Elsevier

Ecological Informatics

Volume 14, March 2013, Pages 2-8
Ecological Informatics

Remote sensing image data and automated analysis to describe marine bird distributions and abundances

https://doi.org/10.1016/j.ecoinf.2012.12.001Get rights and content

Abstract

Advances in image data capture with airborne digital cameras and in object-based image analysis (OBIA) have provided a basis for new arenas of applied remote sensing, one of which is the direct counting and mapping of animal individuals. The derived data represents significant inputs to population size estimation and study of animal–habitat interactions. One growing application is bird distributions and abundances in relation to EIAs for marine installations such as offshore wind farms and sea crossing transportation infrastructure. The ornithological surveys in relation to the EIA process concerning the Burbo Bank Extension and Walney Extension offshore wind farms that are being undertaken jointly by Aarhus University and BLOM-UK are an example of the former, with monthly surveys of two offshore areas (total area ca. 670 km2) in the Irish Sea. Image coverage comprises acquisition of 4 cm and 3 cm image data. Object based image analysis methods are used to localise marine birds. This paper describes the data processing and analysis procedures used. It presents examples of representations in these image data of five bird species. Examples of non-bird ‘background’ image patterns that can impact the bird localisation possibilities are presented along with developed countering-strategies. Results of initial assessments of levels of bird under-mapping by the OBIA method are given; the reported overall success level from 18 image frames with respect to under-mapping is over 92%.

Highlights

► Airborne digital imaging provides a basis for describing marine bird numbers and distributions. ► Object based image analysis provides adaptive tools for marine bird detection. ► An integrated GIS and image analysis workflow enables operational marine bird recording.

Introduction

Knowledge of the real-time locations and numbers of organisms present at a site is sometimes a basic requirement for understanding the abundance, distribution and ecology of many taxa, yet such information is often unattainable (Buckland et al., 2004). This knowledge gap has the potential to limit the development of efficient and effective wildlife management plans and conservation strategies and the assessment of the impacts of installations (e.g. bridges, marine wind farms) on animal populations. Considerable ecological research effort has been invested in procedures to sample the natural distribution and abundance of organisms and for improving interpretation of results from different ecological survey procedures and technologies (e.g. capture-mark-recapture, radio-tracking, distance sampling; Elphick, 2008). Pelagic bird abundances require counting from aerial or ship platforms, either attempting total coverage aerial counts (Komdeur et al., 1992) or surveying strip- or line-transects (Camphuysen et al., 2004). However, the associated confidence intervals are often wide (Buckland et al., 2001, Certain and Bretagnolle, 2008, Frederick et al., 2003, Laursen et al., 2008) as the areal abundance estimation requires fitting of the associated detection function (Camphuysen et al., 2004). Low flying aircraft used for off-shore bird surveys (flying at < 80 m above the sea surface) also risk surveyors' safety and disturbance to wildlife (e.g. Mosbech and Boertmann, 1999).

Aerial survey imaging has also been used for animal counting for several decades (Gibbs et al., 1988, Leonard and Fish, 1974, Woodworth et al., 1997), developing from use of photographic film products to digital image files acquired with either handheld cameras (Thompson and Harwood, 1990) or fuselage mounted cameras (Krafft et al., 2006). Ground-scanning videography has also been applied to studying animal distribution (Anthony et al., 1995, Dolbeer et al., 1997). Developments in imaging and related technologies (e.g. positioning and remote control) and in survey platform possibilities such as unmanned aerial vehicles (UAVs, drones) have coincided with numerous applications of imaging for wildlife counting in recent years (Jones et al., 2006, Laliberte and Ripple, 2003, Sardà-Palomera et al., 2012, Watts et al., 2010). The use of very high spatial resolution satellite image data for wildlife counts was reported by Laliberte and Ripple (2003), Trathan (2004), Barber-Meyer et al. (2007) and Sasamal et al. (2008).

The transformation of images to animal count information has in the past involved manual counting from hardcopy material (Woodworth et al., 1997) or computer monitors (Anthony et al., 1995). However, manual counting is often laborious and an inefficient use of resources (Laliberte and Ripple, 2003) due to either too many birds (e.g. many wetland birds will often number > 10 000 at a single location) or too few birds (e.g. solitary seabirds, with < 1 bird per 10 000 m2). In the former case identification and mapping of the areal extents of flocks, colonies or sub-colonies that have relatively uniform individual organism spatial densities (Guinet et al., 1995) or where the size of individuals is known and relatively constant (Gilmer et al., 1988) may be used as a basis for deriving a total count rather than the identification and exact counting of individuals. Manual counting of high numbers (e.g. > 100) of individuals can, as noted by Bajzak and Piatt (1990) and reported by Groom et al. (2011), be associated with significant count variations between image interpreters and even between repeated counts by the same interpreter.

Wildfowl counts made by Gilmer et al. (1988), by Bajzak and Piatt (1990), and by Cunningham et al. (1996) are among the first documented cases of use of automated methods to count animals from image data. Studies made using various forms of simple image data thresholding (Bajzak and Piatt, 1990, Cunningham et al., 1996), spatial domain filtering (Laliberte and Ripple, 2003, Trathan, 2004) and regression analysis (Barber-Meyer et al., 2007) all demonstrated the potential utility of automated approaches. Several of these studies noted that individual birds in image data represent ‘objects’ that have spatial characteristics, such as size and association to other birds, as well as mere image tone properties (Bajzak and Piatt, 1990, Laliberte and Ripple, 2003). Image analysis methods that can exploit complex texture patterns and contextual patterns using object based image analysis (OBIA) (Benz et al., 2004, Blaschke, 2010) and other advanced pattern recognition, signal processing and morphometric analysis approaches, offer new opportunities in the use of image data to count individual organisms. For example, Descamps et al. (2011) have used Markov point processes and simulated annealing to map flamingo individuals in image data. Advances in OBIA (Benz et al., 2004) have exploited synergies between image data segmentation and the organisation and use of object attribute information for object labelling (so-called ‘object relationship modelling’; Burnett and Blaschke, 2003). This object based paradigm for image analysis enables extensive exploitation of image data tone, texture and context for image-based mapping in robust and adaptive ways with controllable algorithms, and has been widely applied for many spatial mapping issues (Blaschke, 2010, Hay et al., 2005, Jacquin et al., 2008). Thus, this paradigm is seen as also relevant for individual bird counting as demonstrated by Milton et al. (2006), Groom et al. (2007) and Groom et al. (2011). These studies note that the use of the object based methods requires the development of robust segmentation and labelling rule-sets that operate invariantly between different organism mapping assignments and different image data and scenes.

Many of the documented cases of waterbird counting using image data are examples of imaging made of relatively high bird abundances with approximately known spatial extents (e.g. Descamps et al., 2011, Laliberte and Ripple, 2003, Trathan, 2004). These represent a contrast to situations encountered by many of the surveys needed for marine bird ecology impact assessments, which encounter relatively very low and unknown bird densities and also cover large spatial extents e.g. > 100 km2. The coastal and offshore waters of NW Europe are seen as having major potential for the generation of electrical power with wind driven turbines, with large marine wind farms already established in the territorial waters of a number of European states. In the UK, COWRIE (www.offshorewind.co.uk) reviewed the results of different approaches to the collection of data on the abundance and distribution of birds in marine environments, including the use of high definition imagery collection via video and frame (still) capture systems. In 2010, the work for the ornithological surveys in relation to the EIA process concerning the Burbo Bank Extension and Walney Extension offshore wind farms was commissioned by DONG Energy to a consortium comprising Aarhus University and Blom-UK, based on use of high spatial resolution frame image data and OBIA. The period for these surveys was November 2010 – October 2012 with restrictions at present on publication of bird abundance and distribution results. The objective of this paper is to demonstrate the ability of image data to capture marine birds and of OBIA methods to localise them, for a range of bird species and under a range of marine conditions.

Section snippets

Methods

Ornithological surveys were conducted in relation to an impact assessment for two offshore wind farms: the Burbo Bank Extension and Walney Extension (Fig. 1). These provided detailed information on the abundance and distribution of birds through a one year period for Burbo (210 km2) and two years for Walney (463 km2), with approximately one month intervals between surveys. The key data input for producing this information was monthly descriptions of bird and mammal densities and distribution

Results

The ornithological surveys of the study areas were conducted between November 2010 and October 2012. The specific bird abundance and distribution results will be published elsewhere. Here, three sets of method related results are presented:

  • 1.

    examples of the representation of different marine bird and other animal species groups in Vexcel image data, with discussion of their image data patterns that enable automated detection and identification,

  • 2.

    examples of non-bird image data patterns associated

Discussion

Variations in sea-state, marine environments, atmospheric conditions and solar illumination angles combine to produce a wide range of sea surface image patterns that form the background to the targets of a bird mapping operation. The frequent low numbers of marine birds in any given area adds to the complexity of developing methods for large scale operational surveys. A methodology that combines the strengths of manual image interpretation and automated image analysis, such as in the work

Acknowledgements

The support of DONG Energy in this work is gratefully acknowledged.

References (42)

  • S.T. Buckland et al.

    Advanced Distance Sampling

    (2004)
  • C.J. Camphuysen et al.

    Towards standardised seabirds at sea census techniques in connection with environmental impact assessments for offshore wind farms in the U.K. A comparison of ship and aerial sampling methods for marine birds, and their applicability to offshore wind farm assessments

    (2004)
  • D.J. Cunningham et al.

    An image-processing program for automated counting

    Wildlife Society Bulletin

    (1996)
  • S. Descamps et al.

    An automatic counter for aerial images of aggregations of large birds

    Bird Study

    (2011)
  • T.D. Dickey et al.

    Shedding new light on light in the ocean

    Physics Today

    (2011)
  • R.A. Dolbeer et al.

    Aerial photography techniques to estimate populations of laughing gull nests in Jamaica Bay, New York, 1992–1995

    Colonial Waterbirds

    (1997)
  • C.S. Elphick

    How you count counts: the importance of methods research in applied ecology

    Journal of Applied Ecology

    (2008)
  • P.C. Frederick et al.

    Accuracy and variation in estimates of large numbers of birds by individual observers using an aerial survey simulator

    Journal of Field Ornithology

    (2003)
  • J.P. Gibbs et al.

    Comparison of techniques for censusing great blue heron nests

    Journal of Field Ornithology

    (1988)
  • D.S. Gilmer et al.

    Goose counts from aerial photographs using an optical digitizer

    Wildlife Society Bulletin

    (1988)
  • G.B. Groom et al.

    Contextual correction: techniques for improving land cover mapping from remotely sensed images

    International Journal of Remote Sensing

    (1996)
  • Cited by (0)

    View full text