such as distance from source, stream
energy, discharge category and types of land-use upstream in the catchment, can be derived or calculated.
Other data such as valley shape could be calculated by more complex routines from contours or Digital
Terrain Model data. It may also be possible to estimate acid-neutralizing capacity of the water using the
proportions of calcium- or magnesium-bearing rocks in the catchment when comparisons are completed
with ‘ground-truth’ data. Other important environmental factors influencing aquatic biological
communities could be derived from existing spatial equations which include air temperature and
precipitation data.
The RHS system is a good example of the importance of map-derived information for a river
classification scheme. RHS is a survey technique combining field sampling and map-derived data (Raven
et al., 1998a). For each 500 m long survey unit, information on channel substrates, flow features,
morphological and vegetation structure on the banks, land use in the river corridor and artificial
modifications is collected by means of 10 ‘spot-check’ transects and an overview or ‘sweep-up’ summary.
Assessment of habitat quality, which includes the diversity of physical features, is based on a ‘nearest
neighbour’ approach using a UK baseline database of 5348 geographically representative sites derived from
stratified random sampling. A dynamic typology is expressed using a principal component analysis
ordination based primarily on slope, altitude, distance from source and height of source (Jeffers, 1998).
These data are derived from maps. The underlying principle and presumptions are that sites on similar
watercourses will have broadly similar habitat characteristics and comparison of these could provide a basis
for assessing habitat quality and predicting habitat features. Traditionally, map-derived and environmental
data have been obtained manually, a time-consuming exercise with inherent random and systematic errors.
The main objective of this study was to improve the speed and accuracy of data capture by deriving an
automated system. In doing so, it was necessary to compare manual and automated extraction of spatial
environmental variables, test the data for reliability, and determine potential applications.
METHOD
The approach taken for automating the extraction of spatial environmental data for rivers included: linking
the digitized river data to form a network connected from the sources to the mouth of each river; correcting
and pre-processing data to optimize speed of extraction; comparing the reliability and replicability of
automated data extraction with manual methods; and determining and extending the uses by integrating
different spatial datasets.
Linking the network
All the rivers in Great Britain marked on Ordnance Survey 1:50 000 scale maps were digitized by the
Institute of Hydrology, Wallingford, UK, now the Centre for Ecology and Hydrology. Instream water
Copyright # 2002 John Wiley & Sons, Ltd.
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
AUTOMATED EXTRACTION OF RIVER VARIABLES
393
bodies, e.g. lakes or lochs, were replaced by a centre-line during digitization to allow full connectivity along
the river channels (Moore et al., 1994). These data were incorporated into a geographical information
system (GIS) layer in a Windows program, ArcView. The network was then formed from ‘vector’ data –
essentially a series of points, each with a location, direction and length, and joined in series to form a noncrossing line (‘arc’) with a designated start and end point, flow direction and a unique number. Series of
these lines were normally linked from each river mouth to all of its river sources. Programming routines
(‘scripts’) were written in the ArcView language, ‘Avenue’. Algorithms or program module scripts
developed in ArcView to carry out these operations are available on www.ceh.ac.uk/arcscripts and on the
ESRI ARC-VIEW site http://gis.esri.com/arcscripts/index.cfm.
Before the availability of fast electronic database programs, river catchments in Britain were grouped, for
convenience, into 105 ‘hydrometric areas’. The digital data for the national river network were sub-divided
into either hydrometric area groups for mainland rivers or into individual island groups, to speed
processing. The original boundaries were derived from lower resolution maps prior to the availability of
GIS systems and they do not correspond exactly to actual river catchment areas. As a result, some parts of
small streams in the upper catchment were associated with adjacent catchments when spatial data on rivers
were separated into individual hydrometric areas. These ‘arcs’ needed to be reinstated and a script was
written to re-associate them if more than half of their length lay within the correct hydrometric area. A
similar correction was needed for river mouths in estuaries when they did not lie within the digital coastline.
Since ArcInfo ‘generalization’ routines can cause general data disruption within the data layer each river
course was initially tested by a tracing routine until an interruption was detected, reviewed and then
individually connected by the shortest logical route using an inbuilt ArcInfo ‘snapping’ routine (Morris,
personal communication). Although some of these breaks resulted from digitization errors, others were
present on the original maps. For example, when a river flows under a road bridge the ‘blue line’
representing the river on the map will be interrupted. Similarly, some features may be misplaced or omitted
on a map to avoid cluttering. Recent editing of kinks, crossovers and other misplaced arcs resulting from
slight unintentional hand movements during digitization and lines were again joined by the shortest logical
route. Reaches that had been double-digitized, in error, were also corrected by removal of one of the arcs.
Separations had to be introduced between, for example, canals and rivers which had been joined in the
digitization but were, in fact, separate. The arcs in the corrected vector dataset were only allowed to join at
river confluences (nodes), each of which was given a unique number. The arcs were linked in ‘routes’ from
each individual river mouth to all of its river sources by tracing and labelling nodes. During this process a
number of incorrect connections of nodes to arcs, which had been introduced during automated ArcView
processing, were identified and corrected.
Pre-processing to optimize data extraction speed
The river layer was pre-processed to increase the speed of data extraction. River mouths and sources were
linked in ‘routes’ of shortest distance from each individual river mouth to all of its tributary sources. Each
node was numbered sequentially and labelled with data on the river mouth and the most distant source.
Breaks had to be temporarily introduced into the river network where catchments were linked, such as at
‘saddles’ in upper areas, or by man-made channels. This was needed because such linked connections
shared all mouths and sources and could not be sensibly processed. The breaks were reconnected after
processing. Generating shortest routes meant that nodes within braided channels were missed, and these
had to be manually reinstated into the network with the correct source although this problem occurred in
50.1% of the network. Drainage channel grids, such as the Somerset Levels, were marked as ‘not
resolvable’ and treated separately (Figure 1). Each hydrometric area was checked to ensure full connectivity
and the ability to trace both upstream and downstream.
Copyright # 2002 John Wiley & Sons, Ltd.
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
394
F.H. DAWSON ET AL.
Figure 1. Examples of catchment areas with un-resolvable drainage networks: (a) a network of artificial channels in the Great Ouse
catchment, lowland eastern England and (b) a circular pattern near Telford, Shropshire, West England, showing apparent sources
which are within networks but which are most probably ends of artificial channels.
Additional datasets
A number of additional sets of line (‘vector’) or areal grid data (‘raster’ rectangles or ‘tiles’) at a variety of
original scales have been added to increase the utility of the system. These include:
*
*
*
*
*
Altitude data for points along the course of the stream, supplied by CEH Wallingford. These were used
to determine the height of river sources and sample sites, together with their local site gradient. The data
were produced from a layer of arcs derived from the mean heights of the 50 50 m squares or ‘pixels’ of
the digital terrain dataset derived from interpolation of Ordnance Survey of Great Britain contour data
(Morris and Flavin, 1990, 1994).
A flow direction grid identifying the direction in which water falling on any particular 50 50 m square
would flow. The dataset was derived by CEH Wallingford from a Digital Terrain Model of Britain.
Solid and drift geology, supplied under licence from British Geological Survey (BGS) at 1:62 5000 scale.
Catchment land-cover from the 1990 ITE Land Cover Map, supplied under licence from CEH,
Monkswood (Fuller and Parsell, 1990; Fuller et al., 1994a, b; Wyatt et al., 1994).
The extent of flood defence works carried out between 1930 and 1980, to provide an historical
perspective of channel modification (Brookes et al., 1983).
In addition, several more datasets have been utilized manually prior to full integration, including:
*
*
*
*
*
General Quality Assessment classification of water quality of river reaches in England and Wales at
1:250 000 scale (Environment Agency, 1997).
River discharge categories of river reaches in England and Wales at 1:250 000 scale (Environment
Agency, 1997).
Extent of the Indicative (1 in 100 year) Floodplain Map for England and Wales.
Demographic data for UK.
Location of local and national areas of wildlife conservation interest including national parks.
Given the different scales used to generate the original data, techniques were developed to adjust the data
so that comparisons could be made automatically.
Copyright # 2002 John Wiley & Sons, Ltd.
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
AUTOMATED EXTRACTION OF RIVER VARIABLES
395
System operation
Semi-automated, menu-driven processing of data from the network in the GIS program is determined by a
choice of environmental variables displayed on screen (Appendix). A National Grid Reference (NGR)
number can be entered, or alternatively the relevant hydrometric area can be chosen from a display screen
of Britain, followed by more detailed location of a site chosen interactively from the river network for
individual hydrometric areas. The GIS display provides a numeric position of the chosen site location and
this initiates an automated search for the appropriate ‘arc’ from the nearest upstream node. A unique code
contains the source and mouth of the river from which an ArcView network function can find and generate
the route between them. Using this information, the slope and altitudes of the site and source can be
calculated, along with a selection of other attributes associated with the section (Figure 2).
Automated operation from a list of NGRs can also be initiated, with the derived data returned to a
database for later inspection and use. In automatic mode a GIS routine is used to find a single target river
within a preset proximity (3 screen pixels) to the given NGR. If more than one river or sections of the same
river are located in the target area, the preset proximity is progressively decreased from 1:120 000, 1:80 000
to 1:20 000, i.e. equivalent to decreasing the search area to circles of approximately 100, 60, and 15 m
diameter, until either a single river results from the search or, if two rivers are still options at the highest
resolution, a confluence is assumed to be close by and the longer river is chosen. If no river is located by this
search, the computation for this site ceases and the routine continues with the next NGR. Following the
#
R. Piddle
#
R. Frome
Figure 2. A 1:50 000 digitized river network of the rivers Frome and Piddle in southern England, showing the upstream course to the
river source and the catchment area created for a sample site. Based on Ordnance Survey1 1:50 000 mapping with the permission of
Ordnance Survey on behalf of The Controller of Her Majesty’s Stationery Office, # Crown copyright, NERC GD272191.
Copyright # 2002 John Wiley & Sons, Ltd.
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
396
F.H. DAWSON ET AL.
identification of each appropriate river, extraction of the chosen variables continues in semi-automatic
mode.
The upstream catchment boundary is calculated by an inbuilt ArcView routine, from the flow direction
grid, which allows additional variables to be generated, including: the upstream catchment area; the
percentage of the different types of deep geology in the catchment; the percentage cover of quaternary, drift
or sedimentary geology in the catchment; and the percentage of different types of land cover in the
catchment (Plates 1a,b). Data may also be included in equations to generate the required environmental
variables such as total stream length in the catchment, air or water temperature, or relative unit streampower (Walker et al., 2002).
Comparison of automated with manual extraction of map-based data
The altitude of source and site, distance from source and slope at site for each of the 5348 baseline sites of
the 1994–1996 RHS survey of Britain with most islands, were selected using a stratified random selection
procedure and compared with data manually extracted from maps three times by different people.
RESULTS
Comparison between manually and automatically derived data
Results of the automated extraction were repeatable and consistent with the limits of extraction accepted
for manual measurement of the four variables used in the test: for altitude of site, 98% were within 10 m;
for slope of site, 95% were within 25 m km1; for height of source, 86% were within 50 m; and for
distance to source, 93% were within 10 km. However, both extraction techniques had their own
peculiarities (Table 1). Overall the fully automated extraction technique was only able to return data for
97% of sites at which full spatial data were available. However, this was because of insufficiently accurate
map references, and could be resolved by the semi-automatic method of interactively choosing sites. At sites
which could be resolved, returned data were not always available because, for example, sources could not
be found for sites in lowland drainage networks. Other data were not necessarily incorrect but the values
were unlikely such as slopes of zero (2.9%) or negative (0.8%), heights of zero (0.4%), or distance to
sources of zero (0.7%). These need further investigation (Figures 3 and 4).
Table 1. A comparison of four main attributes derived manually from maps and by automated processing from the river network for
the RHS baseline of 5348 sites in Britain
Variable
Difference between manual and automated
Manual extraction limit of difference
Altitude of site
97.8% 510 m
98.8% 520 m
95% 5 25 m km1
98% 5 40 m km1
10 m
Slope of site
Height of source
Distance to source
61% 5 10 m
72% 5 20 m
86% 5 50 m
83% 5 5 km
93% 5 10 km
97.5 5 20 km
Copyright # 2002 John Wiley & Sons, Ltd.
520 m 3 m km1
20–70 m 5 m km1
>70 10 m km1
10 m
510 km 1 km
>10 km 10%
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
#
#
#
(a) Solid geology illustrated for the upstream catchment area. Source, Centre for Ecology and Hydrology, NERC.
#
#
#
(b) Drift geology illustrated for the upstream catchment area. Source, Centre for Ecology and Hydrology, NERC.
Plate 1. An example of automated extraction of map data for the River Welland (Eastern England) showing the route from a point on
the river to its mouth and to its source. Artificial sections of the river network which are non-traceable are indicated in grey.
Copyright # 2002 John Wiley & Sons, Ltd.
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
397
AUTOMATED EXTRACTION OF RIVER VARIABLES
% frequency
% frequency
15
10
5
5
0
0
-40
-30
-20
-10
0
10
20
30
40
difference in altitude of site (m)
(a)
-40
-30
-20
-10
0
10
20
30
40
difference in height of source (m)
(b)
30
10
% frequency
% frequency
10
20
10
0
5
0
-40
-30
(c)
-20
-10
0
10
20
difference in slope (m per km)
30
-10
40
(d)
0
10
difference in distance to source (km)
Figure 3. Frequency histogram showing the difference in results obtained from automated and manual extraction from maps of: (a)
altitude, (b) height of source, (c) slope and (d) distance to source for RHS baseline survey sites in Britain (n ¼ 5347).
Site identification
The results show that automated extraction of data, although reliable and repeatable, was completely
dependent on the presentation of accurate NGRs. The problem could be avoided by better field survey
training. Almost 3% of sites either had incorrect NGRs (i.e. they were more than 100 m from any digitized
river) or were not sufficiently accurate to locate automatically the precise tributary in the network. Manual
processing from maps did allow for this by visual choice and expert knowledge. This knowledge could also
be incorporated into the semi-automatic mode of operation by visually locating the site on the river. It
should be possible to reduce the occurrence of both these problems by adding river names or numbers to
the network, which should also increase the ability to make choices between tributaries.
Distance to source
Distance to source requires resolution of the route to the correct source, which is taken as the furthest
distance upstream as opposed to the highest altitude or even the largest discharge. In several low-lying areas
of England and Wales, the artificial drainage patterns cannot at present be integrated automatically to
identify sources and, hence, derive the required environmental variables. This is because the artificial grid
structure is too complex and has insufficient valley slope to identify a single source. This applies equally to
the manual method. In artificial systems it is questionable if such a ‘source’ has any geomorphological
meaning since the direction of flow can change during the operation of drainage pumps and tidal cycles.
Copyright # 2002 John Wiley & Sons, Ltd.
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
398
F.H. DAWSON ET AL.
Figure 4. Scatter plot showing the difference in results obtained from automated and manual extraction from maps of: (a) altitude, (b)
height of source, (c) slope at site and (d) distance to source for RHS baseline survey sites in Britain (n ¼ 5347).
Manual measurements of distance to source were often unrepeatable because of slippage on the mapmeasuring wheel. As a result, manual extraction almost always underestimated correct length of stream,
especially in sinuous streams (Figures 3(d) and 4(d)). The distance also varied significantly between manual
operators according to which source was considered to be the true one. The measurement of distance to all
sources in the automatic method resulted in a consistent choice of the furthest location.
A problem common to both methods is measurement using the shortest route. This may not necessarily
be the main channel, particularly in multiple-thread rivers, but this problem is relatively minor in Britain. A
similar problem exists with tracing routes, and thus distances to sources, for sites on tributaries arising from
braided sections but not on the shortest or main course. Extending the analysis to additional datasets may
also result in different outputs and make comparison difficult. For example, the original scale of digitization
of a map at, say, 1:625 000 or 1:250 000 scale will not have sufficient detail of bends and meanders when
lengths are compared with 1:50 000 digitization. Data transfer is required to resolve this, and this has been
undertaken for the Environment Agency flood defence works dataset.
Altitude of site and source
In general, there was no tendency for the slope, or for the altitude of either source or site, to be significantly
biased between methods (contrast Figures 3(a) and 3(b) with Figures 4(a) and 4(b)). However, data on
height of source may differ as a consequence of choice or designation of the source by manual operators.
Copyright # 2002 John Wiley & Sons, Ltd.
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
AUTOMATED EXTRACTION OF RIVER VARIABLES
399
Problems were encountered in both methods where altitude data were missing, particularly in areas with
lowland drainage networks but also in sections with lakes.
Slope
Slope at sites within 500 m of river source can vary with either method because of relatively short distances.
Slopes estimated manually were not always repeatable in steep areas because discrimination was difficult
when contour lines were densely packed. There was a greater frequency of whole numbers in the manual
extraction (Figures 3(c) and 4(c)). The difference in altitudes available for calculation of slope between the
manual and the automated extraction created some differences. Integrating databases for instream
structures and water bodies will assist in resolving this for automated calculation, but consideration will
need to be given to the geomorphological relevance of slope to the function and form of the channel
downstream.
APPLICATIONS OF THE SYSTEM
There are gains to be made in reliable and repeatable standardized data processing now that the major
effort in digitizing, producing and validating a coherent linked network has been completed. Integrating
these data with other datasets will allow rapid and comprehensive derivation of new combinations of data
to meet continuing needs of the EC Water Framework Directive (WFD), and in research on rivers. Other
uses could include predicting the impact of changes in rural land management on rural economies, and
flood modelling to alter the intensity of flood hydrographs, both of which have already been tested.
Reach boundaries needed for determining physical quality objectives for rivers
Non-statutory physical quality objectives are proposed for rivers in England and Wales (Walker et al.,
2002). This means that reaches of similar physical character need to be identified for designation and
reporting in a similar manner to that for the biological and chemical quality of water by the Environment
Agency in England and Wales. The use of map-derived data is an essential element of this process, with
automation allowing various options to be determined and compared. As part of a feasibility study, initial
attempts at defining the boundaries of these reaches have used six attributes:
(1)
(2)
(3)
(4)
(5)
(6)
change in channel slope;
change in geology defined in three broad categories of hardness;
change in stream order;
change in predominant land-use defined in four broad categories;
presence of major structures in channel, such as major weirs and dams;
presence of floodplain as defined by the 1 in 100 year flood.
The river was divided into 500 m units from the mouth to its sources. The slope for each section was
calculated from the best available data; changes in slope greater than 50% were used to test the approach.
Initial analysis indicates that this change in bed slope could be a useful starting point for defining reach
boundaries especially as changes in other variables also occur at or near these points (Walker et al., 2002)
(Figure 5).
Copyright # 2002 John Wiley & Sons, Ltd.
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
400
F.H. DAWSON ET AL.
Figure 5. The Sankey catchment (north-west England) showing those 500 m sections of river with changes in slope of > 50% in
black taken from 1:50 000 scale digitized river network shown in grey. Based on Ordnance Survey1 1:50 000 mapping with the
permission of Ordnance Survey on behalf of The Controller of Her Majesty’s Stationery Office, # Crown copyright, NERC
GD272191.
DISCUSSION
This paper has illustrated how a valid and checked coherent linked river network can deliver reliable and
repeatable data in a standard manner. Integration with other datasets has already been shown to allow
rapid and comprehensive derivation of new combinations of data to meet the needs of the Water
Framework Directive (WFD) for defining catchment and river reaches and for continued research on rivers.
Automated extraction is a major advance in supporting river-based classifications, assessment and decisionsupport systems (e.g. Raven et al., 1998b; Wright, 2000; Clark, 2002; Clark and Richards, 2002). The effort
needed to make necessary corrections and validation has yielded benefits because the data can be displayed
and combined with other datasets almost at the ‘press of a button’. A simple example of this network is the
calculation of stream length and catchment area upstream of any site. Reaches of similar physical character
can be defined by using and comparing combinations of changes in slope, geology, stream order, land-use,
presence of major instream structures or a floodplain. Initial analysis on a sub-catchment indicates that a
significant change in bed-slope would produce separation into habitat and ecological units, especially as
several other environmental variables such as geology coincide with this change. This analysis will also have
to examine other stresses in the river system such as the degree of modification to physical or water quality,
both for refining reach division and also in relation to species or habitats of importance.
Copyright # 2002 John Wiley & Sons, Ltd.
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
AUTOMATED EXTRACTION OF RIVER VARIABLES
401
The extraction of a range of environmental variables for any site on a river in Britain will require a
significant improvement in the accuracy of map references. Many of those which were originally transferred
from maps by hand were more than 100 m from any river on the digitized network or did not allow the
system to differentiate between alternative sites. Improvement could be achieved by better training in
producing map references, the use of 10 m resolution instead of 100 m resolution, and the addition of river
and tributary names or numbers to the network. Survey forms should be amended to contain an accurate
grid reference, river name and if the site is located near a confluence a brief explanation of site location.
Such information would avoid any ambiguity for manually assigning a site to a river.
Analysis of the various surrogate relationships used by RHS could be further tested and refined using
additional parameters such as catchment area and rainfall periodicity (Jeffers, 1998). Further analysis could
involve information on groundwater storage, rock porosity and base flow index. These comparisons are
feasible because datasets are now available, although the various assumptions would need to be tested and
reviewed by experts in hydrology, geomorphology, and other disciplines. The analysis should identify
potentially better or new predictor variables for river conditions after suitable statistical testing, together
with better probability estimates for predominant substrate, flow type or erosion and deposition regime
currently used by the RHS system. Such changes may include the use of general slope instead of height of
source and distance, which may, for example, allow a more widely applicable use of RHS in Europe (Raven
et al., 2002).
ACKNOWLEDGEMENTS
This project was funded by NERC CEH with contributions from the Environment Agency. The original ‘blue line’ river
network was supplied by CEH Wallingford (formerly the Institute of Hydrology, Wallingford). We are grateful to staff
at CEH Wallingford, in particular Dave Morris, who was generous with the provision both of advice and additional
data at various stages during the project; to other CEH staff, especially Peter Scarlett, for helpful comments and
contributions, and to Gordon Irons for preliminary work on the blue line network; to John Jeffers, and particularly to
Marc Naura and Jim Walker of the Environment Agency for their help, input, guidance and continuing support.
REFERENCES
Brookes A, Gregory KJ, Dawson FH. 1983. An assessment of river channelization in England and Wales. Science of the
Total Environment 27: 97–111.
Clark MJ. 2002. Dealing with uncertainty: adaptive approaches to sustainable river management. Aquatic
Conservation: Marine and Freshwater Ecosystems 12: 347–363.
Clark MJ, Richards KJ. 2002. Supporting complex decisions for sustainable river management in England and Wales.
Aquatic Conservation: Marine and Freshwater Ecosystems 12: 471–483.
Dawson FH, Szoszkiewicz K. 2000. Relationships of some ecological factors with the associations of vegetation in
British rivers. Hydrobiologia 415: 117–122.
Environment Agency. 1997. The Quality of Rivers and Canals in England and Wales 1995. Environment Agency:
London.
European Commission. 2000. Directive 2000/60/EC, Establishing a framework for Community action in the field of
water policy. Official Journal of the European Communities L 327: 1–72.
Fuller RM, Parsell RJ. 1990. Classification of TM imagery in the study of land use in lowland Britain: practical
considerations for operational use. International Journal of Remote Sensing 11: 1901–1917.
Fuller RM, Groom GB, Wallace SM 1994a. The availability of Landsat TM images for Great Britain. International
Journal of Remote Sensing 15: 1357–1362.
Fuller RM, Groom GB, Jones AR. 1994b. The land cover map of Great Britain: an automated classification of Landsat
thematic mapper data. Photogrammetric Engineering and Remote Sensing 60: 553–562.
Hornby DD, Clarke RT, Wright JF, Dawson FH. 2002. Testing and Further Development of RIVPACS Phase 3: An
evaluation of procedures for acquiring environmental variables for use in RIVPACS from a GIS. Environment
Agency R&D Technical Report E1-007/TR1.
Copyright # 2002 John Wiley & Sons, Ltd.
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
402
F.H. DAWSON ET AL.
Jeffers JNR. 1998. Characterisation of river habitats and the prediction of habitat features using ordination techniques.
Aquatic Conservation: Marine and Freshwater Ecosystems 8: 529–540.
Lanfear KJ. 1990. A fast algorithm for automatically computing Strahler stream order. Water Resources Bulletin 26:
977–981.
Morris DG, Flavin RW. 1990. A digital terrain model for hydrology. Proceedings of the Fourth International
Symposium on Spatial Data Handling, Zurich, Switzerland, 1. 250–262.
Morris DG, Flavin RW. 1994. Subset of UK digital 50 m by 50 m hydrological digital terrain model grids. NERC
Institute of Hydrology, Wallingford.
Moore RV, Morris DG, FIavin RW. 1994. Subset of the UK digital 1:50,000-scale river centre-line network. NERC
Institute of Hydrology, Wallingford.
Raven PJ, Boon PJ, Dawson FH, Ferguson AJD. 1998b. Towards an integrated approach to classifying and evaluating
rivers in the UK. Aquatic Conservation: Marine and Freshwater Ecosystems 8: 383–393.
Raven PJ, Fox P, Everard M, Holmes HTH, Dawson FH. 1997. River Habitat Survey: a new system to classify rivers
according to their habitat quality. In Freshwater Quality: Defining the Indefinable?, Boon PJ, Howell DL (eds). The
Stationery Office: Edinburgh; 215–234.
Raven PJ, Holmes HTH, Dawson FH, Fox PJA, Everard M, Fozzard I, Rouen KJ. 1998a. River Habitat Quality: The
Physical Character of Rivers and Stream in the UK and Isle of Man. River Habitat Survey Report No. 2. Environment
Agency: Bristol.
Raven PJ, Holmes NTH, Charrier P, Dawson FH, Naura M, Boon PJ. 2002. Towards a harmonized approach for
hydromorphological assessment of rivers in Europe: a qualitative comparison of three survey methods. Aquatic
Conservation: Marine and Freshwater Ecosystems 12: 405–424.
Strahler AN. 1957. Quantitative analysis of watershed geomorphology. Transactions of the American Geophysical Union
38: 913–920.
Walker J, Diamond M, Nauria M. 2002. The development of Physical Quality Objectives for rivers in England and
Wales. Aquatic Conservation: Marine and Freshwater Ecosystems 12: 381–390.
Wright JF. 2000. An introduction to RIVPACS. In Assessing the Biological Quality of Freshwaters: RIVPACS and other
Techniques. Wright JF, Sutcliffe DW, Furse MT (eds). Freshwater Biological Association: Ambleside; 1–24.
Wyatt BK, Greatorex-Davies NG, Bunce RGH, Fuller RM, Hill MO. 1994. Comparison of land cover definitions.
Countryside (1990 series): 3. Department of the Environment, London.
APPENDIX: ENVIRONMENTAL PARAMETERS WHICH CAN BE DERIVED USING THE
AUTOMATED EXTRACTION OF MAP-BASED DATA FOR ANY POINT ON A RIVER
NETWORK IN BRITAIN
These variables were originally derived for use with RHS and as such relate primarily to the 500 m sample reaches.
Site position
*
*
*
coordinate of site used in calculation on river network (8-figure UK National Grid Reference, NGR, to
within 10 m) with selected scale of map at visualization.
geology at site } solid, drift and surface of BGS type and in one of 8 RHS groups.
hydrometric number.
RHS-PCA ordination analysis variables
*
*
*
*
altitude of site (m) from the mean of the altitude nearest a point 500 m upstream along river and a point
500 m downstream of the sample point.
distance from river source (km) by the shortest route to the site from furthest point upstream.
height of river source (m) at the same point or the first occurrence of an altitude downstream along
course of the river.
slope at site (m km1) from the difference in altitude between sites at approximately 500 m upstream and
500 m downstream divided by their actual distance apart.
Copyright # 2002 John Wiley & Sons, Ltd.
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
AUTOMATED EXTRACTION OF RIVER VARIABLES
403
River data
*
*
*
*
*
coordinates of river mouth
distance from site (km) by the shortest route to the defined tidal limit or estuary mouth
coordinates of river source
stream order (Strahler, 1957) system using the Lanfear (1990) algorithm
channelization in England and Wales (Brookes et al., 1983)
Catchment data
*
*
*
*
catchment area above site (m2)
total length of watercourses upstream of site (km)
geology of catchment upstream of site as total of each type (by area and percentage) for: solid (129
types), drift (13 types) and surface (i.e. drift or if absent, solid) for each BGS type and RHS group
land cover by category
Copyright # 2002 John Wiley & Sons, Ltd.
Aquatic Conserv: Mar. Freshw. Ecosyst. 12: 391–403 (2002)
Документ
Категория
Без категории
Просмотров
3
Размер файла
608 Кб
Теги
environment, rivers, variables, extraction, method, automaten, classification, britain, help
1/--страниц
Пожаловаться на содержимое документа