Professional Documents
Culture Documents
This report represents the views of the authors and has not been subject to
a full peer review process
1
Marine Scotland Science Freshwater Laboratory
Faskally, Pitlochry, Perthshire, PH16 5LB
*
k.millidine@marlab.ac.uk
2
Northern Rivers Institute, School of Geosciences
University of Aberdeen, Aberdeen, AB24 3UF
Executive Summary
• This study investigated the viability of using in-stream close range digital
photographs and an automated image analysis program (Sedimetrics photographic
package ‘Digital Gravelometer’). The report is divided into 4 sections: (1) obtaining
digital images of streambed substrate; (2) sediment characterisation using image
processing applications; (3) comparison of results using quadrat and digital analysis
methods; and (4) discussion of the limitations and opportunities for the use of digital
image analysis in fish habitat studies.
• There were significant challenges obtaining realistic sediment distributions using the
sedimetrics software. However, providing a lot of care and time were taken during
the image processing procedure, the digital analysis method provided good results
and was comparable with standard visual assessments.
1
Introduction
Instream habitat modelling is potentially useful for assessing the physical processes
influencing the distribution of different fish species and age classes (Lamouroux et al.,
1999; Guay et al., 2000; Kennard et al., 2006; Buisson et al., 2008). Typically fish habitat
models require the characterisation of habitat that is used (utilised) and present
(available) but not used by fish. In the context of riverine fish, habitat utilisation describes
the use of locations by individuals and, requires characterisation of depth, velocity,
substrate and cover. The techniques used to characterise the streambed grain-size
composition vary greatly in both accuracy and effort (Whitman et al. 2003). One
traditional method of rapidly assessing and quantifying particle size is by using a quadrat
and estimating, by eye, the percentage cover of the dominant and sub-dominant
substrate (Mäki-Petäys et al., 1997). However, results from these traditional methods
can vary substantially between observers, introducing significant bias where use of the
same observer is not always possible. Moreover, the results are usually reported in the
form of ‘dominant’ and ‘subdominant’ size-classes or as percentages of classes. These
metrics are rather coarse and are not necessarily the most biologically relevant metric
for describing substrate characteristics. Consequently these approaches may be of
potentially limited value in the development of habitat models for fish.
Another frequently used method for quantifying coarse particle size composition is the
Wolman pebble count (Wolman 1954). This involves walking along a transect and
selecting random clasts at the toe point, which are then measured along the
intermediate axis (b-axis). A large sample can be acquired rapidly and used to generate
a size-frequency distribution. However, Wolman is a reach- or morphological unit-scale
method which is not appropriate for patch-scale analysis of microhabitat use by fish.
Also, it underestimates the proportion of fine material (<8mm); this is a significant issue,
given the importance of fines in determining the quality of salmonid habitat (Greig et al.
2005; Levasseur et al. 2006). Bulk samples of material from patches of bed can be
collected and taken back to the laboratory for sieving and accurate determination of
2
grain size distributions (GSDs). However, bulk sampling is destructive (i.e. it disturbs the
bed) and this affects the availability and, potentially, subsequent use of bed locations by
fish. Other limitations relate to the fact that large volumes of material are needed for
accurate characterisation (particularly where the bed is coarse) and so bulk sampling
becomes impractical in some circumstances.
Given this context, a series of methods were field trialled in order to develop a protocol
for streambed sediment size characterisation that could be used for habitat
characterisation and modelling. The objectives of this study were to (1) obtain high
quality digital images of streambed substrate, (2) characterise grain-size composition
using image processing applications and (3) identify limitations and opportunities for
using digital image analysis in fish habitat studies. Field trials were carried out in the
Newmills Burn, Donside, as part of a broader research project on the use of habitat by
fish in agricultural streams (SFO273). However, the limitations and benefits of the
approach identified in this report are generic and consequently will be applicable to other
sites and locations.
3
1. Obtaining Digital Images of Streambed Substrate
Most habitat modelling studies attempt to locate fish and characterise habitat use at
similar spatial scales. Various methods are available for determining the location of fish,
including snorkelling, PIT tag detection antennae and electrofishing. Each of these
methods has its benefits and limitations. However, for the purposes of this assessment
we assumed the use of a modified electrofishing method with an estimated locational
accuracy of ca. 0.5 m2.
In order to use the same technology in studies of fish habitat it is necessary to obtain
photographs from below the water surface. This is problematic due to constraints
associated with reflection and lighting. Initial efforts to overcome these issues focussed
on the use of standard digital SLR cameras fitted with polarising filters, together with
perspex-bottomed basins to see into the stream (Plate 1). Both of these approaches
proved unable to provide photographs of adequate quality, owing to the following
problems:
1. A flash was required to illuminate the bed sufficiently. However this resulted in
reflection back from the perspex and made the photographs unusable.
2. Without the flash it was necessary to use an extremely high ISO speed and low
shutter speed (to allow for poor light conditions). This combination resulted in
grainy / noisy images (due to high ISO) and a lack of sharp focus on the images
due to camera shake (resulting from the slow shutter speed). These images were
too poor in quality to use in digital analysis.
4
Plate 1
Perspex-bottomed box used in initial trials
Subsequently a series of under water compact cameras were trialled. While these were
able to obtain good quality photographs of the sediment (Plate 2), it was necessary for
the camera lens to be fully submerged. Therefore the area of streambed that could be
photographed largely depended on water depth. This limits the use of underwater
cameras to deeper areas; for example, to obtain a 0.5 m2 image using a 6.4 – 19.2 mm
camera, the minimum distance away from the substrate is 90 cm.
Plate 2
Digital images of bed substrate using a compact underwater camera.
5
In order to overcome these problems a perspex bottomed box was constructed within a
frame that allowed height adjustment (Figure 1). With the bottom of the box in contact
with the water surface, it was then possible to fill the box with water to a desired height
and obtain photographs of the streambed using compact under water cameras below the
water surface (Plate 3). This equipment enabled good quality photographs of the
streambed at the required scale even in very shallow waters.
Figure 1
Schematic of substrate-viewing box
0.7m
0.7m
0.7m
External frame
The moveable Perspex box within the external
frame enables height adjustment, allowing for
variable substrate size.
6
Plate 3
Apparatus for obtaining photographs of the streambed
Once the photographs had been obtained it was necessary to process the photographs
within the analysis software in order to produce a grain size distribution.
7
2.2 The Grain Measurement Procedure
8
Stage 8: Convert grain sizes to millimetres
The final stage is to convert the measured grain sizes into millimetres using the known
scale of the image.
To illustrate the full range of image processing stages, Plate 4 shows the series of
images produced using the default settings within the software (Stages 1, 3, 4, 5 and 6).
Plate 4
Digital image of bed substrate processed for particle size analysis
River beds present a particularly complex problem because grains are highly variable in
shape and may be partially hidden or inclined relative to the plane of the image. In
addition there may be significant heterogeneity in colour and grain-surface texture
between and within individual grains, and the surface has elevation variations which may
result in uneven lighting and shading across individual grains and across the image
(Graham et al. 2005b). Also, algae and macrophytes can obscure sediment and make it
difficult for the software to correctly discriminate between adjacent clasts.
9
Differences in Grain-surface texture between and within individual grains were evident in
the digital photographs taken from the Newmills Burn and this prevented the software
adequately identifying each individual substrate particle using the default settings
(Plate 5). To rectify this problem, the image analysing parameters used within the grain-
identification procedure were modified in an attempt to capture an improved
representation of the digital sample. Plate 6 shows the effects of modifying the default
parameters within the Digital Gravelometer software package, which shows a
reasonable discrimination between the stones and would be more suitable for further
analysis, although it can be seen that erroneous identification of individual clasts
remains.
Plate 5
Converted greyscale image (left) and the same image processed for particle size
analysis using default parameter settings (right)
Plate 6
Converted greyscale image (left) and the same image processed for particle size
analysis using modified parameter settings (right)
10
Digital Gravelometer offers the user considerable flexibility in producing, viewing and
exporting the sediment report. In addition to the basic / default size-distribution
parameters, it enables customisable reports containing a number of statistical
parameters along with graphical grain-size distribution curves and histograms. These
reports include:
• User defined or default percentile estimates (in mm, phi and psi)
• Information about the project (client, operator etc.) and each individual image
(filename, sample ID, collection date etc.)
Figure 2
Example of a grain-size distribution obtained from Digital Gravelometer software
15 Cumulative % 60
% in class
40
10
20
5
0
4 2 0 -2 -4 -6
Phi
11
3. A Comparison of the Results Obtained from the Quadrat and Digital
Analysis Methods
Basic output from the sedimetrics application (dominant and sub-dominant substrate
size) should be comparable with conventional characterisation techniques (i.e. using a
quadrat and estimating the percentage dominant and sub-dominant substrate
composition) if the software is capable of adequately identifying the grains within the
image. Using a quadrat, the substrate particles in the Newmills Burn were classified
using an index developed by Brusven (1977) with the modifications of Bovee (1982)
(Table 1).
Table 1
The Brusven substrate code (as modified by Bovee 1982) used for substrate
classification
In order to assess the comparability of the digital analysis method to the more
conventional quadrat method, a small number of photographs with dominant and sub-
dominant information were analysed using both the default and modified parameter
settings (Table 2). The output settings of the Digital Gravelometer were customised in
order to obtain the proportion of grains as defined in Table 1. The largest and second
largest proportions of grains were respectively classed as dominant and sub-dominant
substrate.
12
Table 2
A comparison of digital analysis and quadrat substrate sampling methods (The numbers
represent the code from Table 1)
Size-class
Size-class using
Size-class using using
Sample Category modified image
quadrat default image
settings
settings
Dominant 1 1 1
1
Sub-dominant 2 2 2
Dominant 2 1 2
2
Sub-dominant 1 2 1
Dominant 3 2 2
3
Sub-dominant 2 4 3
Dominant 4 2 4
4
Sub-dominant 2 1 2
Dominant 2 2 3
5
Sub-dominant 3 1 2
Dominant 2 2 2
6
Sub-dominant 1 1 1
The individual clasts were poorly defined in all samples using the standard setting
(similar to Plate 5) (Table 2). The software was particularly poor at identifying larger
substrate sizes (size-class 3 or more) due to the grain-surface texture which tended to
break up the image into smaller particles. After parameter modification to enhance
substrate discrimination, the results were generally closer to those obtained using the
quadrat method. There was a slight disagreement in a two samples (samples 3 and 5),
but it is uncertain whether this was due to observer or analysis error owing to the fact
that the percentage composition of both the dominant and sub-dominant were very
similar.
4.1 Obtaining Digital Images of Streambed Sediments using the Perspex Box
The time taken to obtain the initial photographs can be substantial for shallow water
areas due to the size and weight of the apparatus involved (shown in Plate 3). In order to
obtain images of the necessary substrate area, the perspex box was large (see
section 1). Furthermore, because a considerable volume of water was needed to fill the
13
apparatus, heavy Perspex sheeting was needed to withstand the forces involved.
Consequently the overall weight of the apparatus was substantial, requiring at least two
people to move and position it; this made deployment in areas surrounded by steep-
sided banking difficult. In addition, there were further difficulties with its deployment at
some locations; for example it was not possible to deploy it near to the banks, especially
if highly vegetated or undercut due to the physical size of the apparatus.
The practicality of using the box may also be limited by the size of substrate to be
photographed, since its elevated height was restricted to 0.5 m (elevation height was
limited to 0.5 m to minimise overall weight of apparatus). Other considerations include
circumstances where substrate mobility reduces the quality of the photographs, or where
coarse irregular emergent substrate prevents deployment.
The use of the perspex box in wider yet shallow streams would be easier, particularly
when access in and out of the water is not impeded by steep, highly vegetated banks.
Although digital image analysis is novel and potentially offers major advantages in terms
of the detail of information provided, in practice it proved to be extremely time consuming
given that the default settings did not adequately represent the grain sizes within the
digital image. The main reason for this appeared to be due to the surface-grain texture of
the substrate which caused the image processing software to break-up larger clasts into
smaller ones. In this circumstance, the user has to manually manipulate the settings for
each of the images. This manipulation can be considerable – e.g. it took approximately
one hour to adequately process each of the samples in table 1. The manual
manipulation of the measurement process also results in observer bias, given that how
effectively the substrate is defined is dependant on the person using the software.
Despite these problems the Digital Gravelometer software was able to identify the
dominant and sub-dominant substrate type present within the images. However,
14
comparisons with other more detailed sediment analysis methods have not been carried
out and we are unsure how they would compare.
This method may work better where sediments are more uniform in colour or where you
need more detailed grain size information e.g. spawning studies or spawning habitat
assessment.
The use of digital image analysis (Digital Gravelometer) has advantages for the
characterisation of exposed gravel surfaces. However, it has many limitations when
used for characterising areas inundated with water, most of which have been identified
in the trial summarised in this report. The full benefits of using this sampling approach
are only likely to be realised in streams that contain substrate with less variable surface-
grain texture, where access in and out of the water is not impeded by steep banking or
where water depths are sufficient for direct use of underwater photography. The
potential to acquire a range of sediment size metrics which can then be tested for
biological value is itself a clear benefit of using digital photogrammetry; however these
benefits would have to outweigh the substantial constraints and limitations identified in
this report. The value of this approach in understanding fish-habitat relationships, where
fish sample sizes are very small, may be very limited and costs too excessive. However,
its value may be greater in other studies, such as spawning habitat use or assessment.
References
Bovee, K. D. 1982 A guide to stream habitat analysis using the instream flow
incremental methodology. Instream Flow Information Paper 12. U.S. Fish and Wildlife
Service, Fort Collins, CO. FWS/OBS-82/86.
15
Butler, J. B. Lane, S. N. and Chandler, J. H. 2001 Automated extraction of grain-size
data from gravel surfaces using digital image processing. Journal of Hydraulic
Research 39: 1–11.
Graham, D. J. Rice, S. P. and Reid, I. 2005b A transferable method for the automated
grain sizing of river gravels. Water Resources Research 41: W07020
16
Jensen, D. W. Steel, E. A. Fullerton, A. H. and Pess, G. R. 2009 Impact of fine
sediment on egg-to-fry survival of pacific salmon: A meta-analysis of published studies.
Reviews in Fisheries Science 17: 348–359.
Johnson, J. H. and Douglass, K. A. 2009 Diurnal stream habitat use of juvenile Atlantic
salmon, brown trout and rainbow trout in winter. Fisheries Management and Ecology 16:
352–359.
17
Moir, H. J. Gibbins, C. N. Buffington, J. M. Webb, J. H. Soulsby, C. and Brewer, M. J.
2009 A new method to identify the fluvial regimes used by spawning salmonids.
Canadian Journal of Fisheries and Aquatic Sciences 66: 1404–1408.
Valdimarsson, S. K and Metcalfe, N. B. 1998 Shelter selection in juvenile Atlantic
salmon; or why do salmon seek shelter in winter? Journal of Fish Biology 52: 42–49.
18
© Crown Copyright 2011
Copies of this report are available from the Marine Scotland website at
www.scotland.gov.uk/marinescotland